i

R O U T L E D G E HA N D B O O K O F
B O U N D E D R AT I O NA L I T Y

Herbert Simon’s renowned theory of bounded rationality is principally interested in cogni-

tive constraints and environmental factors and influences which prevent people from thinking
or behaving according to formal rationality. Simon’s theory has been expanded in numerous
directions and taken up by various disciplines with an interest in how humans think and behave.
This includes philosophy, psychology, neurocognitive sciences, economics, political science,
sociology, management, and organization studies.
The Routledge Handbook of Bounded Rationality draws together an international team of
leading experts to survey the recent literature and the latest developments in these related fields.
The chapters feature entries on key behavioural phenomena, including reasoning, judgement,
decision making, uncertainty, risk, heuristics and biases, and fast and frugal heuristics. The text
also examines current ideas such as fast and slow thinking, nudge, ecological rationality, evolu-
tionary psychology, embodied cognition, and neurophilosophy. Overall, the volume serves to
provide the most complete state-of-the-art collection on bounded rationality available.
This book is essential reading for students and scholars of economics, psychology,
neurocognitive sciences, political sciences, and philosophy.

Riccardo Viale is Full Professor of Cognitive Economics and Behavioural Sciences in the
Department of Economics at the University of Milano-Bicocca, Italy. He is also the founder
and General Secretary of the Herbert Simon Society.
ii
iii

R O U T L E D G E HA N D B O O K
OF BOUNDED
R AT I O NA L I T Y

Edited by Riccardo Viale

iv

First published 2021

by Routledge
2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN
and by Routledge
52 Vanderbilt Avenue, New York, NY 10017
Routledge is an imprint of the Taylor & Francis Group, an informa business
© 2021 selection and editorial matter, Riccardo Viale; individual chapters, the contributors
The right of Riccardo Viale to be identified as the author of the editorial material, and of the
authors for their individual chapters, has been asserted in accordance with sections 77 and 78
of the Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or utilised
in any form or by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying and recording, or in any information storage or
retrieval system, without permission in writing from the publishers.
Trademark notice: Product or corporate names may be trademarks or registered trademarks,
and are used only for identification and explanation without intent to infringe.
British Library Cataloguing-in-Publication Data
A catalogue record for this book is available from the British Library
Library of Congress Cataloging-in-Publication Data
A catalog record has been requested for this book
ISBN: 978-1-138-99938-1 (hbk)
ISBN: 978-1-315-65835-3 (ebk)
Typeset in Bembo
by Newgen Publishing UK
v

CONTENTS

Notes on contributors x
Preface xv

1 Why bounded rationality? 1

Riccardo Viale

2 What is bounded rationality? 55

Gerd Gigerenzer

PART I
Naturalizing bounded rationality 71

3 Towards a critical naturalism about bounded rationality 73

Thomas Sturm

4 Bounded rationality: the two cultures 90

Konstantinos V. Katsikopoulos

5 Seeking rationality: $500 bills and perceptual obviousness 103

Teppo Felin and Mia Felin

6 Bounded rationality, distributed cognition, and the computational

modeling of complex systems 120
Miles MacLeod and Nancy J. Nersessian

7 Bounded rationality and problem solving: the interpretative function of

thought 131
Laura Macchi and Maria Bagassi
v
vi

Contents

8 Simon’s legacies for mathematics educators 155

Laura Martignon, Kathryn Laskey, and Keith Stenning

9 Bounded knowledge 170

Cristina Bicchieri and Giacomo Sillari

PART II
Cognitive misery and mental dualism 183

10 Bounded rationality, reasoning and dual processing 185

Jonathan St. B. T. Evans

11 Why humans are cognitive misers and what it means for the
Great Rationality Debate 196
Keith E. Stanovich

12 Bounded rationality and dual systems 207

Samuel C. Bellini-Leite and Keith Frankish

13 Models and rational deductions 217

Phil N. Johnson-Laird

14 Patterns of defeasible inference in causal diagnostic judgment 228

Jean Baratgin and Jean-Louis Stilgenbauer

15 Attribute-based choice 242

Francine W. Goh and Jeffrey R. Stevens

PART III
Occam’s razor: mental monism and ecological rationality 255

16 Bounded reason in a social world 257

Hugo Mercier and Dan Sperber

17 Rationality without optimality: bounded and ecological rationality

from a Marrian perspective 268
Henry Brighton

18 The winds of change: the Sioux, Silicon Valley, society, and simple
heuristics 280
Julian N. Marewski and Ulrich Hoffrage

vi
vii

Contents

19 Ecological rationality: bounded rationality in an evolutionary light 313

Samuel A. Nordli and Peter M. Todd

20 Mapping heuristics and prospect theory: a study of theory integration 324

Thorsten Pachur

21 Bounded rationality for artificial intelligence 338

Özgür Şimşek

22 Psychopathological irrationality and bounded rationality: why is autism

economically rational? 349
Riccardo Viale

PART IV
Embodied bounded rationality 375

23 Embodied bounded rationality 377

Vittorio Gallese, Antonio Mastrogiorgio, Enrico Petracca, and Riccardo Viale

24 Extending the bounded rationality framework: bounded-resource

models in biology 391
Christopher Cherniak

25 How rationality is bounded by the brain 398

Paul Thagard

26 Building a new rationality from the new cognitive neuroscience 409

Colin H. McCubbins, Mathew D. McCubbins, and Mark Turner

PART V
Homo Oeconomicus Bundatus 421

27 Modeling bounded rationality in economic theory: four examples 423

Ariel Rubinstein

28 Bounded rationality, satisficing and the evolution of economic

thought: diverse concepts 437
Clement A. Tisdell

29 Beyond economists’ armchairs: the rise of procedural economics 448

Shabnam Mousavi and Nicolaus Tideman

vii
viii

Contents

30 Bounded rationality and expectations in economics 459

Ignazio Visco and Giordano Zevi

31 Less is more for Bayesians, too 471

Gregory Wheeler

32 Bounded rationality as the cognitive basis for evolutionary economics 484

Richard R. Nelson

33 Beyond “bounded rationality”: behaviours and learning in complex

evolving worlds 492
Giovanni Dosi, Marco Faillo, and Luigi Marengo

PART VI
Cognitive organization 507

34 Bounded rationality and organizational decision making 509

Massimo Egidi and Giacomo Sillari

35 Attention and organizations 522

Inga Jonaityte and Massimo Warglien

36 The bounded rationality of groups and teams 535

Torsten Reimer, Hayden Barber, and Kirstin Dolick

37 Cognitive biases and debiasing in intelligence analysis 548

Ian K. Belton and Mandeep K. Dhami

PART VII
Behavioral public policies: nudging and boosting 561

38 “Better off, as judged by themselves”: bounded rationality and nudging 563

Cass R. Sunstein

39 An alternative behavioural public policy 570

Adam Oliver

40 Against nudging: Simon-inspired behavioral law and economics

founded on ecological rationality 578
Nathan Berg

41 Bounded rationality in political science 597

Zachary A. McGee, Brooke N. Shannon, and Bryan D. Jones

viii
ix

Contents

42 Layering, expanding, and visualizing: lessons learned from three

“process boosts” in action 610
Valentina Ferretti

43 Cognitive and affective consequences of information and

choice overload 625
Elena Reutskaja, Sheena Iyengar, Barbara Fasolo, and Raffaella Misuraca

44 How much choice is “good enough”?: moderators of information

and choice overload 637
Raffaella Misuraca, Elena Reutskaja, Barbara Fasolo, and Sheena Iyengar

Index 650

ix
x

CONTRIBUTORS

Maria Bagassi, University of Milano-Bicocca, Italy

Jean Baratgin, Université Paris VIII, France

Hayden Barber, Purdue University, West Lafayette, IN, USA

Samuel C. Bellini-Leite, State University of Minas Gerais, Brazil

Ian K. Belton, University of Strathclyde, Glasgow, UK

Nathan Berg, University of Otago, Dunedin, New Zealand

Cristina Bicchieri, University of Pennsylvania, Philadelphia, PA, USA

Henry Brighton, Tilburg University, the Netherlands

Christopher Cherniak, University of Maryland, College Park, MD, USA

Mandeep K. Dhami, Middlesex University, London, UK

Kirstin Dolick, Purdue University, West Lafayette, IN, USA

Giovanni Dosi, Scuola Sant’Anna, Pisa, Italy

Massimo Egidi, LUISS, Rome, Italy

Jonathan St. B. T. Evans, University of Plymouth, Plymouth, UK

Marco Faillo, University of Trento, Trento, Italy

Barbara Fasolo, London School of Economics, London, UK

x
xi

Contributors

Mia Felin, University College, London, UK

Teppo Felin, Oxford University, Oxford, UK

Valentina Ferretti, London School of Economics, London, UK

Keith Frankish, University of Sheffield, Sheffield, UK

Vittorio Gallese, University of Parma, Parma, Italy

Gerd Gigerenzer, Max Planck Institute for Human Development, Berlin, Germany

Francine W. Goh, University of Nebraska, Lincoln, NE, USA

Ulrich Hoffrage, University of Lausanne, Lausanne, Switzerland

Sheena Iyengar, Columbia Business School, New York, USA

Phil N. Johnson-Laird, Princeton University, Princeton, NJ, USA

Inga Jonaityte, Università Ca’ Foscari, Venice, Italy

Bryan D. Jones, University of Texas, Austin, TX, USA

Konstantinos V. Katsikopoulos, University of Southampton Business School, Southampton, UK

Kathryn Laskey, George Mason University, Washington, DC, USA

Laura Macchi, University of Milano-Bicocca, Milan, Italy

Miles MacLeod, University of Twente, the Netherlands

Luigi Marengo, LUISS, Rome, Italy

Julian N. Marewski, University of Lausanne, Lausanne, Switzerland

Laura Martignon, Ludwigsburg University, Ludwigsburg, Germany

Antonio Mastrogiorgio, Università degli Studi G. d'Annunzio Chieti e Pescara, Italy

Colin H. McCubbins, Walmart Media Group, Stanford University, CA, USA

Mathew D. McCubbins, Duke University, Durham, NC, USA

Zachary A. McGee, University of Texas, Austin, TX, USA

Hugo Mercier, Institut Jean Nicod, CNRS, Paris, France

xi
xii

Contributors

Raffaella Misuraca, University of Palermo, Sicily

Shabnam Mousavi, Johns Hopkins University, Baltimore, MD, USA

Richard R. Nelson, Columbia University, New York, USA

Nancy J. Nersessian, Harvard University, Cambridge, MA, USA

Samuel A. Nordli, Indiana University, Bloomington, IN, USA

Adam Oliver, London School of Economics, London, UK

Thorsten Pachur, Max Planck Institute for Human Development, Berlin, Germany

Enrico Petracca, University of Bologna, Italy

Torsten Reimer, Purdue University, West Lafayette, IN, USA

Elena Reutskaja, IESE Business School, Barcelona, Spain

Ariel Rubinstein, School of Economics, Tel Aviv University, Tel Aviv, Israel

Brooke N. Shannon, University of Texas, Austin, TX, USA

Giacomo Sillari, LUISS, Rome, Italy

Özgür Şimşek, University of Bath, Bath, UK

Dan Sperber, Central European University, Vienna, Austria

Keith E. Stanovich, University of Toronto, Toronto, Canada

Keith Stenning, University of Edinburgh, Edinburgh, UK

Jeffrey R. Stevens, University of Nebraska, Lincoln, NE, USA

Jean-Louis Stilgenbauer, Facultés Libres de Philosophie et de Psychologie (IPC), Paris,

France

Thomas Sturm, ICREA & Universitat Autònoma de Barcelona, Barcelona, Spain

Cass R. Sunstein, Harvard University, Cambridge, MA, USA

Paul Thagard, University of Waterloo, ON, Canada

Nicolaus Tideman, Virginia Polytechnic Institute and State University, Blacksburg, VA, USA

Clement A. Tisdell, School of Economics, the University of Queensland, Brisbane, Australia

xii
xiii

Contributors

Peter M. Todd, Indiana University, Bloomington, IN, USA

Mark Turner, Case Western Reserve University, Cleveland, OH, USA

Riccardo Viale, University of Milano-Bicocca, Milan, Italy

Ignazio Visco, Bank of Italy, Rome, Italy

Massimo Warglien, Università Ca’ Foscari, Venice, Italy

Gregory Wheeler, Frankfurt School of Finance & Management, Frankfurt, Germany

Giordano Zevi, Bank of Italy, Rome, Italy

xiii
xiv
newgenprepdf

xv

PREFACE

The Routledge Handbook of Bounded Rationality takes its inspiration from the seminal work of
Herbert Simon, who introduced the concept and developed it in a number of areas: from eco-
nomics to administrative science; from philosophy to cognitive science; from political science
to Artificial Intelligence. I started collaborating with Herbert Simon after I first met him in
1988. One product of this collaboration was the birth of the journal, Mind & Society in 2000.
After his death in February 2001, I was one of a group of economists and cognitive scientists
influenced by his work who founded the Herbert Simon Society in Turin. Its aim was to pro-
mote and update some of the many topics of Simon’s research, but in particular his realist phil-
osophy of science for an empirically based microeconomics and social science. The Routledge
Handbook of Bounded Rationality is inspired by this work and develops a critical assessment of
varied and often contradictory interpretations and applications of bounded rationality.
The project of the book stems from the meetings held at the Max Planck Institute for
Human Development in Berlin and the annual workshops of the Herbert Simon Society held
in New York and Turin between 2011 and 2019. I would like to thank Dan Kahneman, Joe
Stiglitz, Gerd Gigerenzer, Alvin Goldman, Edward Feigenbaum, Ned Block, Dan Sperber,
Colin Camerer, Peter Todd, Ralph Hertwig, Laura Martignon, Pat Langley, Denis Hilton,
Barbara Fasolo, Shabnam Mousavi, Hersh Shefrin, Laura Macchi, Massimo Egidi, Giovanni
Dosi and Luigi Marengo in particular, as well as the many other researchers involved in the
discussions on bounded rationality that took place in these meetings.
Special thanks to Konstantinos Katsikopoulos for his precious advice and suggestions on
topics and contributors.
I am indebted to Laura Gilardi and Anna Mereu for their assistance in the initial organiza-
tion of the work.
I thank Sarah Cuminetti for her text editing and review of some chapters.
This Handbook would not have been possible without the support of Routledge, and in par-
ticular Andy Humphries and his collaborators, in particular, Emma Morley, and Susan Dunsmore.
Last but not least, I wish to thank the Collegio Carlo Alberto of Turin and the Herbert
Simon Society for their contributions to the making of this Handbook.
Riccardo Viale
Turin,
16 April 2020

xv
xvi
1

1
WHY BOUNDED
RATIONALITY?
Riccardo Viale

In the philosophy of science why-questions are a way to address the foundations of scientific
explanation. For Bas van Fraassen (1980), a theory of explanation is essentially a theory of why-
questions. Why-questions and their answers are individuated only relative to a context.1 The
pragmatic problem of the context determines the relevance of the answer and ultimately the
choice made therein.
Daniel Dennett (2017) breaks down the sense of why-questions into two options: “How
come?” and “What for?”. The former refers to the factual description, to the illustration of the
physical or chemical process that explains a phenomenon. For example, wondering “Why did
we witness a total eclipse of the moon on … July?” would refer to the earth’s positioning in
the orbit of the sun and the moon and the shadow projected by the earth on the moon. The
latter refers instead to the reasons that explain a phenomenon. For example, asking “Why did
Britons vote for Brexit?” refers to “for which reason”, that is to the actual motivations – such
as fear of immigration, hostility towards Brussels Eurocrats, communication errors made by
the Remainers, etc. – that led the majority of Britons to vote in favour of this unexpected
choice.
In some instances, a question can be answered using either of the alternatives, depending
on the context of reference. Dennett asks himself the question concerning evolution by natural
selection. One part of the answer may be described using the “How come?” option, that is
through a description of the physico-chemical processes that led to the emergence of the first
eukaryotic being. At this point, the question becomes “What for?”, that is what are the reasons
that led to the selection of some species over others? The question “Why bounded rationality?”
can be addressed in the same way. In this case, the order of the alternatives is reversed, compared
to evolutionary selection. The first question is not about why the term was introduced, that
is to say, what is the reason that in 1950 led Herbert Simon to bestow this name on a series of
empirical phenomena he had observed while studying administrative organization? The first
question is, “For what reason?” is human rationality non-maximizing and non-optimizing, but
rather bounded, heuristic, and satisficing. The answer provided in this volume indicates that
it was the capacity to better adapt to the uncertainty and complexity of the environment that
led to the evolutionary selection of this type of rationality over Olympic or unbounded ration-
ality. The second question is then, “How come?”, that is to say through which neurocognitive
and corporal processes did bounded rationality emerge? This Handbook will try to answer

1
2

Riccardo Viale

this question by identifying the cognitive aspects that characterize the style of human decision
making and analysing the neural and motor, sensorial and visceral embodiment correlates that
explain the heuristic characters of bounded of rationality.

Part I Naturalizing bounded rationality

Rationality or reason holds an important role in philosophical tradition. In brief, we can dis-
tinguish two main philosophical schools of thought on rationality (Viale, 2012): the first affirms
the superiority of reason compared to intellect; the second asserts the opposite. In the former,
the classical philosophical tradition, reason is the strength that liberates us from prejudices, from
myth, from established but false opinions, and from appearances, and which allows a universal
or common criterion to be established for all areas of human conduct. In the tradition that
spans from Heraclitus, Parmenides, Plato and Aristotle to St Augustine, Descartes, Spinoza,
Leibniz and the Enlightenment, reason is man’s fundamental and universal guide; it is what
differentiates us from animals and allows us to tell truth from falsehood, because, as Seneca
states, “Reason is nothing other than a part of the divine spirit descended (or sunk) into a
human body.” It is universal because it is present in all persons, who, as Descartes affirms, think
differently only because they apply reason differently.
In opposition to this Olympic and universal vision of reason, we find the second pos-
ition. According to this minority tradition, expressed by Neoplatonism, St Thomas, medieval
scholastics, Francis Bacon and, to a large extent, Kant, reason is subject to the primacy of
intellect. Intellect should be regarded as superior because it is equipped with that intuitive and
immediate character that allows it to gain a direct understanding of empirical reality, unlike
reason which is limited by its discursive and a priori nature. While affirming the discursive
nature of both, Kant also maintained that only that of the intellect was valid since its concepts
are immediately derived from experience.
Although severely criticized by Kant, the primacy of reason and its discursive nature
remains the dominant position in philosophy. This discursive and linguistic character, as
summarized in Aristotle’s syllogistic or in the Cartesian ideal of the chains of reasoning in
geometry, leads to the formal development of the theory of rationality in the past century.
The resulting logic of rationality will always maintain, as in Aristotle, the dual descriptive
valence of its own procedures of reason, and normative valence in the sense of the rule for
its correct use.
The philosophical tradition of rationality can be considered a subsector of the larger epis-
temology one. In particular, the development of the Naturalizing Epistemology programme
introduced by Quine in the 1960s shaped the normative framework for the contemporary con-
cept of bounded rationality. Let us see why (see Viale, 2013):
Among the most fundamental questions which epistemology has sought to answer are the
following:

1. How ought we to arrive at our beliefs?

2. How do we arrive at our beliefs?
3. Are the processes by which we do arrive at our beliefs the ones by which we ought to arrive
at our beliefs?

Traditionally, the answers to these questions were as follows: both epistemology and psych-
ology should carry out their research independently and separately, and then, once they have
answered questions 1 and 2 respectively, they will attempt to answer question 3.

2
3

Why bounded rationality?

For example, suppose that psychological studies were to demonstrate that people arrive at
their beliefs by some kind of non-conscious mechanism that measures the coherence of new
beliefs with the body of beliefs already held, and which accepts only those that cohere and
rejects those that do not, this would have no bearing on the merits on the epistemological
coherence theory of justification which states that one can only adopt beliefs cohering with
beliefs one already has. The normative questions that epistemologists ask are completely inde-
pendent of the descriptive questions psychologists ask.
However, there is another way to answer the three questions. This is the approach used by
the project for naturalizing epistemology: question 1 cannot be answered independently of
question 2. The question of how we actually arrive at our beliefs is therefore relevant to the
question of how we ought to arrive at our beliefs.
This position is well summed up by the following passage from Quine:

Epistemology becomes as a chapter of psychology and hence of natural science. It

studies a natural phenomenon, viz. a physical human subject. This human subject is
accorded a certain experimentally controlled input – certain patterns of irradiation
in assorted frequencies, for instance – and with a little of time the subject delivers
as output a description of the three-dimensional external world and its history. The
relation between the meagre input and the torrential output is a relation that we are
prompted to study for somewhat the same reasons that always prompted epistem-
ology; namely, in order to see how evidence relates to theory, and in what ways one’s
theory of nature transcends any available evidence for it.
Quine, 1985, p. 24

What prompted this reversal of approach? Largely it was the failure of the foundationalist pro-
ject which tried to show that there is a class of beliefs – typically beliefs about our own sensory
experience – about which it is impossible to be wrong. Moreover, these beliefs were held to
be sufficient to justify the rest of our beliefs. Carnap’s project was aimed at the translation, the
rational reconstruction of every assertion about the world in terms of sensory data, logic and
set theory.
If the project had succeeded from a “conceptual” point of view, namely, the “technical” pos-
sibility of achieving this translation, it would in any case have failed to overcome the “doctrinal”
barrier, namely, the problem of preserving the content of truth within the translation. Merely
by translating an assertion in terms of sensory data, logic and set theory does not mean that it
can be verified by this translation. The most modest of observational generalizations will always
cover more cases than those observed by the observer. Therefore, any attempt to found beliefs
on immediate experience is hopeless, from a logical point of view, even if this is the simplest
empirical generalization.
Also from a conceptual point of view, this translation programme produced few results. It
attempted to reduce every scientific assertion to a neutral language of observational data, logic
and set theory. First, an attempt was made using “direct definitions”, then with “contextual
definitions”, by which sentences containing the term were translated into equivalent sentences
lacking the term. Lastly, with the “reduction modules” of Carnap’s liberalized programme,
hope was given up of translating a sentence into an equivalent and it ended up by explaining
a new term by specifying some sentences which are implied by the sentences containing the
term, and other sentences which imply sentences containing the term.
As Quine states, this minimal objective renounces the last remaining advantage of a pro-
gramme of rational reconstruction, namely, the advantage of reduction by translation.

3
4

Riccardo Viale

If all we hope for is a reconstruction that links science to experience in explicit ways,
short of translation, then it would seem more sensible to settle for psychology. Better to dis-
cover how science is in fact developed and learned than to fabricate a fictitious history of how
our ancestors introduced those terms through a succession of Carnap’s “reduction modules”
(Quine, 1985, p. 21).
Does this mean that the empirical foundation of knowledge, the empirical meaning of
sentences about the world, is no longer founded on solid bases? Quite the contrary. Our
knowledge of the external world is based and founded precisely on the empirical meaning
of language, as is actually attained in the process of each individual’s learning of language.
The common meaning that we attribute to words and to sentences about the external world,
namely, the basis for our possibility of communicating and understanding, and also the empir-
ical meaning of science, rest in the last instance on the common empirical basis of the common
meaning we attribute to our assertions about the world, and this empirical basis can only be
described and explained by empirical psychology.
A further shift towards a naturalization of epistemology occurs at the moment when the
meaning of the three questions is further examined by focusing attention on the cognitive
mechanisms of rationality, the various internal processes of the cognitive elaboration of beliefs,
on the processes whereby from one belief we reach a different belief, namely, on the processes
of deductive and inductive reasoning and inference. This is the decision-making moment of
what action to take; the assessment of assumptions and hypotheses; the following of arguments
and reasoning; deciding what weight and importance to give to the evidential data; the solution
of problems.2
In the programme of naturalizing epistemology, therefore, an important chapter is occupied
by a study of the natural mechanisms of reasoning, judgement, and decision making which
must be immune from any kind of logicist preconception and apriorism.
To sum up, a broad concept of rationality has always been linked to normativity. The realm
of a priori demonstrative proof, that is, the things that people were absolutely certain about,
remained the goal of epistemology and science until the beginning of the twentieth century.
Scientific rationality fell into a definitive crisis with the failure of the neopositivist programme
and the emergence of relativism and socio-historical reductionism or, in other words, the
dominance of the context of discovery over the context of justification. Likewise, in the twen-
tieth century, the inclusive area of epistemological rationality had its radical change with the
Naturalizing Epistemology programme of Willard O. Quine. There are no ways of analytically
founding the truth. The only possibility is to discover the natural processes that humans follow
in generating knowledge.
The realm of certainty and demonstrative proof had already experienced a strong downsizing
during the seventeenth century with the acknowledgement of the irreducible uncertainty of
human life and the emergence of the theory of probability. The impact of this revolution was
at once directed to morality and human choice theory. Blaise Pascal and Pierre Fermat defined
the border of reasonableness as the choice of the alternative that maximizes expected value. The
various paradoxes from St. Petersburg to Allais and Ellsberg led to change in the theory of rea-
sonableness, attempting to resolve the discrepancy between description and normativity. This
was made by tinkering with the utility or probability function while at the same time retaining
the ideal of maximization or optimization. In any case, the concept of reasonableness, after a
decline of interest during the end of the nineteenth and the beginning of the twentieth century,
re-emerged in the 1950s and 1960s in the form of the concept of “rationality” in economics,

4
5

Why bounded rationality?

psychology and other social sciences. This rationality was assimilated to the calculation of prob-
abilities, utilities and optimal decisions. The ideal of optimization entered not only the behav-
ioural sciences but also animal biology (e.g. optimal foraging theory) and artificial intelligence
(e.g. optimal artificial agents) (Gigerenzer and Selten, 2001).
As for the epistemology and also for the philosophy of science, the crisis of the a priori
concept of rationality of choice took a major step forward in the mid-twentieth century with
the introduction by Herbert Simon of Bounded Rationality theory. In Simon, the naturalistic
approaches find a common interpreter: both directly in the theory of rational choice, and also
in the philosophy of science where Simon, in opposition to Karl Popper, is the major sup-
porter of the logic of discovery as the psychology of discovery; and indirectly, in epistemology
where he contributed to the naturalistic theory of knowledge through his works on inductive
reasoning and problem solving.
Thomas Sturm (in Chapter 3 in this volume) recognizes that naturalism has drawn a variety
of connections between rationality and empirical sciences, mainly cognitive sciences. It started
out as a reaction to “armchair” methods such as conceptual analysis, thought experiments or
appeal to intuition. In rationality the target was to substitute the a priori standard substantive
view of rationality with a procedural view that relies on the empirical behavioral models of
reasoning and decision making. Sturm (Chapter 3 in this volume) writes: “One of Simon’s
most influential ideas was to replace the goal of maximizing expected utility by “ ‘satisficing’”
(p. 78). Understood this way, bounded rationality questions the sharp divide between the
descriptive and the normative – quite different from Kahneman’s understanding of the con-
cept. However, no Is-Ought fallacy, no psychologism needs to be implied thereby. Instead,
there is an underlying and legitimate principle of normativity: Ought implies Can (also already
recognized by Kant). “The intractability of many reasoning tasks restricts the applicability of
certain formal and optimizing norms, and makes it reasonable to look for different, more
feasible norms” (Sturm, p. 78). Sturm emphasizes this third position between descriptive and
normative, making reference to Alvin Goldman’s theory of reliabilism. Goldman’s approach to
epistemic rationality is instrumentalist. Given certain epistemic goals, people ought to use reli-
able procedures to attain them. Sturm writes:

Goldman’s approach is an excellent example for this. Among other things, he shows
that formal rules of the standard picture of rationality often do not map straight-
forwardly onto reasoning tasks. For instance, the logical law of noncontradiction,
¬(¬p∧p), states that a proposition and its negation cannot both be true. From
contradictions, anything whatsoever can be derived. We should, it seems, therefore
avoid inconsistencies. What do you do when you discover inconsistencies in your
belief set? Are you obliged to give up the whole set? Certainly not; we should separate
the wheat from the chaff. However, that can prove to be difficult, especially in the
complex belief systems of science, when several beliefs are all tenuous but part of a
systematic theory. Similar examples can be construed for other logical rules, such as
deductive closure, or for rules of probability. There is, therefore, no simple derivation
of norms of epistemic rationality from formal rules of the standard picture; more
support for the gap mentioned earlier on.
To bridge that gap, Goldman maintains that a distinctively philosophical or
“analytic” epistemology has to determine the criteria or goals of our epistemic
endeavors. He defends reliabilism: our belief-forming mechanisms need not always

5
6

Riccardo Viale

guarantee truth, but to safeguard a good ratio of true and false beliefs. From psych-
ology, we can learn what mental processes or mechanisms are reliable and which
ones are not.
Sturm, p. 82

As pointed out by Rubinstein (Chapter 27 in this volume) and Gigerenzer (Chapter 2 in this
volume), there are several views of bounded rationality. This is not only because bounded
rationality is present in various fields such as economics, psychology and management. Even
within a single field such as economics, there are clear differences. There are at least three
different positions: (1) optimization under constraints; (2) failures to follow the prescriptions of
the optimization model; and (3) adaptive behaviour to the environment.
Katsikopoulos (Chapter 4 in this volume) distinguishes between two cultures, which he calls
“idealistic” and “pragmatic”:

At a first approximation, the idealistic culture pursues a minimum departure from the
neoclassical-economics framework of unbounded rationality which assumes the ideals
of omniscience and optimization of a utility function and adds factors such as inequity
aversion or probability weighting to the utility function. On the other hand, the prag-
matic culture holds that people sometimes ignore information and use simple rules of
thumb in order to achieve satisfactory outcomes.
Katsikopoulos, p. 90

Is the epistemic dimension of the two cultures an important philosophical feature? Katsikopoulos
follows the tripartition proposed by Alan Musgrave in characterizing the views of the relations
between facts and theory. Musgrave discusses three views of when an empirical fact lends
support to a model. In the logical view, it matters only if the fact is consistent with the model’s
implications. In the historical view, it also matters if the model’s implications were derived
before or after the fact was observed. In the third view, a variant of the historical view, it is
additionally relevant what the implications of the best competing model are. More support is
provided for the model if its best competitor does not imply the observed fact. Thus, the logical
view accepts as an epistemic aim the explanation of known facts. On the other hand, the historical
view rejects this and aims at the prediction of new facts. For Katsikopoulos:

It may be argued that the idealistic culture espouses the logical view whereas the prag-
matic culture is aligned with the historical view, and in particular Musgrave’s variant.
For example, the development of prospect theory and other risky choice models
which follow the utility-times-probability mathematical form, has been following the
empirical violations of the axioms of expected utility theory … On the other hand,
pragmatic models such as the priority heuristic, have not been developed in order to
account for these violations … but rather in order to predict new facts.
p. 95

Concerning bounded rationality, pragmatic cultures underline the importance of the environ-
ment. The importance of the environment in defining bounded rationality is expressed radic-
ally by the following quotation from Simon: “an ant [or human being], viewed as a behaving
system, is quite simple. The apparent complexity of its behavior over time is largely a reflection
of the complexity of the environment in which it finds itself ” (1981: 63–65). This example
is even more radical than the scissors metaphor: “Human rational behaviour… is shaped by a

6
7

Why bounded rationality?

scissors whose blades are the structure of task environments and the computational capabilities of the actor”
(Simon, 1979b; emphasis added). The environment becomes the main barycentre of rationality.
This polarization is reminiscent of Karl Popper’s situational logic. The environment dictates
its bounds to organisms, be they ants or humans, that have species-independent and universal
reactions to the structure of the environment. For example, Simon writes that we can “dis-
cover, by a careful examination of some of the fundamental structural characteristics of the
environment ... the mechanisms used in decision-making” (Simon, 1956, p. 130). Simon’s pos-
ition tends to overlook the importance of the decision maker’s biological constraints. It would
appear that the embodied structure of decision making does not tell us anything interesting
about the interaction between the environment and the decision maker. This is very difficult to
accept because we know how fundamental the biological bounds of neural, sensorimotor and
visceral systems are in driving the style of decision making. From this point of view, it would
be like setting out to screw in a Phillips screw without caring if the screwdriver is a flathead
or a Phillips. The topic will be analysed in depth in the following sections, and in particular
in the sections on ecological rationality and embodied bounded rationality (BR). Teppo Felin
and Mia Felin (Chapter 5 in this volume) focus on the BR of perception and support the
importance of focusing on the biological features of the perceiver. They show how cues and
stimuli coming from the environment are not species-independent but are instead specific to
the organisms. They write:

The fields of ethology and comparative biology offer the best evidence for the fact
that environments are species-specific rather than general … Each organism has its
own, unique “Umwelt” (surrounding world), where awareness and perception are a
function of the nature of the organism itself …
Felin and Felin, p. 110

A frog does not see an insect if it does not see it moving. They add:

This highlights that what “stimulates”—or which cues become salient and natur-
ally assessed …—has less to do with the stimulus itself (or the objective or general
‘amount’ of some cue: Gigerenzer and Gaissmaier, 2011). Rather, awareness of par-
ticular stimuli has more to do with the nature of the organism … Only those stimuli
or cues that are species-specific “light up,” or are processed and attended to, while
many other things are ignored.
Felin and Felin, p. 110

From this point of view, BR has to do with environments that are species-specific. Felin and
Felin cite the fathers of ethology, Nikko Timbergen and Konrad Lorentz, and particularly
Jacob von Uexkull, when they affirm that “every animal is surrounded with different things,
the dog is surrounded by dog things and the dragonfly is surrounded by dragonfly things”,
that is, “each environment forms a self-enclosed unit, which is governed in all its parts by its
meaning for the subject”. Moreover there are pragmatic goals beyond the biological bounds
which drive the selection of perception. Our perception is focused on what counts for us con-
tingently. We are blind to things that are not part of our pragmatic attention. Therefore, BR
should rely on a perception that selects cues and information from the environment according
to species-specific reasons and pragmatic interests. To understand the Umwelt of frogs and
men, we should focus on the embodied adaptation to the environment: a frog developed
species-perception in relation to the possibility of its body and the specific affordances from

7
8

Riccardo Viale

the environments (see locusts and flies). Every human being X has his or her own X things
environment. Pragmatic goals and special and particular embodied cognition (think of percep-
tual variability among individuals) select an individual X environment. This X environment is
not constant, but it may also change during the life of the individual. Because pragmatic goals
and embodied cognition change.
Bounded rationality played an important role in modelling scientific discovery. Discovery
as the psychology of problem-solving was Herbert Simon’s answer to the deductive ration-
ality of the context of justification developed by neopositivism and post-neopositivist authors
like Karl Popper and, to some extent, Imre Lakatos. Psychological feasibility and proced-
ural rationality are the building blocks of the naturalizing philosophy of science that aims
to establish a descriptive and prescriptive model of scientific discovery. Miles MacLeod and
Nancy Nersessian (Chapter 6 in this volume) analyse the bounded rationality concept in a
particular field of life sciences, computational systems biology. This field aims to produce
large-scale models of complex biological systems. Since this field relies on computer simu-
lation, it would be safe to assume that the computational power of automated modelling
practices would overcome the limitations of the human mind. On the contrary, MacLeod
and Nersessian write:

Given the rhetoric of “big data” and automated modelling practices, one might expect
that computer simulation is a resource for side-stepping human cognitive capacities
and constraints. Yet even these practices still require human cognitive engagement
with their products for the purposes of validation and interpretation. And, as we
have discovered in our investigations of computational systems biology, much problem
solving still involves researchers building their own models in situations of scarce or
inadequate data, using laptop computers. Rather than freeing scientists of their cog-
nitive limitations, using computation to extend control and understanding over ever
more complex systems requires developing sophisticated problem-solving practices
which push them towards these constraints.
MacLeod and Nersessian, p. 120

They monitored two laboratories where the aim was to create high-fidelity computational
models of metabolic, cell signalling, and gene-regulatory networks. These models are dynamic,
tracking the changes in concentration of each metabolite, or the activity of each gene, in a
network over time. This goal was never attained. The modellers should have relied on strat-
egies to build midlevel models called “mesoscopic” that are simplified both at the system
level, with only few relationships represented, and at the network level, with quite abstract
non-mechanistic representations of biochemical interactions. These strategies serve to sim-
plify running simulations, parameter fitting and mathematical analysis. The modellers tried to
improve these inaccurate and unreliable models by using a sequence of increasingly complex
heuristics. This was done through inferences drawn by processing information in the system
comprising memory and environment through a distributed model-based reasoning. This coupled
cognitive system reduces the cognitive load by off-loading memory to the environment and
increases the bounded rational success of the models. MacLeod and Nersessian write: “The
practice of mesoscopic model-building and incremental modification through building-out
strategies provides an exemplar of what Simon, in a later analysis of bounded rationality, called
‘procedural rationality’” (p. 126).
An important aspect of Herbert Simon’s naturalistic view of rationality is linked to the
psychology of problem solving. Every decision is the solution to a problem and therefore the

8
9

Why bounded rationality?

analysis of the procedural rationality of a decision must take into account the ways in which
the space of the decision making/problem solving is represented as well as the operators avail-
able, the goal, the related sub-goals, and the various steps leading to the final decision/solution.
In Human Problem Solving (2019), Newell and Simon clearly define what is meant by problem
solving according to the approach of Information Processing Psychology. This dimension of
problem solving applies to everyday individual or organizational routine decisions as well as
to the solution of ill-structured problems in scientific and technological research and in any
creative thinking situation. The tradition that prevailed in the psychology of reasoning and
decision making was based on the analysis of the conformity of the product of reasoning and
decision making with the norms of deductive and inductive logic and the theory of utility. It
was noted that human reasoning and decision making do not conform to those canons and
this was labelled a pathology of rationality in the form of bias, error and irrational propensities.
Human beings have been depicted as obtuse reasoners and decision makers predisposed to
fallacies and therefore are ill-adapted to their environment. The tradition of pathological ration-
ality focused on the formal aspect of thinking and neglected to consider the actual adaptive
weight of reasoning and decision-making processes. Moreover, it remained silent and defence-
less before the matters of creativity, scientific discovery, invention, and in general before the
answers given to ill-structured and complex situations. The problem-solving tradition, first
with Karl Duncker and then with Allen Newell and Herbert Simon, followed a remarkably
different path, as clearly explained by Laura Macchi and Maria Bagassi (Chapter 7 in this
volume) with reference to the Gestalt tradition. Simon writes:

The simulation models of the 1950s were offspring of the marriage between that had
emerged from symbolic logic and cybernetics, on the one side, and Wurzburg and
Gestalt psychology, on the other … From Wurzburg and Gestalt psychology were
inherited the ideas that long-term memory is an organization of directed associations,
and that problem solving is a process of selective goal-oriented search.
Simon, 1979a, pp. 364–365

Macchi and Bagassi write that, concerning Gestalt psychology in particular, “Duncker was not
only interested in the moment when the solution was found, but also in the process of thought,
the phases by which the restructuring is reached” (p. 135) and they add

The solution process involves a selective research in phases, over a delimited area; it requires
that paths explored previously that turned out to be dead ends be discarded and new
paths adopted. The new path can even have been evaluated in a previous phase.
p. 135

And, “According to Duncker, the problem solver never blunders along blindly, seeking the
solution by trial and error; he proceeds by applying heuristic strategies and methods” (p. 137).
This situation is particularly evident in what is defined as insight problem solving, that is to say
the solutions that are perceived as an unexpected act of creativity. The problem is familiar to
Gestaltists like Duncker for whom the sudden comprehension of the Aha! moment of the new
is linked to a sudden restructuring of the thought material.

When in 1990, Simon, in collaboration with Kaplan, explored in depth the peculiar-
ities of the search and the steps of the solution process for insight problems, he dealt
with the issue of a change in representation, explaining that the shift to an alternative

9
10

Riccardo Viale

problem space occurs because a heuristic has been adopted, initiated by boundaries
of salience.
Macchi and Bagassi, p. 140

Macchi and Bagassi also state:

According to the authors, the subject who finds himself at an impasse must first of all
make a conscious decision to change his way of looking at the problem, by the “Try
a Switch” meta-heuristic, given that the initial representation of the problem did not
provide him with the means of reaching the solution.
p. 140

Macchi and Bagassi write that once this decision has been made, the subject

must be on the lookout for the important cues that will help him select the represen-
tation that will lead to the solution, from the infinite number available. In his search
for these decisive cues to switch the representation in the representation metaspace, he
is assisted by adopting a heuristic that fits the problem, also called the Notice Invariants
Heuristic.
Macchi and Bagassi, p. 140

However, in this type of process it would appear evident that, unlike Simon’s position, the
dynamics of thought mainly occur at the subconscious level, with no reference to working
memory or the conscious retrieval of data from long-term memory. Several psychological and
neurocognitive studies (for example, on the presence of the Default Mode Network) on mind
wandering and incubation show that some associative processes are unconscious and at the same
time they are rule-based (Viale, 2016).
A particular perspective from which to analyse bounded rationality is the representation
of mathematical reasoning tasks and the consequent educational problems. Laura Martignon,
Kathryn Laskey and Keith Stenning (Chapter 8 in this volume) analyse the legacies in educa-
tion left by Simon, many of which are elucidated in his Sciences of the Artificial (1981). Simon
writes: “Solving a problem simply means representing it so as to make the solution transparent”
(1996,. p. 132). Martignon, Laskey and Stenning write:

A solution becomes transparent if it emerges from the structure in terms of which the
problem has been modelled. The structure of the problem is the result of an effort
of adaptation between the problem itself and the conceptual constructions of the
mind. This adaptation is successful if it is ecologically rational. Ecological rationality
is a fundamental characteristic of successful representations. It refers to behaviors and
thought processes that are adaptive and goal-oriented in the context of the environ-
ment in which an organism is situated, thus giving survival advantage to the organism.
Discovering a representation that makes a problem easily solvable gives evolutionary
advantage in terms of time resources to adopters of the new representation, because
they are able to solve the problem more quickly and easily than those holding to the
old representation.
Martignon, Laskey and Stenning, p. 157

10
11

Why bounded rationality?

Different representations are not equivalent from a cognitive point of view. Some are more
adaptive and advantageous than others, because they correspond better to the cognitive features
and possibilities of the agent. For example, Martignon, Laskey and Stenning write:

An ecologically rational strategy is one our cognitive apparatus can perform natur-
ally and easily, and that has adaptive value in a given environment. In other words,
ecological rationality involves the discovery of a representation that makes an optimal
or nearly optimal solution transparent. For reasoning from evidence to hypothesis,
because the natural frequency representation is ecologically rational, it allows humans
to perform an otherwise difficult task quickly and easily. Use of the natural fre-
quency representation in educational settings facilitates understanding and improves
performance.
p. 160

Boundedly rational agents are limited in their computational, processing, and storage abil-
ities. Chapter 9 in this volume by Cristina Bicchieri and Giacomo Sillari focuses on epistemic
bounds, that is limits to agents’ knowledge:

Not all knowledge a boundedly rational agent possesses is explicitly stored in memory
…, nor does the agent need full epistemic capacities in order to do her part in inter-
active situations … Indeed, there is a gulf between knowledge that is available to an
agent and knowledge that is accessed by an agent …
Bicchieri and Sillari, p. 189

In their chapter, they discuss logical models of awareness that precisely draw the distinction
between unbounded (implicit) and bounded (explicit) knowledge through a formalized notion
of accessibility. A specific aspect of bounded rationality is the limit to human logical abilities and
in particular limits to the way real agents reason about knowledge. A useful way to understand
its limits is differential accessibility

Part II Cognitive misery and mental dualism

According to many authors (Evans, Chapter 10 in this volume; Stanovich, Chapter 11 in this
volume), our brains are made up of two systems: one automatic and associative (System 1) and
the other reflective and analytical (System 2). The first is responsible for biases, and the other
is able to correct them. They maintain that the difference between the two systems lies at the
heart of bounded rationality. System 1, that prevails in numerous decisions, does not follow
the normative principles of rationality. It is only System 2 that, when successfully overriding the
choices of System 1, has the possibility to process normatively rational answers. The problem
lies in the fact that this occurs fairly rarely. Most individuals are cognitive misers, that is, they
tend to make the smallest possible effort in terms of attention and processing. Very few are able
to seriously engage their analytical capabilities. However, in many everyday situations, when
the time to make decisions is limited or the problems are too complex, even those few become
cognitive misers. They limit their analytical capability and by so doing, they detach from the
normative requirements of deductive, probabilistic and economic rationality.
There are numerous differing positions on dual-process or dual-system theories and on their
relative properties. The question is, which of the 23 different theories identified by Stanovich

11
12

Riccardo Viale

can be deemed acceptable? An initial selection can be made by considering the neuroscientific
data that seem to show that there is no single System 1, with a single set of attributes, but rather
multiple cognitive and neural systems. Therefore, it is preferable to use the terminology of
dual types of processing. There are at least two sets of theories: the parallel-competitive theories
(Sloman, 1996; 2002) assume that Type 1 and 2 processing proceed in parallel, each having
their say with conflicts resolved if necessary. Criterion S refers to situations where individuals
are led to respond in a manner consistent with Type 1 but then come to realize an alternative
responding consistent with Type 2.
In contrast, there are the default-interventionist theories (Evans, 2008; Evans and Stanovich,
2013). They assume that fast Type 1 processing generates intuitive default responses on which
subsequent reflective Type 2 processing may or may not intervene. The list of the properties
contains at least 6–10 features. Usually it is expected that these are co-occurring properties. In
this case, if all of these features do not always co-occur, then the dual-process view is incorrect.
In fact, Kruglanski and Gigerenzer (2011) claim that dual-process views fail because “these
dimensions are unaligned rather than aligned” (p. 98). That is to say, there are no examples in
which all the features co-occur.
Evans and Stanovich (2013) reply that a minimal part of the paired properties usually
attributed to the two processes is defining. The majority are correlates that might or might not
be present.

In summary, our view is that the defining feature of Type 1 processing and of Type
2 processing is not the conjunction of eight different binary properties (“2 cells out
of 256!”). Research has advanced since the “suggestive list of characteristics” phase
of over a decade ago. Our view of the literature is that autonomous processing is the
defining feature of Type 1 processing. Even more convincing is the converging evi-
dence that the key feature of Type 2 processing is the ability to sustain the decoupling
of secondary representations. The latter is a foundational cognitive requirement for
hypothetical thinking.
Stanovich and Toplak, 2012, p. 7

In conclusion, there are only two defining properties: the Type 1 processes are autonomous
while the Type 2 processes are characterized by cognitive decoupling.3 “The defining feature
of Type 1 processing is its autonomy – the execution of Type 1 process is mandatory when
their triggering stimuli are encountered, and they are not dependent on input from high-level
control systems” (Stanovich and Toplak, 2012, p. 7). Cognitive decoupling is the defining fea-
ture of Type 2 processing. When we reason hypothetically, we create temporary models of
the world and test out actions in that simulated world. In order to reason hypothetically, we
must be able to prevent our primary representations of the real world from becoming confused
with the secondary representations of hypothetical situations, that is, we should decouple
our representations (Stanovich and Toplak, 2012). Leslie (1987) was the first to introduce the
concept when she was modelling pretence by positing a secondary representation that was a
copy of the primary representation but that was decoupled from the real world. Decoupling
allows it to be manipulated, that is, be a mechanism for simulation. Decoupling secondary
representations from the world and then maintaining the decoupling while simulations is
carried out is the defining feature of Type 2 processing. Decoupling seems to be a special
features of our race. Since we were becoming the first creatures to rely on cognitive simulation,
it was a matter of vital importance not to become “unhooked” from the world too much of
the time. Therefore, the salience of the primary representations of the worlds remained difficult

12
13

Why bounded rationality?

and costly to overcome. Therefore, the ability to sustain mental simulations while keeping the
relevant representations decoupled is likely a key aspect of the brain’s computational power
(Stanovich and Toplak, 2012). Decoupling is the feature of Type 2 processing to override and
to reverse Type 1 processing that produces responses that are nonoptimal. To summarize, Type
1 processes are autonomous because they do not require “controlled attention and they make
minimal demand on working memory and central processing resources”. They are mandatory
when their triggering stimuli are encountered and they are not dependent on input from high-
level control systems. Type 2 processes are characterized by decoupling, that is the ability to
prevent our representations of the real world from becoming confused with representations of
imaginary situations. Such processing is necessary for hypothetical thinking, mental simulation,
counterfactual reasoning and consequentialist decision making.
In reality, humans are cognitive misers. The tendency is to default to processing mechanisms
of low computational expense (Stanovich, Chapter 11 in this volume). They engage the brain
only when all else fails, and usually not even then. Cognitive misers have three rules: (1) default
to Type 1 processes whenever possible; (2) when that is not possible and analytic processing
is necessary, default to serial associative cognition with focal bias (as in the case of matching
bias in a selection task); and (3) when the reflective mind wants to start cognitive simulation
by decoupling, not complete it, that is, override failure (Stanovich, 2009, p. 69). Moreover if
the analytic processes want to override the autonomous processes, it is possible that they won’t
succeed because the mindware is not available or it is contaminated. What is the mindware? It
includes rules, procedures and strategies that can be retrieved by Type 2 analytic processes and
used to transform decoupled representations to override the autonomous mind. It is mainly
the product of past learning experiences. If a rule is not learned or is not well learned or is not
appropriately applied, this is the cause of an override failure. For example, knowledge of the
gambler fallacy for a pathological gambler (Stanovich, 2009, p. 73). But not all mindware is
helpful for overriding autonomous processes. Some acquired mindware can be the direct cause
of errors and biases. For example, egocentric thinking, the my-side perspective, the evaluation-
disabling memes (the memes against critical thinking and for consensual dogmatic faith) or false
lay psychological theories (e.g. the personal immunity to bias or the personal knowledge of the
causes of own actions, etc.).
In the same direction, Evans (Chapter 10 in this volume) proposes the Default Interventionist
theory as a framework for bounded rationality:

The model assumes that Type 1 processing leads to a default intuitive answer A1 which
appears in consciousness without effort or reflection. A1 is then subject to scrutiny
by Type 2 processes which are effortful and require working memory. However, such
scrutiny can be shallow or deep which depends on a number of factors. Motivational
factors include how people are instructed, whether they have a rational thinking dis-
position (which inclines people to check intuitions more carefully) and the feeling or
rightness (FOR) or subjective confidence that they have in A1 … Cognitive resources
included time available (speeded tasks will reduce Type 2 checking), individual
differences in working memory capacity (highly correlated with IQ) and “mindware”.
Evans, p. 189

When A1 is scrutinized, it may or may not be deemed satisfactory. Whether the intuition
can successfully be replaced by an answer based on explicit reasoning and knowledge, how-
ever, depends on the cognitive resources available, as Stanovich has argued. Decision making
is affected by individual differences in both cognitive ability and rational thinking dispositions.

13
14

Riccardo Viale

The latter measures the extent to which people will rely on intuitions or check them out by
reasoning. On the basis of this work, Stanovich argues that a number of conditions must be
met for someone to successfully engage in Type 2 reasoning and solve a demanding and novel
experimental problem. First, people must be aware of the need for intervention by Type 2
reasoning, which is where rational thinking dispositions and feelings of rightness come into
play. However, if they do intervene, then this will only be successful both if (1) they have suffi-
cient cognitive capacity for the hypothetical reasoning involved; and (2) they have the relevant
mindware, that is, formal or procedural knowledge required by the context.
An important aspect of bounded rationality is linked to the feeling of rightness (FOR) of
default intuitive answers that come to mind. When the FOR is high, we are happy with our
intuitions that we deem satisfactory (satisficing) for our decision making. According to Evans:

The FOR research seems to fit a neat story of evolution and bounded rationality. Our
intuitions come helpfully packaged with feelings of confidence which tell us whether
we need to expend effort or not in checking them out. However, there is a snag.
None of this research – on reasoning and decision-making – has provided evidence
for even a partial accuracy of the FOR judgement, as observed in other fields where
metacognition is studied.
p. 191

There are two comments on these results:

The first is that there is no clear reason to think that FOR corresponds to a stage of
cognitive processing in the manner that researchers in the field assume. They could
simply be conscious feelings that are by-products of the processes that determine the
behaviour measures. In other words, a brain process that leads to a quick and firmly
held decision, also generates conscious feelings of rightness and confidence, which in
themselves have no functional significance.
Evans, p. 192

The second is that the evolution of metacognitive feeling (like FOR) might be used simply to
justify some bias in given adaptive contexts. It is well known (e.g. Stich or Sperber) that norma-
tive canons like truth and consistency are not always evolutionary adaptive. A belief bias might
increase more significantly the individual’s fitness before real and present dangers than a precise
and correct but slow and time-consuming syllogistic reasoning. Natural selection provides the
reproductive advantage of one organism over the next, not for the optimality of any one char-
acteristic (for example, deductive reasoning).
If the mind is structured into two parts or process types,4 Stanovich (Chapter 11) wonders
which evolutionary representation can explain the emergence and the conflict between these
two components of mental activity. He relies on Dawkins’ discussion of replicators and vehicles:

Replicators as entities (e.g. genes) that copy themselves and vehicles as the containers
(e.g. organisms) in which replicators house themselves. It is vehicles that interact with
the environment, and the differential success of the vehicles in interacting with the
environment determines the success of the replicators that they house. Humans have
proven to be good vehicles for genes. As have bees.
Stanovich, p. 197

14
15

Why bounded rationality?

But the goal structures of bees and humans are very different (Dennett, 2017). The bee’s goal
structure shows a predominance of the replicator’s goals over those of the vehicle. In case of
conflict, the vehicle’s goals are sacrificed. A given bee will sacrifice itself as a vehicle if there is
greater benefit to the same genes by helping other bees (for example, causing its own death when
it loses its stinger while protecting its genetically-related hive-queen). In humans, however, the
replicator’s goals that are mainly oriented towards reproductive fitness can be neutralized by the
vehicle’s goals or may not even be useful to achieve reproductive fitness. In the former case,
the vehicle’s goal, for example, the fulfilling of pleasure, leads us to prevent pregnancy through
the adoption of contraceptives. Therefore, the reproductive goal is reduced for reasons linked
to the vehicle’s goals. In the latter, the goals that were effective for reproductive goals in the
past are no longer useful because of the speed of environmental change. For example, the con-
sumption of excessive fats, that was adaptively useful in the past, no longer fulfils the goals of
either the vehicle or the replicator. Goals such as economic rationality and any other type of
optimality are no longer among the replicator’s goals, only among the vehicle’s.
How does this framework relate to the dualism of the mind and bounded rationality? The
mind generates, through System 1, behaviours that are mainly instinctive, linked to the her-
editary component of our nervous system. These are a set of behaviours led by the replicator’s
long-term goals, that is to say the human genome. System 2, however, has introduced new
short-term goals and new forms of reasoning and decision making aimed at fulfilling those
goals. System 2 is primarily a control system focused on the interests of the whole person. It is
the primary maximizer of an individual’s personal goal satisfaction. The problem lies in the mis-
match between the two systems. The goals of System 2 often correspond to interests that are
common to both the vehicle and the replicator. Other times, they are in contrast with those of
the replicator. In those cases, System 1 tends to prevail and rational optimality goals are there-
fore set aside (Viale, 2019). Stanovich (Chapter 11) is more optimistic:

Because System 2 is more attuned to the person’s needs as a coherent organism than
is System 1 (which is more directly tuned to the ancient reproductive goals of the
subpersonal replicators), in the minority of cases where the outputs of the two systems
conflict, people will often be better off if they can accomplish an override of the
System 1-triggered output. Such a system conflict could be signaling a vehicle/repli-
cator goal mismatch and, statistically, such a mismatch is more likely to be resolved in
favor of the vehicle (which all of us should want) if the System 1 output is overridden.
When humans live in complex societies, basic goals and primary drives (bodily
pleasure, safety, sustenance) are satisfied indirectly by maximizing secondary symbolic
goals such as prestige, status, employment, and remuneration. In order to achieve
many of these secondary goals, the more directly-coded System 1 responses must be
suppressed—at least temporarily. Long-leashed derived goals create the conditions
for a separation between the goals of evolutionary adaptation and the interests of the
vehicle.
Stanovich, p. 201

Samuel Bellini-Leite and Keith Frankish (Chapter 12 in this volume) criticize a version of
the dual-process theory that seems closed to the Fodorian division of mind into encapsulated
input systems and unencapsulated central system. According to Fodor, the central system is
represented as an unbounded central processor, sensitive to global assessments of context and
relevance. Bellini-Leite and Frankish write:

15
16

Riccardo Viale

It is tempting to think of conscious thought as open-ended and unconstrained, and

of nonconscious processes as inflexible and encapsulated. Cognitive psychology and
computational neuroscience show that this is wrong. Conscious cognition is in fact
severely limited in capacity, while nonconscious processes are tuned to the heavy
demands of contextual and relevance processing. It is thus vital that modern dual-
process theories do not carry over the conception of rationality implicit in Fodor’s
precursor theory. In thinking about the functions and capacities of the posited dual
systems, the perspective of bounded rationality is essential.
p. 214

Recent models of brain functioning that undermine the rigid Fodorian distinction between
strictly encapsulated systems, the modules, and free unencapsulated systems, the central pro-
cessor, are the Predictive Processing (PP) theories (Clark, 2016). The brain’s function is to pre-
dict proximal stimuli not only to process them. Bellini-Leite and Frankish write:

These theories see the brain as having a hierarchical structure, with lower levels being
close to perceptual input mechanisms and higher levels receiving information from
diverse multimodal regions. Signals flow both up and down the hierarchy, with top-
down signals predicting lower-level activity and bottom-up signals flagging errors in
those predictions. Predictions are based on probabilistic generative models distributed
through the hierarchy. These models monitor statistical patterns in the layer below,
generating nonconscious hypotheses in an attempt to accommodate incoming data.
Within this architecture, the influence of different neural regions is modulated to
reflect their success in prediction through a mechanism known as precision weighting.
p. 213

As outlined in the following section on embodied bounded rationality, there are no sharp bound-
aries between perception, judgement and action. The limited System 2 receives representations
that have already been filtered and processed. Lower layers pass on only stimuli that generate
high prediction errors and that may be analysed and corrected by S2.
Stanovich (Chapter 11) writes that researchers working in the heuristics and biases trad-
ition tend to see a large gap between normative models of rational responding and descriptive
models of what people actually do. These researchers have been termed Meliorists because they
assume that human reasoning is not as good as it could be, and that thinking could be improved.
However, over the last several decades, an alternative interpretation of the findings from the
heuristics and biases research programme has been championed. This group of theorists have
been termed the Panglossians. These theorists often argue either that the normative model being
applied is not the appropriate one because the subject’s interpretation of the task is different
from what the researcher assumes it is, or that the modal response in the task makes perfect
sense from an evolutionary perspective. According to Stanovich, the contrasting positions of
the Panglossians and Meliorists define the differing poles in what has been termed the Great
Rationality Debate in cognitive science – the debate about whether humans can be systemat-
ically irrational. The three contributions that follow outline the arguments for the Meliorist
and Panglossian approaches with regard to deductive reasoning, causal reasoning and decision
making. Part III on ecological rationality corresponds to what Stanovich calls the Panglossian
position, as opposed to the Meliorist one that follows the tradition of Daniel Kahneman and
Amos Tversky’s heuristic and biases and the application to public policy by Richard Thaler and
Cass Sunstein.

16
17

Why bounded rationality?

Phil Johnson-Laird (Chapter 13 in this volume), together with Wason, started working on
mental dualism in connection with logical reasoning. Johnson-Laird’s theory maintains that
when dealing with a deduction, the use of mental models happens intuitively, corresponding
to System 1 activity. The same deduction can also be analysed deliberately according to the
rules of logic with greater difficulty and effort. This corresponds to the activation of System
2. In the case of reasoning based on mental models, the results are not logically complete,
because not all the model possibilities are made explicit compared to a deliberate analysis.
Johnson-Laird writes that “human reasoners are equipped with two systems for reasoning,
which interact with one another. Intuitions depend on mental models, which represent only
what is true” (p. 222). Bounded rationality in deduction is based on dependence on mental
models. He adds:

Our reasoning is bounded. We make deductions in domains that are undecidable

or intractable in their demands on time and memory. Yet, we are rational in prin-
ciple. We grasp the force of counterexamples, and spontaneously use them to refute
invalid inferences. Our deductions from fully explicit models yield valid conclusions
based on all the information in the premises. Alas, some inferences in life violate
the desirable logical properties: their logic is incomplete, and they lack a deci-
sion procedure or a tractable one. In any case, to make a rational deduction is at
least to maintain the information in the premises, to simplify, and to reach a new
conclusion.
p. 217

One of the ways in which the individual represents the environment and interacts with it is
through causal attribution. When faced with any social phenomenon, for example, a fall in
your stock’s market value, or a natural one, like the onset of a number of physical ailments, the
individual will try to understand the causes of that effect, in other words, will engage in diag-
nostic reasoning. Diagnostic reasoning consists in going back to the causes that triggered one or
multiple effects. Formally, it means estimating the probability Pr (cause|effect). It is well known
that the estimation of this probability deviates from the rational norm defined by the Bayes rule.
Recent results show that people follow very varied strategies in order to estimate the diagnostic
probability. Most often, the encountered strategies consist in combining the quantities within
the Bayes rule in a sub-optimal fashion. Jean Baratgin and Jean-Louis Stilgenbauer (Chapter 14
in this volume) suggest

the existence of additional strategies based on patterns of defeasible reasoning. In

particular, two schemes appear to be psychologically plausible. The first is a form of
defeasible deduction based on the Modus Ponens (DMP), the second is a scheme of
defeasible abduction based on Affirming the Consequent (DAC).
p. 232

These two reasoning patterns represent two formally different strategies for estimating the
diagnostic probability Pr(cause|effect), as shown in Figure 14.3 on p. 233. They correspond
respectively in the A.I. and philosophical literature to two modes of inference: abduction for
DAC and deduction for DMP. First, the defeasible deduction is a weakened form of deduction
that permits the production of provisionally true and/or probable conclusions. Second, abduc-
tion is a heuristic reasoning that serves to identify explanatory causes or hypotheses. Baratgin
and Stilgenbauer show:

17
18

Riccardo Viale

the choice of the scheme to estimate the diagnostic probability Pr(cause|effect) depends
on the value of the predictive probability Pr(effect|cause). When Pr(cause|effect) <
Pr(effect|cause), participants do not report any particular preference between DMP
and DAC. Yet, when Pr(cause|effect) > Pr(effect|cause), participants report a clear
preference for DMP to estimate the diagnostic probability.
p. 234

Decision making can be represented according to two main approaches: the alternative-based
approach and the attribute-based approach. The former includes compensatory decision
making. Overall values for each option in a choice set are generated before comparing options
to arrive at a decision. According to the second category, attributes across options are compared
and these attribute comparisons are used to make a decision. Because they compare attributes,
they may not use all available information to make a choice, which categorizes many of them
as heuristics. Francine Goh and Jeffery Stevens (Chapter 15 in this volume) write:

Choice data have suggested that attribute-based models can better predict choice
compared to alternative-based models in some situations (e.g., when there are many
options in the choice set, when calculating an overall value for an option is too cog-
nitively taxing).
p. 243

As Part III on ecological rationality will show, the situation where attribute-based decision
making is more successful is characterized by uncertainty and complexity, which are the envir-
onmental features of bounded rationality. In fact, Goh and Jeffrey write:

While the study of decision making has historically focused on alternative-based

models, attribute-based models have experienced a resurgence of interest from
researchers for a number of reasons. First, they follow from Simon’s notion of bounded
rationality because they often reflect real-world limitations faced by decision makers
by using less information and simpler computations. This is especially pertinent in
instances where decision makers have to make a choice from myriad options or when
there is risk involved in the decision making process. Second, they capture choice data
quite well, predicting multi-attribute, risky, intertemporal and strategic choices while
accounting for, or bypassing, anomalies regularly encountered in the use of alterna-
tive-based models. Third, in addition to capturing choice data, attribute-based models
can also capture the decision process by making predictions about eye tracking and
information acquisition data.
p. 250

Part III Ockam’s razor: mental monism and ecological rationality

The strong version of the dual-process account of mind is untenable. The dual list of the 6–10
properties are not co-occurring, therefore, the dual-process view is incorrect. However, even
the minimal dual account of mind, based on only one defining property, autonomy for Type 1
processes and decoupling for Type 2 processes, seems untenable. Psychological and neural data
seem a direct empirical falsifier (Viale, 2019). The mind or psychological processes would not
be split into the first part that is prone to vices, impulsiveness and error, and the second part,

18
19

Why bounded rationality?

wise, rational and ready to correct the errors of the first part. The mind seems to be better
represented by a unified model, that is, by a single account of mental processes.
What are the features of the single account of mind and what are the implications for the
bounded rationality theory? The two main anti-dualistic approaches are that of the “cogni-
tive continuum” put forward by Hammond (1996) and Cleerman and Jimenez (2002; see also
Osman, 2004) and the “rule-based processing unified theory of decision making” argued by
Kruglanski and Gigerenzer (2011).
First, according to the dynamic graded continuum (DGC), different types of reasoning are
dependent on the representations from which participants reason (Osman, 2004, p. 1002).
The same underlying production rule can generate a variety of responses as a result of the
features of the task that an individual considers relevant (Osman, 2004, p. 1003). The proper-
ties of reasoning (control, awareness, speed, etc.) vary in degree, depending on the structural
features of the tasks that induce different cognitive activity (Hammond, 1996). The cognitive
continuum ranges from intuition to analysis. The better-structured a task is, the more ana-
lytically induced will be the decision-making mode. Conversely, with an ill-structured task,
decision making is likely to be intuition-induced (Hammond, 1996). These theories seem to
fit better than dual-process theories to a set of criteria like the Criterion S,5 the individual
differences in cognitive ability, the dissociation between implicit and explicit processing, as
analysed in a series of tasks as the selection task, the conjunction problem and belief bias in
syllogistic reasoning (Osman, 2004, pp. 996–1005). What matters is whether in this architec-
ture of the mind, there can be correction and reversibility after the bias has been generated.
From what can be deduced from the “cognitive continuum” thesis, this type of processing
appears to be triggered by the structure of the problem. If a problem is opaque and unnatural
in its logical structure or if it is linguistically and pragmatically confused and deviant or if
it is unfamiliar, namely, the quality of representations is poor, this will stimulate the intui-
tive, fast and implicit part of the mind responsible for biases. In this case, contrary to what
is proposed by dual-process theories, there is no possibility of correction by the analytical
component which is only stimulated by well-structured problems and the relative quality of
representations (Viale, 2019).
Second, according to the “rule-based processing unified theory of decision making” argued
by Kruglanski and Gigerenzer (2011), a cognitive judgement is mediated by rules that are in
the individual’s adaptive toolbox. How is the selection made? It follows a two-step process: first,
the task and individual memory constrain the set of applicable rules, resulting in a consider-
ation set, and, second, difficulty of instantiation, individual processing potential and (perceived)
ecological rationality of the rule guide the final choice of a rule from a set (Kruglanski and
Gigerenzer, 2011, p. 102).
Difficulty in instantiation means difficulty in recognizing the rule-matching cue in given
circumstances. The presence of noises in the judgemental environment and weak and faint
cues are responsible for this difficulty. The individual might overcome this difficulty and try to
retrieve from memory rules that are difficult to access if he or she possesses sufficient processing
potential. The processing potential are characterized by two features: attentional capacity and
motivation. When they are low, the individual is not able to conduct an extensive memory
search to retrieve the suitable rules for the structure of the task. Moreover, in such conditions,
the individual may be sluggish and incapable of applying rules whose implementation requires
effortful computations. The individual tends to base his or her judgement on relatively simple
inferential rules and may be less able “to carefully assess the ecological rationality of a rule, that
is, to properly estimate its validity in a given environment” (Kruglanski and Gigerenzer, 2011,
p. 103). With limited attentional capacity and motivation, easy-to-use heuristics are used to a

19
20

Riccardo Viale

greater extent than more complex inferential rules. For example, it was found that the “higher
the magnitude of individuals’ accuracy and motivation, or their need for cognition, the greater
their readiness to apply complex rules and to digest compound information” (Kruglanski and
Gigerenzer, 2011, p. 102). To sum up, in situations where the suitable rules are complex, the
signal to noise ratio is low, the individual has low resources, low conditions of processing poten-
tial and a low need for cognition (Petty and Cacioppo, 1986), and where there is a need for fast
cognitive closure, unsuitable, easy-to-use heuristics are used. The consequence are biases and
not ecologically rational decisions. This kind of decision environment is most frequent and it
limits individual behaviour to choices that are not adaptive. In these cases, the reversibility of
decisions taken is very difficult and it may only occasionally happen, ex post, when the indi-
vidual experiences the negative consequences of his choice. But also in this case, beyond the
constraints of adverse initial conditions, such as low processing potential and signal-to-noise
ratio, emotional factors such as inertia, the status quo and present-day bias can also block any
reversibility of choice.
To summarize, when faced with ill-structured, opaque problems and situations, there are two
possible styles of decision making, according to the unified theory of mind: (1) the “need for
cognition” individual may decide to invest cognitive resources and time analysing the problems;
and (2) the “lack of cognition” individual is not endowed with cognitive resources and he or
she takes fast decisions based on “gut feeling”, and, soon afterwards, inertia, the status quo and
procrastination phenomena prevent her or him from analysing the situation and improving the
structure and representation. As many empirical studies show, the need for cognition people
are a small minority. Generally, people lack attention, motivation and cognitive resources to
analyse ill-defined and opaque problems. They rely on fast, intuitive, emotional answers and
they lack the cognitive and emotional resources to analyse the problems and review the answers
afterwards (Viale, 2019; forthcoming).
Hugo Mercier and Dan Sperber (Chapter 16 in this volume) hold a similar position.
They write:

In light of these results, and of the fundamental problems that affect standard dual
process models, we have suggested an alternative theory that is much more in line
with the ideas of bounded rationality … Instead of dividing the mind between system
1—which would mostly abide by the dictates of bounded rationality—and system 2—
which wouldn’t—we suggest that it is intuitions all the way up (see also Kruglanski
and Gigerenze, 2011; Osman, 2004). However, some of these intuitions—which form
our ability to reason—would bear on reasons, as explained presently. In this theory,
reason heavily relies on its cognitive and social environment to solve problems—when
to kick in, how to figure out if something is a good reason, and how to find relevant
reasons.
Mercier and Sperber, p. 259

According to Mercier and Sperber, the theory of reason concerns the central aspect of
bounded rationality, that is the interactive and recursive relation between inferential activity
and the environment. The theory is dubbed the interactionist view of reason because the reasons
supporting one’s theories or used to evaluate other individuals’ arguments are generated at the
metarepresentational level (reasons on reasons) and dialogically with ourselves (more rarely), but
mostly with regard to others. The key feature of this theory is that these reasons are generated
adaptively following the heuristic of satisficing:

20
21

Why bounded rationality?

The production of reasons, as we describe it, is a paradigmatic example of a satisficing

process. The criterion to be reached is that of producing reasons good enough to
convince one’s interlocutor. For the interactionist account, reason evolved by being
used chiefly in dialogic contexts, in which we can benefit from the back and forth
of discussion to refine our arguments, instead of attempting to anticipating through
extraordinary computational force what the silver bullet might be. This means, how-
ever, that people are not well prepared to produce strong reasons in the absence of
feedback (except in the relatively rare circumstances where such feedback can be reli-
ably anticipated).
Mercier and Sperber, p. 263

An important aspect of this theory concerns its implications in environment design, in the
architecture of choice that promotes the generation of good reasons. The relevance of Mercier
and Sperber’s analysis is clear with regard to choice architecture in the theory of Nudge and
Boost. As Part VII of this volume will show, one way to empower the citizens in decision
making is to promote environments that boost their capacity to analyse and make conscious
choices, through increasing environmental feedbacks, the heuristic simplification of informa-
tion and a critical dialogue with the members of social context of reference.
What is the role of bounded rationality in the cognitive sciences? The main model explaining
how information is processed and how decision are made is David Marr’s (1982) model on the
three levels of analysis. Originally developed with Tomaso Poggio, these three levels map onto
fundamental distinctions in the study of rationality. At the highest level, termed the computa-
tional level, the aim is to identify an appropriate formal system, a calculus, used to define the
problem and derive its solution. Moving down to Marr’s second level, the algorithmic level,
the aim is to specify the algorithms and data structures needed to compute the solution. Marr’s
third level of analysis is the implementation level, which considers the constraints of physically
instantiating the proposed algorithms and data structures in, say, biological machinery or a con-
ventional digital computer. Henry Brighton writes (Chapter 17 in this volume):

Marr’s levels of analysis are seen as a core tenet of cognitive science … defining the
de facto categories used to orientate cognitive models in relation to rationality claims
… From this Marrian perspective, bounded rationality is typically seen as an attempt
to inform computational-level theory development by importing constraints arising
from algorithmic-level concerns. These constraints revise the problem specification,
and consequently, what constitutes a rational solution.
p. 270

For example, it has been proposed that simple heuristics might be used to approximate a full
Bayesian computation specified at the computational level. Ecological rationality involves a
conjecture about simple heuristics, that belongs to Marr’s algorithmic rather than the com-
putational level. And given that rationality claims are traditionally made at the computational
level, the assumption is that ecological rationality must therefore inherit principles of Bayesian
rationality to explain the success of simple heuristics. Brighton writes:

The claim here is that ecological rationality is an adaptation to the problem of deci-
sion making in large worlds. These are environments that we can observe but our
partial ignorance precludes them from being probabilistically quantifiable … and a key

21
22

Riccardo Viale

implication of this claim is that optimality ceases to be a meaningful characteristic of

rational decisions under these conditions.
p. 277

In a large world characterized by uncertainty, problems cannot be faced through optimal

computations made by a formal system, but rather through exploratory heuristics characterized
by the “less-is-more” principle. Therefore, the three Marrian levels of analysis are not suitable
to characterize ecological rationality, while a two-level analysis is useful to pursue a form of
functionalist inquiry.
When Herbert Simon introduced the concept of BR, most economists showed little interest
in Simon’s question and preferred theories assuming perfect foresight and optimal computations.
According to Gerd Gigerenzer (Chapter 2 in this volume), today a different phenomenon has
gained momentum among economists:

Instead of being rejected as worthless or unimportant, his concept of bounded ration-

ality was hijacked and radically reframed, making his revolutionary ideas no longer
recognizable. Neo-classical economists absorbed the term into the orthodoxy of per-
fect foresight and concluded that satisficing is quintessentially optimizing. Proponents
of the heuristics-and-biases program in psychology, in contrast, appropriated the term
for their own focus on human lack of rationality. This is why bounded rationality now
has three faces that could hardly be more dissimilar.
Gigerenzer, p. 59

According to Gigerenzer, there are three principles that define Simon’s programme of bounded
rationality:

1. Uncertainty. To study decision making under uncertainty, not only risk.

2. Process. To study the actual process of decision making, as opposed to as-if expected utility
maximization.
3. Scissors. To study how the structure of an environment, together with the cognitive process,
produces the resulting behavior.

Gigerenzer writes:

The rationality in neoclassical economics typically refer to three pillars: consistency,

maximization of expected utility, and—if learning is involved—Bayesian updating
of probabilities. Leonard J. Savage (1954), known as the father of modern Bayesian
decision theory, defined two conditions necessary for these three pillars of rationality:

1. Perfect Foresight of Future States: The agent knows the exhaustive and mutually
exclusive set of future states S of the world.
2. Perfect Foresight of Consequences: The agent knows the exhaustive and mutually
exclusive set C of consequences of each of his or her actions, given a state.
p. 56

The term ‘risk’ may be applied to the previous situations with perfect foresight of future
states and consequences. These are also called ‘small worlds’ by Savage (1954). What Simon
noticed, however, was that the managers he observed and humans in general mostly have to

22
23

Why bounded rationality?

deal with situations that are unlike small worlds with perfect foresight and that do not meet the
assumptions for expected utility maximization and Bayesian updating (Savage, 1954, p. 16).
Simon called for an empirical study of how humans reason when perfect foresight is not pos-
sible. The real-life problems are inside a complex environment. They are typically ill-defined
problems, that is, the goals are not definite; we don’t know what counts as an alternative and
how many alternatives there are; it’s unclear what the consequences might be and how to
estimate their probabilities and utilities (Viale, 2018). This environment might be called also
large world (Savage, 1954) and it is characterized by uncertainty. Small worlds instead are in
principle predictable and without surprises and they are characterized by the knowledge of
all relevant variables, their consequences and probabilities. The conditions of small world are
the requirements of Neoclassical Rationality as Simon stressed in his Nobel Lecture (1979,
p. 500). In these worlds the problems may be well defined but they can be also computation-
ally intractable. As is well known, an example of a computational tractable problem is the dice
game or the roulette game. Instead a well-defined problem such as a chess game is computa-
tionally intractable. In any case, the real world is most of the time large and these conditions
of knowledge are rarely met.6 Since they are rarely met, the normative rational requirements
of neoclassical economics are unjustified and the application of their theories can easily lead to
disaster (Stiglitz, 2010). Unfortunately, behavioural economics, while it criticizes the descrip-
tive side of neoclassical economics, without really proposing an alternative realist model of
decision making, retains the normative one. In fact, the heuristic and biases programme is
developed to cope with what is called human irrational behaviour, characterized by biases
and formal errors caused by psychological mechanisms as the heuristics. Thaler (1991, p. 138)
writes very clearly about:

A demonstration that human choices often violate the axioms of rationality does not
necessarily imply any criticism of the axioms of rational choice as a normative idea.
Rather, the research is implied intended to show that for descriptive purposes, alter-
native models are sometimes necessary.

In a large world, the axioms of rationality cannot be applied. Therefore, they cannot be
considered feasible normative canons of rationality. Behavioural economists, who decades ago
defined their critical contribution to the neoclassical mainstream a purely descriptive enter-
prise (ibid.) now advocate using behavioural concepts for prescriptive policy purposes as in the
Nudge theory (Thaler and Sunstein, 2008).
When Herbert Simon began his attempt to empirically change economics, his methodo-
logical and epistemological coordinates were realist (Simon, Egidi, Viale and Marris, 1992;
Simon, 2000). His main critical target was the instrumentalist as-if approach of Milton Friedman
(1953). A descriptive enterprise in economics had to overcome the unbounded rationality
assumptions of neoclassical economics as unbounded self-interest, unbounded willpower and
unbounded computational capacity. The behavioural economics programme initiated by Simon
had the goal of replacing these a priori assumptions with more realistic ones. How much psy-
chological realism has been brought into economics by behavioural economists? Unfortunately
very little because there are barriers to psychological realism that are common to neoclassical
economics and that are the son of the shared reliance on Friedman’s as-if principle (Berg and
Gigerenzer, 2010). All relevant behavioural theories suffer from the same shortcomings of neo-
classical economics: assuming that risky choice always emerges from a process of weighting
and averaging all the relevant pieces of information; the decision maker knows the objectively
feasible action set; the decision maker know the list of outcomes associates with lotteries or the

23
24

Riccardo Viale

probabilities of the known outcomes (Berg and Gigerenzer, 2010). The shift from neoclassical
economics to behavioural economics and in particular, after the impact of Allais Paradox, from
expected utility theory to prospect theory, appeared to be based on the introduction of more trans-
formations with additional parameters to square the basic operation of probability-weighted
averaging with observed choices over lotteries (Berg and Gigerenzer, 2010). The weighting-
and-adding objective function is used as-if it is a model of mind. But it’s not. It is a fic-
tional mind, a valid instrument to make a posteriori inferences through the introduction of
suitable parameters in order to reach a better R-squared. The same methodological model is
observed in many other behavioural theories (Berg and Gigerenzer, 2010). For example, the
Fehr and Schmidt social preference model (1999) recognizes the insight that people care about
others’ payoffs. Therefore, they modify the utility function through the addition of at least two
additional free parameters. People are assumed not to maximize a utility function depending
only on their own payoffs but a behavioural or other-gathering utility function. To do this, a
decision maker assigns benefits and costs to each element of the choice space based on weighted
sum of the intrinsic benefits of own payoffs together with the psychic benefits of being ahead
of others and the psychic costs of falling behind others. The decision maker will select the
action with the largest utility score based on the weighted summation. Another as-if model is
Laibson’s (1997) model of impulsiveness in consumption, a psychological bias that over-weights the
present over the future. He puts more weight on the present by reducing weight on all future
acts of consumption. In other words, he reduces the weight of all terms in the weighted sum
of utilities except for the term representing the utility of current consumption. The unrealistic
pretension is evident: the decision maker, after an exhaustive search of all possible acts of con-
sumption, computes the weighted sum of utility terms for each act and chooses the one with
highest weighted utility score. The deviation between the value that recovers the neoclassical
version and the new parameter that reduces the weight on the future is considered the empirical
confirmation of the model (Berg and Gigerenzer, 2010).
The instrumentalist methodology of behavioural economics uses the addition and man-
aging of free parameters to improve the realism of the models. In so doing, it improves the
within-sample fit and improves the R-squared. Most of the philosophers of science, both in
the realist tradition (e.g. Hacking, 1983) and in the antirealist tradition (e.g. van Fraassen,
1980), agree on the empirical adequacy by successful prediction, particularly of novel facts, as
the first principle in deciding between competing hypotheses (Viale, 2013). A large number
of free parameters allow the model to fit many sets of data without proving it generates
successful out-of-sample prediction. On the contrary, the most challenging test of a theory is
in prediction using a single set of fixed parameters. Something that few models of behavioural
economics dare to do.
In conclusion, we live in a large world where the canons of neoclassical rationality are unjus-
tified both descriptively and normatively. Therefore, the reasoning errors, fallacies and biases
that behavioural economists are engaged to overcome most of the time are not irrationalities.
Moreover, the decision-making models that behavioural economists have introduced most of
the time are an as-if instrumentalist tool to fit the observed choice data.
Since we live in large world characterized ontologically by complexity, recursivity,
non-linearity and uncertainty, the rationality of choices should be judged by their adap-
tivity and problem-solving ability. In fact, bounded rationality is not confined only to the
constraints of the computational power of human mind. As in Simon’s scissors metaphor
(1990), rationality should be judged by the matching or mismatching of the relationship
of mind-environment or in other words choice-task structure. What kind of reasoning
processes are able to match the environmental tasks and solve the problems? This is an

24
25

Why bounded rationality?

empirical question that some years ago confronted some cognitive scientists, such as Herbert
Simon, Vernon Smith, Richard Selten and, in particular more directly, Gerd Gigerenzer and
the ABC Research Group (Gigerenzer, Todd, and the ABC Research Group, 1999). The
adaptive toolbox of formalized heuristics is the result of these empirical investigations. In a
number of problems, simple heuristics were more accurate than standard statistical methods
that have the same or more information. The results became known as the “less-is-more
effect”. There is a point where more is not better, but harmful. There is an inverse-U-
shaped relation between level of accuracy and amount of information, computation, or
time (Gigerenzer and Gassmaier, 2011, p. 453). For example, “starting in the late 1990s it
was shown for the first time that relying on one good reason (and ignoring the rest) can
lead to higher predictive accuracy than achieved by a linear multiple regression” (Gigerenzer
and Gassmaier, 2011, p. 453). Herbert Simon himself spoke, in his appraisal of the volume
by Gigerenzer, Todd and the ABC Research Group (1999), of a “revolution in cogni-
tive science, striking a great blow for sanity in the approach to human rationality”. The
toolbox is composed of many heuristics that have been tested successfully against statistical
algorithms of rationality, not in the easy task of fitting a closed sample of data but in the
much harder task of prediction. They have proved to be both a better description of deci-
sion making and a better prescription on how to decide. Obviously, the adaptive success of
any given heuristic depends on the particular given environment. In which environments
will a given heuristic succeed, and in which will it fail? Gigerenzer, Todd et al. (2012) have
identified a number of environmental structure variables:

Uncertainty: how well a criterion can be predicted.

Redundancy: the correlation between cues.
Sample size: number of observations (relative to number of cues).
Variability in weights: the distribution of the cue weights.

How do we assess the adaptive success in ecological rationality? Gigerenzer and Gassmaier
(2011, p. 457) write: “The study of ecological rationality results in comparative statement of
the kind ‘strategy X is more accurate (frugal, fast) than Y in environment E’.”
According to Julian Marewski and Ulrich Hoffrage (Chapter 18 in this volume):

The fast-and-frugal heuristics framework asks four basic questions: descriptive, eco-
logical, applied, and methodological …

1. Descriptive: What heuristics do people rely on to make decisions, and when do

they rely on which heuristic?
2. Ecological: To what environmental structure is each heuristic adapted, such that
relying on that heuristic can aid people to make clever decisions; and reversely, in
which environments will that heuristic fail?
3. Applied: How can human performance be improved so that environments match
heuristics––be it by changing the environment in which people act or by chan-
ging the heuristics people rely upon?
4. Methodological: How can people’s use of heuristics and the heuristics fits to different
environments be studied?

Candidate answers to those four questions have been formulated in numerous fields
and disciplines to which the fast-and-frugal heuristics program has been applied …

25
26

Riccardo Viale

How can those four question aid understanding drastic environmental change––not-
ably, the ways in which individuals think, decide, and act?
p. 284–285

They analyse how to cope with the radical environmental changes caused by the digital revo-
lution. They proceed in six steps:

First, we will speculate how future aversive environments might look, and second
we will discuss how such environments might then influence behavior, for instance,
by leading people to adopt different heuristics. Third, we will speculate how those
heuristics might, in turn, shape environments, namely, the societies in which they are
enacted. Subsequently, we will explore how aversive environmental change can be
managed––both by an individual navigating through a changing environment (fourth)
and by societies as a whole undergoing environmental change (fifth). We will close this
chapter by, sixth, turning from changes in life-history to changes in evolutionary history.
Marewski and Hoffrage, p. 292

For example, they propose three heuristic principles to create more resilient digitalized societies
that avoid the growing perverse effects of interconnectedness, influenceability and traceability:

1. Principle of disconnectedness: Implement policies that disconnect people from

each other.
2. Principle of deceleration: Implement policies that slow down the spread of informa-
tion among people.
3. Principle of information-loss: Implement policies that prevent the unlimited storage
of behavioral data.
Marewski and Hoffrage, p. 297

These principles aim to answer the questions leading to the sixth step, that concerns changes
in evolutionary history. The digital world differs from the predigital environment. Its features
make the adaptation of human beings less easy. These principles aims at shaping the digital
environment in a way that increases the ecological rationality of human beings.
An important feature of ecological rationality is that it situates BR within the frame-
work of natural selection. Ancestral environments and fitness pressure shaped the evolved
capacities of human decision making. Samuel Nordli and Peter Todd (Chapter 19 in this
volume) write:

But the evolutionary perspective enables us to do more, looking at the ultimate selec-
tion pressures that operated over time in our and other species’ evolutionary trajec-
tories—that is, the study of any organism’s decision making as interactions between
(A) its evolved capacities, needs, and desires/goals, in terms of (B) how those aspects
fit the structure of the past environmental decision-making contexts, and (C) how
they match or mismatch the structure of the present environmental contexts.
Nordli and Todd, p. 313

Ecological rationality expands the concept of BR by emphasizing the role of past environments
to which we have adapted, the present environment in which we make decisions, and the
structure of information as it is processed through decision making. In this way, “the ecological

26
27

Why bounded rationality?

rationality approach follows in the investigative tradition of behavioral ecology as championed

by Tinbergen …, who argued that proximate (mechanistic and developmental) and ultimate
(adaptive and evolutionary) analyses are each required in order to fully understand observed
behaviour” (Nordli and Todd, p. 313). The weakness of current research in decision making lies
in the fact that it overlooks the explanation coming from evolutionary psychology and behav-
ioural ecology. It is focused only on the description of the mechanisms of choice. But

this lack of evolutionary contextualization can leave unrealistic notions of rationality

unchallenged, engendering a tendency to blame irrational biases or systematic inad-
equacies when human behavior does not match the optimal performance of nor-
mative behavioral prescriptions … From this proximate perspective, the failure of
boundedly-rational behavior to reach classically-rational levels of performance is just
that: a failure.
Nordli and Todd, p. 314

On the contrary, ecological rationality does not predefine rationality based on a priori analyt-
ical canons on how people should ideally behave. Instead “it seeks both to describe how people
actually act in specific contexts, and to understand how decisions and actions fit together with
particular environmental contexts in terms of their evolutionary provenance and adaptive signifi-
cance” (Nordli and Todd, p. 314). Obviously, if past and present environments differ in important
ways, this may result in a mismatch between the old mechanisms of decision making suitable for
the old times and the new environment, which is responsible for non-adaptive decisions. Humans’
evolved preferences for unhealthy sugary, salty and fatty foods are examples of this mismatch.
Conversely (as will be analysed also in Viale, Chapter 22 in this volume), “there are
examples where ostensibly-irrational behavior (according to normative prescriptive models)
may be considered perfectly reasonable when evaluated from an ecological rationality per-
spective” (Nordli and Todd, p. 319). There are many examples in this sense: the influence of
emotions on cognition to achieve goals that are important for survival and reproduction; or the
“sour grapes” effect when

the logically unfounded belief that grapes are not any good because they cannot be
reached may simply be the product of an effective proximate mechanism (i.e., emo-
tional biases in specific contexts) that ultimately serves to discourage the pursuit of
goals in situations where contextual cues indicate that goal achievement is unlikely.
Nordli and Todd, p. 319

If we consider the matter from an evolutionary perspective, this is an adaptive behaviour that
reflects “an ecologically rational strategy for balancing the risks, rewards, and probabilities that
characterize the environmental structure of contexts in which motivated reasoning occurs—
even in complex modern environments” (Nordli and Todd, p. 319).
Kahneman and Tversky’s prospect theory and the other heuristic and biases contributions
claim to be a consequence of the BR approach in decision making. In fact, Herbert Simon has
inspired two directions in the modelling of decision making. According to Thorsten Pachur
(Chapter 20 in this volume), “the first is prospect theory … and its subsequent elaboration and
formal specification in cumulative prospect theory …”:

They assume that people’s sensitivity to differences in the outcomes and probabilities of
risky options diminishes, the further away those magnitudes are from natural reference

27
28

Riccardo Viale

points (such as zero, impossibility, or certainty) and that losses carry more psycho-
logical weight than gains. These notions are described algebraically with psycho-eco-
nomic functions that translate objective magnitudes of outcomes and probabilities
into subjective ones. CPT is typically considered as an “as-if ” model, in the sense
these functions are not meant to represent the actual cognitive processes underlying
choices. The second modelling tradition rooted in bounded rationality, by contrast,
has developed heuristics, that are intended as cognitive process models.
Pachur, p. 324

Is it possible and convenient to provide some form of integration between the two kinds of
modelling? Pachur explains the pros and cons of this integration.
Herbert Simon was one of the founders of Artificial Intelligence. AI was born in 1956 at
the Dartmouth Conference and the dominant model was the computational model of the
mind. The mind was seen as the software of the brain and AI was meant to simulate the discov-
eries of cognitive sciences about mental activity. According to John McCarthy, Marvin Minsky
and others at the Dartmouth Conference, the goal was to “make a machine that behaves in
a way that would be considered intelligent if it was a human being”. However, AI developed
differently than the founders intended it to. Instead of simulating the cognitive features of
the human mind, AI tried to replicate a simplified model of the brain. Neural networks and
multi-layered neural networks are the dominant tools of machine learning and deep learning.
Deep learning programs like AlphaGo, that are so successful in the game Go, are considered
the prototypes of what AI will be in the future. This is controversial. These programs are
clever in games characterized by risk and computational complexity. Real life, however, is
characterized mainly by uncertainty, unpredictability, and surprises. If AI wants to simulate
how human beings adapt to the uncertain environment it has to change its course. The right
course might be:

1. To create a map of human heuristics that are ecologically rational in given tasks and envir-
onmental settings.
2. To incorporate heuristics in evolutionary computation techniques, which are stochastic
algorithms whose search methods are modelled on genetic inheritance and Darwinian strife
for survival.
3. To develop robots that mirror the structure of the human body not only at the perceptual
or motor level, but also at the visceral and sensorial level.7

Özgür Şimşek (Chapter 21 in this volume) maintains a similar position:

It is unlikely that the performance gap between people and AI can be closed by
advances in computational speed only. Just like people and animals, our machines will
need to learn how to be boundedly rational, in other words, how to achieve their
objectives when time, computational resources, and information are limited. Models
of bounded rationality ….. provide fertile ground for developing algorithmic ideas for
creating such AI systems.
Şimşek, p. 339

Her chapter provides an overview of some of this research and discuss its implications for
AI: “The discussion will start with one-shot comparison problems and two families of simple
heuristics … that can be used in that context: tallying and lexicographic decision rules”

28
29

Why bounded rationality?

(Şimşek, p. 339). She describes how these heuristics perfectly match certain characteristics of
natural environments that allow them to make accurate decisions using relatively small amounts
of information and computation, and how this knowledge can be applied in the context of
sequential decision problems, of which the game of Go is an example, to reduce the branching
factor in a simple and transparent manner.
Finally, another aspect of the ecological dimension of rationality is found in the analysis of
psychopathological disorders in reasoning and decision making. According to Riccardo Viale
(Chapter 22 in this volume):

Over the years, the evaluation of what is considered reason and the equation “nor-
mality equals rationality” seems to have undergone a paradoxical change. Initially,
the rationalist legacy of the eighteenth-century Enlightenment resulted in a very
clear separation of reason from madness. The madman was characterized by a failure
of deductive and inductive logical thinking. In the 1950s, however, this distinction
disappeared when cognitive science brought to light a whole series of supposed failures
of rational behaviour in the normal individual as well. What dealt a serious blow to the
theories of human decision-making and judgement formation was the fact that, quite
surprisingly, several reasoning flaws that had been attributed solely to the psychiatric
patient were, in fact, also found in the normal individual. Over the past few years, the
theory that started with the Enlightenment seems to have reversed in a circular way. If
we consider rationality as based on the logical coherence adopted in economic theory,
then rationality is more apparent in individuals suffering from certain neurological and
psychiatric disorders than among people without a disease. Paradoxically, normality
seems characterized by irrationality, and abnormality by rationality.
Viale, p. 353

But what characterizes abnormality? Normal individuals are characterized by limitations of

logical rationality, which do not make them impeccable logicians, statisticians and maximizers,
but that allow them to successfully adapt to the environment to solve problems and to learn
from their mistakes. From this standpoint, a behaviour can be defined as pathological. This
is not because it deviates from normative standards, but because it is unable to successfully
interact with the social and physical environment, in other words, to have cognitive success.
The irrationality of the mental disorder is such that it differs from bounded ecological ration-
ality because it would not allow some of the following adaptive cognitive functions (depending
on the mental disorder):

1. Learning from mistakes, environmental feedbacks, in particular, advice coming from the
social context.
2. Using social imitation heuristics, such as that of successful people or of the crowd that allow
the individual to speed up the decision-making processes and to find solutions that are
readily available to be used.
3. Realistically representing the terms of a problem in the context of a decision, without
affective or emotional hyperpolarization only on some salient variables and not on others
that are more relevant for the decision.
4. Having a correct non-distorted perception of spatial and temporal coordinates in the con-
text of a decision.
5. Anticipating at the corporal level the affective effects of a future choice in a way that cor-
responds to the reality of the phenomenon not in a distorted or dissociated way.

29
30

Riccardo Viale

According to Viale:

Any rigidity and impermeability to the signals from the environment and the lack of
a realistic representation of the variables at work do not allow adequate fitness of the
structure of a decisional task to allow us to be successful in our choices. It is therefore
from an ecological point of view rather than from a normative one that we can con-
sider a behaviour irrational in the context of some mental disorders.
p. 361

Part IV Embodied bounded rationality

Bounded rationality in human reasoning, judgement and decision making mainly results from
the adaptive evolution of our central nervous system. The way in which the brain and the sub-
cortical structures have developed provides the foundation of the features that characterize the
relation between mind and environment. It is, however, not enough to refer to the brain to
explain cognitive activity and decision making. As will be illustrated later, in the past few years,
growing importance has been attached to other parts of the body as well, such as the peripheral
nervous system, the motor apparatus, the endocrine system, the cardiovascular, respiratory and
digestive systems. In interacting with the environment, decision making cannot be fully grasped
without analysing the bodily parts that allow us to interact and to physically put in place our
decisions. From this perspective, it is possible to talk about embodied cognition applied to deci-
sion making, or, in other words, embodied bounded rationality. The metaphor of the scissors
must therefore be integrated to add one missing element. The two blades cannot function
unless they are coordinated by something else: the pivot, that is the body that allows cognition
to interact with the environment.
According to Vittorio Gallese, Antonio Matrogiacomo, Enrico Petracca and Riccardo Viale
(Chapter 23 in this volume):

The idea that cognition is embodied is evidenced by a number of experiments

connecting body states to judgement, decision making, problem solving, attitude
formation, etc. Experimental evidence shows that body variables decisively direct
and affect decision making … Further, problem solving is non-trivially dependent
on body correlates such as, for instance, eye movement ... Other various experi-
mental evidence shows that people judge steepness depending on the weight of their
backpacks … that environmental temperature affects social attitudes … that imagined
food consumption makes people satiated … or that physical weight induces the per-
ception of importance.
Gallese et al., p. 378–379

Embodied cognition can help us to reconsider such a fundamental notion in bounded ration-
ality as heuristics. Ecological rationality and the functioning of simple heuristics, in particular,
would greatly benefit from inputs from neurobiological and embodied cognition studies. In
fact, something is changing among ecological rationality scholars, in particular with reference
to deeper analysis of the neurobiological dimension of heuristics decision making. According
to Nordli and Todd (Chapter 19 in this volume), neurophysiological studies could promote
a new theoretical framework for ecological rationality. For instance, those studies may con-
tribute to understanding strategy selection in decision making, where strategy selection means
the selection of a given heuristic for a particular context. The selection is successful when the

30
31

Why bounded rationality?

selected heuristic matches the structure of a given situation. That may happen by developing
a mapping between context structure and the appropriate heuristics to use. The mapping may
be generated by a “strategy selection learning” procedure. According Nordli and Todd, the
reinforcement learning-based mechanism of strategy selection “is consistent with accounts that
tie reinforcement learning processes to the recurrent cortico-basal circuit which is the very
same circuitry that is critical to the exploitation of past behavior in the form of habits and fixed
action patterns” (p. 320). In particular, the basal ganglia seem to lie behind most ecologically
rational behaviour. Basal ganglia seem to evaluate context and select actions based on either
motor or cognitive past behaviour. This neural mechanism allows the strategy selection of
adaptive heuristics based on evaluation of past behaviour.
Christopher Cherniak (Chapter 24 in this volume) analyses an interesting dimension of the
boundedness of rationality in connection with brain activity. According to him. the most fun-
damental psychological law is that we are finite beings. The classical paradoxes of semantics and
set theory “can be reexamined not as odd pathology, but instead as indications of use of ‘quick
and dirty heuristics’––that is, the ultimate speed-reliability tradeoffs of correctness and com-
pleteness for feasibility” (Cherniak, p. 391). Cherniak introduces an interesting parallel between
psychological boundedness and neural boundedness, between rationality and brainwiring hard-
ware. In the first case, as well as in the second, the limitation of neural resources has generated
strong evolutionary pressure to employ them efficiently. He writes: “Much of higher central
nervous systems operates at signal propagation velocities below the 60 mph speed limit. Since
connectivity is in limited supply, network optimization is quite valuable” (p. 392). And he
adds: “Connection minimization seems a first law of brain tractography, an organizing prin-
ciple driving neuroanatomy … with hitherto unreported precision” (p. 392). Thus, according
to Cherniak:

“Save wire” turns out to be a strongly predictive “best in a billion” model. Wiring
minimization can be detected at multiple levels, e.g., placement of the entire
brain, layout of its ganglia and/or cortex areas, subcellular architecture of dendrite
arbors, etc. Much of this biological structure appears to arise for free, directly
from physics.
p. 392

A wire-saving Adjacency Heuristic may be applied to the placement of interconnected func-

tional areas of cerebral cortex: “If components are connected, then they are placed adjacent
to each other” (p. 393). Cherniak wonders how the intentional level of mind meshes with the
hardware level of brain. “Prima facie, that relationship appears in tension: In some aspects,
the brainwiring appears virtually perfectly optimized, yet the rationality has layers of impossi-
bility between it and perfection …” (p. 394). Cherniak may be right in assessing the distance
of rationality from perfection if he considers environments as Savage’s small worlds – e.g., dice
games, roulette, and so on – where optimal results may be achieved. The possibility of optimal
rationality is evident since the probability of the outcomes is cognizable. But in the large worlds –
the real world of society, climate, finance, politics, and so on – the uncertainty, complexity and
unpredictability of future events do not allow us to evaluate the distance of our rationality from
perfection. This is because there is no possible perfection at all. In these worlds, as in the world
of constrained neural resources, frugality and simplicity are the heuristic tools to pursue opti-
mality, that is to say better adaptive fitting.
Which features of the brain can contribute to explaining bounded rationality? Paul Thagard
writes (Chapter 25 in this volume):

31
32

Riccardo Viale

Human brains have numerous strengths that have enabled people to spread all over
the planet and increase in population to more than 7 billion. The most impressive
features of human brains are not the special-purpose adaptations touted by evolu-
tionary psychologists, but rather the general adaptability furnished by the flexible
ways in which humans can learn from experience … Nevertheless, the brain has
numerous limitations that forestall optimal rationality. Our brain’s assemblage of 86
billion neurons provides a lot of computing power, but elephants have three times as
many. In order to have more neurons, people would need to have bigger brains that
require bigger heads, but childbirth is already often a difficult procedure. Human
brain size reflects a trade-off between the benefits of more processing power and ease
of delivery through a pelvis that also must function for bipedal locomotion. Another
constraint on human brain size concerns energy. Even though the roughly 1.4 kg of
the human brain take up less than 3 percent of the average human weight, the brain
uses up to 20 percent of the energy available to the body. Larger brains would require
more energy, which either requires less energy available for other functions such as
metabolism and reproduction, or greater sources of food. The evolution of human
brains requires a trade-off between size and energy efficiency, as well as between size
and birth delivery.
Thagard, p. 400–401

The size of the brain and the limited amount of information that people can store are accom-
panied by another weakness in its computational power. Thagard writes:

Human brains also come with limitations in speed of processing. Billions of neurons
allow for massively parallel operation, but the neurons themselves are slow. A typical
neuron fires up to 200 times per second, whereas current computers have operations
at the rate of trillions of times per second. Why are neurons so slow? Most neural
connections are chemical, requiring the movement of neurotransmitters such as glu-
tamate and GABA from one neuron to another. This chemical transmission is slower
than purely electrical signalling, which occurs rarely in brains, but it allows for flexi-
bility in timing and the development of different kinds of pathways.
p. 401

And he adds:

If brains were faster, they would still not be able to do an infinite amount of pro-
cessing, but they would be able to better approximate some of the standards required
for the rational norms of deductive logic and probability and utility theory. The brain
lacks the speed to be able to do all of the calculations that are acquired for absolute
standards of rationality.
p. 401

Nevertheless, the brain can perform computations that are important for the survival and repro-
duction of organisms:

Perception, inference, and decision-making can all be modelled as processes of par-

allel constraint satisfaction, in which a brain or computer considers a range of possible
interpretations of a conflict situation and comes up with a good but not necessarily

32
33

Why bounded rationality?

optimal solution … For example, recognizing a moving object as an instance of prey

or predator should consider alternative hypotheses about the animal, constrained
by perceptual and environmental information. Parallel constraint satisfaction is effi-
ciently computed by neural networks that implement constraints by excitatory and
inhibitory links …
Thagard, p. 402

Another important feature of the brain influencing decision-making processes and lying at
the heart of bounded rationality is the integration of cognition and emotion. This integration
occurs between high-level areas, such as the prefrontal cortex, and the emotional systems, such
as the limbic system consisting of primitive areas such as the amygdala. This interconnection
interferes with the requests of decision-making rationality. For example, according to expected
utility theory, probability and utility must be first calculated independently and then combined
at a mathematical level. The profound integration in the brain between the cognitive and
emotional centres does not allow for this separation. As Thagard maintains, utility operates on
the estimation of probability and probability operates on the estimation of utility: “The brain
has no firewall between cognition and emotion, so it is not surprising that people often adopt
beliefs that they find emotionally appealing, in domains that range from politics to relationships”
(Thagard, p. 403). A case in point concerns time discounting. People have a tendency to go for
short-term small gains in neglect of long-term large gains. This is because the decision on what
to do immediately stimulates areas of the ventral striatum and orbitofrontal cortex connected
to the emotional sphere, while long-term evaluations occur through computations that take
place in the prefrontal cortex and parietal areas that do not have strong emotional implications.
Hence the integration of cognition and emotion in the brain contributes as much as size and
speed limitations to the boundedness of rationality. The neural constraints of information pro-
cessing prevent us from satisfying the desires of formal rationality, but appear to have an adaptive
function when faced with complex and uncertain environments and scenarios. Since we have
to decide mainly about an epistemic or ontological uncertain reality, no optimization based on
knowledge of the options and their probability can work. Thus, some failures and biases are
not really failures and biases but they are rather the only possible heuristic way to cope with
uncertainty. The same limitation of attention processes connected to Miller’s magic number
for working memory can be viewed from an adaptive perspective. It may be an adaptive con-
straint to cope with uncertainty and to decide in a fast, unaware, intuitive way when faced with
potential but uncertain danger rather than losing time in slow, analytic and aware thinking and
decision making. Therefore, what seems to be human failures to follow normative principles
(introduced only recently in the history of humankind) is, on the contrary, an adaptive and
ecological rational feature and not a defect of the mind-brain.
The flexible and dynamic adaptability of the brain in various decisional contexts is highlighted
also by Colin McCubbins, Mathew McCubbins and Mark Turner (Chapter 26 in this volume).
They note that a number of recent discoveries provide an overview of the neural bases of deci-
sion making that differs significantly from the picture painted by neuro-computation theory
in the past. The salient features of bounded rationality are explained by the way the brain
functions. This is particularly true for the dynamic instability of self-expression or, in other
words, constantly evolving preferences and beliefs due to experience and learning, depending
on emotional and external changes. In particular,

Cognitive neuroscientists have proposed that the human brain is constantly scanning
over a range of often-conflicting alternatives and collapsing that range to an action

33
34

Riccardo Viale

only in the moment of decision … Consider someone who wants to pick up a

coffee cup. There are many ways to do so successfully. The brain may explore
many of those approaches simultaneously. The neurobiological basis of action can
be varied, with simultaneous but conflicting lines, each with some probability of
being given, at the final instant, control over skeletal and motor programs. One of
those possibilities will precipitate in the moment of action. Our enacting only one
action suite does not mean that the brain was exclusively focused on that one suite;
it means only that, in the moment of action, one coherent action was executed, as
others were forsaken.
McCubbins et al., p. 415

Part V Homo Oeconomicus Bundatus

The history of economic thought shows that there was an inverse relationship between the
articulated and empirical development of the economic agent’s theory of mind and the attri-
bution of rational capacity to the said agent. The greater the demand for rationality, the less
the agent’s psychological characterization was developed. As the previous examples show,
the maximum rational capacity now required by contemporary neoclassical economics (sat-
isfaction of the computational requisites of Bayesian decision theory) means that the agent
has minimal psychological content. Instead, the opposite occurs in Hayekian subjectivism.
This inverse proportionality appears to be a natural consequence of the relationship between
reason and mind. Gifts of unlimited reasoning, like those ascribed to the neoclassical eco-
nomic agent, are not combined with the empirical representation of a mind characterized by
cognitive limits and weaknesses. The unlimited reason attributed, a priori, to the economic
agent compresses and suffocates any space for human psychological expression. This dualism
between reason and mind is not a novelty, but instead has deep, philosophical roots. It is linked
to a precise tradition that can be seen as the progenitor of the model of rationality that was in
vogue in economic science for two centuries (Viale, 2012). The theory of rationality inherited
by the tradition that spans from Heraclitus, Parmenides, Plato and Aristotle to St Augustine,
Descartes, Spinoza, Leibniz and the Enlightenment and which underlies the theory of eco-
nomic action in the nineteenth century is therefore characterized by unlimited, a priori and
linguistic-intentionalist attributes. Its supposed descriptive property is based on the a priori
presumption of the universal rationality of the human species and not on the a posteriori justifi-
cation of people’s real rational capacities. Neoclassical economics and the model of rationality
contained in von Neumann and Morgenstern’s game theory are the realization of this ideal
(well represented by Laplace’s metaphor of the demon endowed with unlimited computational
rationality).
Gerd Gigerenzer (Chapter 2 in this volume) outlines three competitive alternatives that have
been using the concept of BR. Ariel Rubinstein (Chapter 27 in this volume) seems to propose
a fourth one. He starts with a definition of economic theory:

Economic Theory is a collection of stories, usually expressed in formal language,

about human interactions that involve joint and conflicting interests. Economic
Theory is not meant to provide predictions of the future. At most, it can clarify
concepts and provide non-exclusive explanations of economic phenomena. In many
respects, a model in Economic Theory is no different than a story. Both a story and
a model are linked to reality in an associative manner. Both the storyteller and the
economic theorist have in mind a real-life situation but do not consider the story or

34
35

Why bounded rationality?

the model to be a full description of reality. Both leave it to the reader to draw their
own conclusions, if any.
Rubinstein, p. 423

There is no room for a realist methodology to establish an empirically-based microeconomics.

This conceptual and analytical position8 is reflected in the concept of BR: “Models of Bounded
Rationality are for me …. models that include explicit references to procedural aspects of deci-
sion making, which are crucial for the derivation of the analytical results” (p. 423). Procedural
aspects do not refer to the same phenomena here as they do for Simon and Gigerenzer. They
are not derived experimentally by behavioural and cognitive studies on problem-solving and
decision making. They are, on the contrary, introduced as in Millian methodology by quali-
tative observation, intuition and analytical deduction: “A good model of bounded rationality
should include a procedure of reasoning that ‘makes sense’ and is somewhat related to what we
observe in real life” (Rubinstein, p. 423). What Rubinstein and Gigerenzer have in common
is a lack of interest in the inconsistency of the model with respect to the normative aspects of
rationality (as in the heuristics and biases approach) and prefer instead to analyse the procedural
aspects of behaviour: “A model in which rational agents ignore some aspect of rationality is a
bad model rather than a model of bounded rationality” (Rubinstein, p. 423).
Rubinstein’s is the fourth model of BR together with optimization under constraints,
heuristics and biases and ecological rationality. It is the model less anchored to empirical data
coming from the cognitive sciences. Are the first two models useful to develop BR? According
to Clement Tisdell (Chapter 28 in this volume) and, contrary to Gigerenzer, they are:

However, it can be argued that the first set of models are relevant to the study of
bounded rationality. They highlight limits to the neoclassical vision of unrestricted
rationality. While their knowledge and rationality assumptions are still too strong,
they can help to identify factors that ought to influence behaviors under conditions
of restricted rationality. As for the second class of models (which include behavioral
ones), most (but not all) identify limits to perceptions of states of nature and common
faults in reasoning, both of which can be considered to be a consequence of bounded
rationality. These classes of models (mostly behavioral economic ones) do demonstrate
some of the limits to unrestricted rationality.
Tisdell, p. 443

The analytical and conceptual approach to economic theory of Rubinstein is also common to
a great part of mainstream economics. Shabnam Mousavi and Nicolaus Tideman (Chapter 29
in this volume) write: “the paradigm of mainstream economics is a coherent one built logically
on a substantive notion of rationality, with expected utility theory as its crowning achievement”
(p. 448). And they quote Gary Becker (1962): “Now, everyone more or less agrees that rational
behavior simply implies consistent maximization of a well-ordered function, such as utility or
profit.” And they add:

In this framework, an actor seeks the best or optimal outcome, and specifying the
criteria for its existence and (preferably) uniqueness occupies theorists and modellers,
who rely mainly on deductive methods. Issues of how to collect data to test the
theory, or building models to achieve concrete real-world goals are usually not of
primary concern. The search for information is also assumed to be optimal. That
is, the rational agent uses an optimal stopping rule, continuing to calculate marginal

35
36

Riccardo Viale

costs and marginal benefits of further search, and stops when they are equal. This is
acknowledged to be an artificial search process, with no claim to represent the actual
search process. Understanding reality is thus pursued through “stylized facts” instead
of through observation.
Mousavi and Tideman, p. 448

This as-if approach that pretends to transform economics in a scientific discipline is rooted in
how physics was developed in the twentieth century.
In Viale (1997, pp. 18–21; 2012, pp. 179–181), I analyse the philosophy of science followed
by neoclassical economics:

The model is clearly based on Newtonian mechanics but it carries two serious
drawbacks. As Rosenberg pointed out (1983, reprinted in Hausman, 1994, pp. 378–
380), since the nineteenth century, economists have been elaborating a theory
whose form is identical to the great theoretical breakthroughs made in science from
the sixteenth century onwards. The strategy is to view the behaviour economists
seek to explain as reflecting forces that always move towards a stable equilibrium,
which maximizes or minimizes some theoretically crucial variable. In the case of
microeconomics, this crucial variable is utility (or its latter-day surrogates) and the
equilibrium is provided by a level of price across all markets that maximizes this vari-
able. This strategy is most impressively exemplified in Newtonian mechanics and in
the Darwinian theory of natural selection. In Newtonian mechanics, the system’s
behaviour always minimizes or maximizes variables that reflect the state of the system
which is mechanically possible, while in Darwinian theory, it is the environment
that maximizes the fitness of individuals of a species. This strategy is crucial to the
success of these theories because of the way it directs and shapes the research. If we
believe that a system always acts to maximize the value of a mechanical variable – for
example, total energy – and our measurement of the observable values of that vari-
able diverges from the predictions of the theory and the initial conditions, we do
not infer that the system described is failing to maximize the value of the variable
in question. We do not falsify the theory, but we assume that we have incompletely
described the constraints under which the system is actually operating. The axioms
of these theories do not embody even implicit ceteris paribus clauses. With these
theories, the choice is always between rejecting the auxiliary hypotheses and test
conditions or rejecting the theory altogether. Hence, the only change that can be
made to the theory is to deny that its subjects invariably maximize or minimize its
chosen variable.
In Newtonian mechanics, attempts to describe the systems under study more
completely resulted in the discovery of new planets and new laws, like those of
thermodynamics. In biology, the assumption that fitness is maximized led to the
discovery of forces not previously recognized to affect genetic variation within a
population and led to the discovery of genetic laws that explain the persistence in a
population of apparently non-adaptive traits, like sickle-cell anaemia. But what about
microeconomics? The success of this strategy in other disciplines may justify the
attempts made by economists to make recalcitrant facts about human behaviour and
the economic systems humans have constructed fit the economic theory. Moreover,
this strategy allows the use of many powerful formal tools, such as differential calculus,
topology and differential geometry. But many years of work in the same direction

36
37

Why bounded rationality?

have produced nothing comparable to the physicists’ discovery of new planets or new
technologies, or the biological understanding of the mechanism of adaptation and
heredity. Therefore, it is time to question the merit of applying this approach to eco-
nomics, since it carries all the disadvantages of empirical non-falsifiability without the
advantages of the discovery of novel facts and applications.
There is also another general reason “not to ape physics” that stems from a recent
philosophical debate in physics compared with the other sciences. In contemporary
physics, there is a trade-off between explanatory power and factual content. As pointed
out by Cartwright (1983, pp. 135–139), fundamental laws like Maxwell’s equations
or Schroedinger’s equations have great explanatory power but do not describe the
true facts about reality. The aim of these equations is to cover a wide variety of
different phenomena with a small number of principles. If the fundamental laws set
out to fit reality into the highly constrained structures of their mathematical formulae,
they will have to distort the true picture of what happens. Quantum mechanics is a
good example.9 In fact, the explanatory power of quantum theory comes from its
ability to deploy a small number of well-understood Hamiltonians to cover a wide
range of cases. But this explanatory power has its price. By limiting the number of
Hamiltonians (that is the mathematical representation of the kinetic and potential
energies of the real system), we set constraints on our ability to represent situations
realistically and we lose the true representation of the system. In physics, if we want to
have a realistic representation of the phenomena, we must abandon the fundamental
laws and look at the phenomenological laws like those of Galileo, Ohm or Kepler.
Fundamental laws usually explain phenomenological laws not in a deductive way, as
was asserted until recently – think of the Hempel-Popperian deductive nomological
model – but using a model which fits the phenomenon into the theory (Cartwright,
1983, pp. 131–134). This picture of science fits Friedman’s “as-if ” instrumentalist
methodology well: the theory is only a fictitious deductive machine to produce
explanations and predictions. This model of science seems to work only for the fun-
damental laws of physics. If we climb up the levels of aggregation of reality, we find
that already at the level of biology we are not able to find explanatory structures
based on the three components – theory, model, facts – that are present in physics.
Instead we find phenomenological laws with many exceptions that aim to describe
complex phenomena. The same consideration applies a fortiori to the explanation of
human behaviour. The great complexity and variability of behavioural phenomena
preclude the discovery of genuine general phenomenological laws relating to par-
ticular domains of psychological reality. This absence does not allow any serious uni-
fying mathematical abstraction, comparable to that of the fundamental laws in physics,
which has the power to explain reality at the phenomenological level. The moral is
that instead of “putting the cart before the horse”, or in other words playing the game
of the theoretical physicists and trying to elaborate fictitious, but useless, theoretical
formal models of human decision-making, it would be wiser to study, patiently, the
empirical phenomenology of human decision making and to try to elaborate some
useful, and genuine, empirical laws that are valid, locally and contextually, for a small
fragment of human reality.

Mousavi and Tideman (Chapter 29 in this volume) emphasize this point when they refer to the
economists’ detachment from observed behaviour as a serious flaw in their methodology. In this
sense they quote an interview of Simon in Challenge (1986, pp. 22–23):

37
38

Riccardo Viale

They [economists] don’t talk about evidence at all. You read the pages where Lucas
talks about why businessmen can’t figure out what’s happened to prices and it is just
what he feels as he sits there smoking his cigarettes in his armchair. I don’t know what
Keynes smoked, but when you look at the pages where he talks about labor’s money
illusion, no evidence is cited. So the real differences in economics, as compared with
psychology, is that almost everybody operates within the theoretical logic of utility-
maximization in the neoclassical model.
When economists want to explain particular phenomena in the real world, they
have to introduce new assumptions. The distinctive change in the behaviour of the
economic actors comes from a change in the behavioural assumptions. No empirical
evidence is given to support those changes. They emerge from the mind of the eco-
nomic theorist sitting in his armchair.
Mousavi and Tideman, p. 449

Another contribution against armchair economics and in favour of the empirical foundation
of economic theory and research on the agents’ cognitive features comes from the analysis by
Ignazio Visco and Giordano Zevi (Chapter 30 in this volume) on the weaknesses of rational
expectations theory. They consider some of the links between concepts of bounded ration-
ality and the approaches followed by economists in their analysis of the role played by eco-
nomic agents’ expectations in driving the evolution of the economy through time. Visco and
Zevi write:

Indeed, the relevance of expectations has been repeatedly underlined in macroeco-

nomic theory and policy making. However, the economists’ degree of attention on
how they are actually formed and on how they interact with the economic observables
has followed high and low cycles. In recent years, the increasing availability of survey
data and the failings of models based on purely rational representative agents have
prompted renewed interest in inquiries into the direct measurement of expectations
and empirical studies of their formation.
p. 465

Robert Lucas introduced analytically the theory of rational expectations without caring about
the empirical dimension of decision making. Aside from the bounded rationality features of
economic agents, there are also other empirical data to consider. For example, learning implies
that a rational expectations equilibrium is only one of the possible outcomes when agents con-
tinuously update their expectations based on the comparison between past expectations and
actual realizations. Uncertainty limits the possibility of rational expectations because it does not
allow for any form of rational prediction. Visco and Zevi write:

In line with Simon’s 1955 seminal contribution, later developed with March, agents
use expectations based on simple heuristics as a device to willingly ignore part of the
available information in order to reach local optima which, in this particular setting,
beat the outcomes of fully rational choices. This strategy proves to be superior to
more sophisticated ways of forming expectations, grounded for example in recur-
sive least squares as in mainstream learning literature, due to the highly unstable and
uncertain, in the Knightian sense, environment.
p. 465

38
39

Why bounded rationality?

What should be done in uncertain environments? Is it better to collect as much information

as possible to reduce uncertainty or, on the contrary, is it preferable to end a costly and useless
data collection that has no chance of increasing our knowledge of the phenomena?10 The more-
is-more principle has been a central dogma of Bayesian rationality in dealing with uncertain
environments. Gregory Wheeler (Chapter 31 in this volume) writes:

Lore has it that a fundamental principle of Bayesian rationality is for decision makers
to never turn down the offer of free information. Cost-free information can only help
you, never hurt you, and in the worst case will leave you at status quo ante. Purported
exceptions to this principle are no exceptions at all, but instead involve a hidden cost
to learning. Make those costs plain and the problem you face is one of balancing the
quality of a choice against the costs to you of carrying it out, a trade-off that Bayesian
methods are ideally suited to solve.
p. 471

The implications of this principle in economic thinking are evident: a delay may occur in
making the final decision if there is the opportunity to acquire some information (for example,
from an experiment) relevant to the decision. According to Wheeler, in situations of Knightian
uncertainty when a probability assessment does not improve after learning the outcome of an
experiment, “the commonplace that ‘knowledge is not disadvantageous’ is false, even when the
cost of obtaining the information is zero” (p. 474).
BR is recognized as an important feature by many heterodox approaches in economics.
One is the new Schumpeterian programme of evolutionary economics. According to Richard
Nelson (Chapter 32 in this volume), there are at least three BR factors that are fundamental
for evolutionary economists: (1) it is important to distinguish between choice contexts which
are familiar to the economic actor and who responds to them more or less automatically, and
contexts that induce the actor to engage in serious contemplation of alternatives; (2) it is
important to recognize that actors differ in the capabilities that they bring to various choice
contexts; and (3) it is important to relate the perceptions of individual actors about the contexts
they face, the courses of action that they understand and are competent to employ, and their
judgements about which of these actions are appropriate and likely to be effective, to the beliefs
and understandings and know-how of the broader community of which the actor is a part.
An important feature that connects evolutionary economics to ecological rationality is the
function of routines. In stable environments when actors have time to learn the kind of actions
that work, they tend to perform these actions automatically. There are routines that work in
an organization as well as in consumer behaviour. For example, pricing routines for household
shopping and firms are mostly automatic. Nelson writes:

Viable routines generally have a reasonable amount of flexibility built into them to
enable them to adjust to the kind of variable circumstances that are to be expected
in the broad context where they are operative. Household shopping routines need to
be sensitive to what is and is not available at the store, and to some degree to prices.
Firm pricing routines need to take costs into account. But my argument is that in
established shopping routines these adjustments generally are made relatively rou-
tinely. There may be some conscious consideration of alternatives, but so long as the
context remains in the normal range, wide search and intensive deliberation are highly
unlikely. Similarly, the pricing routines of firms almost always are sensitive to costs,

39
40

Riccardo Viale

with much of that sensitivity, if not necessarily all, built into a formula used relatively
routinely.
p. 487

He proposes a decision-making model that is very close to ecological rationality. And in

fact he uses the term “adaptively responsive” to denote the sensitivity of routines to broadly
experienced and thus anticipated variation in the details of the context that invokes their use.
The evolutionary economics approach is very sympathetic to BR. Since its focus is on
innovation, technical change, dynamics and uncertain environments, it has always criticized
the normative static dimension of economic rationality. Like Richard Nelson, Giovanni Dosi,
Marco Faillo and Luigi Marengo (Chapter 33 in this volume) take a step forward in criticizing
the “enchanted” BR of behavioural economists like Kahneman and Tversky. They conclude:

A multi-millennial tradition of Western thought has asked “How do people behave?”

and “How do social organizations behave?”, from Aristotle to St Augustin, Hume,
Adam Smith, Kant, to name just a few giants. However, modern economics – and,
more recently, social sciences colonized by modern economics – have taken up the
answer by one of the shallowest thinkers, Bentham: people decide their courses of
action by making calculations on the expected pleasures and pains associated with
them. And, indeed, this Weltanschauung has spread all the way to the economics of
marriage, of child bearing, of church going, of torture … Our argument is that the
Benthamian view is misleading or plainly wrong concerning the motivations, decision
processes and nature of the actions. First, the drivers of human motivation are many
more than one. As Adam Smith masterly argues in his Theory of Moral Sentiments,
utility (what he called “prudence”) is just one of them, and in a lot of social contexts,
not the most important one. Second, the decision processes are very rarely explicit
calculations and comparisons of outcomes. Third, the ensuing decisions very seldom
look like a “rational” (“as…if ”) outcome of the foregoing decision processes, even
when the latter would be possible to calculate. And in the real word, complex and evolving
as it is, they rarely are. In such circumstances, we suggest, a positive theory of indi-
vidual and collective behaviours has to entirely dispose of the max U(…,…, …) appar-
atus, either as an actual descriptive device, and as a yardstick, whatever that means.
If we are right, then also relaxations of the paradigms involving varying degrees of
“bounded rationality” in the decision process and an enlargement of the arguments
in the utility function (e.g. adding “intrinsic motivations”, or even “altruism”) are
quite misleading. They are a bit like adding epicycles over epicycles in a Ptolemaic
astronomy. The radical alternative we advocate is an anthropology of a homo heuristicus
…., socially embedded, imperfectly learning in a complex evolving environment, and
with multiple drivers of his actions.
p. 501

Part VI Cognitive organization

Herbert Simon introduces the conceptual premises of BR in his PhD thesis in political science.
Administrative Behaviour (1945) is the book based on his PhD thesis that started a new chapter
in the theory of organization, management, public administration and policy. BR is a fun-
damental concept that explains how individual decision making gives rise to organizations
and how institutions constrain individual decisions. In particular, it is evident that, in public

40
41

Why bounded rationality?

administration and policy making, the complexity of choices and the overload of information
are the determinants of the organizational and institutional changes that aim to adapt to the
environment through specialization and differentiation. How does an organization adapt to
the external environment? Through an internal differentiation of tasks similar to the problem-
solving models that envisage breaking down problems into sub-problems. Massimo Egidi and
Giacomo Sillari (Chapter 34 in this volume) write:

Simon identified the main characteristics of managerial decision making by analysing

the structure of the organizational process. He recognized that the core of every
organization is the pattern of division of tasks and their coordination. The organiza-
tion is thought of as a goal-oriented structure based on internal tasks that must be
coordinated in order to achieve the organization’s overall objectives. Behavior within
organizations is thus goal-oriented, and goals are by and large complex and hier-
archically structured, as many intermediate sub-goals need to be realized, often in
a specific order, for the final goal to be achieved. The dynamics of organizational
decision-making may therefore be very complex, presenting two main aspects. First,
goals are often defined in very general and ambiguous terms, thus necessitating con-
tinuous revising of the sub-goals’ hierarchy. Second, hidden conflicting objectives can
be unearthed during various organizational decisions, and this may, again, make it
necessary to revise both sub-goals and their hierarch.
p. 509–510

How do organizational structure and problem solving relate to bounded rationality? Egidi and
Sillari write:

Indeed, in 1956, Cyert, Simon and Trow carried out an empirical analysis of man-
agerial decision contexts that highlighted how search and learning were at the core
of human rationality. The study revealed an evident “dualism” of behavior: on the
one hand, there is behavior guided by a coherent choice among alternatives typ-
ical of structured and repetitive conditions; on the other, behavior characterized by
highly uncertain and ill-defined conditions, where the predominant role was played
by problem-solving activities. The dualism between repetitive and well-known deci-
sion contexts and ill-defined decision contexts proved to be a key distinction for our
understanding of decision processes. In situations of the former kind, it highlights
the process of decision-making routinization. In the latter, the necessary conditions
for applying standard rational choice theory are lacking, and the most important
decision process is the ability of the subjects to formulate and solve problems. This
suggests that the real restrictions on rational decisions happen during the process of
construction of the context of the decision. The notion of bounded rationality refers
mostly to these conditions, and hence it is implicitly intertwined with the notion of
problem solving.
p. 511–512

One of the features of bounded rationality in organization is attention. From different points of
view, an organization can be regarded as the answer to the limits and the constraints of human
attention processes. The organization’s hierarchical structure, communication channels, agenda
of priorities and performance evaluation can be viewed as a recursive product of selective
attention. As is known, attention is linked to the limits of executive memory and to the related

41
42

Riccardo Viale

computational processes that can only focus on limited information at a time. Inga Jonaitye and
Massimo Warglien (Chapter 35 in this volume) write:

The allocation of attention is commonly seen as a fundamental problem for

organizations … Developing Barnard’s … intuition that “narrowing choice” is a
central function of executive decisions, Simon … early on identified the process of
directing and channeling managers’ attention as a key function of organizations …
The foundations for such perspectives rely on the central role that limited attention
provides in defining bounded rationality … The “limits” of decision-making ration-
ality are to a large extent the results of the attentional bottleneck …
p. 522

For Simon, March and Cyert, decisional constraints and the division of tasks create a close
connection between attention and organizational sub-goals. Since individuals cannot pay
attention to all the organizational problems in parallel, there is a need for an internal organiza-
tion of attention allocating each problem or part thereof to different individuals, for the pur-
pose, however, of ultimate coordination. Two interesting organizational models explicitly refer
to attention. Jonaitye and Warglien write:

The garbage can model … explores its coordination implications in non-routine deci-
sion making. Taking a radical departure from traditional models of organizational
decision making, the garbage can model explores “organized anarchies” in which
preferences are problematic, technologies are unclear, and participation is fluid … The
second model was introduced by Ocasio … as the attention-based view of the firm
(ABV). Ocasio describes organizations as systems of structurally distributed attention
and defines attention as a cognitive process that encompasses the noticing, encoding,
interpreting, and focusing of time and effort by organizational decision-makers on
issues and answers.
p. 523–524

No rational choice theory is able to explain or prescribe proper policy making and organiza-
tional development. On the contrary, bounded rationality emphasizes the adaptive model-
ling of successful organization and policy making that do not follow any optimal rules other
than the recursive adaptation to the ever-changing and uncertain environment. Heuristic
decision making seems the best way to cope with the goal of successful adaptation and eco-
logical rationality. Torsten Reimer, Hayden Barber and Kirstin Dolick (Chapter 36 in this
volume) exemplify the adaptive dimension of bounded rationality by a funny story about
the Beatles:

Like many teenage boys in Liverpool in 1960, John Lennon and Paul McCartney
wanted to start a rock band. However, as their fledging band went from record com-
pany to record company with their manager Brian Epstein, it became clear: Britain
was getting tired of rock and roll … On the other hand, America at that time had
never seen anything like the Beatles. The Beatles received a very different reaction in
the US than in Britain … It was not just the Beatles, but the pairing of the Beatles
with a particular time and place that made their music so successful. The story of
the Beatles points to an important aspect of the bounded rationality of groups and
teams: The success or failure of a group cannot be understood without looking at the

42
43

Why bounded rationality?

environment in which the group is embedded. The same strategy and behavior may
be successful in one environment but result in a failure in another environment.
p. 535

How can the role of adaptation be studied in group decision making? Group research that
subscribes to the approach of bounded rationality focuses on the match between group strat-
egies and characteristics of both the information and social environment. Historically, group
research has often focused on group performance and productivity. The history of group
research using outcome and performance measures is, by and large, a history of demonstrating
group failure. This kind of result seems to be based on information environments that have
been used in laboratory experiments. It may well turn out that several process losses that are
described in the literature reflect group behaviors that are functional in many environments.
It is the match in strategy and environment that leads to successful group outcomes. Reimer,
Barber and Dolick presented research on hidden profiles that provides evidence for that claim.
The architecture of choice inside the organization is an old Simonian concept. Simon
believed that organizations can place members in a psychological environment that will provide
them with the information needed to make decisions correctly. Ian Belton and Mandeep Dhami
(Chapter 37 in this volume) write: “organizations can establish standard working practices, train
individuals, and structure the work environment so that it encourages rational thinking and
consistency or regularization of practice” (p. 549). In their chapter, they evaluate the solutions
that a particular kind of organization, the intelligence analysis organization, has offered to
combat cognitive bias in their intelligence analysts. They identify the cognitive biases that
may affect the practice of intelligence analysis and review debiasing strategies preferred by the
intelligence community as opposed to those developed and tested by psychological research.
However, they refer to mainstream behavioural economics and therefore do not differentiate
between adaptive and non-adaptive heuristic decision making. In analysing the adaptive aspect
of decision making, they have recognized that many biases are not really errors but the best
possible ways to solve problems in uncertain environments such as the subject of intelligence
analysis. For example, they quote fluency as a bias, whereas fluency is often an effective heur-
istic tool in uncertain environments such as international political situations. In any case,

Psychologically informed and empirically tested debiasing interventions often focus

on improving an individual’s ability to identify tasks/situations where an intuitive
response is likely to be biased, so that they can override their intuition with an appro-
priate deliberative strategy. Typically, interventions of this kind involve training or
instruction that aim to increase understanding and awareness of cognitive biases.
Other debiasing interventions aim to fill mindware gaps by teaching relevant rules
(e.g., probability or logic), or specific strategies to use in a given task. Finally, some
debiasing interventions involve restructuring the task environment to reduce biased
behavior, either by encouraging more deliberative thinking or by inducing unbiased
intuition.
Belton and Dhami, p. 553

Part VII Behavioral public policies: nudging or boosting?

The architecture of choice is present in Herbert Simon’s cognitive theory of organization. The
same concept seems to inspire his political science analyses when he refers to the bounded
rationality of government and citizens. In this sense, for Simon, bounded rationality and the

43
44

Riccardo Viale

foundation of behavioural economics may be considered the precursors of recent theories on

the behavioural economics approach to public policy. In particular, Richard Thaler and Cass
Sunstein (2008) have recently introduced a particular behavioural approach to public policy
known as nudge or libertarian paternalism. A public policy intervention is classified as a nudge
when it is not a coercive measure, retains freedom of choice, is based on automatic and reflex
responses, does not involve methods of direct persuasion, does not significantly alter economic
incentives, and revises the context of choice according to the discoveries of behavioural eco-
nomics (Sunstein, Chapter 38 in this volume; Oliver, Chapter 39 in this volume). What is
proposed is therefore a form of libertarian paternalism that has a dual valency. As paternalism, it
aims to make up for citizens’ irrational and self-harming tendencies by “gently nudging them”
to decide rationally for their own good. In its libertarian form, it aims to give the last word to the
outcome of the conscious and deliberative processes of the individual citizen who can always
choose to resist the nudge (Sunstein, Chapter 38).
Thaler and Sunstein’s thesis is that citizens are subject to many deviations of rationality
that bring them to make a decision that is against their own interests. This widespread
irrationality is provoked by the automatic judgement and decision-making behaviour enabled
by heuristics. In this way, the individual is incapable of following a series of basic principles
of rationality at a probabilistic and logical level. By doing so, he makes the wrong choice
and goes against his own interests.11 Thaler and Sunstein’s theses derive in part from the pro-
gramme of behavioural economics commenced in the post-war period by Herbert Simon
and continued in the 1970s by Daniel Kahneman, Amos Tversky and their School. However,
their approach is based on a controversial concept of bounded rationality focused only the
study of the correspondence between judgement and decision-making performance in tests
and laboratory simulations and canonical models of probabilistic and deductive rationality.
It therefore highlights a constant and systematic irrational misalignment between behaviour
and norm. The Behavioral Economics Nudge approach discounts a series of weaknesses
in various analytical dimensions of an epistemic, epistemological, methodological and eth-
ical nature (Viale, 2018). As is clear from the Simonian framework and from subsequent
developments (Gigerenzer, Chapter 2 in this volume), there are some attributes of the con-
cept of bounded rationality that contrast with the current approach of behavioural eco-
nomics that inspired Nudge theory. This is not a rationality that is interested in questions of
formal coherence, but rather in the question of ecological adaptation to the environment of
choice and problem solving. It centres on the procedural and realistic attributes of ration-
ality and not on instrumental and conventionalist ones. It recognizes that the complexity of
the real environment of choice brings the player face to face with decisions in conditions
of uncertainty (for example, “ill-structured problems” such as financial markets or political
forecasting) rather than risk (for example, “well-structured problems”, such as dice throwing
or chess) (Viale, forthcoming).
The relation between bounded rationality and libertarian paternalism is controversial. For
example, in Chapter 39, Adam Oliver writes:

Underpinning libertarian paternalism is the assumption that of the many decisions

that each of us make quickly and automatically each day – decisions that are guided
by simple rules of thumb (i.e., the heuristics famously associated with Herbert
Simon …) and influenced by various innate behavioural influences (e.g., present
bias, loss aversion) – a few will lead us to act in ways that if we deliberated a little
more, we would prefer not to do (e.g., present bias might lead us to eat more
doughnuts than we would ideally consume if we thought about our decisions a little

44
45

Why bounded rationality?

more). It is noteworthy that these heuristics may have evolved because most of the
time they guide us efficiently through our daily lives … but the basic idea in liber-
tarian paternalism is that with knowledge of the behavioural influences, the “choice
architecture” (i.e., the context or environment) that people face can be redesigned
such that their automatic choices are even more likely to better align with their
deliberative preferences.
p. 571

Oliver’s model of bounded rationality clearly emerges from the heuristics and biases tradition
of Kahneman and Tversky. People are limited in their rationality and their biases may generate
negative internalities. Therefore, libertarian paternalism should help to improve internalities.
In his chapter, Oliver proposes to take this approach one step forward towards “a political
economy of behavioural public policy that sits alongside the liberal economic tradition of
John Stuart Mill, albeit being somewhat more interventionist than Mill would have allowed”
(p. 570). In other words, he supports a view that focuses also on social negative externalities
caused by biases, errors and free-riding behaviour. He proposes a view that aims to improve
social externalities and not only internalities. The political economy of behavioural public
policy proposed by Oliver “has two arms: to nurture reciprocity so as improve the stock of
human flourishing, and to regulate against harm-inducing egoism to protect the capacity of
people to flourish as they themselves see fit” (p. 576).
Nathan Berg (Chapter 40 in this volume) proposes a series of arguments that undermine
the theoretical foundation and ethical and social consequences of nudge. For example, the
important feature of diversity and heterogeneity – which is fundamental to generate know-
ledge, creativity, innovation, social change – is undermined by homogeneous goals of nudge
policy making. Berg writes: “If populations respond to nudges as they are designed to and bring
the population’s profile of beliefs and behaviors into closer conformity, then multiple benefi-
cial streams of belief and behavioral heterogeneity also risk being reduced” (p. 580). “There are
risks and unintended consequences from policies that reduce heterogeneity, especially pater-
nalistic policies motivated by the goal of ‘correcting’ alleged ‘bias’ with respect to prescribed
beliefs and behaviors” (Berg, p. 580). Furthermore, why nudge people to correct their biases?
The reasons behind this paternalistic policy-making programme are untenable. The propensity
to violate the axioms of rationality in uncertain environments is not pathological but adaptive.
The “behavioural law and economics” that justify the paternalistic aims of nudging are valid
only in risky environments and not in uncertain ones. Berg writes:

There is an alternative behavioral law and economics research program based instead
on the insights of Herbert Simon. Well-performing individuals and organizations
equip themselves to adapt to unstable and complex reward-generating environments
by utilising decision processes that sometimes violate axiomatic logical consistency.
When well-matched to the environment in which they are used, logically incon-
sistent decision processes—as well as beliefs and actions that deviate from economists’
prescriptions—can be purposeful, in addition to providing both individual and
external social benefits. Analogous to biodiversity, when heterogeneity of beliefs and
decisions generates positive externalities, it should be considered a public good. The
risk of encroaching on this heterogeneity (inadvertently inducing behavioral and
belief monocultures) is worthwhile to consider in social-welfare analyses of nudging
and other paternalistic policies.
p. 583

45
46

Riccardo Viale

Decision making may be adaptive – that is, ecologically rational – if it generates a reward from
the environment. Berg writes:

One theme in the ecological rationality research program is to clearly define sets
of reward-generating environments and performance metrics with respect to which
a given decision process performs well. Another theme relevant to Simon-inspired
behavioral law and economics is how law and regulation, in effect, can construct an
environment where a heterogeneous ecology of decision rules employed by many
different individuals can perform well, by an appropriately specified aggregation rule
or social welfare function.
p. 581–582

Finally, the impact of bounded rationality is found also in supporting “satisficing” policy
making. Let us consider a society of “satisficers” vs. a society of “optimizers”. The former
requires less paternalistic and costly policies. Satisficers are free to decide without any coer-
cive normative best choice canons. On the contrary, in the latter, behavioural economics
concerns a society of optimizers who do not fall into bias errors and are nudged towards
optimal rational choices. The implied paternalistic policy making is obviously much stronger
and costly.12
Zachary McGee, Brooke Shannon and Bryan Jones (Chapter 41 in this volume) write
that one of the first sets of scholars to implement bounded rationality as a microfoundation
of political science were scholars studying public budgets. “Budgetary considerations are suit-
able subjects for exploring institutional decision making, since there is a clear budgetary pro-
cess, mimicking the policy process itself ” (McGee, Shannon and Jones, p. 602). Budget policy
requires various decision-making phases: “In the pre-decision-making phase, even preceding
agenda setting, elites identify problems and then are able to offer solutions. When problems
are identified, they are placed on the agenda, and then solutions can begin to be sought and
offered” (McGee, Shannon, and Jones, p. 603). Budgets are a consistent example of trade-offs
and outputs characteristic of bounded rationality and the information-processing abilities of
individuals and institutions.

They inherently hold prioritization of issues and values, and reveal the capacity
for attention in policy makers. The institutional constraint on policy makers in
Washington for budgets is strong, as interest groups, rival parties, and a scarce amount
of resources frame the debate and compromise prior to policy output, two parameters
for the budgetary process.
McGee, Shannon, and Jones, p. 603

Behavioral rationality acknowledges that individuals and institutions have similar characteristics:

While both individuals and institutions work towards fulfilling their agendas and are
goal-oriented, their goals are impeded by a limited capacity for processing infor-
mation. Since attention-space is limited, agendas must be set to change policy on
the most pressing issues first in case time expires before reaching the end of the
agenda. These shared characteristics make policy change not incremental, but instead,
in bursts.
McGee, Shannon, and Jones, p. 603

46
47

Why bounded rationality?

Information processing provides easy links between individual and organizational choice:

Information processing, in the broadest sense, examines the supply and prioritiza-
tion of information. Usually congressional committees will receive information and
incrementally adjust the policies they deal with. Sometimes, however, new infor-
mation or shifts in issue definitions might cause rapid changes in the problem and
solution definitions. These changes can result in rapid changes in the proposed policy
solutions. Taken together, these conditions are known as the general punctuation thesis
… Information is not scarce within government; in fact, information is in oversupply.
This oversupply of information leads actors to be overwhelmed by their choice envir-
onment. To deal with the oversupply, they must winnow out the information that is
not useful; this winnowing process is boundedly rational. How actors search for and
weight the information they receive (e.g. members in a committee hearing) is cru-
cial to whether or not a policy problem is resolved or a specific solution is chosen.
Ultimately, the decisions made about what information is important are agenda-
setting decisions. Therefore, to think about agenda setting, attention allocation, or
information processing is to confront bounded rationality and its influence on the
policy process literature.
McGee, Shannon, and Jones, p. 607

Chapter 42 by Valentina Ferretti outlines three examples of multi-methodology boosts designed

to support three key stages of the decision-making process: (1) the framing of the problem/
opportunity to be addressed; (2) the expansion of the original decision makers’ mental model
about the objectives to be achieved; and, finally, (3) the elicitation of preferences about the
worthiness of the performances of the alternatives.
There are three common denominators to the behavioural decision analysis interventions
discussed in her chapter:

First, multi-methodology boosts require time and facilitators skilled in multiple

methods. Second, a key advantage of the proposed boosts is their applicability to
many different domains. Third, they stretch the bounds that the uncertain and often
constrained external environment places on human rationality, thus mitigating some
of the associated cognitive consequences such as poor framing of decisions, insuf-
ficient thinking about relevant objectives and scope insensitivity. In particular, the
presence of limited information on alternatives and their consequences can be tackled
through informed value judgments supported by the integration of sound preference
elicitation protocols and visualization analytics, as well as through thought-provoking
questions to expand the set of objectives and create better alternatives.
Ferretti, p. 622

The techniques proposed in this chapter should thus be considered as “process boosts” and in
particular as “framing boosts” and as “preference elicitation boosts”, which could be added to
the first taxonomy of long-term boosts proposed by Hertwig and Grune-Yanoff (2017), i.e.,
risk literacy boosts, uncertainty management boosts and motivational boosts.
One of the neoclassical economics-informed policy recommendations to protect the con-
sumer is to supply most of the information available to allow them to choose from among
many difference options. Many governments around the world follow this recommendation

47
48

Riccardo Viale

that is aimed to reduce information asymmetry and to overcome market failures. Unfortunately,
though, too much information can cause information and choice overload, which have nega-
tive consequences on the accuracy of decision making and on the emotional state of the deci-
sion maker. Chapters 43 and 44 by Elena Reutskaja, Barbara Fasolo, Raffalella Misuraca and
Sheena Iyengar summarize evidence collected by researchers for more than half a century on
the topic of information and choice overload, exploring how people deal with large amounts
of information and how they make choices from sets with multiple alternatives:

Although economics and psychology have both traditionally emphasized the benefits
of more information and more choice, a sizable body of research has demonstrated that
having too much information or too many choices can lead to negative consequences.
Researchers have found that too much information and choice hinders information
processing and usage, motivation to act, and quality and accuracy of decisions, and
that it also negatively impacts affective states of the decision makers. All in all, empir-
ical evidence suggests that both too little and too much choice and information are
“bad” and there is a “golden mean” of how much choice is enough. However, the
optimal information and choice offering is not universal and depends on a vast variety
of contextual and individual factors.
Reutskaja, Fasolo, Misuraca, and Iyengar, p. 633

The results of these studies provide policy makers with some useful bounded rational advice: make
the citizens’ context of choice as transparent and simple as possible. Information should be
limited and relevant to allow citizens to make better political, social and economic decisions.
In an interview in Pittsburgh, Herbert Simon was asked whether simple decision making could
be achieved by reducing the presentation of alternatives. His response was: “Partly. I think the
difficulty of decision making centers very much around the degree of uncertainty and the gaps
in our knowledge.” Moreover, Simon (1971, p. 40) put it:

In an information-rich world, the wealth of information means a dearth of some-

thing else: a scarcity of whatever it is that information consumes. What informa-
tion consumes is rather obvious: it consumes the attention of its recipients. Hence
a wealth of information creates a poverty of attention and a need to allocate that
attention efficiently among the overabundance of information sources that might
consume it.

Smart and ethical choice architecture should direct and preserve decision makers’ attention and
respect decision makers’ freedom. How we manage our attention in our ever more information-
rich world will ultimately dictate our future.
To conclude, a real libertarian paternalism, according bounded rationality, is aimed to supply
the cognitive and educational tools to people to better process information and to improve
their deliberate problem solving in the large world. In other words, to increase their ecological
rationality. Therefore, the only justified libertarian paternalisms seem to be the cognitive and
the educational ones (Viale, 2019; forthcoming). Both aim at empowering the citizens’ decision
making. Cognitive paternalism aims to help citizens neutralize their errors, increasing the envir-
onmental feedbacks of their choice, structuring and simplifying the complex structure of tasks
and filling the gap between the contingent utility of their choice and their future utility and
well-being. Educational paternalism aims at supplying knowledge and skills that change not only
the performance but also the autonomous competence of the decision maker. This applies, for

48
49

Why bounded rationality?

example, to financial, health and environmental education or the teaching of risk literacy and
heuristic decision making.
It is important to consider how to help human reasoning. For example, the mind’s statis-
tical reasoning processes evolved to operate on natural frequencies and Bayesian computations
are simpler to perform with natural frequencies than with probabilities. It is well known that
if information is presented as the outcome of learning from experience, known as natural fre-
quencies, and not as conditional probabilities, the proportion of people reasoning by Bayes rule
increases a lot. Statistics expressed in terms of natural frequencies improve Bayesian inferences
in finance as in many other kinds topics. Therefore, a bounded rationality-inspired architecture
of choice (BRAN Bounded Rational Adaptive Nudge, in Viale, 2017; forthcoming) should
change information formats in probabilistic reasoning from probabilities to natural frequencies.
The importance of nudging people by the natural frequency format to reason correctly in stat-
istical tasks is crucial in many environments. In particular, the frequency format improves the
statistical and the Bayesian reasoning in many financial and medical judgements to correctly
predict, for example, the probability of a disease according to prior probability and new evi-
dence (supplied, for example, by a test with some false positives) or the probability of a fall of
a Stock Exchange according to prior probability and new evidence (the bankruptcy of a big
global bank). The same argument can be applied to many public policies with dramatic future
implications for human life such as financial defaults, natural disasters, terrorist attacks, micro
criminalities, epidemics, but also quieter social phenomena where people have prior probability
and some new evidence, such as the choice of a bank in relation to various rankings or the
choice of a university for their son in relation to the labour market or the choice of a hospital
for a surgical operation in relation to the success rate of similar medical institutions.
This topic is related to another important component of cognitive libertarian pater-
nalism: how to increase the knowledge of feedback from our choices. One of the reasons to
increase the feedback is not only that we can learn from our errors and not to fall another
time in the same choice. It is also that we can improve inductively our theories of the world.
That is, we can improve our prediction on future states of the world, for example, our future
choice of an investment or of a party or of a school. In the experiments on Bayesian learning,
people learn probabilities from experience and are subsequently tested as to whether they make
judgements consistent with Bayes’ rule. Often the tests are successful. Therefore, many cogni-
tive scientists conclude that people’s judgements are largely consistent with this theory. This
kind of test is the cognitive justification for an ecological rational role of the nudge that manages
to increase the knowledge of the feedback from people’s choices.
It seems possible also to design ecologically sound mapping of choice for future welfare. For
example, when an individual has to make a choice about different mortgages or credit
agreements, it is possible to simulate future simple environments with few cues in frequency
formats (for example, the monthly rate) and ask him or her to imagine that situation. In this
case, the attempt is to create a situational rationality dimension and to trigger embodied cog-
nition aspects of the choice. This situation would allow him or her to better understand the
future effects of his or her choice, trying to make subjective present utility converge with future
utility. In other words, this architecture of choice should foster people’s competence to vary
their sense of psychological connectedness, that is their sense of connection with their future
self. In the context of saving, this could mean that the more aware someone is of being the
future recipient of today’s savings, the more prepared that person will be to save for retirement.
In some experiments, participants who interacted with their virtual future selves, and pre-
sumably overcame disconnectedness, were more likely to accept later monetary rewards over
immediate ones.

49
50

Riccardo Viale

Another proposal of cognitive paternalism is designing environments that nudge the util-
ization of a proper suitable heuristic. It is possible, for example, to design an environment
that exploits so-called social intelligence by relying on heuristics designed for social information.
Imitate-the-successful heuristic, for instance, speeds up learning of cue orders and can find orders
that excel take-the-best’s validity orders. Other heuristics include imitation heuristics, tit-for-tat,
the social-circle heuristic, and averaging the judgements of the others to exploit the “wisdom of crowds”
(Gigerenzer and Gassmaier, 2011).
Simplifying and structure complex choice is also a good challenge for BRAN heuristic choice.
For example, trying to make choices in environment that present high redundancy and vari-
ability in weights of their structure. High redundancy means structure where cues are highly
connected (for example, the market value of a business). High variability means structure where
there is great difference in weight between some cues and the others (for example, the weight
of APRC13 compared to other information in the choice of a credit contract). In this struc-
ture, when there is also high uncertainty, it is likely that one-reason decision making as take-the-best
heuristic is able to allow successful inferences that can be superior to those based on algorithms
such as classification and regression tree or conjoint analysis. In most of the choices linked
to your well-being, such as finance, education, health, food, consumption goods, housing,
and so on, you have to search for more than one cue. In these cases also, you may follow a
sequential heuristic that is based on one-reason decision making. An example is elimination by
aspects of lexicographic heuristics to nudge proper choices in a large world. How? Structuring, for
example, with proper software the information given to families, by fast-and-frugal trees in which
is incorporated the lexicographic logic. This is the typical non-compensatory strategy for
choosing in an ecological rational way. In this strategy, people order the cues relying on recall
from the mental sample. A person does not need to learn cue orders individually but instead can
learn from others, such as through teaching and imitation (Gigerenzer and Gassmaier, 2011).
This is an example of BRAN nudges.
To conclude, by fostering an inspired educational libertarian paternalism, competence in
risk literacy, in uncertainty management and in managing motivations and cognitive control
should be boosted (Hertwig and Grune-Yanoff, 2017; 2020) with:

1. The first competence is about understanding statistical information. This competence can
be achieved through (Hertwig and Grune-Yanoff, 2017): (a) graphical representations;
(b) experience-based representations as opposed to description-based representations;
(c) representations that avoid biasing framing effects relying, for example, on absolute
instead of relative frequencies; and (d) learning how to transform opaque representation
(e.g., single-event probabilities) into transparent ones (e.g. frequency-based representations).
2. When people have no access to actuarial information, they should make decisions under
uncertainty, with no explicit risk information available. This is the case of most decisions
in the world of finance. The competence for uncertainty management fosters proced-
ural rules for making good financial decisions, predictions and assessments under uncertain
conditions with the help of simple rules of collective intelligence: fast and frugal trees;
simple heuristics; rule of thumb and procedural routines (Hertwig and Grune-Yanoff,
2017; 2020). For example, Drexler et al. (2014) found that providing microentrepreneurs
with training in basic accounting heuristics and procedural routines significantly improved
their financial practices and outcomes. The impact of rule-of-thumb training was signifi-
cantly greater than that of standard accounting training.
3. People often lack self-control and have weak attention during financial decision making.
The results are often suboptimal. An educational libertarian paternalism may foster the

50
51

Why bounded rationality?

competence to autonomously adjust one’s motivation, and cognitive control in deci-

sion making through growth-mindset or sense-of-purpose exercises; attention state
training; psychological connectedness training; reward-bundling exercises; the stra-
tegic use of automatic processes and harnessing simple implementation intentions;
and training in pre-commitment strategies and self-control strategies (Hertwig and
Grune-Yanoff, 2017).

Notes
1 How to identify why-questions? Consider the stock example: (*) Why did Adam eat the apple? The
why-question asked by (*) is determined by:
(1) A Topic: a proposition expressing the fact whose explanation we are asking. (e.g., in the case of (*),
that Adam ate the apple). Neither having the same topic nor being expressed by the same sentence
constitutes a criterion of identity for why-questions. In other words, the same sentence can express
different why-questions. For Van Fraassen, there are various ways in which this can happen. This
leads him to bring in two more contextual factors:
(2) A Contrast Class: a set of propositions, including the topic, that determines the range of alternatives
against which a why-question is asked. For example, Why did Adam eat the apple (as opposed to
Eve)?, why did Adam eat the apple (as opposed to, say throwing it away because it was old)?, why
did Adam eat the apple (as opposed to the oranges, the grapes, etc.)?
(3) A Relevance Relation: “a respect-in-which a reason is requested”. An example of how
(3) discriminates among different why-questions: consider the question ”Why was Mr. Jones killed
by his wife?”. On a natural understanding, it is irrelevant to answer ”Because she stuck a knife in
his throat, thereby causing such and such events that led to Mr. Jones’s death”.
2 In the past, it was believed that man was a rational animal because his reasoning was thought to comply
aprioristically with the precepts of classical logic. The answer to question 2 was therefore taken as
being non-problematic and established a priori within the terms of a positive response to question
3. What seemed to be merely a conjecture which could explain the reason for this belief in the ancients
instead now seems to reveal a widespread prejudice even among the most sophisticated and modern
psychologists. Even a demolisher of rationalistic certainties like Phil Johnson-Laird justifies, in a some-
what aprioristic way, human logical rationality (also if not in the sense of classical logic, but in that of the
psychological theory of “mental models”) by stating “that if people were intrinsically irrational, then
the invention of logic, mathematics, and much else besides, would be inexplicable” (Johnson-Laird,
1983, p. 66).
3 I call this proposal Minimal Dual Account of Mind (Viale, 2019).
4 As will be noted later, the theory of mental dualism has attracted some criticism (Kruglanski and
Gigerenzer, 2011; Viale, 2019) and the unified account of the human mind appears to have become a
more credible alternative.
5 Sloman’s (1996) Criterion S is met when a person simultaneously holds two contradictory beliefs. The
idea is that this occurs when the associative and rule-based systems arrive at conflicting responses to a
situation.
6 Science aims to transform large world problems into small world problems and solutions. This is
possible only when large world problems are characterized by epistemic uncertainty and not by
fundamental or ontological uncertainty. The first kind of uncertainty occurs when, ideally, empir-
ical research and the collection of data are able to supply statistical figures that characterize relevant
variables, their consequences and probabilities. The second kind of uncertainty deals with events
that empirical research is not able to represent probabilistically because of complexity or unpredict-
able surprises. The first kind of uncertainty usually applies to most biomedical research (for example,
trials for a new drug) whereas the second applies to macro-political, environmental and financial
phenomena (for example, the prediction of a financial crisis). For example, Covid-19, like any other
infection, is typically characterized by epistemic uncertainty. In a few years biomedical research will be
able to define its viral behaviour and possible treatments (ceteris paribus with likely mutations). On the
contrary, financial markets are typically characterized by ontological uncertainty. No-one may predict
their future behaviour.

51
52

Riccardo Viale

7 According to authors such as Dreyfus (1992) and Winograd (1984), no computational device can
imitate the human singularity of being situated in the world. Therefore, the project of Artificial
Intelligence is doomed to failure.
8 It is not clear what is the difference for Rubinstein between a story and a model. In philosophy of
science, stories and models are often considered similar. According to Alan Gibbard and Hal Varian
(1978), a model is similar to a story. As in a caricature, it emphasizes and exaggerates some special
features and neglects other features that are closer to the real phenomenon.
9 The central principle is Schroedinger’s equation. This equation describes how systems subject to
various forces evolve in time, but in order to know the evolution of a real system, we also need to
know the mathematical representation of the kinetic and potential energies of that real system. This is
called Hamiltonian representation. In principle, if we wanted to have a realistic picture of the various
systems, we ought to calculate a Hamiltonian representation of each real system we wish to study.
Instead, if we look at physics text-books, we discover that they contain only a limited number of
Hamiltonians, which do not fit real objects, such as the hydrogen atom, but only highly fictionalized
objects, such as free particle motion, the linear harmonic oscillator, piecewise constant potentials,
diatomic molecules, central potential scattering, and so on.
10 I refer to ontic or structural uncertainty that is characterized by unpredictability. In situations
characterized instead by epistemic uncertainty, the collection of data may increase our knowledge and
predictive power. The only danger in these cases is of falling into the variance phenomenon when the
data to be collected are too vast and should therefore be limited to only a small, non-representative
subset.
11 The identification of biases and formal errors of judgement is not a recent phenomenon. Adam
Smith highlighted both the phenomena of “loss aversion” and “hyperbolic discounting” as explanatory
factors for human behaviour. Niccolò Machiavelli also drew attention to the decision-making power
of both the “endowment effect” and “loss aversion”. Lastly, David Hume underlined the danger of
“present bias” and the myopic nature of human judgement.
12 Paradoxically, the society of satisficers might need a stronger paternalism than that of optimizers.
Think of an ecological rational policy maker who wants to apply the normative side of ecological
rationality. He or she will oblige people to learn all the heuristics adapted for each separate task and
environment. He or she will try to nudge or oblige people to use the proper heuristics adapted for
each particular environment. The result: much more paternalism, cost and complexity compared to
behavioural economics nudging.
13 APRC is the Annual Percentage Rate of Charge.

References
Becker, G. (1962). “Investment in human capital: A theoretical analysis”. Journal of Political Economy,
70(9), 9–49.
Berg, N. and Gigerenzer, G. (2010). “As-if behavioral economics: neoclassical economics in disguise?”,
History of Economic Ideas, XVIII, 1, 133–165.
Cartwright, N. (1983). How the Laws of Physics Lie. Oxford: Oxford University Press.
Challenge interview with Herbert A. Simon (1986). “The failure of armchair economics”. Available
at: http://digitalcollections.library.cmu.edu/awweb/awarchive?type=file&item=34037.
Clark, A. (2016). Surfing Uncertainty. Oxford: Oxford University Press.
Cleermans, A., and Jimenez, L. (2002). “Implicit learning and consciousness: A graded, dynamic per-
spective”. In R. M. French and A. Cleermans (eds), Implicit Learning and Consciousness: An Empirical,
Philosophical and Computational Consensus in the Making (pp. 1–40). Hove, UK: Psychology Press.
Dennett, D. (2017). From Bacteria to Bach and Back: The Evolution of Minds. New York: Norton.
Drexler, A., Fischer, G., and Scholar, A. (2014). “Keeping it simple: Financial literacy and rules of thumb”,
American Economic Journal: Applied Economics, 6, 1–31.
Dreyfus, H. (1992). What computers still can’t do: A critique of artificial intelligence. Cambridge, MA: MIT Press.
Evans, St. B. T. J. (2008). “Dual-processing accounts of reasoning, judgement and social cognition”.
Annual Review of Psychology, 59, 255–278.
Evans, St. B. T. J., and Stanovich, K. E. (2013). “Dual-process theories of higher cognition: advancing the
debate”, Perspectives on Psychological Science, 8(3), 223–241.
Fehr, E. and Schmidt, K. (1999). “A theory of fairness, competition and cooperation”. Quarterly Journal
of Economics, 114, 817–868.

52
53

Why bounded rationality?

Friedman, M. (1953). Essays in Positive Economics. Chicago: University of Chicago Press.

Gibbard, A., and Varian, H. (1978). “Economic models”. Journal of Philosophy, 75, 664–677.
Gigerenzer, G. and Gassmaier, W. (2011). “Heuristic decision making”. Annual Review of Psychology,
62: 451–482.
Gigerenzer, G., and Selten, R. (2001). Bounded Rationality. Cambridge, MA: MIT Press.
Gigerenzer, G., Todd, P. M. and the ABC Research Group (eds) (1999). Simple Heuristics that Make Us
Smart. New York: Oxford University Press.
Hacking, I. (1983). Representing and Intervening. Cambridge: Cambridge University Press.
Hammond, K. R. (1996). Human Judgement and Social Policy. Oxford: Oxford University Press.
Hertwig, R., and Grune-Yanoff, T. (2017). “Nudging and boosting: Steering or empowering good
decisions”. Perspectives on Psychological Science, 12(6), 973–986.
Hertwig, R., and Grune-Yanoff T. (2020). “Boosting and nudging: Two paths toward better financial
decisions”. In R. Viale, S. Mousavi, U. Filotto and B. Alemanni (eds), Financial Education and Risk
Literacy. Aldershot: Elgar.
Kruglanski, A. W., and Gigerenzer, G. (2011). “Intuitive and deliberative judgements are based on
common principles”. Psychological Review, 118, 97–109.
Leslie, A. M. (1987). “Pretense and representation: The origins of ‘theory of mind’”. Psychological Review,
94: 412–426.
Laibson, D. (1997). “Golden eggs and hyperbolic discounting”. Quarterly Journal of Economics,
112(2): 443–477.
Marr, D. (1982). Vision. San Francisco CA: Freeman.
Newell, A., and Simon, H. (2019). Human Problem Solving. Brattleboro, VT: EPBM.
Osman, M. (2004). “An evaluation of dual-process theories of reasoning”. Psychonomic Bulletin and Review,
11(6), 988–1010.
Petty, R. E., and Cacioppo, J. T. (1986). “The Elaboration Likelihood Model of persuasion”. In L.
Berkowitz (ed.), Advances in Experimental Social Psychology. New York: Academic Press.
Quine, W. O. (1985). “Epistemology naturalized”. In H. Kornblith (ed.), Naturalizing Epistemology.
Cambridge: Cambridge University Press.
Rosenberg, A. (1983). “If economics is not science, what is it?” In D. M. Hausman (ed.), The Philosophy
of Economics. Cambridge: Cambridge University Press.
Savage, L. J. (1954). The Foundations of Statistics. 2nd edn. New York: Dover.
Simon, H. (1956) “Rational choice and the structure of the environment”. Psychological Review, 63(2),
129–138.
Simon, H. (1971). “Designing organizations for an information-rich world”. In M. Greenberg (ed.),
Computers, Communication and the Public Interest. Baltimore, MD: The Johns Hopkins University Press.
Simon, H. (1979a). “Information processing models of cognition”. Annual Review of Psychology,
30: 363–396.
Simon, H. (1979b). “Rational decision making in business organizations”. American Economic Review,
69(4), 493–513.
Simon, H. (1981). Sciences of the Artificial. Cambridge, MA: MIT Press.
Simon, H. (1990). “Invariants of human behaviour”. Annual Review of Psychology, 41, 1–19.
Simon, H. (2000). “Bounded rationality in social sciences: Today and tomorrow”. Mind & Society, 1, 25–41.
Simon, H. A., Egidi, M., Viale, R. and Marris, R. (1992). Economics, Bounded Rationality and the Cognitive
Revolution. Aldershot: Elgar.
Sloman, S. (1996). “The empirical case for two systems of reasoning”. Psychological Bulletin, 119, 3–22.
Sloman, S. (2002). “Two systems of reasoning”. In T. Gilovich, D. Griffin and D. Kahneman (eds),
Heuristics and Biases: The Psychology of Intuitive Judgement (pp. 379–398). Cambridge: Cambridge
University Press.
Stanovich, K. E. (2009). “Distinguishing the reflective, algorithmic and autonomous minds: Is it time for a
tri-process theory?” In J. Evans and K. Frankish (eds), In Two Minds. Oxford: Oxford University Press.
Stanovich, K. E., and Toplak, M. (2012). “Defining features versus incidental correlates of Type 1 and
Type 2 processing”. Mind & Society, 11: 3–13.
Stiglitz, G. (2010). Freefall: America, Free Markets, and the Sinking of the World Economy. New York: Norton.
Thaler, R. H. (1991). Quasi Rational Economics. New York: Russell Sage Foundation.
Thaler, R., and Sunstein, C. (2008). Nudge. London: Penguin.
Todd, P. M., Gigerenzer, G. and the ABC Research Group (eds) (2012). Ecological Rationality: Intelligence
in the World. New York: Oxford University Press.

53
54

Riccardo Viale

Van Fraassen, B. (1980). The Scientific Image. Oxford: Oxford University Press.
Viale, R. (1997). “Cognitive economics”. Working Paper 1. LaSCoMES, Bocconi University.
Viale, R. (2012). Methodological Cognitivism: Mind, Rationality and Society. Heidelberg: Springer.
Viale, R. (2013). Methodological Cognitivism: Cognition, Science, and Innovation. Heidelberg: Springer.
Viale, R. (2016). “Brain driven creativity”. In L. Macchi, M. Bagassi, and R. Viale (eds), Cognitive
Unconscious and Human Rationality. Cambridge, MA: MIT Press.
Viale, R. (2017). “How to decide in the large world of finance: The Bounded Rational Adaptive Nudges
(BRAN)”. Quaderni di Finanza, 84, 99–109.
Viale, R. (2018). “The normative and descriptive weaknesses of behavioral economics-informed
nudge: Depowered paternalism and unjustified libertarianism” Mind & Society, 1–2(17), 53–69.
Viale, R. (2019). “Architecture of the mind and libertarian paternalism: Is the reversibility of system 1
nudges likely to happen?” Mind & Society, 2(19).
Viale, R. (forthcoming). Nudging. Cambridge, MA: MIT Press.
Winograd, T. (1984). “Computer software for working with language”. Scientific American, September.

54
55

2
WHAT IS BOUNDED
RATIONALITY?
Gerd Gigerenzer

Bounded rationality has, surprisingly, three faces. The original is by Herbert A. Simon, who
coined the term. The other two arrived later and bent Simon’s term into meanings that could
hardly be more different. Equally puzzling, the two new meanings contradict each other, each
having absorbed the term into their own framework. This double take-over has been so subtle
that few people notice that, when talking about bounded rationality, they are often talking
about different things. The three faces of bounded rationality are not simply a matter of ter-
minology; they reflect fundamentally dissimilar research programs and visions about the nature
of human rationality. In this chapter, I will describe the three faces, explain how we got there,
and outline how to develop Simon’s original program.

Simon’s bounded rationality

Herbert A. Simon (1916–2001) coined the term bounded rationality. While still a student, fresh
out of a price theory class at the University of Chicago, he tried to apply the perspective of
utility maximization to budget decision problems in his native Milwaukee’s recreation depart-
ment in the mid-1930s. To his surprise, he learned that managers did not try to compare the
marginal utility of a proposed expenditure with its marginal costs, but simply added incre-
mental changes to last year’s budget and engaged in other rules of thumb. During that concrete
encounter, Simon realized what managers could and could not measure, and concluded that
the framework of utility maximization “was hopeless” (Simon 1988, p. 286). This discrepancy
between theory and reality marked the beginning of what he later called the study of bounded
rationality: “Now I had a new research problem: How do human beings reason when the
conditions for rationality postulated by the model of neoclassical economics are not met?”
(Simon, 1989, p. 377).
What is rationality in neoclassical economics? Although not all economists agree, they typ-
ically refer to three pillars: consistency, maximization of expected utility, and—if learning is
involved—Bayesian updating of probabilities. Leonard J. Savage (1954), known as the father of
modern Bayesian decision theory, defined two conditions necessary for these three pillars of
rationality:

55
56

Gerd Gigerenzer

1. Perfect Foresight of Future States: The agent knows the exhaustive and mutually exclusive set
S of future states of the world.
2. Perfect Foresight of Consequences: The agent knows the exhaustive and mutually exclusive set
C of consequences of each of his or her actions, given a state.

Savage called the pair (S, C) a small world. States and consequences must necessarily be described
at some limited level of detail—hence the adjective small. The prototype of a small world is
a lottery where all possible future states (the tickets) along with all possible outcomes (the
payoffs) and their probabilities are known for certain, or a game such as roulette where all states
(in roulette: the 36 numbers plus zero), consequences, and probabilities are known. In a small
world, nothing new or unexpected can ever happen. These conditions have been variously
called unbounded rationality, full rationality, constructivist rationality, or the Bayesian ration-
ality approach. What Simon noticed, however, was that the managers he observed and humans
in general mostly have to deal with situations that are unlike small worlds with perfect foresight
and that do not meet the assumptions for expected utility maximization and Bayesian updating.
For instance, if the exhaustive and mutually exclusive set of future states of the world and their
consequences is not known, one cannot maximize expected utility or assign prior probabilities
to states that add up to one, which makes Bayesian updating unfeasible. Simon called for an
empirical study of how humans reason when perfect foresight is not possible. In his lifetime,
most economists showed little interest in Simon’s question and preferred theories assuming
perfect foresight.

Risk and uncertainty

I will use the term risk for situations in which agents have perfect foresight of future states and
their consequences, as defined above, and also certain knowledge about the probabilities of
the states. For the many situations in which this is not the case, I will use the term uncertainty.
This distinction between risk and uncertainty goes back to Frank Knight (1921). Uncertainty is
sometimes used to denote a small world without known probabilities (Luce & Raiffa, 1957),
a situation known as ambiguity. Yet ambiguity contains only a minor degree of uncertainty; it
still assumes perfect foresight of future states and consequences. When speaking of uncertainty,
I also refer to situations that do not meet the definition of small worlds. Uncertainty is some-
times called fundamental uncertainty or radical uncertainty. It includes ambiguity and intractability
(such as in chess), ill-defined problems such as budget decisions where optimal solutions cannot
be known because perfect foresight is an illusion, and, in general, all future events that are
not perfectly foreseeable. In short, uncertainty includes most of the interesting problems that
humans face in real life.
Now we can formulate the first principle of Simon’s bounded rationality program: to study
how human beings make decisions under uncertainty, not only under risk.
Decision making under uncertainty is obviously relevant for understanding how managers,
business people, or other human beings make choices. In contrast, decision making under risk
is rare, mostly restricted to human-made environments such as lotteries, games, and gambling.
Nevertheless, lotteries remain the stock-in-trade of decision research, defining risk aversion,
regret, and loss aversion, and are the domain of modifications of expected utility theory, such
as prospect theory.
Savage himself explicitly limited his theory, currently known as Bayesian decision theory, to
small worlds only. For instance, he pointed out that his theory does not apply to planning a
picnic or playing a game of chess (Savage, 1954, p. 16). When planning a picnic, one cannot

56
57

What is bounded rationality?

know ahead all future states that might happen. And when playing chess, where a finite number
of states (sequences of moves) indeed exists, the problem is intractability because no human or
machine can enumerate the exhaustive set of possible moves and determine the optimal one.
As a consequence, one cannot assign probabilities to these states that add up to one. To illus-
trate, chess has approximately 10120 different unique sequences of moves or games, which is a
number greater than the estimated number of atoms in the universe. Savage believed it would
be “utterly ridiculous” to apply rational choice theory to situations without perfect foresight
(1954, p. 16). Both Simon and Savage were fully in agreement that expected utility maximiza-
tion is a local but not a universal tool.

As-if and real decision-making processes

Simon’s question is about the process of decision making: “How do human beings reason?” Simon
criticized the fact that neo-classical economists showed surprisingly little interest in studying how
people actually make decisions, but instead engaged in armchair speculation and relied mainly on
their intuition. When confronted with the same kind of evidence as Simon once was, namely,
that business professionals did not behave as postulated by utility theory, Milton Friedman reacted
differently: He dismissed the empirical evidence as irrelevant. In his as-if defense, Friedman (1953)
famously argued that the goal of “positive economics” is prediction, not psychological realism.
The theory says that people behave as if they maximized their expected utility, not that they
actually do. All that counts are sufficiently good predictions. I do not doubt that expected utility
theory and its variants can explain behavior after the fact, due to its great flexibility; whether
it can actually predict behavior is less clear. True predictions must be out-of-sample, not by
fitting a model to known data. After reviewing 50 years of research for evidence about how well
utility functions, such as utility of income functions, utility of wealth functions, and the value
function in prospect theory actually predict behavior, D. Friedman, Isaac, James, and Sunder
(2014) concluded: “Their power to predict out-of-sample is in the poor-to-nonexistent range,
and we have seen no convincing victories over naïve alternatives” (p. 3).
Now we can formulate the second principle of Simon’s bounded rationality program: To
study how human beings make decisions, as opposed to relying on as-if expected utility theories.
In my opinion, Simon’s insistence on studying the process and Friedman’s goal of pre-
diction are not incompatible opponents; they can be reconciled. My hypothesis is that by
constructing realistic process models, one can, on average, make better predictions than by
using as-if models. I will provide evidence for this argument below using the example of the
prediction of customer purchases.
But what are the tools humans use to make good decisions under uncertainty? Knight
(1921) had proposed “judgment” and “experience,” but not much more. In his General Theory,
John Maynard Keynes (1936) proposed “animal spirits,” without telling us what these exactly
are. In their book Animal Spirits, Akerlof and Shiller (2009) remedied this by moving a step
forward and distinguishing five such spirits—confidence, fairness, corruption, money illusion,
and stories. The authors argue that these “restless and inconsistent” elements were the principal
reasons for the financial crisis of 2008 (pp. vii, 4).
Simon (1955, 1956) made a more concrete start, proposing that humans make decisions
under uncertainty by relying on aspiration levels, limited search, and heuristics such as satisficing.
In its simplest version, the satisficing rule is:

Step 1: Set an aspiration level α.

Step 2: Choose the first option that satisfies α.

57
58

Gerd Gigerenzer

To illustrate, consider a study of how real estate entrepreneurs decide in which location to invest
in order to develop a new commercial high-rise or a residential area. Berg (2014) reports that
of 49 professional investors, each and every one relied on a version of satisficing: If I believe I can
get at least x return within y years, then I take the option. Here, “x return in y years” is the aspiration
level. In general, an aspiration is a goal, and an aspiration level is a goal value. The psycholo-
gist Kurt Lewin (1935), who promoted the concept of aspiration, considered successful people
to be those who set goals that are within their ability to reach. Using a satisficing heuristic,
people can deal with uncertainty because the rule does not require perfect knowledge about all
options, such as possible locations, marriage partners, or professions. How did Simon decide to
study political science and economics? In his own account, he “simply picked the first profes-
sion that sounded fascinating” (1978, p. 1).
In general, satisficing can handle situations where the exhaustive and mutually exclusive set
of options is not knowable, and where options need to be searched one by one. In such situ-
ations, optimizing is not an option.

Behavior = f (cognition, environment)

The third principle that defines Simon’s bounded rationality is captured by his scissors ana-
logy: behavior is a function of both cognition and environment (Simon, 1990, p. 7): “Human
rational behavior (and the rational behavior of all physical symbol systems) is shaped by a scissors
whose two blades are the structure of task environments and the computational capabilities of
the actor.”
Simon did not flesh out the scissors principle in detail; I will provide an example below in
the section on ecological rationality. In general, it requires an analysis of both the heuristics
people use and the structure of the environments in which they act. Economic theory analyzes
the incentive structure in environments but not the actual decision strategies, assuming as-if
utility maximization. Thus, it deals with only one blade. Similarly, much of psychological the-
orizing restricts itself to cognitive processes and is mute about environmental structure. Thus,
it deals solely with the other blade. Debates in psychology tend to center around “internal”
questions: whether the mind is a Bayesian or not, whether information is integrated in an
additive or multiplicative way, whether the mind works by mental models or mental logic, or
whether a person is risk averse or risk seeking. By analyzing only one blade of Simon’s scissors,
one cannot understand the ecological rationality of a heuristic, a belief, or other strategies.
To summarize, there are three principles that define Simons’ program of bounded rationality:

1. Uncertainty. To study decision making under uncertainty, not only risk.

2. Process. To study the actual process of decision making, as opposed to as-if expected utility
maximization.
3. Scissors. To study how the structure of an environment, together with the cognitive process,
produces the resulting behavior.

For many of Simon’s contemporaries, this program was asking too much. It clashed with the
practice of reducing all uncertainties to risks by assuming perfect foresight, such as by converting
real-world problems into lotteries. It also clashed with the ideal of explaining all behavior after
the fact by saying that it maximizes some utility, regardless of Savage’s small-world restriction.
Needless to say, Simon’s revolution did not happen during his lifetime.
The rejection of novel ideas is nothing new in the history of science. However, the reaction
to Simon’s ideas was rather unusual. Instead of being rejected as worthless or unimportant, his

58
59

What is bounded rationality?

concept of bounded rationality was hijacked and radically reframed, making his revolutionary
ideas no longer recognizable. Neo-classical economists absorbed the term into the orthodoxy
of perfect foresight and concluded that satisficing is quintessentially optimizing. Proponents of
the heuristics-and-biases program in psychology, in contrast, appropriated the term for their
own focus on human lack of rationality. This is why bounded rationality now has three faces
that could hardly be more dissimilar.

Bounded rationality as optimization under constraints

Satisficing involves search. Simon (1955, 1956) was one of the first to raise the topic of search
in the economics literature. Note that in situations of risk, search is not an issue, given that
all alternatives are already known. However, it was Stigler (1961) who popularized the topic
by changing satisficing search to optimal search. In his classical example of the purchase of
a second-hand car, the buyer is assumed to stop searching when the costs of further search
exceed its benefits. Yet to factor costs and benefits into the equation, the buyer needs to know
all alternatives and distributions in the first place, in addition to the costs of search. Stigler’s
approach poured search theory back into the old bottle of expected utility maximization, intro-
ducing perfect foresight once again, and reducing uncertainty to risk. With this move, satis-
ficing is equivalent to optimizing with the constraint of search costs. In fact, Simon (1955)
had ventured in the same direction earlier. There now exists a large literature in economics in
which satisficing is modeled as optimization under constraints, mostly in the expected utility
maximization tradition, whereas models of satisficing under uncertainty are rare. In a memorial
book for the late Simon, Arrow (2004, p. 48) argued that “boundedly rational procedures are
in fact fully optimal procedures when one takes account of the cost of computation in addition
to the benefits and costs inherent in the problem as originally posed.” Bounded rationality is
thereby seen as nothing but constrained optimization in disguise.
In 1993, Thomas Sargent published his book Bounded Rationality in Macroeconomics, where he
re-interprets “the idea of bounded rationality as a research program to build models populated
by agents who behave like working economists or econometricians” (1993, p. 22). As he points
out, by adding more and more constraints to optimization, models of bounded rationality
become larger and mathematically more demanding, which is why there is no rush among
econometricians to implement such models. Econometricians try to reduce the number of
parameters, and that “is not what bounded rationality promises” (p. 5). In this view, bounded
rationality makes optimization more difficult, and unbounded rationality ends up as the sim-
pler and tractable alternative. This interpretation turned the idea of simple heuristics such
as satisficing upside down. In personal conversation, Simon playfully remarked that he had
considered suing authors who misused the term bounded rationality to mean optimization
under constraints (Gigerenzer, 2004). In 1998, Ariel Rubinstein published Modeling Bounded
Rationality and admirably included some of Simon’s comments (Rubinstein, 1998). Simon
criticized Rubinstein for showing no awareness of psychological research apart from that of
Kahneman and Tversky. Rubinstein responded that, unlike some other economists, he was
concerned neither with predicting behavior nor with giving normative advice. Rather, “by
modeling bounded rationality, we try to examine the logic of a variety of principles that guide
decision makers, especially within interactive systems (markets and games)” (1998, p. 191). But
these logical armchair analyses were exactly what Simon was battling.
During his lifetime, Simon was highly influential in many fields, from artificial intelligence
to psychology to political science. The major exception was in economics, the field in which
he was awarded the Nobel Memorial Prize. In his own assessment, his program of bounded

59
60

Gerd Gigerenzer

rationality was received with “something less than unbounded enthusiasm” and “largely ignored
as irrelevant for economics” (Simon, 1997, p. 269). I believe that economics has missed an
important call to extend its theoretical and methodological toolbox.

Bounded rationality as irrationality

Beginning in the 1970s, psychologists within the heuristics-and-biases program ventured to
show that the assumptions of neo-classical economic theory are descriptively incorrect: the
consistency axioms, the maximization of expected utility, and the updating of probabilities by
Bayes’ rule (e.g., Tversky & Kahneman, 1974). These assumptions are also known as rational
choice theory. This program produced a long list of discrepancies between theory and actual
behavior. A discrepancy could be due to flaws either in the theory or in behavior. Unlike
Simon, the program maintained rational choice theory as normatively correct and blamed the
discrepancies on people’s lack of rationality. Its proponents have even defended utility theory as
a universal norm against critics such as Lola Lopes (e.g., Tversky & Bar-Hillel, 1983, p. 713).
Accepting rational choice theory as a universal norm and attributing deviations from this norm
to people’s lack of rationality is clearly not what Simon had in mind. “Bounded rationality is
not irrationality” (Simon, 1985, p. 297).
Nevertheless, the heuristics-and-biases program also appropriated the term bounded ration-
ality for itself. “Our research attempted to obtain a map of bounded rationality, by exploring
the systematic biases that separate the beliefs that people have and the choices they make
from the optimal beliefs and choices assumed in rational agent models” (Kahneman, 2003,
p. 1449). Now we have a third meaning of the term: human errors, defined as whenever human
judgment deviates from rational choice theory. That contradicts both Simon’s original meaning
and economists’ re-interpretation of the term as economic rationality. Moreover, these alleged
biases were attributed to people’s use of heuristics. As a consequence, the term heuristic became
associated with bias, contrary to a long tradition in psychology and computer science and to
the definitions listed in the Oxford English Dictionary (n.d.).
The observation that people sometimes violate rational choice theory eventually led to dual
process theories that “explain” this apparent flaw by assuming an error-prone “System 1” that
intuitively uses heuristics and a “System 2” that operates according to rational choice theory.
So far, these systems have been characterized by lists of general dichotomies (heuristic versus
logical, unconscious versus conscious, and so on), without precise models of the processes, such
as formal models of heuristics. Moreover, aligning “heuristic” with “unconscious” and “biases”
is not even correct; every heuristic I have studied can be used both consciously and uncon-
sciously, and can lead to better or worse decisions than what would be considered “rational”
(Kruglanski & Gigerenzer, 2011).
How did bounded rationality come to be equated with people’s irrationality? A key to
finding an answer is found in the classical papers of Kahneman and Tversky, which are reprinted
in the anthology by Kahneman, Slovic, and Tversky (1982). In these articles, neither Simon nor
bounded rationality is cited. But in the Preface to the anthology, Simon is briefly mentioned,
apparently more as a nod to a distinguished figure than an acknowledgment of a significant
intellectual link (Lopes, 1992). Thus, the re-interpretation of bounded rationality as irration-
ality likely occurred as an afterthought.
Does the heuristics-and-biases research implement Simon’s three principles? First, like
neo-classical theory, the program does not make the distinction between risk and uncer-
tainty. For most problems investigated, it assumes that logic or probability theory provides the
correct answer. With respect to the study of the process of decision making, it has made some

60
61

What is bounded rationality?

contributions in its work on heuristics, but these are all associated with biases. Moreover, the
heuristics proposed in the 1970s have never been defined so that their predictions could be
tested but instead were used to explain deviations between theory and behavior after the fact,
as with the availability heuristic in the original letter-R study (Tversky & Kahneman, 1973).
When availability was later more clearly defined and its predictions tested for the very same
study, it was shown not to predict people’s judgment (Sedlmeier, Hertwig, & Gigerenzer,
1998). Hindsight is easier than foresight. The use of labels instead of models differs from the
earlier work by Tversky (1972) on elimination-by-aspect. In terms of the scissors analogy, the
heuristics-and-biases program focused solely on internal explanations of behavior, without
an analysis of the environmental conditions under which a heuristic is likely to succeed in
reaching a goal or not. In this program, the use of heuristics is the problem and the supposed
cause of errors. For decisions under uncertainty, however, heuristics are the solution, not the
problem.

Homo heuristicus, Homo economicus, and Homer Simpson

The three faces of bounded rationality are not complementary, but conflicting. Let us give
each of the three faces a name. Homo heuristicus personifies Simon’s original program to study
decision making under uncertainty. Homo economicus stands for the neo-classical program
assuming situations of risk. Finally, Homer Simpson is an apt name for the program studying
the deviations of people’s behavior from the ideal of Homo economicus, attributing these to flaws
in people rather in the theory (the name is from Thaler & Sunstein, 2008).
These three meanings of bounded rationality are not the only ones in circulation. Every time
I enter the US to give a talk, the immigration officer asks me about the topic of my talk. Usually
I respond by giving a two-sentence summary of bounded rationality. One officer nodded and
remarked: “Oh, that’s when you get old!” Another one got the point and said: “Oh, uncer-
tainty, that’s what I do, trying to make sense of people. Would you please write down the title
of your book?” Most of the time, however, explaining bounded rationality just speeds up the
process: “Oh, thanks, just go through.”
I end this chapter with a short introduction to the program of ecological rationality, the
modern extended version of Simon’s bounded rationality.

The ecological rationality program: Homo heuristicus

Simon’s question was descriptive: What are the tools humans use to make decisions under
uncertainty? The program of ecological rationality starts from that question and extends it
to three altogether, a descriptive, prescriptive, and engineering one (Gigerenzer, Hertwig,
& Pachur, 2011; Gigerenzer & Selten, 2001). The descriptive question concerns the reper-
toire of tools that individuals and institutions rely on to make decisions under uncertainty.
Simon suggested one such tool, satisficing. This repertoire is called the adaptive toolbox.
The term toolbox signifies that there is more than one tool, and adaptive that the tools are
adapted to specific classes of problems, just as a hammer is intended for nails, and a screw-
driver for screws. The prescriptive questions fleshes out the scissors analogy: What envir-
onmental structures can a heuristic exploit in order to make better decisions? This question
is addressed using analysis and computer simulation. The results to the first two questions
provide answers to the engineering question: How can transparent, intuitive, and efficient
decision systems be designed? This is the study of intuitive design (not covered here; see
Gigerenzer et al., 2011).

61
62

Gerd Gigerenzer

The ecological rationality program is based on three methodological principles:

1. Algorithmic models of heuristics. Early candidates of heuristics such as availability and repre-
sentativeness were largely undefined. Algorithmic models of heuristics such as satisficing
replace vague labels.
2. Competitive testing. Models of heuristics should be tested against the best competing models,
not against a null hypothesis of no effect.
3. Test of predictions, not of data fitting. Prediction is about the future, or about data that have not
yet been observed. Fitting takes place when the data have already been observed and the
parameters of a model are chosen so that they maximize the fit (such as R2). Fitting alone
is not a proper test of a model: The more free parameters are added to a model, the better
it fits the data, but this does not necessarily hold for prediction, which involves uncertainty.
In fact, simple heuristics can outperform more complex strategies in prediction.

The adaptive toolbox

The study of the adaptive toolbox of an individual, an organization, or a species proceeds by
observation and experiment. It aims at algorithmic models of the heuristics that people actu-
ally use under uncertainty rather than aiming at as-if models of expected utility maximization.
Models of bounded rationality describe not merely the outcome of a judgment or decision
but also how it is reached (that is, the heuristic processes or proximal mechanisms). They also
describe the building blocks of heuristics, which allow heuristics to be combined or extended.
Consider once again the satisficing heuristic. In situations where one is unsure about the proper
aspiration level, the basic satisficing rule can be extended to satisficing with adaptation:

Step 1: Set an aspiration level α.

Step 2: Choose the first option that satisfies α.
Step 3: If after time β no option has satisfied α, then change α by an amount γ and continue
until an option is found.

The first two steps are identical to the basic form of satisficing, but the third allows the aspir-
ation level to be adapted if no option is found after a certain amount of time.
To illustrate, consider the question of how car dealers should price used cars. There is no
unique answer in economic theories. Building on Stigler’s (1961) optimization under con-
straint theory, some theories assume that prices should be fine-tuned to the changes in market
conditions, such as supply and demand; others assume two kinds of customers, informed
shoppers who seek the lowest price and naïve customers who go for a brand name, repu-
tation, or other surrogates for quality. To address this variability among customers, dealers
should respond with mixed strategies where prices are randomized, just as some supermarkets
change prices in a quasi-random way. An analysis of 628 German car dealers and over 16,000
cars showed, however, that none of these theories predicted the actual prices (Artinger &
Gigerenzer, 2016). There was virtually no fine-tuning, despite drastic changes in supply, and
randomizing prices was also absent.
How then do dealers set prices? In Artinger and Gigerenzer’s (2016) study, almost all of the
dealers (97 percent) used a satisficing heuristic. Nineteen percent of the dealers relied on the
basic satisficing heuristic: They chose an initial price (the aspiration level), typically below
the average price on the market, and kept it constant until the car was sold. The majority
(78 percent) of dealers relied on satisficing with adaptation. Typically, they chose an initial price

62
63

What is bounded rationality?

above the middle of the price range on the market and, if the car was not sold after about four
weeks, lowered the price, and so on (Artinger & Gigerenzer, 2016). Do dealers leave money on
the table by relying on satisficing? A comparison with the economic models mentioned above
indicated that both versions of satisficing used by the dealers led to higher, in fact, more than
double the profit. Heuristics can perform well under uncertainty.
The study of the adaptive toolbox has documented several classes of heuristics besides satis-
ficing. One class consists of one-reason heuristics, which rely on a single powerful cue and
ignore all others (see below). Another class contains lexicographic heuristics, including fast-
and-frugal trees and take-the-best. Lexicographic rules also base the final decision on one
reason only, but unlike one-reason heuristics, they may search through several reasons before
finding the decisive one. A fourth class consists of social heuristics, designed for coordination,
competition, and co-operation. An overview of these heuristics can be found in Gigerenzer
and Gaissmaier (2011) and Gigerenzer et al. (2011)

Ecological rationality of heuristics

The goal of the study of ecological rationality is to determine the match between heuristics and
environment, that is, the structure of environments that a given class of heuristics can exploit
(Todd, Gigerenzer, & the ABC Research Group, 2012; Gigerenzer & Selten, 2001). As a first
approximation, “a heuristic is ecologically rational to the degree it is adapted to the structure of
the environment” (Gigerenzer & Todd, 1999, p. 13). In his Nobel lecture, Smith (2003) used
this definition and generalized it from heuristics to markets. I will illustrate the study of eco-
logical rationality with the prediction of customer behavior.
A large apparel retailer sends special offers and targeted information to customers. The
retailer’s goal is to target offers at “active” customers who are likely to make purchases in
the future. The overall aim is to remove inactive, unprofitable customers from the customer
base; identify profitable, inactive customers who should be reactivated; and determine active
customers who should be targeted with regular marketing activities. But how can the marketing
department predict which of their previous customers will make a purchase in the future? In the
tradition of optimizing models, stochastic customer base models such as the Pareto/NBD model
(NBD stands for “negative binomial distribution”) have been proposed as models for repeat
purchase behavior. These models try to fine-tune predictions by making complex assumptions
about individual behavior and the heterogeneity across customers, assuming that customers
buy at a steady albeit stochastic rate until they eventually become inactive (see Wübben & von
Wangenheim, 2008). When experienced managers from a large apparel retailer were studied,
however, it was found that they did not use these complex models. Instead, the managers relied
on a simple rule:

Hiatus heuristic: If a customer has not made a purchase for nine months or longer, classify them
as inactive, otherwise as active.

The hiatus can vary, depending on the kind of business, but the decision is based on a
single reason, the recency of purchase. All other cues, such as the frequency of purchase, are
ignored. The hiatus heuristic is an instance of the class of one-reason heuristics (Gigerenzer &
Gaissmaier, 2011).
Why would experienced managers rely on only a single reason? The traditional heuristics-
and-biases view might conclude that relying on one reason is naïve and may be due to man-
agers’ limited cognitive capacities. From this viewpoint, managers should better use stochastic

63
64

Gerd Gigerenzer

customer base models such as the Pareto/NBD model. But note that the managers are in a
situation of uncertainty, not in a small world of risk. The methodology outlined above—
algorithmic models of heuristics, competitive testing, and tests of predictions—explains how
to perform a competitive test between the complex Pareto/NBD model and the simple hiatus
heuristic. The result of such a test showed that the Pareto/NBD model did predict future
purchases for 75 percent of the customers correctly, but the hiatus heuristic did so for 83 per-
cent (Wübben & von Wangenheim, 2008). Is this apparel company an exception? Subsequent
tests showed otherwise: In 60 further tests, the hiatus heuristic was, on average, better at
predicting behavior than were several stochastic customer base models. Or are these sto-
chastic customer models substandard? In the same 60 tests, the hiatus heuristic on average
also better predicted behavior in comparison with logistic regression and with random forests,
one of the most powerful machine learning techniques (Artinger, Kozodi, Wangenheim, &
Gigerenzer, 2018). Under uncertainty, a single powerful reason can lead to more accurate,
more transparent, and less costly predictions than those made by highly complex machine
learning algorithms.
This existence proof of the predictive power of the heuristic does not yet tell us when the
heuristic will succeed or not. After all, it predicted better on average, but not in every individual
case. What environmental structures does the hiatus heuristic exploit when making accurate
predictions with little effort? The study of the ecological rationality of heuristics provides
answers to such questions (Gigerenzer et al., 2011). To illustrate, I briefly describe a condition
for the ecological rationality of the hiatus heuristic, which also holds for similar one-reason
heuristics.
Instead of the hiatus heuristic, consider now a linear strategy that uses and combines more
cues. It uses n binary cues x1, …, xn, with values of either +1 or -1, where the positive value
indicates future purchases. The weights of the cues are w1, …, wn, all of which are positive:

y = w1x1 + w2x2 + w3x3 + … + wnxn

The linear rule makes the inference “active” if y > 0, otherwise “inactive.” Linear algorithms
are standard approaches to prediction, which I use as reference point for a comparative analysis
with one-reason heuristics. Let us give the single cue used by the heuristic the number 1, and
the remaining cues the numbers 2, …., n. Like the beta weights in a logistic regression, the
weights of each of the remaining cues reflect their additional contribution to the higher ranked
cues. If the following dominant cue condition holds, one can show that the linear rule cannot
lead to more accurate inferences than a one-reason heuristic:

Dominant Cue Condition. The weights w1, w2, w3, … wn form a dominant cue struc-
ture if they satisfy the inequality constraint:

n
w1 > ∑ w i
i=2

In words, the condition is that the weight of the first cue is larger than the sum of the weights
of all other cues. The weights 1, ½, ¼, and ⅛ are an example. If this condition holds, the linear
rule will always make the same inference as the hiatus heuristic. For instance, if the dominant
cue has a positive value (such as “made a purchase in the last nine months”), the hiatus heuristic
leads to the inference that the customer is “active,” as does the linear algorithm, whatever the
values on the remaining cues are—even if all are negative. The reason is that a dominant cue

64
65

What is bounded rationality?

cannot be overturned or compensated by the sum of all lower-ranking cues. In this situation,
despite more effort, the linear strategy cannot be more accurate than the heuristic. If instead the
weights were, say, 1, ¾, ½, and ¼, the dominant cue condition would not hold. For a more sys-
tematic treatment of the ecological rationality of heuristics, see Gigerenzer (2016), Martignon
and Hoffrage (2002), Martignon, Katsikopoulos, and Woike (2008) and Şimşek (2013).
This analysis makes clear that the rationality of a heuristic cannot be evaluated by looking
at the heuristic and concluding that it is overly simple, but only by looking at its match with
the environment. Dominant cues and even noncompensatory cues (a stronger condition where
dominance must hold for every cue relative to the lower-ranked ones, not only for the top cue)
appear to be the rule rather than the exception in many real-world problems (Şimşek, 2013).
The dominant cue condition explains when it is rational to rely on a single cue and ignore the
rest. At a more general level, which I cannot cover here, the bias–variance dilemma explains in
what situations simple heuristics can predict better than more complex strategies, and vice versa
(Gigerenzer & Brighton, 2009).

Ecological rationality of beliefs

The study of ecological rationality is equally relevant when evaluating the rationality of beliefs.
Beliefs have often been evaluated using only logic or probability theory as the gold standard,
assuming situations of risk instead of uncertainty. As a result, beliefs that are correct under
uncertainty have been mistaken as systematic biases and cognitive illusions (Gigerenzer, 2015;
2018; Gigerenzer, Fiedler, & Olsson, 2012). Consider intuitions about chance, where one trad-
ition of experimental studies concludes that people in general have good intuitions (e.g., Kareev,
1992; Piaget & Inhelder, 1975) and where researchers in the heuristics-and-bias program have
diagnosed systematic misconceptions. Consider a classical example.

A fair coin is thrown four times. Which of the two strings of heads (H) and tails (T) is
more likely to be encountered?
HHH
HHT

Most people think that HHT is more likely. This belief has been deemed a fallacy, based on the
argument that each of the two strings has the same probability of occurring. The alleged fal-
lacy was attributed to people’s illusory belief in the “law of small numbers,” that is, that people
expect the equal probability of H and T to hold in small samples as well: HHT is more “repre-
sentative” than HHH (Kahneman & Tversky, 1972).
Surprisingly, however, this popular belief is actually correct if a person observes a sequence
of n coin tosses that is longer than the length k of the string (here, k = 3). For instance, for n = 4
four tosses, there are 16 possible sequences, each equally likely. Yet 4 of those contain at least
one HHT, and only 3 an HHH (see Figure 2.1; Hahn & Warren, 2009). Similarly, the expected
waiting time for encountering HHT is 8 tosses of a coin, compared with 14 tosses for HHH.
Now we can specify the general condition under which people’s belief is ecologically rational:

If k < n, a string of Hs with a single alternation such as HHT is more likely to be

encountered than a pure string such as HHH.

The belief is only a fallacy in the special case of k = n, or with an infinite sample, which corres-
ponds to the population probability (Hahn & Warren, 2009). Moreover, the ecological analysis

65
66

Gerd Gigerenzer

Figure 2.1 A fair coin is flipped four times, and the result is recorded. H = heads, T = tails. Which
string of flips is more likely: at least one HHH or HHT? Most people believe that HHT is more likely,
whereas some researchers say that both strings are equally likely and that people’s intuition are therefore
biased. In fact, to encounter HHT is more likely than HHH. There are 16 possible outcomes, each
equally likely; four of these have strings of HHT (“cross”), but only three strings of HHH (“check
mark”). The general principle is: If the number n of tosses (here n = 4) is higher than the size k of the
string (k = 3), then pure strings of heads (or tails) are less likely than those with an alteration. In this
situation, people’s intuition is correct (Hahn & Warren, 2009).

clarifies that the belief that HHT is more likely than HHH is not a case of the gambler’s fallacy,
as often assumed. The gambler’s fallacy refers to the intuition that after witnessing a string of,
say, two heads, one expects that the next outcome will be more likely tails than heads. This
would be a true fallacy because it corresponds to the condition k = n. In other words, a total of
three throws is considered, either HHH or HHT, and there is no sample k with the property
k < n (Gigerenzer, 2018).
The general point is that the statistics of a sample are not always the same as the probabilities
of the population. By the same kind of ecological argument, it could be shown that the belief in
the hot hand in basketball is not a fallacy, as claimed by Gilovich, Vallone, and Tversky (1985).
Rather, a reanalysis of the original data showed that the belief is correct but that the researchers
overlooked the difference between sample statistics and population probabilities and used the
wrong gold standard (Miller & Sunjuro, 2016). Labeling a belief a bias even if no bias exists is
an example of a broader phenomenon that I call the “bias bias”: the tendency to see systematic
biases in behavior even when there is no verifiable error at all (Gigerenzer, 2018).
The study of ecological rationality is a general approach for evaluating the rationality of
heuristics and beliefs. It differs from logical rationality in that it does not compare behavior
with some principle of logic or probability theory that is taken to be universally true, assuming
that the situation is always one of risk. Rather, heuristics and beliefs should be evaluated against
the structure of the environment, where small samples may systematically differ from a world
of risk.

Guidelines for the study of decision making under uncertainty

Let me end with a summary in the form of three theoretical guidelines for studying bounded
rationality:

1. Take uncertainty seriously. Theories of human behavior should take the distinction between
risk and uncertainty seriously. What is rational under risk is not necessarily rational under
uncertainty, and what is a biased belief under risk is not necessarily biased under uncer-
tainty. In situations of risk, fine-tuned behavior is likely to be adaptive, such as continuous
Bayesian probability updating. In situations of uncertainty, by contrast, simple heuristics are
likely to be adaptive.

66
67

What is bounded rationality?

2. Take heuristics seriously. Under uncertainty, heuristics are not the problem but the solution.
We need more studies of the adaptive toolbox of individuals and institutions and of their
development over time.
3. Take ecological rationality seriously. Study the ecological rationality of heuristics, beliefs, and
other behavior instead of their logical rationality alone.
These three insights flow from Simon’s original program of bounded rationality. They
open up the study of decision making to the real world in place of small-world problems
where we always know the optimal solution. They also make clear that people’s deviations
from rationality in small worlds do not imply irrationality in the large world we live in.
Under uncertainty, the bounds of rationality are both inside the mind and outside in the
world. The challenge is to find out how these bounds work together. As the three insights
also show, the models we build to study decision making should assume that agents are nei-
ther omniscient nor irrational. Bounded rationality is the study of rationality for mortals.

References
Akerlof, G. A., & Shiller, R. J. (2009). Animal spirits. Princeton, NJ: Princeton University Press.
Arrow, K. J. (2004). Is bounded rationality unbounded rational? Some ruminations. In M. Augier &
J. G. March (Eds.), Models of a man: Essays in memory of Herbert A. Simon (pp. 47–56). Cambridge,
MA: MIT Press.
Artinger, F. M., & Gigerenzer, G. (2016). The cheap twin: From the ecological rationality of heuristic
pricing to the aggregate market. In J. Humphreys (Ed.), Proceedings of the Seventy-sixth Annual Meeting
of the Academy of Management. Online ISSN: 2151–6561. doi:10.5465/ambpp.2016.206.
Artinger, F. M., Kozodi, N., Wangenheim, F., & Gigerenzer, G. (2018). Recency: Prediction with
smart data. In J. Goldenberg, J. Laran, & A. Stephen (Eds.), American Marketing Association Winter
Conference Proceedings: Vol. 29. Integrating paradigms in a world where marketing is everywhere (L2–L6).
Chicago: American Marketing Association.
Berg, N. (2014), Success from satisficing and imitation: Entrepreneurs’ location choice and implications of
heuristics for local economic development, Journal of Business Research, 67, 1700–1709. doi:10.1016/
j.jbusres.2014.02.016.
Friedman, D., Isaac, R. M., James, D., & Sunder, S. (2014). Risky curves. On the empirical failure of expected
utility. New York: Routledge.
Friedman, M. (1953). Essays in positive economics. Chicago: University of Chicago Press.
Gigerenzer, G. (2004). Striking a blow for sanity in theories of rationality. In M. Augier & J. G. March
(Eds.), Models of a man: Essays in memory of Herbert A. Simon (pp. 389–409). Cambridge, MA: MIT Press.
Gigerenzer, G. (2015). On the supposed evidence for libertarian paternalism. Review of Philosophy and
Psychology, 6, 361–383.
Gigerenzer, G. (2016). Towards a rational theory of heuristics. In R. Frantz & L. Marsh (Eds.), Minds, models,
and milieux: Commemorating the centennial of the birth of Herbert Simon (pp. 34–59). New York: Palgrave
Macmillan.
Gigerenzer, G. (2018). The bias bias in behavioral economics. Review of Behavioral Economics, 5, 303–336.
doi:10.1561/105.00000092.
Gigerenzer, G., & Brighton, H. (2009). Homo heuristicus: Why biased minds make better inferences.
Topics in Cognitive Science, 1, 107–143. doi:10.1111/j.1756-8765.2008.01006.x.
Gigerenzer, G., Fiedler, K., & Olsson, H. (2012). Rethinking cognitive biases as environmental
consequences. In P. M. Todd, G. Gigerenzer, & the ABC Research Group, (Eds.), Ecological ration-
ality: Intelligence in the world (pp. 80–110). New York: Oxford University Press.
Gigerenzer, G., & Gaissmaier, W. (2011). Heuristic decision making. Annual Review of Psychology, 62,
451–482. doi:10.1146/annurev-psych-120709-145346.
Gigerenzer, G., Hertwig, R., & Pachur, T. (Eds.) (2011). Heuristics: The foundations of adaptive behavior.
New York: Oxford University Press.
Gigerenzer, G., & Selten, R. (Eds.). (2001). Bounded rationality: The adaptive toolbox. Cambridge,
MA: MIT Press.

67
68

Gerd Gigerenzer

Gigerenzer, G., & Todd, P. M. (1999). Fast and frugal heuristics: The adaptive toolbox. In G. Gigerenzer,
P. M. Todd, & the ABC Research Group (Eds.), Simple heuristics that make us smart (pp. 3–34).
New York: Oxford University Press.
Gilovich, T., Vallone, R. & Tversky. A. (1985). The hot hand in basketball: On the misperception of
random sequences. Cognitive Psychology, 17, 295–314. doi:10.1016/0010-0285(85)90010-6.
Hahn, U., & Warren, P. A. (2009). Perceptions of randomness: Why three heads are better than four.
Psychological Review, 116(2), 454–461. doi:10.1037/a0015241.
Kahneman, D. (2003). Maps of bounded rationality: Psychology for behavioral economics. American
Economic Review, 93(5), 1449–1475.
Kahneman, D., Slovic, P., & Tversky, A. (1982). Judgment under uncertainty: Heuristics and biases.
Cambridge: Cambridge University Press.
Kahneman, D., & Tversky, A. (1972). Subjective probability: A judgment of representativeness. Cognitive
Psychology, 3, 430–454.
Kareev, Y. (1992). Not that bad after all: Generation of random sequences. Journal of Experimental
Psychology: Human Perception and Performance, 18, 1189–1194. doi:10.1037/0096-1523.18.4.1189.
Keynes, J. M. (1936). The general theory of employment, interest and money. London: Macmillan.
Knight, F. (1921). Risk, uncertainty and profit. Boston, MA: Houghton Mifflin Co.
Kruglanski, A. W., & Gigerenzer, G. (2011). Intuitive and deliberative judgments are based on common
principles. Psychological Review, 118, 97–109.
Lewin, K. A. (1935). A dynamic theory of personality. New York: McGraw-Hill.
Lopes, L. L. (1992). Three misleading assumptions in the customary rhetoric of the bias literature. Theory
& Psychology, 2, 231–236. doi:10.1177/0959354392022010.
Luce, R. D., & Raiffa, H. (1957). Games and decisions. New York: Dover.
Martignon, L., & Hoffrage, U. (2002). Fast, frugal, and fit: Lexicographic heuristics for paired com-
parison. Theory and Decision, 52, 29–71. doi:10.1023/A:1015516217425.
Martignon, L., Katsikopoulos, K. V., & Woike, J. K. (2008). Categorization with limited resources: A
family of simply heuristics. Journal of Mathematical Psychology, 52(6), 352–361. doi:10.1016/
j.jmp.2008.04.003.
Miller, J. B., & Sanjuro, A. (2016). Surprised by the gambler’s and hot hand fallacies? A truth in the law of
small numbers. IGIER Working Paper No. 552. doi:10.2139/ssrn.2627354.
Oxford English Dictionary (n.d.). Heuristic. In OED Online. Available at: www.oed.com/view/Entry/
86554?isAdvanced=false&result=1&rskey=C4NVLA& (accessed July 8, 2020).
Piaget, J., & Inhelder, B. (1975). The origin of the idea of chance in children. New York: Norton.
Rubinstein, A. (1998). Modeling bounded rationality. Cambridge, MA: MIT Press.
Sargent, T. J. (1993). Bounded rationality in macroeconomics. Oxford: Clarendon Press.
Savage, L. J. (1954). The foundations of statistics. New York: Wiley.
Sedlmeier, P., Hertwig, R., & Gigerenzer, G. (1998). Are judgments of the positional frequencies of
letters systematically biased due to availability? Journal of Experimental Psychology: Learning, Memory, and
Cognition, 24, 754–770.
Simon, H. A. (1955). A behavioral model of rational choice. Quarterly Journal of Economics, 69, 99–118.
doi:10.2307/1884852.
Simon, H. A. (1956). Rational choice and the structure of the environment. Psychological Review, 63,
129–38. doi:10.1037/h0042769.
Simon, H. A. (1978). Rationality as process and as product of thought. American Economic Review, 68, 1–16.
Simon, H. A. (1985). Human nature in politics: The dialogue of psychology and political science. American
Political Science Review, 79, 293–304.
Simon, H. A. (1988). Nobel laureate Simon “looks back”: A low-frequency mode. Public Administration
Quarterly, 12, 275–300.
Simon, H. A. (1989). The scientist as problem solver. In D. Klahr & K. Kotovsky (Eds.), Complex informa-
tion processing: The impact of Herbert A. Simon (pp. 375–398). Hillsdale, NJ: Erlbaum.
Simon, H. A. (1990). Invariants of human behavior. Annual Review of Psychology, 41, 1–19. doi:10.1146/
annurev.ps.41.020190.000245
Simon, H. A. (1997). Models of bounded rationality, Vol. 3: Empirically grounded economic reason. Cambridge,
MA: MIT Press.
Şimşek, Ö. (2013). Linear decision rule as aspiration for simple decision heuristics. Advances in Neural
Information Processing Systems, 26, 2904–2912.

68
69

What is bounded rationality?

Smith, V. L. (2003). Constructivist and ecological rationality in economics. The American Economic Review,
93(3), 465–508.
Stigler, G. J. (1961). The economics of information. Journal of Political Economy, 69, 213–225. doi:10.1086/
258464.
Thaler, R. H., & Sunstein, C. R. (2008). Nudge: Improving decisions about health, wealth, and happiness. New
Haven, CT: Yale University Press.
Todd, P. M., Gigerenzer, G., & the ABC Research Group (Eds.) (2012). Ecological rationality: Intelligence in
the world. New York: Oxford University Press.
Tversky, A. (1972). Elimination by aspects: A theory of choice. Psychological Review, 79, 281–299.
doi:10.1037/h0032955.
Tversky, A., & Bar-Hillel, M. (1983). Risk: The long and the short. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 9(4), 713–717. doi:10.1037/0278-7393.9.4.713.
Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability.
Cognitive Psychology, 5, 207–232.
Tversky, A., & Kahneman, D. (1974). Judgement under uncertainty: Heuristics and biases. Science, 185,
1124–1131. doi:10.1126/science.185.4157.1124.
Wübben M, & von Wangenheim F. (2008). Instant customer base analysis: managerial heuristics often “get
it right.” Journal of Marketing, 72, 82–93. doi:10.1509/jmkg.72.3.82.

69
70
71

PART I

Naturalizing bounded rationality

72
73

3
TOWARDS A CRITICAL
NATURALISM ABOUT
BOUNDED RATIONALITY
Thomas Sturm

Introduction
Rationality (or reason1) has often been declared to be a central topic of philosophy: as the
source of principles either for metaphysics and ethics, or for epistemology and philosophy
of science. Some authors even identify philosophy with the theory of rationality (Habermas,
1981, p. 16; Putnam, 1981, p. 113; Nozick, 1993, p. xi; Nida-Rümelin, 1996, p. 73; Grice,
2001, p. 4). However, most philosophers nowadays understand that it is not only philosophers
who have developed our understanding of rationality. Beginning in the Enlightenment, and to
an ever greater extent over the last century, sciences such as mathematics, statistics, economics
or psychology have (partly) conquered this territory (Gigerenzer et al., 1989; Erickson et al.,
2013; Sturm, forthcoming). The contributions on bounded rationality from Herbert Simon
through to Gerd Gigerenzer, and their collaborators and followers, belong to this trend too.
The idea that theories of rationality obtain support from science should be grist to the mill
of the philosophical doctrine of naturalism. While there are many versions of naturalism, they all
share the view that the relation between philosophy and the sciences is (at least) one of close allies,
and that the sciences contribute through their theories and methods to the solution of philo-
sophical problems. Anti-naturalists, in turn, defend a separation of philosophy and the sciences,
insisting that philosophy has sui generis methods for dealing with its own distinctive problems.
Both these opposing positions have been applied to rationality (Sober, 1981; Putnam, 1982;
Stich, 1985, 1990, 1993; Hauptli, 1995; Pacho, 1995; Stein, 1996; Chiappe & Verwaeke, 1997;
Bermúdez & Millar, 2002). Meanwhile, very few philosophers have drawn on Simon’s (1956,
1957) work on bounded rationality (Giere, 1985, 1988; Cherniak, 1986; Gigerenzer & Sturm,
2012; deLanghe, 2013; Hahn, 2013, Chapter 8).2 This is surprising, since bounded rationality (1) is
an empirically grounded account of reasoning; and (2) does not merely have descriptive or explana-
tory aspirations, but normative or prescriptive ones as well. Thus, this notion of bounded rationality
offers a path towards a naturalistic account of rationality—one that addresses the important issue of
the normativity of rationality, which is often seen as a major challenge to any such naturalization.
No comprehensive study of the philosophical aspects of bounded rationality exists. Such a
study would, at least, have to (1) explicate and scrutinize the presuppositions of the concept

73
74

Thomas Sturm

(such as its descriptive as well as normative aspirations, or its difference to non-bounded

conceptions of rationality); (2) determine areas in philosophy where models of bounded ration-
ality might be used, and assess such uses where they exist; and (3) address the fundamental
debates over its theoretical and methodological adequacy in psychology, economics, sociology,
and political science, as well as the potential and limits of its interdisciplinary uses. While such
a study cannot be carried out here, I shall include selected aspects of (1) and (2). I first out-
line three systems of reasoning that are constitutive of the “standard picture” of rationality, and
highlighting two major criticisms of that picture, both associated with bounded rationality.
Then, I introduce some major assumptions of naturalism and the challenges they face. Finally,
I discuss the prospects and limits of bounded rationality for what I call “critical” naturalism in
epistemology, with occasional considerations concerning the philosophy of science.

The “standard picture”: three normative systems of rationality

We try to reason well, whether to convince others or ourselves of a point of view, to discover
new knowledge that allows us to defend our pet beliefs, or to make good decisions. Our
reasoning certainly can be either good or bad. However, this can only be the case if there are
standards or norms of reasoning that have been agreed upon. What are they? According to
Stein’s (1996) useful term, the “standard picture” of rationality consists of a set of three norma-
tive systems of judgment and decision making that are dominant nowadays in numerous areas,
and often opposed by defenders of bounded rationality.
One of those systems is derived from the revolution in logic that started in the nineteenth
century with the work of the philosopher-mathematician Gottlob Frege (1848–1925). If we
consider this, it will quickly lead us to the dominant view concerning the relation between
reasoning and normativity. Frege’s achievements influenced subsequent work in logic, from that
of Bertrand Russell and Alfred N. Whitehead through to that of Alfred Tarski and Alan Turing.
Thus, Frege’s work even paved the way for modern computers—and for Simon and Newell’s
“Logic Theorist” program (Newell et al., 1958). Frege’s main project was to show that a part of
mathematics, namely, arithmetic, could be reduced to logical laws. This is the doctrine of logi-
cism. Since the existing logics, as Frege found them, were insufficient for such a reduction, he
radically overhauled them by creating an axiomatized predicate calculus with innovative tools
for complex quantification. While his logicism was—as he himself later admitted—unsuccessful,
the central achievements of his new logic are still alive and well today.
Frege was aware of views that drew no sharp distinction between logic and psychology.
He rejected them by claiming that the validity of logical laws is independent of how people
actually think and reason (Frege, 1966). This is his anti-psychologism, which has remained a
forceful reason for anti-naturalism until today. Frege’s main basis for this stance was that actual
reasoning is subject to mistakes and fallacies, and so cannot ensure what logic guarantees.
Logical rules determine whether reasoning from premises to conclusions is truth-preserving.
Whether the premises of an argument are true or not cannot be decided by logic alone; but
logic tells us which patterns of inference guarantee that if the premises are true, then the con-
clusion will also be true. The validity of logical laws has to do, among other things, with the
meaning of logical connectives such as ‘→’ (interpreted as a specific version of the ordinary
“if-then” conditional) and ‘¬’ (negation). Thus, modus ponens (1. p→q; 2. p; 3. Therefore, q.)
or modus tollens (1. p→q; 2. ¬ q; 3. Therefore, ¬ p) are deductively valid under any interpret-
ation of the propositional variables p and q. Of course, humans sometimes violate such rules.
For instance, there is the fallacy of “denying the antecedent”: 1. p→q; 2. ¬ p; 3. Therefore,
¬ q. Consider for example:

74
75

Towards a critical naturalism about BR

If you are a banker, then you have a regular income. You are not a banker. Therefore,
you do not have a regular income.

Clearly, you do not have to sell your soul to monetary businesses to have a regular income. In
other inferences, the mistake is less easy to detect. Alan Turing considered the following argu-
ment: “If each man had a definite set of rules of conduct by which he regulated his life he
would be no better than a machine. But there are no such rules, so men cannot be machines”
(Turing, 1950, p. 452). Again, this is an example of the same fallacy. But maybe what those who
propose such an argument really mean is:

Only if each man had a definite set of rules of conduct by which he regulated his life
he would be no better than a machine. But there are no such rules, so men cannot
be machines.

In this case, the inference would not embody “denying the antecedent”. But surely people can
mix up p’s and q’s, neglect logical meanings, and thus end up committing a fallacy.
While logicism concerns the foundations of arithmetic, anti-psychologism concerns the
basis of logic. Even psychologists such as Peter Wason (1966) and many others have accepted
anti-psychologism by testing human reasoning against logical laws taken as unquestioned
normative yardsticks. Neither did Simon and Newell question the validity and applicability
of formal logic—the Logic Theorist was constructed to prove numerous theorems in Russell
and Whitehead’s Principia Mathematica (the so-called artificial intelligence thesis), and Simon
and Newell also viewed the program as a model of human reasoning (the information-
processing thesis): the Logic Theorist “provides an explanation for the processes used by
humans to solve problems in symbolic logic” (Newell et al., 1958, p. 163). Thus, the logical
model was projected into the mind, not the other way around (Gigerenzer & Sturm, 2007,
pp. 325–327).
Anti-psychologism is often described as being based on a rejection of the Is–Ought fallacy
(noted in Hume, emphasized by Kant, used and discussed further in later debates on theoret-
ical and practical rationality; see Sturm, forthcoming): one cannot infer normative judgments
from descriptive ones. For instance, from the descriptive claim that people often engage in tax
evasion, it does not follow that one should engage in tax fraud. From the claim that people
often do not blame others for engaging in tax fraud, it does not follow that one should not
be blamed for engaging in tax fraud. Frege’s anti-psychologism, however, is not derived from
rejecting the Is–Ought fallacy. Contrary to what is sometimes asserted (e.g., Notturno, 1985),
Frege did not view the laws of logic as inherently normative, but as descriptive. However,
descriptive of what? He claimed that logical rules or patterns of inference are universally and
timelessly valid. His objection to psychologism, then, addressed a specific version of this doc-
trine: attempts to explain logical “laws of thought” as laws of actual thinking, with the latter
being understood as a chain of “subjective” mental representations. Against this version of
psychologism, Frege argued that such representations are too variable to constitute a possible
basis of logical laws, given their universality and timeless validity. That is why, in order to jus-
tify such laws, Frege (1966, 43) required that we accept a “third realm” (drittes Reich) beyond
the material world and our subjective mental representations. This argument is disputed, but
it cannot be discussed here (cf. e.g., Sober, 1978; Carl, 1994, Chapter 2).3 What matters here
is a straightforward point: since we can make mistakes in our reasoning, we ought to correct
them, and that may involve logical laws. To deny this force of logic by claiming that our actual
reasoning sometimes works in different ways is to fall prey to various kinds of psychologism.

75
76

Thomas Sturm

Still, we will see that under certain conditions, the charge of psychologism can be avoided, and
assumptions of bounded rationality play a role in this.
Anti-psychologism became popular beyond logic, especially through the epistemology and
philosophy of science of the Vienna Circle and Popper’s Falsificationism (Peckhaus, 2006).
After all, not all reasoning, whether ordinary or scientific, is deductive. Many non-deductive
inferences, for instance, about scientific hypotheses, use probabilities. This leads to a second
normative system that is important for the standard picture: theories of probability as developed
by, e.g., Rudolf Carnap or Hans Reichenbach, alongside with mathematicians and statisticians.4
Their theories tried to do for “inductive” logic what Frege had achieved for deductive logic: an
axiomatic system with clear tasks, limits, and structure, only one that normatively guides prob-
able reasoning. While the mathematics of probability was axiomatized by the mathematician
Andrej Kolmogorov (1903–1987), the meaning of probability remains disputed, and thereby
so do its proper applications. There are, for instance, logical theories aimed at determining
degrees of confirmation above 0 for scientific hypotheses, relative to given empirical evidence
(Carnap); the so-called frequentism, which views probabilities as the limiting frequency of
outcomes in long-run or potentially infinite series of similar events (Reichenbach); or the pro-
pensity theory, which takes probabilities to be properties that are inherent to sets of repeatable
conditions (Karl Popper). However, none of these theories is without its problems; and none
of them can claim to represent “the” meaning of ‘probability’: a term which is ambiguous and
can serve different functions (Gigerenzer et al., 1989; Gillies, 2000). A fundamental pluralism
with regard to probability is widespread today. What is more, while Carnap and others assumed
that Fregean anti-psychologism concerning logic could be expanded into epistemology or phil-
osophy of science, today it is far from clear that such a move comes without costs (Sober, 1978;
Peckhaus, 2006). It would require all philosophical claims about knowledge or science to be
formal or logical truths, which is highly implausible.
Third, and finally, since the mid-twentieth century, theories of rational choice have emerged,
beginning with John von Neumann and Oscar Morgenstern’s landmark Theory of Games and
Economic Behavior (1944), which was itself influenced by the rise of logic. Moreover, rational
choice owes much to economic theories of utility maximization that emerged in previous cen-
turies. It seems plausible to say that if you are a rational agent, you ought to maximize your
(expected) advantage, given what you believe the probability of certain outcomes to be.
So, these normative systems coalesced into the standard picture of rationality: if you are
perfectly rational, you have to reason in line with logic, probability, and decision theory. Even
today, these formal and optimizing theories are viewed even by many as (1) a descriptive model
for human reasoning and, perhaps more importantly, (2) the normative yardsticks for such
reasoning.5 However, questions concerning the normative validity of the standard picture
arise: What justifies the rules of each of these theories? And why should we reason in line with
them in particular situations?

Two objections to the standard picture

I now consider two influential objections to the standard picture, both related to bounded
rationality. As the term “bounded rationality” is often meant to include all empirical theories
that try to account for informal and non-optimal reasoning using both lazy and hard constraints
(Oaksford & Chater, 1991), the objections differ considerably. This, in turn, leads to different
connections that one can draw between bounded rationality and naturalism.
Objection 1: Insofar as the formal theories were interpreted descriptively—not in the Fregean
sense of describing a realm of logical entities and their formal relations, but as being descriptive

76
77

Towards a critical naturalism about BR

of human thinking and decision-making—one could find empirical counterexamples to them.

This led to the demand to build an alternative account of judgment and decision-making.
Thus, Wason’s (1966) four-card test aimed to refute Piaget’s theory of cognitive develop-
ment, according to which we master propositional calculus from age 12 onwards. Similarly,
the program of “heuristics and biases” (henceforth, HB), developed by Daniel Kahneman and
Amos Tversky, attempted to show that the view of “man as an intuitive statistician” (Peterson &
Beach, 1967) whose behavior approximates the canons of probability was mistaken. Reasoning
fallacies and mistakes that the HB program detected include conjunction errors (the “Linda
problem”), neglecting the obligation to use sufficiently large statistical samples (“law of small
numbers”), the base rate fallacy, and so on (Kahneman, Slovic, & Tversky, 1982; Tversky &
Kahneman, 1983).
Kahneman has claimed that Simon’s work is in line with the HB approach:

[Tversky and I] explored the psychology of intuitive beliefs and choices and examined
their bounded rationality. Herbert A. Simon … had proposed much earlier that deci-
sion makers should be viewed as boundedly rational, and had offered a model in
which utility maximization was replaced by satisficing. Our research attempted to
obtain a map of bounded rationality, by exploring the systematic biases that separate
the beliefs that people have and the choices they make from the optimal beliefs and
choices assumed in rational-agent models.
Kahneman, 2003, 1449; see also Kahneman, Slovic,
& Tversky, 1982, pp. xi–xii

Kahneman’s view that humans are often irrational results from adopting rules of the standard
picture as yardsticks. Our intuitive judgments and decisions then have to be explained in
terms of so-called heuristics (e.g., “representativeness” or “availability”). Understood this way,
“bounded rationality” implies that standard “rational models are psychologically unrealistic”,
and “maps of bounded rationality” are supposed to account for deviations from the “rational
agent model” (Kahneman, 2003, p. 1449).
Objection 2: Formal and optimizing rules set substantive norms for ideal reasoners, i.e.
reasoners equipped with infinite time, memory, and other resources. But, Simon said, we must
distinguish between substantive and “procedural” rationality (a distinction that paved the way
towards bounded rationality; see Erickson et al., 2013, Chapter 2). It is one thing to judge
whether an outcome is reasonable in the light of a standard norm; it is quite another to judge
whether an outcome is reasonable in the light of whether the process that leads to it is overly
costly.
Importantly, there are stubborn problems of computational intractability. Cognitive processes
which require resources that increase at an exponential rate (i.e., 2n, or more) are regarded as
creating such problems. Consider the traveling salesman problem, i.e. finding the optimal route
between a number of locations when one travels to each location only once and the whole tour
is as short as possible. Calculating the optimal route between e.g., 50 cities requires calculating
no less than 300,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,00
0,000,000 possible routes. To date, no human, and no computer can do that. There is a similar
situation for tasks such as finding the optimal path for winning in chess, or making investment
choices, and other daily and scientific problems (Michaelewicz & Fogel, 2000). Hard compu-
tational intractability may make it impossible to use the norms of the standard picture. Even if
the formalistic and optimizing approach to rationality produced rules that were valid within the
axiomatic systems to which they belong, we may have to give up the close connection between

77
78

Thomas Sturm

the formal validity and the rationality of patterns of inference. There is, then, a gap between
formal (including optimizing) rules and rationality.
Simon and others have seen this gap as a major reason for developing models of bounded
rationality, in order to figure out how people actually do reason, but also how they can
reason well:

Bounded rationality dispenses with the notion of optimization and, usually, with
probabilities and utilities as well. It provides an alternative to current norms, not an
account that accepts current norms and studies when humans deviate from these
norms. Bounded rationality means rethinking the norms as well as studying the actual
behavior of minds and institutions.
Gigerenzer & Selten, 2001, p. 6

One of Simon’s most influential ideas was to replace the goal of maximizing expected utility
by “satisficing”. In Gigerenzer’s view, reasoning (often) uses fast and frugal heuristics (hence-
forth, FFH) that also work with little information and computation, such as the “recognition
heuristic”, the “fluency heuristic”, or “take the best” (for an overview of ten well-studied
heuristics, see Gigerenzer & Sturm, 2012, p. 249f.). That FFH do not merely describe behavior
but should, in a number of cases, also guide our judgments and decisions is justified if and
insofar as reasoning processes express a certain fitness function between mind and environment
(Gigerenzer, Chapter 2 in this volume). That function needs to be discovered empirically.
Understood this way, bounded rationality questions the sharp divide between the descrip-
tive and the normative—quite different from Kahneman’s understanding of the concept.
However, no Is-Ought fallacy thereby needs to be implied. Instead, underlying this blurring of
the boundary is a legitimate principle of normativity: ought implies can (also already recognized
by Kant). The intractability of many reasoning tasks restricts the applicability of certain formal
and optimizing norms, and thus makes it reasonable to look for different, more feasible norms.
The difference between these two objections leads to two disputes. The first, purely empir-
ical objection has prompted discussion on the extent to which humans are rational, asked in
the light of norms of the standard picture of rationality. The results range from “completely
irrational” to “sometimes irrational”. In contrast, the second objection has raised the question
of which norms should guide good reasoning in the face of widespread intractability and
uncertainty: those of the standard picture, or those of bounded rationality understood norma-
tively (i.e., Simon’s and Gigerenzer’s approach)? Or maybe both? As we will see next, philo-
sophical projects of naturalizing rationality disagree here too.

Naturalism: its aims, scope, assumptions, and problems

Although it is considerably older, naturalism has become popular since the 1960s (Quine, 1969;
Kitcher, 1992; Kornblith, 1994; Rosenberg, 1996).6 Naturalists object to “armchair” methods
such as conceptual analysis, thought experiments, or appeals to intuition (DePaul & Ramsey,
1998; Kornblith, 2007), claiming that science provides better resources with its methods or the-
ories, and viewing philosophy that isolates itself as sterile and obsolete. Some naturalists have
even claimed that the traditional tasks and methods of philosophy will sooner or later be replaced
by science (e.g., Quine, 1969).
Naturalism is a metaphilosophical thesis: a view about philosophy’s questions, methods, and
aims. To be convincing, naturalism has to prove itself by being specific as to its domains, as to
what claims it makes on what grounds, and as to how it defends itself against objections. Thus,

78
79

Towards a critical naturalism about BR

naturalism has been developed in metaphysics (Ladyman, 2007), philosophy of mind (Dennett,
1991; Dretske, 1996), ethics (Harman, 1977, Lenman, 2009), epistemology (Goldman 1986;
Kornblith, 1993, 2002), philosophy of science (Giere, 1985, 1988, 1992; Laudan, 1987,
1990; Kitcher, 1993; Mi & Chen, 2007; Viale, 2013), and other domains as well (De Caro
& MacArthur, 2004, 2010; Galparsoro & Cordero, 2013). Cutting across these divisions of
domains, naturalists make different general claims. Two of their basic convictions are:

(ON) Reality consists only of natural objects, their properties, and relations: there are
no unexplainable or irreducible supernatural entities or powers.
(MN) Philosophical questions can and should be answered by relying on the methods
of the empirical sciences: there is no a priori philosophical knowledge and there are no
special philosophical methods.

“ON” stands for “ontological naturalism”, and “MN” for “methodological naturalism” (De
Caro & MacArthur, 2010; Galparsoro & Cordero, 2013). Naturalists usually have additional
commitments, such as to Darwinism (Rosenberg, 1996) or physicalism (Papineau, 2015).
Without them, (ON) might be so uncontroversial that most non-naturalistic philosophers
would accept it too. We do not know what (ON) comes down to unless we have a specific
scientific theory that explains the relevant objects and their properties and relations. If one
defends (ON) about the mind, one could use theories rooted in biology or neuroscience. In the
area that concerns us here, rationality, naturalists usually look to psychology, but also to other
sciences, such as those just mentioned. Such an expansion also helps to distinguish between
psychologism and naturalism. In any case, it all depends on the quality or validity of the specific
theory the naturalist offers. Naturalists acknowledge the necessity of this task of specification,
and many have invested much effort in connecting scientific theories to philosophical problems.
(MN) is controversial too. To begin, can all philosophical questions be answered by scien-
tific methods? Or only some? If so, which ones? And how can the sciences deliver? Questions
concerning the nature of time, free will, knowledge, or the normative basis of morality cer-
tainly do not seem to be suited to empirical methods. With few exceptions (despite the recent
development of “experimental philosophy”), philosophers do not perform observations,
experiments, statistical analysis, and so on. As Nozick (1993, p. xi) has noted, what philosophers
really love is not wisdom but reasoning as such: thinking about it as well as practicing it.
As if all that were not enough, naturalists have to face straightforward anti-naturalistic
arguments, e.g., the claims that we have a priori knowledge that cannot be explained natural-
istically (Putnam, 1982), that naturalism falls prey to naturalistic fallacies (Kim, 1988), that it
leads to circularity or self-destructiveness (BonJour, 1994), triviality (Stroud, 1996), or that it
lacks clarity (Sklar, 2010). We will see that similar points come up when we look at naturalism
about rationality. In particular, a central challenge naturalists have to face in our case should
be clear: to provide a scientific account that treats the normativity of rationality as something
within the natural world, and explainable by scientific theory. This is not an easy task.

Naturalism about (bounded) rationality

Naturalists have drawn a variety of connections between rationality and the sciences. They
appeal to the HB approach in order to address problems of epistemic rationality (Goldman,
1986; Stich, 1990; Kornblith, 1993; Bishop and Trout, 2005) or decision making concerning
scientific theories (Solomon, 1992; Kitcher, 1993). They deal with claims of evolutionary
psychology to explain the nature and foundation of rationality (Stein, 1996); and occasionally

79
80

Thomas Sturm

research on satisficing and FFH has been exploited as well (Giere, 1985, 1988; Cherniak, 1986;
Gigerenzer & Sturm, 2012; Sturm, 2019; Hey, 2016.
Before we can move on, some clarification is needed. If naturalism about rationality merely
required some sciences to be used in the detection, explication, and justification of the norms of
rationality, then the standard picture might already count as naturalistic, since it involves results
of mathematicians or economists. However, that would contradict (MN). The normative
systems of the standard picture are all a priori, not the result of observation or experiment.7 One
might, of course, suggest dropping the reference to empirical science, thus replacing (MN) by:

(MN*) Philosophical questions can and should be answered by relying on the methods
and results of the sciences.

Stated differently, “naturalism” would just mean “scientism”. This move, however, trivializes
naturalism about rationality. Since the non-empirical, a priori character of logical rules has
always been a central point of conflict between psychologism and anti-psychologism, nat-
uralism and anti-naturalism, we should not take it for granted that a formal-rules account of
rationality is naturalistic enough. The same line of reasoning works, mutatis mutandis, for the-
ories of probability and decision making. The fact that Kolmogorov was a mathematician and
not a philosopher does not mean that his work provided a naturalistic account of probability
theory.
(MN*) might, furthermore, undermine (ON): If naturalists simply accept what comes from
the formal sciences, without giving a naturalistic explanation of the relevant abstract entities
and rules, then they might be asked how they deal with Frege’s talk of a “third realm” of logical
laws. (ON) must do better. An interesting naturalism about rationality should therefore build
on genuinely empirical theories. No surprise, then, that many naturalists about rationality look
to evolutionary and cognitive psychology, or to biology and neuroscience. For reasons of space,
I cannot further discuss (ON) about rationality in this chapter (cf. Sober, 1981; Chiappe &
Verwaeke, 1997).
Now, as seen above, the term “bounded rationality” has been used for both the HB program,
and the Simon-Gigerenzer program of satisficing and FFH. Let us now consider each of these
in turn.
Once the HB program had become popular in science, it was adopted by naturalists too.
Many viewed it as the scientific version of the mundane view that humans are prone to errors
and fallacies. What is more, some followed the interpretation according to which HB studies
showed that humans are deeply irrational (e.g., Stich, 1985). Unfortunately, such naturalists
had to recognize that results of the HB approach became increasingly contested. The criti-
cism not only came from philosophical armchair arguments (though these could be influen-
tial: Cohen, 1981). Psychologists objected that the data of HB studies were not stable enough
to claim that human beings completely violated the standard picture of rationality, that theor-
etical concepts such as “representativeness” or “availability” were not precise enough to permit
interesting, testable predictions, or that the normative assumptions used in the experiments
could be questioned as well (Gigerenzer, 1991, 1996; Lopes, 1991; Cosmides and Tooby,
1996). This psychological debate is called the “rationality wars” (Samuels, Stich, & Bishop,
2002; Sturm, 2012a).
Thus, even naturalists have to make hard choices: Do the empirical results really warrant
the conclusion that humans are highly irrational? What methods, concepts, and normative
assumptions should psychologists use? What do we even mean when we call judgments, or
processes, rational or irrational? Hilary Kornblith (1993, Chapter 5), another naturalist, has

80
81

Towards a critical naturalism about BR

discussed such questions for the “law of small numbers” (Kahneman, Slovic, & Tversky, 1982,
Chapter 2). He argues that the alleged irrationality of human beings is not supported by the
relevant studies, and that we can be more optimistic if we apply a background assumption from
evolutionary theory:

Just as our perceptual mechanisms are well adapted to the environment in which they
typically operate and build in presuppositions about the environment which are typ-
ically true, so our inferential mechanisms may also be built around presuppositions
about standard environments which allow us to gain information about those envir-
onments quickly and accurately.
Kornblith, 1993, p. 86

He also immediately warns his reader: “There is of course, no a priori guarantee that this is
the right perspective on human inference, and we should not try to force the data into such
a mould” (Kornblith, 1993). Thus, naturalists do what they should abhor, given (MN): they
reflect on science from a critical, normative point of view––as Kornblith says, “we should not
try to force the data into such a mould” (emphasis added). It is, of course, the existence of
conflicting approaches in cognitive psychology, or disputes over the interpretation of data that
prompt Kornblith to take sides. Still, a strict and consistent naturalist should want to explain this
critical, normative perspective itself again in purely naturalistic terms, by looking to e.g., a bio-
logical or psychological theory of such a critical, normative point of view. Kornblith does not
do that, nor do other naturalists who engage critically with the empirical literature. Thus, while
(MN*) trivializes naturalism about rationality, (MN) leads either into a regress or it implies the
acceptance of an unexplained non-naturalistic perspective.
One reasonable response here is that naturalism should give up the aim of ultimately replacing
philosophy by science, and should settle for cooperation between the two (cf. Feldman, 2001):

(CMN) Philosophical questions can and should be pursued in cooperation with the
sciences (e.g., by using their methods and/or results).

(CMN) does not demand that all normative claims be explained in non-normative terms.
Moreover, it is compatible with a certain normative use of HB results. Consider that epistem-
ology has different traditional tasks (Stich, 1993): the definition of the concept of knowledge
(as attempted in Plato’s Theaetetus or in much current analytic epistemology), the refutation
of radical skepticism about knowledge (as pursued in Descartes’ Meditations), or also genu-
inely normative tasks, such as justifying or correcting our beliefs, or providing methods for
discovering new knowledge. Naturalists, when appealing to cognitive psychology, often hope
to profit for the third, normative tasks.8
Goldman’s (1986, Chapters 13, 14) approach is an excellent example of this. Among other
things, he shows that formal rules in the standard picture of rationality often do not map onto
reasoning tasks in a straightforward manner. For instance, the logical law of non-contradiction,
¬(¬p∧p), states that a proposition and its negation cannot both be true. From contradictions,
anything whatsoever can be derived. We should, it seems, therefore avoid inconsistencies. But
then, what do you do when you discover inconsistencies in your belief set? Are you obliged to
give up the whole set? Certainly not: we should separate the wheat from the chaff. However,
that can prove to be difficult, especially in the complex belief systems of science, where several
beliefs are all tenuous but part of a systematic theory. Similar examples can be found for other
logical rules, such as deductive closure, or for rules of probability. There is, therefore, no simple

81
82

Thomas Sturm

derivation of norms of epistemic rationality from formal rules of the standard picture. This is
more support for the gap mentioned earlier on.
To bridge that gap, Goldman (1986; 2008) maintains that a distinctively philosophical or
“analytic” epistemology has to determine the criteria or goals of our epistemic endeavors.
He here defends reliabilism: our belief-forming mechanisms need not always guarantee truth,
but must safeguard a good ratio of true and false beliefs. We can learn from psychology
which mental processes or mechanisms are reliable, and which ones are not. Heuristics are
rules of thumb that often lead to true beliefs, though they also lead to errors and fallacies;
that is just an epistemic risk we have to live with. For instance, HB studies provide insight
into how not to reason: we can, and we should, learn to control the workings of heuristics so
as to avoid biases (cf. Kahneman, Slovic, & Tversky, 1982, Chapter 30; and “dual systems”
theory, Evans, Chapter 10 in this volume).9 In Goldman’s cooperative naturalism, cognitive
psychology will not replace philosophical epistemology, but neither will the latter thrive
without the former.
What should we think of such an approach? To begin with, Goldman, like other naturalists
(e.g., Giere, 1985), thinks of epistemic rationality as a means–ends notion: given certain epi-
stemic goals, we ought to use such-and-such a method. This instrumentalism about epistemic
rationality helps to justify certain methods over others. Since means–ends relations have to be
discovered empirically, this also adds support to the idea that norms and methods can be justi-
fied empirically. However, while Goldman asserts that the goals have to be set by analytic epis-
temology, and while he favors a reliabilism according to which we ought to select methods that
lead to a good ratio of true and false beliefs, not all naturalists agree. Some argue that we should
not care whether our beliefs are true, or that the goals we should care about are primarily prag-
matic ones (Giere, 1988; Stich, 1990). Furthermore, it is far from clear that instrumentalism is
acceptable at all when it comes to epistemic rationality: whether or not epistemic reasoning is
good or bad might be entirely independent of what goals we pursue. At least some of our beliefs
can be found to be justified, or be rationally warranted, no matter what our specific epistemic
or other goals are (Siegel, 1989; Kelly, 2003).
Another point is that Goldman (1986) does not show that naturalists have no options besides
HB. We will see how the Simon–Gigerenzer conception of bounded rationality provides an
alternative.10 What is more, for HB to actually work, HB studies would have to be uncontested
with respect to their claims concerning cognitive mechanisms. They simply are not. Kahneman
and Tversky have long been challenged to produce process models that have real explanatory
value or that allow for precise, testable predictions. The heuristics they cite, such as represen-
tativeness or availability, look more like mere redescriptions of the behavior they are supposed
to explain (Gigerenzer, 1996).
However, let us assume for the sake of the argument that the HB program could do better
at the explanatory and predictive tasks, and focus again on the normative issue. Goldman him-
self has pointed to the gap between formal rules and rationality. He, therefore, cannot—unlike
Kahneman and Tversky—look back to the standard picture to provide unquestionable nor-
mative yardsticks for reasoning. For Kahneman and Tversky, but also for Goodman and many
other psychologists and philosophers, heuristics are viewed as being justified by an accuracy–effort
trade-off: we use them because looking for information and computation can be too costly; we
trade a loss in accuracy for faster and more frugal cognition. The difference between the HB
approach and Goldman’s reliabilism is that the former views heuristics as second best, whereas
the latter acquiesces in saying that there is nothing better. For the HB approach, heuristics are
good as rules of thumb, but ultimately our epistemic evaluations should be made in the light
of standard, formal, and optimizing rules. For Goldman, we cannot but balance the costs and

82
83

Towards a critical naturalism about BR

benefits of reasoning mechanisms, and we should adopt an instrumentalist attitude towards the
methods which we view as norms of epistemic rationality.11
If one turns to the Simon–Gigerenzer line of thought, then Goldman’s line of argument
can be disputed. Naturalists occasionally cite computational intractability (Goldman, 1986,
p. 282), Simon’s satisficing (Giere, 1985; Cherniak, 1986; DeLanghe, 2013), and attempts to
make use of FFH in scientific theory choice (Hey, 2016; for limits concerning this area, see
Nickles, 2016). However, they do so in less systematic and comprehensive ways. Actually,
around the time when Gigerenzer was developing his work on FFH and “ecological rationality”
(in the early 1990s, that is), Kornblith was independently taking a similar step by suggesting, as
cited above, that “our inferential mechanisms may also be built around presuppositions about
standard environments which allow us to gain information about those environments quickly
and accurately” (Kornblith, 1993, p. 86). FFH work well when they are used in the right envir-
onments; outside them, they may lead to errors. Consider the recognition heuristic (Goldstein
& Gigerenzer, 2002). The surprising result is that subjects estimate, e.g., city sizes or winners
of sports tournaments better when they know less. Thus, German subjects judge the size of US
cities better than US subjects, and vice versa. Recognition by name alone works very well here.
So, the heuristic works if and only if the environment—here, media and other information
channels—guarantee that name recognition is systematically correlated with a correct estima-
tion of the relevant criterion. Such a simple rule works better than trying to search for much
more information. Similarly so, mutatis mutandis, for other FFH.
There should be nothing mechanical in the normative use of heuristics: we should not rely
on them blindly. Consider one of the original recognition experiments, the comparison of the
sizes of San Diego and San Antonio. When the study was carried out, San Diego was larger;
by 2010, San Antonio had overtaken San Diego. That change in the environment was not yet
reflected in German media channels. When German subjects can judge only by name recog-
nition, they will probably still judge San Diego to be the larger city, and thus make a mistake.
While FFH can be extremely useful for many reasoning tasks, an alert reasoner will use them
with caution. We have to continue to think critically for ourselves.
At the same time, FFH can offer to naturalism—specifically, to (CMN)—more than a
merely instrumentalist justification of norms: not only does the connection between means
and (given) ends need to be studied empirically, but the relation of fitness between a heuristic
and the environment in which we can, and should, legitimately use it, is a matter of empirical
investigation too. At the same time, there are limits to the usefulness of FFH:

In some important domains, one can infer from empirical research what norms
of rationality are best, as well as how human reasoning can be improved. In other
domains one cannot; that is, in these the “standard” conception of rationality (Stein,
1996) as being based upon certain rules of logic or probability is not undermined by
our arguments.
Gigerenzer & Sturm, 2012, p. 244

This compatibility between the standard and the bounded conception of rationality can be
deepened in two regards. First, one cannot even formulate FFH without basic concepts and rules
of formal logic. The recognition heuristic requires a minimal grasp of the form of the “if –
then” conditional: “If one of two objects is recognized and the other is not, then infer that the
recognized object has the higher value with respect to the criterion.” Other rules, such as “take
the best”, in addition, require the ability to master disjunction, and so on. Logical notions and
rules are built into the very formulation of FFH. The gap between formal rules and rationality

83
84

Thomas Sturm

notwithstanding, a minimal dependence of bounded rationality on basic logic is unavoidable

(for a similar view, cf. Cherniak, 1986).
Second, consider the claim of Gigerenzer’s program that heuristics are sometimes as good
as, and sometimes even beat, probabilistic norms such as regression or Bayes’ theorem. In
accordance with what standard is such a claim being made? Clearly, each attempt to justify a
heuristic as normatively valid requires some standard against which the heuristic is tested. More
specifically, when defenders of bounded rationality claim that FFH outperform classical rules
of the standard picture of rationality, the only way to prove this is by comparing how FFH fare
in comparison with rules of probability or statistics. Accordingly, “take the best” is checked
against actual frequencies, and other heuristics against, say, the benchmarks of Bayesian rules
(Martignon & Blackmond Laskey, 1999; Martignon & Hoffrage, 1999). FFH can only be nor-
matively valid if they compete successfully with such rules. Both these points mean that there is,
at the normative level, a compatibility and even complementarity between the standard and the
bounded conceptions of rationality (for more on this, see Sturm, 2019).

Conclusion: for a critical naturalism about rationality

Time to take stock. I have tried to show how productive a careful analysis of the perplexing
historical, scientific, and philosophical interactions between the standard picture of rationality,
the psychology of reasoning, models of bounded rationality, and philosophical naturalism can
be. The standard picture arose through the coalescence of logic, probability, and rational choice
theory through the twentieth century. That picture then was criticized by both the Kahneman–
Tversky HB approach, which emphasizes that humans are lazy reasoners, and by Simon and
later on by Gigerenzer and his collaborators for being too mindlessly formal and optimizing in
terms of what the proper norms of good reasoning are. Philosophers who defend naturalism
in epistemology and philosophy of science first quickly took up the HB approach, but did so
in problematic ways that threatened their own naturalism.
Simon’s satisficing and Gigerenzer’s FFH are better allies of philosophical naturalism. Four
points can be highlighted to summarize the results concerning this claim. First, a methodo-
logical naturalism about epistemic rationality can be based on this particular version of bounded
rationality, given that means–ends as well as mind–environment relations have to be studied,
both of which have to be determined empirically. Second, such naturalism about rationality
and, in particular FFH, can be normative, but only if heuristics are used in mindful, delib-
erate ways, at least in principle. Third, such naturalism is limited in its scope and potential
applications, and it should be fully reflexive with regard to these limits. It should live in coexist-
ence with the standard account of rationality by leaving, for instance, computationally tractable
tasks to the latter. It even requires the standard account to some extent, since logical notions
and rules are required to formulate FFH, and theories of probability in order to assess their
validity. Fourth, and finally, bounded rationality cannot be expected to provide a basis for the
naturalists’ ontological claim (ON) concerning rationality. Research into bounded rationality is
not yet mature enough to deliver a systematic theory that could be used to explain the nature
of rationality in general. Maybe it never will be.
Given these four points, we can justifiably call the ensuing position, while tipping our hats
to Kant, “critical naturalism”. It is naturalistic because of the first point, namely, its empir-
ical methodology, and critical in the Kantian sense due to the three other points: We should
use heuristics in reflective and limited ways, and we should avoid rushing towards ontological
claims that are not sufficiently warranted. But even if one does not share this refined version
of naturalism, it seems to me that epistemologists and philosophers of science have, up to now,

84
85

Towards a critical naturalism about BR

all too rarely used bounded rationality as understood by Simon and by Gigerenzer and his
colleagues. More work can and should be done in this area.

Acknowledgments
For comments and discussions, the author is grateful to Riccardo Viale and Gerd Gigerenzer.
Christopher Evans helped to improve the language of this chapter, and gave several highly
useful recommendations concerning content too. This work was supported by the Spanish
Ministry for the Economy, Industry and Competitiveness (MINECO) through the research
project “Naturalism and the sciences of rationality: an integrated philosophy and history”
(FFI2016–79923-P).

Notes
1 I cannot here consider the view that these terms are not synonymous, but see Sturm (2018).
2 One might think that, given its title, Stich (1993) belongs into this camp too. However, Stich only uses
Simon’s work in artificial intelligence (AI) to explain scientific discovery (Langley et al., 1987). AI and
bounded rationality were Simon’s two major research agendas; but they had no internal connection that
would necessitate dealing with the former here as well. Therefore, I will leave aside Simon’s research in
AI, and Stich’s use of it to naturalize epistemology.
3 Frege’s distinction between the material realm, the realm of subjective mental representations, and the
realm of objective thoughts bears similarities with Karl Popper’s equally notorious division into world
1 (the physical world), world 2 (the mental world), and world 3 (the world of objective knowledge)
(Popper, 1972). However, Popper’s division is historically probably more influenced by his teacher,
the psychologist Karl Bühler and his theory of basic functions of language (Sturm, 2012b). Moreover,
while Popper’s theory is clearly intended to be an ontological distinction, it is a disputed question
whether Frege’s claim about the “third realm” was truly intended to be an ontological claim (for a less
demanding, epistemological interpretation of Frege’s view, see Carl 1994, Chapters 2–4).
4 Of course, theories of probability were far longer in the making, famously starting with the Pascal–
Fermat exchanges on games of chance in 1654 (Gigerenzer et al., 1989; Gillies, 2000).
5 To think that even today some experts view (a) to be true may seem false, given the heuristics-and-
biases approach that is explained above. However, things are not so simple. In economics, whether (a) is
taken to be true or not depends a lot on whether standard norms, e.g., for rational choices, are viewed
as descriptive of individual decisions (where it seems implausible), of aggregates of decisions of many
individuals, or of economic structures. On this debate, see Ross (2014).
6 There is no reliable survey showing whether or not naturalism dominates current philosophy. A prelim-
inary attempt was made by Bourget and Chalmers (2013). When asked about their metaphilosophical
position, participants (mostly professional philosophers, mostly though not exclusively from Anglo-
Saxon universities) chose naturalism first (49.8 percent), followed by non-naturalism (25.9 percent) and
“other” (24.3 percent). Somewhat surprisingly, the same participants overwhelmingly denied essential
claims of typical naturalistic positions, such as the rejection of a priori knowledge (71.1 percent said
such knowledge exists), or the rejection of the distinction between analytic and synthetic propositions.
Perhaps this reveals that the questions were problematic, but it is also possible that many philosophers
hold unusual views (to say the least).
7 This may be questioned, e.g., in the light of historical considerations. Gigerenzer et al. (1989) argue that
theories of probability were sometimes revised when they clashed with what educated minds thought
about probabilities. Consider the famous St. Petersburg paradox: recognizing that no reasonable person
was willing to invest infinite sums of money in a coin tossing game where the expected gains were
infinite did not lead to the judgment that such a person was mistaken. Instead, it led Daniel Bernoulli to
substitute the concept of expectation with that of monetary values, “with the awareness that the richer
you are, the more it takes to make you happy” (Gigerenzer et al., 1989, p. 15). However, it would be
questionable to claim that Bernoulli had attempted to justify a normative rule by means of observation
or experiment. One might see this, rather, as an instance of applying the method of reflective equilib-
rium: we develop our normative theories by considering whether instances of good reasoning fit with
our best principles, and vice versa; and we adjust them when conflicts arise (see also Cohen, 1981).

85
86

Thomas Sturm

8 They also try, of course, to profit from psychology for other tasks of epistemology, for instance, in pro-
viding a descriptive account of knowledge or cognition. This is what Quine’s (1969) original proposal
for a naturalistic epistemology comes down to, or what evolutionary epistemologists have developed.
However, such descriptive naturalistic epistemology faces the challenge that it simply changes the game
by not taking seriously enough the normative tasks of epistemology (Kim, 1988). Accordingly, in
order to avoid being charged with changing the topic, a number of naturalists accepted the normative
tasks too.
9 Again, this undermines excessive conclusions (e.g., Stich, 1985) on the basis of HB studies according
to which humans are fundamentally irrational, a claim that is not entirely absent from Kahneman and
Tversky either; see Sturm (2012a).
10 Goodman (2008) offers a more mixed assessment of the competing approaches.
11 In philosophy of science, Solomon (1992) has used the HB program to explain e.g., the geological
revolution of the nineteenth century. She finds that Stich (1985) has argued convincingly against
criticisms of the HB program, a judgment with which I disagree, though I cannot show this here. She
also uses HB studies to explain the normative successes of choices made during the geological revo-
lution, claiming that Alfred Wegner’s choice of continental drift over contractionism was driven by
aspects of representativeness (Solomon, 1992, p. 447), and that belief perseverance or availability played
a role too. However, the HB program of testing how far people follow formal or optimizing rules
makes sense only where such rules can be clearly stated. Only then can judgments of bias be justifiably
made, and explanations in terms of heuristics be given. Suitable norms just do not exist for cases such
as those Solomon discusses. Nor does Solomon explain why geologists came to agree so quickly that
continental drift was the more adequate theory in the 1960s. As Solomon admits, her explanation is
only tentative and perhaps not the whole story anyhow.

References
Bermúdez, J. L., & Millar, A. (Eds.) (2002). Reason and nature. Oxford: Oxford University Press.
Bishop, M., & Trout, J. D. (2005). Epistemology and the psychology of human judgment. New York: Oxford
University Press.
BonJour, L. (1994). Against naturalized epistemology. Midwest Studies in Philosophy, 19, 283–300.
Bourget, D., & Chalmers, D. (2013). What do philosophers believe? Available at: https://philpapers.org/
archive/BOUWDP
Carl, W. (1994). Frege’s theory of sense and reference. Cambridge: Cambridge UP.
Cherniak, C. (1986). Minimal rationality. Cambridge, MA: MIT Press.
Chiappe, D. L., & Vervaeke, J. (1997). Fodor, Cherniak, and the naturalization of rationality. Theory &
Psychology, 7, 799–821.
Cohen, L. J. (1981). Can human irrationality be experimentally demonstrated? Behavioral and Brain
Sciences, 4, 317–331 (comments and responses, 331–359).
Cosmides, L., & Tooby, J. (1996). Are humans good intuitive statisticians after all? Cognition, 58, 1–73.
De Caro, M., & MacArthur, D. (Eds.) (2004). Naturalism in question. Cambridge, MA: Harvard
University Press.
De Caro, M., & MacArthur, D. (Eds.) (2010). Naturalism and normativity. New York: Columbia
University Press.
deLanghe, R. (2013). Satisficing as an account of Kuhnian rationality. Philosophy Study, 3, 398–411.
DePaul, M., & Ramsey, W. (Eds.) (1998). Rethinking intuition: The psychology of intuition and its role
in philosophical inquiry. Lanham, MD: Rowman & Littlefield.
Dennett, D. (1991). Consciousness explained. Boston: Little, Brown & Co.
Dretske, F. (1996). Naturalizing the mind. Cambridge, MA: MIT Press.
Erickson, P., Klein, J., Daston, L., Lemov, R, Sturm, T., & Gordin, M. (2013). How reason almost lost its
mind. Chicago: University of Chicago Press.
Feldman, R. (2001). Naturalized epistemology. In E. N. Zalta (Ed.), Stanford encyclopedia of philosophy.
Available at: http://plato.stanford.edu/archives/sum2012/entries/epistemology-naturalized/
Frege, G. (1966). Der Gedanke (1918–19). In G. Frege, Logische Untersuchungen, ed. G. Patzig (pp. 30–53).
Göttingen: Vandenhoek & Ruprecht.
Galparsoro, J. I., & Cordero, A. (Eds.) (2013). Reflections on naturalism. Rotterdam: Sense Publishers.
Giere, R. (1985). Philosophy of science naturalized. Philosophy of Science, 52, 331–356.

86
87

Towards a critical naturalism about BR

Giere, R. (1988). Explaining science: A cognitive approach. Chicago: University of Chicago Press.
Giere, R. (Ed.) (1992). Cognitive models of science. Minneapolis, MN: University of Minnesota Press.
Gigerenzer, G. (1991). How to make cognitive illusions disappear: Beyond heuristics and biases. European
Review of Social Psychology, 2, 83–115.
Gigerenzer, G. (1996). On narrow norms and vague heuristics: A rebuttal to Kahneman and Tversky.
Psychological Review, 103, 592–596.
Gigerenzer, G., & Selten, R. (2001). Rethinking rationality. In G. Gigerenzer & R. Selten (Eds.), Bounded
rationality (pp. 1–12). Cambridge, MA: MIT Press.
Gigerenzer, G.. & Sturm, T. (2012). How (far) can rationality be naturalized? Synthese, 187, 243–268.
Gigerenzer, G., Swijtink, Z., Porter, T., Daston, L., Beatty, J., & Krüger, L. (1989). The empire of chance.
Cambridge: Cambridge University Press.
Gigerenzer, G., Todd, P. M., & the ABC Research Group (Eds.) (1999). Simple heuristics that make us smart.
New York: Oxford University Press.
Gillies, D. (2000). Probability. London: Routledge.
Goldman, A. (1986). Epistemology and cognition. Cambridge, MA: Harvard University Press.
Goldman, A. (2008). Human rationality: Epistemological and psychological perspectives. In A.
Beckermann & S. Walter (Eds.), Philosophy: Foundations and applications (pp. 230–247). Paderborn,
Germany: Mentis.
Goldstein, D. G., & Gigerenzer, G. (2002). Models of ecological rationality: The recognition heuristic.
Psychological Review, 109, 75–90.
Grice, P. (2001). Aspects of reason. Oxford: Oxford University Press.
Habermas, J. (1981). Theorie des kommunikativen Handelns. 1. Frankfurt/M.: Suhrkamp.
Hahn, S. (2013). Rationalität. Paderborn: Mentis.
Harman, G. (1977). The nature of morality. New York: Oxford University Press.
Hauptli, B. (1995). The reasonableness of reason: Explaining rationality naturalistically. Chicago: Open Court.
Hey, S. (2016). Heuristics and meta-heuristics in scientific judgment. British Journal for the Philosophy of
Science, 67, 471–495.
Kahneman, D. (2003). Maps of bounded rationality: Psychology for behavioral economics. American
Economic Review, 93, 1449–1475.
Kahneman, D., Slovic, P., & Tversky, A. (Eds.) (1982). Judgment under uncertainty. New York: Cambridge
University Press.
Kelly, T. (2003). Epistemic rationality as instrumental rationality: A critique. Philosophy and Phenomenological
Research, 66, 612–640.
Kim, J. (1988). What is “naturalized epistemology”? Philosophical Perspectives, 2, 381–405.
Kitcher, P. (1992). The naturalists return. Philosophical Review, 101, 53–114.
Kitcher, P. (1993). The advancement of science. New York: Oxford University Press.
Kornblith, H. (1993). Inductive inference and its natural ground. Cambridge, MA: MIT Press.
Kornblith, H. (Ed.) (1994). Naturalizing epistemology (2nd ed). Cambridge, MA: MIT Press.
Kornblith, H. (2002). Knowledge and its place in nature. Oxford: Oxford University Press.
Kornblith, H. (2007). Naturalism and intuitions. Grazer Philosophische Studien, 72, 27–49.
Ladyman, J. (2007). Everything must go: Metaphysics naturalized. Oxford: Oxford University Press.
Langley, P., Simon, H., Bradshaw, G., & Zytkow, J. (1987). Scientific discovery. Cambridge, MA: MIT Press.
Laudan, L. (1987). Progress or rationality? The prospects for normative naturalism. American Philosophical
Quarterly, 24, 19–31.
Laudan, L. (1990). Normative naturalism. Philosophy of Science, 57, 44–59.
Lenman, J. (2009). Naturalism without tears. Ratio, 12, 1–18.
Lopes, L., (1991). The rhetoric of irrationality. Theory and Psychology, 1, 65–82.
Martignon, L., & Blackmond Laskey, K. (1999). Bayesian benchmarks for fast and frugal heuristics.
Oxford: Oxford University Press.
Martignon, L., & Hoffrage, U. (1999). Why does one-reason decision making work? A case study in eco-
logical rationality. In G. Gigerenzer, P. M. Todd, & the ABC Research Group (Eds.), Simple heuristics
that make us smart (pp. 119–140). New York: Oxford University Press.
Mele, A., & Rawlings, P. (Eds.) (2004). The Oxford handbook of rationality. Oxford: Oxford University Press.
Mi, C., & Chen, R. (Eds.) (2007). Naturalized epistemology and philosophy of science. Amsterdam: Rodopi.
Michaelewicz, Z., & Fogel, D. (2000). How to solve it: Modern heuristics. New York: Springer.
Newell, A., Shaw, J. C., & Simon, H. A. (1958). Elements of a theory of human problem solving.
Psychological Review, 65, 151–166.

87
88

Thomas Sturm

Nickles, T. (2016). Fast and frugal heuristics appraisal at research frontiers. In E. Ippoliti, F. Sterpetti, & T.
Nickles (Eds.), Models and inferences in science (pp. 31–54). Heidelberg: Springer.
Nida-Rümelin, J. (1996). Zur Einheitlichkeit praktischer Rationalität. In K.-O. Apel & M. Kettner (Eds.),
Die eine Vernunft und die vielen Rationalitäten (pp. 73–90). Frankfurt a/M: Suhrkamp.
Notturno, M. A. (1985). Objectivity, rationality and the third realm. Justification and the grounds of psycholo-
gism: A study of Frege and Popper. Dordrecht: Nijhoff.
Nozick, R. (1993). The nature of rationality. Princeton, NJ: Princeton University Press.
Oaksford, M., & Chater, N. (1991). Bounded rationality in taking risks and drawing inferences. Theory &
Psychology, 2, 225–230.
Pacho, J. (1995). ¿Naturalizar la razón? Madrid: Sigo XXI editores.
Papineau, D. (2015). Naturalism. In E. N. Zalta (ed.), The Stanford encyclopedia of philosophy. Available
at: http://plato.stanford.edu/entries/naturalism/entries/naturalism/
Peckhaus, V. (2006). Psychologism and the distinction between discovery and justification. In J. Schickore
& F. Steinle (Eds.), Revisiting discovery and justification (pp. 99–116). Berlin: Springer.
Peterson, C., & Beach, L. (1967). Man as an intuitive statistician. Psychological Bulletin, 68, 29–46.
Popper, K. (1972). Objective knowledge: An evolutionary approach. Oxford: Clarendon Press.
Putnam, H. (1981). Reason, truth, and history. Cambridge: Cambridge University Press.
Putnam, H. (1982). Why reason can’t be naturalized. Synthese, 52, 3–24.
Quine, W. V. O. (1969). Epistemology naturalized. In W. V. O. Quine, Ontological relativity and other essays
(pp. 69–90). New York: Columbia University Press.
Rosenberg, A. (1996). A field guide to recent species of naturalism. British Journal for the Philosophy of
Science, 47, 1–29.
Ross, D. (2014). Philosophy of economics. London: Palgrave Macmillan.
Samuels, R., Stich, S., & Bishop, M. (2002). Ending the rationality wars: How to make disputes about
human rationality disappear. In R. Elio (Ed.), Common sense, reasoning and rationality (pp. 236–268).
New York: Oxford University Press.
Siegel, H. (1989). Philosophy of science naturalized? Some problems with Giere’s naturalism. Studies in
History and Philosophy of Science, 20, 365–375.
Simon, H. (1956). Rational choice and the structure of the environment. Psychological Review, 63,
129–138.
Simon, H. (1957). Models of man. New York: Wiley.
Sklar, L. (2010). I’d love to be a naturalist-if only I knew what naturalism was. Philosophy of Science, 77,
1121–1137.
Sober, E. (1978). Psychologism. Journal for the Theory of Social Behaviour, 8, 165–192.
Sober, E. (1981). The evolution of rationality. Synthese, 46, 95–120.
Solomon, M. (1992). Scientific rationality and human reasoning. Philosophy of Science, 59, 439–455.
Stein, E. (1996). Without good reason. Oxford: Clarendon.
Stich, S. (1985). Could man be an irrational animal? Some notes on the epistemology of rationality.
Synthese, 64, 115–135.
Stich, S. (1990). The fragmentation of reason. Cambridge, MA: MIT Press.
Stich, S. (1993). Naturalizing epistemology: Quine, Simon, and the prospects for pragmatism. Royal
Institute for Philosophy Supplement, 34, 1–17.
Stroud, B. (1996). The charm of naturalism. Proceedings and Addresses of the American Philosophical Association,
70, 43–55.
Sturm, T. (2012a). The “rationality wars” in psychology: Where they are and where they could go. Inquiry,
55, 66–81.
Sturm, T. (2012b). Bühler and Popper: Kantian therapies for the crisis in psychology. Studies in History and
Philosophy of Biological and Biomedical Sciences, 43, 462–472.
Sturm, T. (2018). Rationalität versus Vernunft? Über eine Unterscheidung bei John Rawls (und anderen).
In D. Sölch (Ed.), Philosophische Sprache zwischen Tradition und Innovation (pp. 211–233). Frankfurt
a.M: Lang.
Sturm, T. (2019). Formal versus bounded norms in the psychology of rationality: Toward a multilevel
analysis of their relationship. Philosophy of the Social Sciences, 49, 190–209.
Sturm, T. (Forthcoming). Theories of rationality and the descriptive/normative divide: A historical
approach. In M. Knauff & W. Spohn (Eds.), Handbook on rationality. Cambridge, MA: MIT Press.
Turing, A. (1950). Computing machinery and intelligence. Mind, 59, 433–460.

88
89

Towards a critical naturalism about BR

Tversky, A., & Kahneman, D. (1983). Extensional versus intuitive reasoning: Conjunction fallacy in prob-
ability judgment. Psychological Review, 90, 293–315.
Viale, R. (2013). Methodological cognitivism, vol. 2: Cognition, science, and innovation. Heidelberg: Springer.
von Neumann, J., & Morgenstern, O. (1944). Theory of games and economic behavior. Princeton, NJ: Princeton
University Press.
Wason, P. (1966). Reasoning about a rule. Quarterly Journal of Experimental Psychology, 20, 273–281.

89
90

4
BOUNDED RATIONALITY
The two cultures

Konstantinos V. Katsikopoulos

Introduction
Bounded rationality does not speak with one voice. This is not only because bounded ration-
ality is researched in various fields such as economics, psychology, and management. Even
within a single field such as economics, there are clear differences. For example, Selten (2001)
rejects the optimization of a utility function as an expression of bounded rationality, contrary
to the standard approach of behavioral economics as in bargaining games by Fehr and Schmidt
(1999). There are multiple views of bounded rationality, as pointed out by Rubinstein (1998).
The first contribution of the present chapter is to analyze the formal modeling used to
describe people’s bounded rationality. At the risk of oversimplifying, I distinguish between two
cultures, which I call “idealistic” and “pragmatic.” At a first approximation, the idealistic cul-
ture pursues a minimum departure from the neoclassical-economics framework of unbounded
rationality which assumes the ideals of omniscience and optimization of a utility function and
adds factors such as inequity aversion or probability weighting to the utility function. On the
other hand, the pragmatic culture holds that people sometimes ignore information and use
simple rules of thumb in order to achieve satisfactory outcomes. Note that I do not use the label
“pragmatic” as used by, among others, Friedman (1953), to emphasize some views of the prac-
tical purposes of economics, such as delivering accurate predictions. A detailed discussion of the
differences in modeling between the two cultures is provided in the next section. The reality
of the cultures and their differences are demonstrated by examples drawn from the literatures
on risky choice and bargaining games. Note that it does not make sense to try to perfectly map
specific researchers or programs of research to one or the other culture; for example, Amos
Tversky worked on both cultures, with prospect theory being an idealistic model and elimin-
ation by aspects being a pragmatic model.
Although the distinction between the idealistic and pragmatic cultures of bounded ration-
ality can be criticized, as all binary distinctions can, it provides food for thought and new
insights. I aim at emulating Breiman’s (2001) analysis of two cultures in statistics. Breiman
argued that there exist two cultures which lead to two very different kinds of statistical theory
and practice: proof-based and data-driven. Analogously, I argue in the third section that the
idealistic and pragmatic cultures tell two very different stories about people’s bounded ration-
ality and how to improve it. This is the second contribution of the present chapter. Echoing

90
91

Bounded rationality: the two cultures

Morgan (2001), I conclude that these stories play a vital role in our understanding of the eco-
nomic world and the policies we develop.

The two cultures: differences in modeling

Table 4.1 presents six key modeling differences between the idealistic and pragmatic cultures of
research on describing people’s bounded rationality. This presentation is epigrammatic. The rest
of the section spells out each difference, as well as their relationships.
I first discuss the labels “idealistic” and “pragmatic,” which are connected to the first diffe-
rence in Table 4.1. I then discuss the second and third differences and the remaining differences.

What do the labels “idealistic” and “pragmatic” mean?

The first difference between the two cultures refers to the building blocks they use in order to
generate their models. This difference is the main reason for the labels “idealistic” and “prag-
matic.” The idealistic culture of bounded rationality is indeed inspired by an ideal, unbound-
edly rational creature. This is a decision maker who possesses all information that can possibly
be gathered and, based on it, makes all possible correct deductions, which she uses to make an
“optimal” decision. For example, in a choice among gambles, this decision maker knows all
possible outcomes of each gamble, is able to assign a numerical utility to each outcome, knows
the probability with which each outcome occurs, and finally calculates the expected utility of
each gamble and chooses a gamble which obtains the maximum.
The choices of an expected utility optimizer can be represented by the logical axioms jointly
equivalent to expected utility theory (von Neumann and Morgenstern 1944). An example
axiom is transitivity where, for all gambles x, y and z, if x is chosen over y and y is chosen over
z, then x is chosen over z. According to some authors, such as Savage (1954), these axioms have
normative status, meaning that a decision maker should satisfy them.
The same kinds of axioms are the building blocks of the idealistic culture of bounded ration-
ality. A researcher can generate new models of bounded rationality by retaining some axioms
of unbounded rationality, taking out others and proposing new ones. For example, Kahneman
and Tversky’s prospect theory (1979) always satisfies transitivity but may violate independence

Table 4.1 Six modeling differences between the idealistic and pragmatic cultures of research on
describing people’s bounded rationality

Bounded-rationality cultures differences in Idealistic Pragmatic

modeling
1. Building blocks, based on which Logical axioms Empirical facts
models are generated (e.g. people make transitive (e.g. people make choices
choices) based on only one reason)
2. Assumptions about people’s goal Optimization of a utility Achievement of a satisfactory
function outcome
3. Treatment of psychological No models of processes; Models of processes as simple
processes instead, as-if optimization rules of thumb
4. Treatment of parameters Let parameters vary freely Fix parameters
5. Epistemic aim Explanation of known facts Prediction of new facts
6. Models tested Only from idealistic culture From both cultures

91
92

Konstantinos V. Katsikopoulos

(for all gambles x, y and z, and probabilities p, if x is chosen over y, then the compound gamble
(x, p; z, 1 – p) is chosen over (y, p; z, 1 – p)). Bounded-rationality models, such as prospect
theory, have also been axiomatized by axioms that can be argued to be normative (Wakker and
Tversky 1993). Thus, the prospect-theory decision maker is also ideal, just a bit less so than her
expected-utility ancestor.
But not everybody is happy with this industry of transforming neoclassical models
into bounded-rationality ones. Güth called it a “neoclassical repair shop” (1995, p. 342).
Dissatisfaction and impatience with it run through the whole volume edited by Selten and
Gigerenzer (2001), who look away from axioms to find the building blocks of bounded ration-
ality. They have the work of Herbert Simon––the father of bounded rationality––to fall back
on, who, throughout his whole career, insisted on first considering what is known about how
real people actually make decisions in the real world (Katsikopoulos and Lan 2011): In the
abstract of Simon’s obituary, his long-standing colleague March wrote: “In particular, he per-
sistently sought to clarify the real processes of human decision making …” (Augier and March
2002, p. 1).
I call “pragmatic” the culture that uses empirical facts as its building blocks. As an example
of a model of the pragmatic culture, take the priority heuristic for choices among risky gambles
(Brandstätter, Gigerenzer, and Hertwig 2006). The heuristic is based on the fact that people
often make choices by using just one reason and consider a second or third reason only if they
have to (Ford et al. 1989). According to the priority heuristic, when choosing between two
gambles (which lead only to gains compared to the status quo), the first reason people look at
is the minimum gains of the two gambles x and y, respectively min(x) and min(y); if |min(x) –
min(y)| > c where c is a fixed threshold, then the gamble with the higher minimum gain is
chosen; otherwise the second reason, which is the probabilities of the minimum gains of the
two gambles, is looked up, and so on, until a reason is found which permits choosing one
gamble. The existence of thresholds that allow for a choice or necessitate more search for infor-
mation is an empirical fact (Tanner and Swets 1954).
It should be noted that while the models of the pragmatic culture are not primarily inspired
or justified by normative axioms, they are amenable to study from a normative or axiomatic
perspective. In the former case, the performance of pragmatic models, in terms of criteria
such as predictive accuracy, is investigated (Katsikopoulos 2011b). In the latter case, it is tested
whether or not pragmatic models satisfy axioms such as transitivity or independence (Manzini
and Mariotti 2007, 2012, 2014; Katsikopoulos and Gigerenzer 2008) and pragmatic models are
shown to be equivalent to a set of axioms (Drechsler, Katsikopoulos, and Gigerenzer 2014).
An analogous point can be made for the models of the idealistic culture. Idealistic models
are subject to empirical study as in the experimental tests of prospect theory. But empirical
facts are not the sole, or in some cases not even the primary inspiration or justification for
the development of idealistic models. For example, a key assumption of cumulative prospect
theory––that people weigh probabilities nonlinearly––was not only inspired by the empirical
fact that people’s risk attitude depends on whether outcomes are gains or losses and on if the
probabilities of gains or losses are large (Tversky and Kahneman 1992). Rather, in addition to
this empirical fact, there is also a crucial influence of a non-empirical factor on the develop-
ment of the probability-weighting assumption. This factor is that the assumption is necessary
to explain the pattern if the modeler sticks to the general mathematical form of utility-times-
probability, common in idealistic models. This assumption is not necessary in other models
(Katsikopoulos and Gigerenzer 2008).
Put another way, the character of the idealistic culture is logical whereas that of the prag-
matic culture is ecological. Ecology here is meant in Simon’s (1955, 1956) sense of the

92
93

Bounded rationality: the two cultures

environment––physical or mental––where decision-making takes place. Simon insisted that

human behavior could be well understood only if it is studied in relation to its environment.
But despite the overall impact of Simon’s work, in economics, his call has been heeded by the
pragmatic culture but not by the much more prevalent idealistic culture.
Other authors have also discussed conceptually the different views on bounded rational.
Gigerenzer (2008) proposes three views: “as-if optimization,” “ecological rationality,” and
“irrationality” (see also Berg and Gigerenzer 2010 and Brighton and Gigerenzer 2012). As-if
optimization is related to what I call idealistic culture and ecological rationality is related to
what I call pragmatic culture. The irrationality view refers to empirical research which has
concluded that people systematically violate axioms of logic and probability as in the heuristics-
and-biases research program of Tversky and Kahneman (Kahneman, Slovic, and Tversky 1982;
Tversky and Kahneman 1974). Here, I see this research as part of the idealistic culture of
bounded rationality. It forms the empirical basis of this culture and gives rise to the story
that people are systematically irrational and the authorities should nudge them toward better
decisions, as I discuss in the third section.
Another author who has discussed different ways of conceptualizing rationality, bounded as
well as unbounded, is Lee (2011). He points out that in neoclassical economics, rationality is
identified with logical consistency and optimization. Here, I argue that this is also the case in
the idealistic culture of bounded rationality. Intriguingly, Lee calls pragmatic the classical eco-
nomic notions of rationality, such as Adam Smith’s.
Finally, consider again the labels used for the two cultures. Instead of “pragmatic,” one may
be tempted to use another label such as “empirical.” But I believe that “pragmatic” is the right
choice for the kind of models represented by the entries in the right column of Table 4.1.
A glance at Table 4.1 shows that these models are “more practical as opposed to idealistic”
which is how the Merriam-Webster online dictionary defines the word “pragmatic.” For
example, in the second row, pragmatic models are defined as those in which a person’s goal
is to achieve a satisfactory outcome as opposed to attempt to optimize. This kind of pragma-
tism is successful in the real world as it has been found that, under some conditions, prag-
matic models outperform optimization models in medicine, management and engineering
(Katsikopoulos 2011b).
On the other hand, a difficulty with the label “idealistic” is that this word has all sorts of
moral and political connotations. I do not wish to have these connotations ascribed to the
bounded-rationality models and stories discussed here. In the present chapter, the idealistic
culture of bounded rationality refers to work inspired by the ideal of an unbounded rational
decision maker who is omniscient and optimizes a utility function.

Optimization
Simon repeatedly questioned the usual assumption of economics that people try to optimize.
Resounding plain common sense, Simon (1947) pointed out that people rarely even think
about how to optimize and instead are content to satisfice. As Klein (2001) argues, in the real
world, satisficing may be the only choice as the optimal outcome may not be calculable or even
well defined. The pragmatic culture takes this point to heart and assumes that people’s goal is to
achieve a satisfactory outcome. For example, Brandstätter et al.’s (2006) priority heuristic does
not necessarily lead to choices that optimize expected utility or value but it does guarantee that
a gamble with a much smaller minimum gain will not be chosen. On the other hand, in ideal-
istic models, such as prospect theory, people are assumed to choose a gamble that optimizes a
utility function.

93
94

Konstantinos V. Katsikopoulos

Now, what exactly does it mean to say that people optimize a utility function? The typical
interpretation in neoclassical economics is that people behave as if they optimize (Friedman
1953). The claim is not that people necessarily perform all the calculations needed in order to
optimize but that their behavior agrees with the behavior that results from these calculations.
That is, optimization is not meant to describe the underlying psychological processes, only their
outcome. This neglect of process dominates the idealistic modeling of bounded rationality
as well. It may seem odd to argue that, say, prospect theory does not model processes, but it
indeed does not in the sense that prospect theory does not specify how exactly it can be that
a person would manage to nonlinearly weight probabilities, calculate nonlinear utilities, and
integrate the two (note that there are elements of a process in prospect theory, as in its initial
stage of setting a reference point). I am aware that behavioral economists routinely call their
models process models, but if one takes the definition of a cognitive process in the light just
described, this is not so. To be fair, this is a topic of considerable dispute (Berg and Gigerenzer
2010; Gintis 2011).
In sum, the third difference in Table 4.1 is that, unlike the idealistic culture, the pragmatic
culture insists on developing process models. Of course, even within the pragmatic culture,
there is often disagreement about what is and what is not a process model. It seems that a large
chunk of process models describe simple rules of thumb which determine how people first
search for information, then stop this search and finally make a decision based on the informa-
tion gathered. For example, this is the case in the priority heuristic (Brandstätter et al., 2006)
as well as in an earlier tradition of models such as elimination by aspects (Tversky 1972) and
satisficing (Simon 1955).

Testing models
The remaining technical differences between the idealistic and pragmatic cultures have to do
with model testing. The fourth difference in Table 4.1 refers to how parameters are treated. In
theory, the parameters should be estimated independently of the data used to test the model. As
Luce (1999, p. 727) wrote, parameters are to be estimated “once and for all … from experiments
designed to do just that.” Gonzalez and Wu (1999), for example, estimated the probability
weighting functions of individual decision makers. Practically, the problems start when model
development in a research area is not cumulative enough in order to build on previous param-
eter estimates. For some researchers, these problems are formidable and they think they have
a “proliferation of free parameters in many types of theories with little success in developing
theories of such parameters” (Luce 1997, p. 79).
Other researchers are not so wary of parameters (the different points of view are discussed
in Katsikopoulos 2011a). Overall, it is modelers within the pragmatic culture who seem to
avoid the use of free parameters. It is advertised as a strength of the priority heuristic that it has
fixed parameters as is also the case in many models developed by Gigerenzer and his colleagues
(Gigerenzer, Hertwig, and Pachur 2011). On the other hand, it is routine in behavioral eco-
nomics to develop models with multiple free parameters.
I will give an example from the literature on bargaining games. Fehr and Schmidt (1999)
have developed an idealistic model where players are assumed to behave as if they optimize
a utility function. This utility function includes a player’s own payoff but it also includes the
player’s aversion to inequity, as when earning a smaller or larger payoff than other players. For
example, in a two-player ultimatum game where the proposer offers a fraction p < ½ of a unit
pie to the responder and the responder accepts it, the utility of the responder equals p – α[(1 –
p) – p] where α > 0 measures the responder’s envy due to earning less than the proposer, and

94
95

Bounded rationality: the two cultures

the utility of the proposer equals (1 – p) – β[(1 – p) – p] where β > 0 measures the proposer’s
discomfort due to earning more than the responder. These functions can be used to identify
which decisions optimize the players’ utilities (for the proposer, which p to offer; and for the
responder, whether to accept each p or not).
Whereas in the Fehr–Schmidt model parameters α and β are allowed to vary freely, in a
pragmatic model of bargaining games players are assumed to use a toolbox of rules of thumb,
each with fixed parameters (Fischbacher, Hertwig, and Bruhin 2013). Examples of the rules
of the proposer are that she offers p = ½ or the largest possible p which is smaller than ½, and
examples of the rules of the responder are that she accepts all p > 0 or only those p such that
p > p*, where p* is what she offers when she is the proposer.
Of course, Fehr and Schmidt (1999) did attempt to estimate the parameters of their model.
But this is not the point. Leaving aside the fact that there is a controversy on whether the esti-
mation was done properly or not (Binmore and Shaked 2010; Fehr and Schmidt 2010), the
point is that a model with free parameters already constituted a precisely defined model for Fehr
and Schmidt (1999), while this is not the case in pragmatic models.
Now, one could argue that it is close to irrelevant, or just a matter of taste, whether a model
uses free or fixed parameters; what matters is if the model can describe empirical facts well.
Interestingly, it turns out that the idealistic and pragmatic cultures understand “describe” and
“well” very differently. This is captured by the fifth and sixth differences in Table 4.1.
In order to understand these differences, it helps to digress and consider the work of
Musgrave (1974). He discusses three views of when an empirical fact lends support to a model.
In the logical view, it matters only if the fact is consistent with the model’s implications. In the
historical view, it also matters if the model’s implications were derived before or after the fact
was observed. More support is provided for the model if the derivation preceded the observa-
tion. Musgrave argues for a third view, a variant of the historical view in which it is addition-
ally relevant what the implications of the best competing model are. More support is provided
for the model if its best competitor does not imply the observed fact. Thus, the logical view
accepts as an epistemic aim the explanation of known facts (here explanation is the consistency of
a model’s implications with the facts, ignoring, for example, whether or not the model proposes
causal factors that lead to the facts). On the other hand, the historical view rejects this and aims
at the prediction of new facts. A second distinction is that Musgrave’s variant of the historic view
considers it a plus to competitively test models, whereas the logical view is silent on that.
It may be argued that the idealistic culture espouses the logical view whereas the prag-
matic culture is aligned with the historical view, and in particular Musgrave’s variant. More
specifically, models that are able to accommodate a wide range of empirical facts are highly
valued in the idealistic culture even if the models were developed after the facts have been
observed. For example, the development of prospect theory and other risky choice models
which follow the utility-times-probability mathematical form, has been following the empir-
ical violations of the axioms of expected utility theory (Starmer 2000). On the other hand,
pragmatic models, such as the priority heuristic, have not been developed in order to account
for these violations––even though later it was shown that they could do so (Katsikopoulos
and Gigerenzer 2008)––but rather in order to predict new facts. This is the fifth difference in
Table 4.1.
The distinction between explaining known facts and predicting new facts is sometimes
acknowledged in work on idealistic models (Blanco, Engelmann, and Norman 2011; De Bruyn
and Bolton 2008; Fehr and Schmidt 1999). Even in this case, one can discriminate between
the idealistic and pragmatic cultures. Idealistic culture only tests models from this same cul-
ture whereas in the pragmatic culture models from both cultures are tested. For example,

95
96

Konstantinos V. Katsikopoulos

Brandstätter et al. (2006) compared the predictive accuracy of the priority heuristic with that of
cumulative prospect theory. On the other hand, I am not aware of studies within the idealistic
culture where the performance of idealistic models is compared to that of pragmatic models.
This is the sixth and final difference in Table 4.1. We next move from modeling to storytelling.

The two cultures: different stories about people’s bounded rationality

and how to improve it
Explanation and prediction are examples of the ultimate services that a scientific model can
offer. As Morgan and Grüne-Yanoff (2013) argue, however, the intermediate services of models
are just as important. Examples of intermediate services of models are to provide “insights”,
“platforms for further discussion” or “coherent stories” for research to continue (p. 145). In
economics, where models are consumed not just by researchers but also by policy makers and
the public and in fact have the potential to affect people’s behavior, intermediate services such
as stories are particularly important (Tuckett 2011).
According to Morgan (2001, p. 379), a story is “the phenomenon of grasping things together
at this intervening level between complete and exhaustive detail and complete generalization.”
For a given bounded-rationality culture, I take this quote to mean that this culture’s story lies
between the empirical evidence and the formal models of the culture. In other words, I see
a story as an amalgam of evidence and modeling. In my view, the function of a story is to
allow the researchers who produced models, as well as the consumers of the models, including
other researchers, policy makers and the public, to start a conversation about people’s decision
making, to keep the conversation going and to give it new twists now and then. This section
analyzes the conversations of the idealistic and pragmatic cultures.

The story told by the idealistic culture

Not only is the idealistic culture inspired by an omniscient and optimizing decision maker, it
never lets go of her, not really. Even though prospect theory and inequity aversion are meant as
models of bounded rationality, they live in the shadow of unbounded rationality. For example,
prospect theory can, when its parameters are chosen appropriately, be reduced to expected
utility theory and so can inequity aversion be reduced to standard game theory. Furthermore,
idealistic models of bounded rationality are meant to be descriptive (what does a real person
do?) but not normative (what should an ideal person do?), so that whatever researchers have
learned from these models has not changed the good old standard of ideal rationality (Bishop
and Trout 2005).
I argue that the story of the idealistic culture goes like this: People are systematically
behaving irrationally, but because they are in principle able to figure out how to behave ration-
ally, they should keep trying to do so. It is clear that a person who buys this story will end up
as frustrated as Tantalus ever was. This frustration is bound to lead to one of two dysfunctional
behaviors: Either deny the reality of making bad decisions and hide in books about ideal ration-
ality in order to get at least some intellectual solace or acknowledge one’s dire prospects and
surrender to the designs of somebody smarter (which you hope are well-meaning). The first
of these behaviors is often seen in neoclassical economics and the second one in behavioral
economics.
The first point above is that, according to the story of the idealistic culture, people are sys-
tematically behaving irrationally. The empirical basis of the idealistic culture is the heuristics-
and-biases research program (Heukelom 2009): This program has concluded that people

96
97

Bounded rationality: the two cultures

systematically violate axioms of logic and probability which the idealistic culture considers to
be normative. The irrationality story is told in best-selling books for the public with titles such
as Predictably Irrational (Ariely 2008) and has been integrated with startling ease in the columns
and blogs of star commentators such as David Brooks of The New York Times.
It is important to understand why the irrationality story became such a hit. Lopes (1991)
provides an insightful analysis. She points out that until the 1970s, most decision researchers
believed that people were pretty good decision makers (Peterson and Beach 1967). She finds it
implausible that people suddenly started making worse decisions––in fact, a bibliographic ana-
lysis showed that at that time there was similar amounts of empirical support for “good” and
“bad” decision making––and attributes the change to the success of the rhetoric of irrationality.
Lopes argues that it was Tversky and Kahneman, who in a series of articles that culminated
in an authoritative summary in Science (Lopes 1974), managed to turn the beat around. This
article opened up the way for the irrationality message to be spread outside psychology and
notably into economics. At the point this chapter is being written, it is cited more than classic
pieces in economics such as the Theory of Games and Economic Behavior by von Neumann and
Morgenstern (1944). I single out some of the reasons Lopes provides for this rhetorical success
which do not have to do with the truth of the message (a subject of intense disagreement that
is beyond the scope of the present chapter, see Gigerenzer 2007 and Kahneman 2011).
To begin with, the experiments of Tversky and Kahneman are interesting puzzles, not dull
drills. For example, in the Linda problem, participants are given the verbal description of a
woman which suggests that she may be a feminist, and are asked to estimate if it is more prob-
able that she is (1) a bank teller or (2) a bank teller and active in the feminist movement.
Your spontaneous answer is likely to be (2). Tversky and Kahneman argued that (1) is the
correct answer because formally the probability of an event A (Linda is a bank teller) is higher
than the probability of the intersection of two events A and B (Linda is a bank teller and active
in the feminist movement). So, if you have some education in probability or statistics, you find
yourself in the interesting position of having made a mistake, having been able to follow the
reasons for it indeed being one and yet still feel somehow drawn to it. As Gould (1988) put
it, “a little homunculus in my head continues to jump up and down, shouting at me—but she
can’t just be a bank teller; read the description” (emphasis added). Furthermore, as you keep
on reading the article, you see that most people are like you and have made the same mistake.
Fortunately, you can probably convince yourself that you are smarter than most of these people
because you do understand what the mistake is. The authors themselves may have fueled your
reactions by calling people’s decisions “ludicrous” and “self-defeating” (Tversky and Kahneman
1971, pp. 109, 107). As Lopes summarizes, “[These] problems effectively engage interest and
attention while massaging professional egos” (1991, p. 79).
As soon as it has been said that people systematically behave irrationally, the story of the
idealistic culture unfolds quite smoothly. Clearly, the story continues, not all of us are irrational
since some of us did come up with logic and probability and many of us have studied and
mastered these tools. So, there is a job here for whoever can make us more “rational”. As Lopes
puts it, “The idea that people-are-irrational-and-science-has-proved-it is useful propaganda for
anyone who has rationality to sell” (1991, p. 78). The questions then become: What does the
idealistic culture have to sell us? And what about the pragmatic culture?

Nudge or boost?
When it comes to promoting rationality, the idealistic and pragmatic cultures could have
converged to a common story. The two cultures do agree that there is a job that needs to be

97
98

Konstantinos V. Katsikopoulos

done: Human decision making should be supported (note, however, that the two cultures
differ drastically in how they view the quality of people’s decision making based on the avail-
able empirical evidence and how they go about generating empirical evidence; see Gigerenzer
1996, and Kahneman and Tversky 1996). But, as I argue below, each culture has its own story
on how to get the job done and so far they are sticking to it.
As discussed earlier, the empirical part of the idealistic story is that people systematically
behave irrationally. Perhaps because of that, the policy part of this story does not put much faith
in people’s ability to ever become “rational” on their own. So, the story goes, if we want people
to behave rationally, we somehow have to steer them into doing so. In the words of Thaler and
Sunstein (2008), authorities have to nudge people toward “better decisions about health, wealth
and happiness.” For example, legislators can set the default option in one’s driving license so
that people are organ donors and cafeteria owners can rearrange menus so that children are
more likely to eat more vegetables. All the public needs to do is surrender to the well-meaning
designs of those who are smarter.
The pragmatic culture tells a different story. This story is based on a different approach
to gathering empirical evidence on people’s rationality from that of the idealistic culture.
The pragmatic culture is indifferent to testing adherence to axioms. It instead focuses on the
impact of providing people with tools for boosting performance on tasks of practical import-
ance, such as Bayesian reasoning (Fong, Krantz, and Nisbett 1986). An example of such a
task is a medical doctor wanting to know the probability of a woman having breast cancer,
given that she is more than 50 years old, the results of a mammography test, and the inform-
ativeness of the test. An example of a tool for doing this calculation is natural frequency
formats for representing probabilities where conditional probabilities, such as the sensitivity
of mammography (i.e., the probability that a mammography is positive given that a woman
has breast cancer) are replaced by the corresponding joint frequencies. For example, sensi-
tivity can be represented by the statement that out of 100 women with breast cancer, 99 have
a positive mammography. This tool can improve the Bayesian reasoning of professionals (e.g.,
medical doctors as well as judges and lawyers) and laypeople (Hoffrage, Lindsey, Hertwig,
and Gigerenzer 2000).
In other words, the story of the pragmatic culture is that people can indeed learn to behave
rationally. Unlike the case of the idealistic culture, this is not a frustrating message for the public
but an empowering one.
Interestingly, the story of the pragmatic culture may not appear particularly empowering
when one considers the first premise of this culture which is that people do not optimize. But
the story is in fact unexpectedly empowering when we also take into account that empirical
studies, computer simulations and mathematical analyses show that pragmatic models can out-
perform optimization models (Katsikopoulos 2011b). What is required in order to reap the
benefits is that people learn, or are taught, what the right pragmatic model is to use in which
situation. The story of the pragmatic culture is centered on education, or as it is often more
aptly called, boost (Hertwig and Grüne-Yanoff 2017).
Table 4.2 summarizes the above discussion by outlining three key differences of the ideal-
istic and pragmatic cultures in story telling about people’s bounded rationality and how to
improve it.
In order to put the differences between the two cultures of bounded rationality into per-
spective, note that their stories are more similar to each other than they are to the story told
by the culture of unbounded rationality. In this culture, which is prevalent in neoclassical eco-
nomics, modeling is idealistic and the empirical evidence is interpreted as showing that people
behave rationally in the sense of conforming to the axioms of logic and probability, except for

98
99

Bounded rationality: the two cultures

Table 4.2 Three key differences of the idealistic and pragmatic cultures in story telling about people’s
bounded rationality and how to improve it

Bounded rationality cultures Idealistic Pragmatic

differences in storytelling

1. Gist of the story People systematically behave People do well if they learn to
irrationally; they should do use the right tool in the right
better situation
2. Psychological reactions of Frustration; surrender to the Empowerment
the public designs of someone smarter
3. Role of the authorities Nudge Educate

some random violations; and the policy part of the story is to let people engage in free-market
activities and reduce the role of the authorities to activities such as providing incentives.
In 2009, Nature published a news feature by freelance writer Michael Bond, covering both
nudge and education stories. Gigerenzer comes across as a champion of education, whereas
Thaler and Kahneman appear skeptical, saying that “our ability to de-bias people is quite
limited” (Bond, 2009, p. 1191) and that “it takes an enormous amount of practice to change
our intuition” (2009, p. 1192). There is, unfortunately, a standstill and no talk of combining
the two stories.

Conclusion
Scientist and novelist C. P. Snow (1959) lamented the schism between the two cultures of the
sciences and the humanities. The present chapter has the much more modest goal of analyzing
the research on bounded rationality. Bounded rationality would have perhaps pleased Snow as
it exhibits technical as well as story-telling aspects. On the other hand, here I showed some-
thing that could have worried Snow: There exist two distinct cultures of research on bounded
rationality, the idealistic and the pragmatic, and they lead to two very different approaches to
economic theory and policy. Time will tell what will come out of this tension. But if we are
not aware that it exists, we cannot hope to make something good out of it.

Acknowledgments
The author wishes to thank Gerd Gigerenzer, Werner Güth, Wasilios Hariskos, Ralph Hertwig,
Oliver Kirchkamp, Özgür Şimşek, Riccardo Viale, and the participants at workshops at the
Warsaw School of Social Sciences and Humanities (Wroclaw campus) and the University of
Paris (Villa Finaly), for their comments and suggestions.
The author acknowledges permission granted by Taylor and Francis (www.tandfonline.com)
to use material from the previously published article: Katsikopoulos, K. V. (2014). Bounded
rationality: the two cultures, Journal of Economic Methodology, 21(4), 361–374.

References
Ariely, Dan (2008). Predictably Irrational, New York: HarperCollins.
Augier, Mie, and James G. March (2002). A model scholar: Herbert A. Simon (1916–2001), Journal of
Economic Behavior and Organization, 49, 1–17.

99
100

Konstantinos V. Katsikopoulos

Berg, Nathan, and Gerd Gigerenzer (2010). As-if behavioral economics: Neoclassical economics in dis-
guise, History of Economic Ideas, 18, 133–166.
Binmore, Ken, and Avner Shaked (2010). Experimental economics: Where next? Journal of Economic
Behavior and Organization, 73, 87–100.
Bishop, Michael, and J. D. Trout (2005). Epistemology and the Psychology of Human Judgment, Oxford: Oxford
University Press.
Blanco, Mariana, Dirk Engelmann, and Hans Theo Normann (2011). A within-subject analysis of other-
regarding preferences, Games and Economic Behavior, 72, 321–338.
Bond, Michael (2009). Decision-making: Risk school, Nature, 461, 1189–1192.
Brandstätter, Eduard, Gerd Gigerenzer, and Ralph Hertwig, R. (2006). The priority heuristic: making
choices without trade-offs. Psychological Review, 113, 409–432.
Breiman, Leo (2001). Statistical modeling: The two cultures, Statistical Science, 16, 199–231.
Brighton, Henry, and Gerd Gigerenzer (2012). Are rational actor models “rational” outside small worlds?
In Samir Okasha and Ken Binmore (Eds.), Evolution and Rationality: Decisions, Co-operation and Strategic
Behavior (pp. 84–109). Cambridge: Cambridge University Press.
De Bruin, Arnaud, and Gary E. Bolton (2008). Estimating the influence of fairness on bargaining behavior,
Management Science, 54, 1774–1791.
Drechsler, Mareile, Konstantinos V. Katsikopoulos, and Gerd Gigerenzer (2014). Axiomatizing bounded
rationality: The priority heuristic, Theory and Decision, 77, 183–196.
Fehr, Ernst, and Klaus Schmidt (1999). A theory of fairness, competition, and cooperation, Quarterly
Journal of Economics, 114, 817–868.
Fehr, Ernst, and Klaus Schmidt (2010). On inequity aversion: A reply to Binmore and Shaked, Journal of
Economic Behavior and Organization, 73, 101–108.
Fischbacher, Urs, Ralph Hertwig, and Adrian Bruhin (2013). How to model heterogeneity in costly
punishment: Insights from responders' response times, Journal of Behavioral Decision Making, 26,
462–476.
Fong, Geoffrey T., David H. Krantz, and Richard E. Nisbett (1986). The effects of statistical training on
thinking about everyday problems, Cognitive Psychology, 18, 253–292.
Ford, Kevin J., Neal Schmitt, Susan L. Schechtman, Brian H. Hults, and Mary L. Doherty, (1989). Process
tracing methods: Contributions, problems, and neglected research questions, Organizational Behavior
and Human Decision Processes, 43, 75–117.
Friedman, Milton (1953). Essays in Positive Economics, Chicago, IL: University of Chicago Press.
Gigerenzer, Gerd (1996). On narrow norms and vague heuristics: A reply to Kahneman and Tversky,
Psychological Review, 103, 592–596.
Gigerenzer, Gerd (2007). Gut Feelings, London: Viking.
Gigerenzer, Gerd (2008). Rationality for Mortals: How People Cope with Uncertainty, New York: Oxford
University Press.
Gigerenzer, Gerd, Ralph Hertwig . and Thorsten Pachur (Eds.) (2011). Heuristics: The Foundations of
Adaptive Behavior, New York: Oxford University Press.
Gigerenzer, Gerd, and Reinhard Selten (Eds.) (2001). Bounded Rationality: The Adaptive Toolbox,
Cambridge, MA: MIT Press.
Gintis, Herbert (2011). Lecture at the School of Economic, Political and Policy Sciences of the University
of Texas at Dallas.
Gonzalez, Richard, and George Wu (1999). On the shape of the probability weighting function, Cognitive
Psychology, 38, 129–166.
Gould, Stephen J. (1988). The streak of streaks, The New York Review of Books, August 18.
Güth, Werner (1995). On ultimatum bargaining experiments – A personal review, Journal of Economic
Behavior and Organization, 27, 329–344.
Hertwig, Ralph and Till Grüne-Yanoff (2017). Nudging and boosting: Steering or empowering good
decisions, Perspectives on Psychological Science 12, 973–986.
Heukelom, Floris (2009). Kahneman and Tversky and the Making of Behavioral Economics, Amsterdam: Tinbergen
Institute Research Series.
Hoffrage, Ulrich, Samuel Lindsey, Ralph Hertwig, and Gerd Gigerenzer (2000). Communicating statis-
tical information, Science, 290, 2261–2262.
Kahneman, Daniel (2011). Thinking, Fast and Slow, New York: Farrar, Strauss and Giroux.
Kahneman, Daniel, Paul Slovic, and Amos Tversky (1982). Judgment Under Uncertainty: Heuristics and
Biases, Cambridge: Cambridge University Press.

100
101

Bounded rationality: the two cultures

Kahneman, Daniel, and Amos Tversky (1979). Prospect theory: An analysis of decision under risk,
Econometrica, 47, 263–291.
Kahneman, Daniel, and Amos Tversky (1996). On the reality of cognitive illusions: A reply to Gigerenzer’s
critique, Psychological Review, 103, 582–591.
Katsikopoulos, Konstantinos V. (2011a). How to model it? Review of “Cognitive Modeling” (J. R.
Busemeyer and A. Diederich), Journal of Mathematical Psychology, 55, 198–201.
Katsikopoulos, Konstantinos V. (2011b). Psychological heuristics for making inferences: Definition, per-
formance, and the emerging theory and practice, Decision Analysis, 8, 10–29.
Katsikopoulos, Konstantinos V., and Gerd Gigerenzer (2008). One-reason decision-making: Modeling
violations of expected utility theory, Journal of Risk and Uncertainty, 37, 35–56.
Katsikopoulos, Konstantinos V., and Dan Lan (2011). Herbert Simon’s spell on judgment and decision-
making, Judgment and Decision Making, 6, 722–732.
Klein, Gary (2001). The fiction of optimization, In Gerd Gigerenzer and Reinhard Selten (Eds.), Bounded
Rationality: The Adaptive Toolbox, Cambridge, MA: MIT Press, pp. 103–121.
Lee, Cassey (2011). Bounded rationality and the emergence of simplicity amidst complexity, Journal of
Economic Surveys, 25, 507–526.
Lopes, Lola L. (1991). The rhetoric of irrationality, Theory and Psychology, 1, 65–82.
Luce, R. Duncan (1997). Several unresolved conceptual problems of mathematical psychology, Journal of
Mathematical Psychology, 41, 79–87.
Luce, R. Duncan (1999). Where is mathematical modeling in psychology headed? Theory and Psychology,
9, 723–737.
Manzini, Paola, and Marco Mariotti (2007). Sequentially rationalizable choice, American Economic Review,
97, 1824–1839.
Manzini, Paola, and Marco Mariotti (2012). Choice by lexicographic semiorders, Theoretical Economics,
7, 1–23.
Manzini, Paola, and Marco Mariotti (2014). Welfare economics and bounded rationality, Journal of
Economic Methodology, 21, 343–360.
Merriam-Webster Online Dictionary (2013). Definition of “pragmatic”. Available at: www.merriam-
webster.com/dictionary/pragmatic
Morgan, Mary S. (2001). Models, stories and the economic world, Journal of Economic Methodology, 8,
361–384.
Morgan, Mary S., and Till Grüne-Yanoff (2013). Modeling practices in the social and human sciences. An
interdisciplinary exchange, Perspectives on Science, 21, 143–156.
Musgrave, Alan (1974). Logical and historical theories of confirmation, British Journal of Philosophy and
Science, 25, 1–23.
Peterson, Cameron R., and Lee Roy Beach (1967). Man as an intuitive statistician, Psychological Bulletin,
68, 29–46.
Rubinstein, Ariel S. (1998). Modeling Bounded Rationality, Cambridge, MA: MIT Press.
Savage, Leonard J. (1954). The Foundations of Statistics, New York: John Wiley and Sons.
Selten, Reinhard (2001). What is bounded rationality? In Gerd Gigerenzer and Reinhard Selten (Eds.),
Bounded Rationality: The Adaptive Toolbox, Cambridge, MA: MIT Press, pp. 13–36.
Simon, Herbert A. (1947). Administrative Behavior, New York: The Free Press.
Simon, Herbert A. (1955). A behavioral model of rational choice. Quarterly Journal of Economics, 69,
99–118.
Simon, Herbert A. (1956). Rational choice and the structure of environments. Psychological Review, 63,
129–138.
Snow, Charles P. (1959). The Two Cultures, London: Cambridge University Press.
Starmer, Chris (2000). Developments in non-expected utility theory: The hunt for a descriptive theory of
choice under risk, Journal of Economic Literature, 38, 332–382.
Tanner Jr., Wilson P., and John A. Swets (1954). A decision-making theory of signal detection, Psychological
Review, 61, 401–409.
Thaler, Richard H., and Sunstein, Cass R. (2008). Nudge, London: Penguin.
Tuckett, David (2011). Minding the Markets: An Emotional Finance View of Financial Instability,
New York: Palgrave Macmillan.
Tversky, Amos (1972). Elimination by aspects: A theory of choice, Psychological Review, 79, 281–299.
Tversky, Amos, and Daniel Kahneman (1971). Belief in the law of small numbers, Psychological Bulletin,
76, 105–110.

101
102

Konstantinos V. Katsikopoulos

Tversky, Amos, and Daniel Kahneman (1974). Heuristics and biases: Judgment under uncertainty, Science,
185, 1124–1130.
Tversky, Amos, and Daniel Kahneman (1992). Advances in prospect theory: Cumulative representation of
uncertainty, Journal of Risk and Uncertainty, 5, 297–323.
von Neumann, John, and Oskar Morgenstern (1944). Theory of Games and Economic Behavior, Princeton,
NJ: Princeton University Press.
Wakker, Peter, and Amos Tversky (1993). An axiomatization of cumulative prospect-theory, Journal of Risk
and Uncertainty, 7, 147–175.

102
103

5
SEEKING RATIONALITY
$500 bills and perceptual obviousness

Teppo Felin and Mia Felin

Introduction
The value of a hypothetical $500 bill is obvious. And because it is obvious, “there are no $500
bills on the sidewalk”—if there were, they would already be picked up (Akerlof and Yellen,
1985: 708–709; also see Frank and Bernanke, 2007).
The $500-bill “axiom” provides a useful—albeit informal—shorthand for explaining market
efficiency and rational expectations. Metaphorically, everything in the economy could be seen
like the proverbial $500 bill. The value of all assets is obvious: everything is correctly labeled,
priced and put to its best use (Muth, 1961; cf. Arrow, 1986).1 This is because “all agents inside
the model, the econometrician, and God share the same model” (as discussed by Thomas
Sargent, see Evans and Honkapohja, 2005: 566; cf. Buchanan, 1959; Frydman and Phelps,
2013). Agents are omniscient and markets are at equilibrium—put differently, objects signal
their own value. This is because there is no heterogeneity in perception or expectations and
therefore there are no above-normal, economic profits to be had. If there were, they would
largely be a function of luck and thus be quickly competed away (Alchian, 1950; Denrell, Fang,
and Winter, 2003).
Behavioral economics has stepped in to fill this vacuum and pointed out many instances
where seemingly obvious value is systematically missed or left on the table, whether at the more
micro level of judgment, perception, and decision making or at the more macro level of markets
(Kahneman, 2003; Thaler, 2015).2 To put this in terms of the above $500-bill axiom: eco-
nomic agents may miss valuable and obvious things, like the $500 bill, because they are blind or
bounded in some fashion. Bounded rationality thus provides “an alternative to classical omnis-
cient rationality” (Simon, 1979: 357; cf. Kahneman, 2003), helping us understand why we see
and attend to some things, but manage to miss and be blind to many other things. Behavioral
economics further builds on these ideas (for a recent review, see Lieder and Griffiths, 2019),
linking this work to empirical findings of blindness, priming, and bias from psychology and
cognitive science (e.g., Bargh and Chartrand, 1999; Simons and Chabris, 1999). This research
emphasizes the “prevalence of bias in human judgment” and more generally how humans “can
be blind to the obvious” (Kahneman, 2011). Individuals and markets (in the aggregate) seem
to routinely violate the axioms of rationality, as obvious things or sources of value are missed.
Because of the pervasiveness of these biases and varied forms of boundedness, opportunities

103
104

Teppo Felin and Mia Felin

for more optimal decision making and nudging abound (Stanovich et al., 2016; Sunstein and
Thaler, 2008; Thaler, 2016).
In this chapter we discuss the outlines of an alternative to both neoclassical and (certain)
behavioral conceptions of rationality. Just as Thaler argues “that models of rational behavior
became standard because they were the easiest to solve” (2015: 1579), so we argue that behav-
ioral models of rationality have now become standard because omniscience is easy to prove
wrong.3 In short, it is easy to show that people miss obvious things. And just as many economic
models build on “bad psychology” (Clark, 1918: 4; Thaler, 2015: 1579), we argue that (some)
behavioral models build on a problematic view of perception. We show how certain versions
of the bounded rationality concept are built on specific perceptual assumptions: empirical
findings of perceptual blindness and deviations from omniscience and full rationality. We discuss
problems with the focus on perceptual blindness and value-related obviousness in behavioral
economics, specifically in the context of both rational expectations and bounded rationality.
We first provide a brief overview of the bounded rationality concept itself, and then discuss its
foundations in the psychology of perception (see Kahneman, 2003; Simon, 1955; for a recent
discussion see Chater et al., 2018). We revisit key insights from biology and the psychology of
perception. We discuss two key issues: (1) perception and the organism-environment relation-
ship; and (2) seeking or “looking for” rationality. We conclude with a discussion of the eco-
nomic implications of our argument, as these relate to perception, belief heterogeneity and the
origins of value in markets.

Economics, bounded rationality, and perception

Herbert Simon introduced bounded rationality as a counterweight to the full rationality
presumed by many in economics. His aim was

to replace the global rationality of economic man with a kind of rational behavior
that is compatible with the access to information and the computational capacities that are
actually possessed by organisms, including man, in the kind of environments in which
such organisms exist”.
1955: 99

In short, Simon introduced a psychologically more realistic conception of rationality, behavior,

and human decision making. Simon’s concept of bounded rationality has become a central
building block across a wide swath of disciplines, including economics, management, psych-
ology and cognitive science (Chater et al., 2018).
What is important to our arguments in this chapter is that Simon anchored his model of
rationality on certain “psychological theories of perception and cognition” (Simon, 1956: 138;
cf. Felin, Koenderink, and Krueger, 2017). Simon frequently used the example of an organism
searching for something of value, like food, in their environments. An organism naturally is not
fully rational or aware in the sense that it knows where the most optimal sources of food might
be in its environment (cf. Todd and Gigerenzer, 2003). The organism instead searches for food
locally within its immediate, visible vicinity. Search activity is driven by the organism’s “per-
ceptual apparatus” and its “length and range of vision” (Simon, 1956: 130). The search is never
complete, exhaustive, or perfectly optimal—rather, it is bounded. Organisms satisfice based on
more proximate aspiration levels, foregoing full optimality and perfect rationality.
The visual and perceptual emphasis associated with Simon’s concept of bounded rationality
is further reinforced in the work of Daniel Kahneman (cf. Chater et al., 2018).4 As noted by

104
105

Seeking rationality

Kahneman in his Nobel speech, behavioral economics “[relies] extensively on visual analo-
gies” and visual perception (2003: 1450–1453; also see Kahneman, 2011). While Kahneman
builds on Simon’s program of research, his focus is even more strongly on the perceptual rather
than computational aspects of bounded rationality. He argues that the behavior of organisms
is “not guided by what [agents] are able to compute”—central to Simon’s approach—“but by
what they happen to see at a given moment” (Kahneman, 2003: 1469). The argument is that
the human visual system is marred by varied forms of blindness and bias, as illustrated by visual
illusions, perceptual priming and concepts like inattentional or change blindness (Simons and
Chabris, 1999; also see Bargh and Chartrand, 1999). These illusions and forms of blindness
also provide the central evidence of the popular press books published by scholars in the cog-
nitive and behavioral sciences (e.g., Chabris and Simons, 2010; Chater, 2018; Kahneman,
2011). The touchstone of this program of research is the suboptimality of much decision
making and pervasiveness of bias and blindness. Metaphorically, humans routinely miss (or
ignore or don’t see) obvious things, like proverbial $500 bills, right in front of them. This has
led to a large program of research to try to nudge decision makers toward better judgments and
more optimal choices (e.g., Sunstein and Thaler, 2008; Thaler, 2015; Sunstein, Chapter 38
in this volume).
Before discussing the perceptual aspects of this argument, it’s worth recognizing that the
behavioral program of research has recently been challenged by scholars who have failed to
replicate many of the key findings that purport to provide evidence of widespread human
irrationality and bias. A significant portion of the empirical studies and evidence used by
Kahneman in his bestselling Thinking, Fast and Slow has failed to be replicated, or the work
has been theoretically questioned in different ways. This includes empirical and theoretical
challenges to a number of key biases and fallacies, including the hot hand fallacy (Miller and
Sanjurjo, 2018), social and perceptual priming (Crandall and Sherman, 2016; Ramscar, 2016),
dual process theory and the idea of System 1 and System 2 thinking (Melnikoff and Bargh,
2018), inattentional and change blindness (Chater et al., 2018; Felin et al., 2019), and so forth.
Furthermore, there is also a long-standing argument about whether biases and blindness are
rational heuristics, as illustrated by the rationality of the anchoring bias (Lieder et al., 2018) or
the rationality of inattention (Matejka and McKay, 2015). This work on heuristics has been
pioneered and further developed by Gerd Gigerenzer and his colleagues over the past decades
(Gigerenzer and Todd, 1999). Furthermore, the visual illusions that are frequently referenced
by behavioral scholars as metaphorical examples of blindness—extended to the context of
rationality (see Kahneman, 2003)—have been shown by perception scholars not to be illusions
at all (Braddick, 2018; Rogers, 2014). That said, scholars of course are actively debating these
issues, and thus it is hard to point to conclusive answers (see van Buren and Scholl, 2018;
Rogers, 2019). However, while these debates will undoubtedly continue, the problem is that
the evidence for pervasive human bias and blindness is often couched as established, scientific
facts to external audiences. For example, related to perceptual priming (a literature currently
under severe empirical scrutiny), Kahneman argues that “disbelief is not an option. The results
are not made up, nor are they statistical flukes. You have no choice but to accept that the major
conclusions of these studies are true” (2011: 57). But in retrospect, it appears that many of
these seemingly solid, empirical findings are in fact not as clear-cut and conclusive as previously
thought (Felin et al., 2019; for more, see Chater et al., 2018).
In this chapter we set these issues aside and specifically discuss how assumptions about per-
ception are essential for the rationality literature, focusing both on rational expectations and
particularly the concept of bounded rationality. We argue that the underlying assumptions
about perception (and what should be visually obvious) need to be carefully revisited, as these

105
106

Teppo Felin and Mia Felin

apply to varied forms of bounded rationality, judgment and decision making. Thus we make
the perceptual assumptions of existing work more explicit and provide the broad outlines of an
alternative.

What do we see and why? What’s obvious?

Perhaps the simplest question regarding perception is, what do organisms (humans included)
see and why? Or put differently, what is perceptually obvious and why? Humans and other
organisms continually encounter stimuli and objects that make up or constitute their visual
scene, environment or situation.5 Some things in the visual scene (somehow) become salient
and obvious, while other things remain in the background, perhaps outside awareness. And to
complicate things even further, different organisms—or even two different people or organisms
looking at the very same visual scene—may also differ in what they perceive. What is it, then,
that we see and become aware of, and why?
A simple way to answer the question of “what do we see?” is to point to the physical
objects that actually constitute any visual scene itself: the things in front of and surrounding
an organism. To provide a practical illustration, you as the reader of this chapter are right now
encountering a visual scene. This sentence and the book chapter you are reading are part of
your visual scene. But if you lift up your gaze and look around, you will also see other things.
If you are in an office, there is likely to be a desk close by, perhaps a lamp or different forms of
lighting, some number of chairs, maybe a whiteboard, perhaps art or office decorations on the
wall, or some number of books on the shelf. And if you direct your gaze out the window, you
might see any number of other things: people walking by on the street, nearby buildings, trees
and vegetation, or the sky. In short, your immediate surroundings are teeming with potential
things to see, many of them obvious. But the central question here is: why—among the many
things right in front of (and around) us—do we see or become aware of certain things, and
not others?
Before addressing this question, a simple response to the “what do we see?” question is: we
see what is actually there. At first glance this approach provides a very straightforward (and in
fact prominent) account of perception. It suggests that there should be no mystery when it
comes to perception. Perception simply records the physical things in front of (and around) us: a
chair, books, computer screen, clouds, and so forth. As put by Marr in his book Vision, vision is
a “true description of what is there” (1982: 29–30; cf. Hoffman et al., 2015). Perception then is
seen as a catalogue of actual objects, things and stimuli in front of us. This in fact is the implicit
assumption of wide swaths of vision science, such as ideal observer theory (Geisler, 2011), psy-
chophysics (Kahneman, 1966; for a recent review, see Kingdom and Prins, 2016), and Bayesian
approaches to vision (Kersten et al., 2004; Yuille and Kersten, 2006). More generally, this ver-
idical approach to perception is the foundation of many (if not most) contemporary theories of
vision and perception (for a review, see Hoffman et al., 2015).
One problem with this veridical, catalogue, or camera-view of perception is that it is
impractical (Chater et al., 2018). Listing or capturing all of these stimuli and objects not only
is not necessary or useful, but also impractical and impossible. This idea in fact metaphorically
provides the visual equivalent of bounded rationality, which is that listing and accounting for all
the obvious visual things in front of us would both take too much time and also defeat the very
purpose of perception (or rationality) itself. Or put differently, everything is not recognizable
or computable or listable, given limited resources or time. This then might yield a concep-
tion of perception as a defective camera, compared to a more omniscient ideal that somehow
captures everything. But beyond this problem, a more essential issue is that the camera-view of

106
107

Seeking rationality

perception doesn’t tell us which of the many possible things in our visual scenes are salient and
why. That is, why do we become aware of certain objects or stimuli and not others? Again, the
problem is that visual scenes are teeming with “things:” potential objects, stimuli and cues. The
fact that we miss some things, even obvious ones, might of course be the basis of calling humans
blind (Simons and Chabris, 1999). Or more productively, it might be the basis of developing a
theory of why we become aware of some things and not others.
This is where Kahneman (2003) and the behavioral program of research build on psy-
chophysics, which focuses on salience as a function of the actual nature of particular stimuli
or cues themselves. What is seen by organisms is tied to the actual, physical nature of envir-
onments confronted by organisms and agents. As put by Kahneman, “the impressions that
become accessible in any particular situation are mainly determined, of course, by the actual
properties of the object of judgment,” and “physical salience [of objects and environments]
determines accessibility” (2003: 1453, emphasis added). Importantly, some stimuli, impressions
and objects are more readily accessible than others. Here Kahneman argues that these percep-
tual objects become salient and accessible because they have particular characteristics, such as
their “size, distance, and loudness” (2003: 1453). To simplify and illustrate: a large, proximate,
and “loud” stimulus or object is more readily seen, and thus obvious, compared to a small, dis-
tant, and “quiet” object. Kahneman calls these “natural assessments” (Kahneman, 2003; also
see Kahneman and Frederick, 2002; Tversky and Kahneman, 1983). Visually salient objects
are more readily and quickly attended to because of these stimulus characteristics. Stimulus
characteristics make certain things “stand out” and become more prominent and thus visually
accessible. The list of stimulus characteristics (beyond size, etc.) also includes “more abstract
properties such as similarity, causal propensity, surprisingness, affective valence, and mood”
(Kahneman, 2003, p. 1453). Thus this program of research offers some practical predictions
arguing that certain, actual characteristics of stimuli (again, size being a particularly salient one)
determine whether we see them or not.
Though the heuristics program of research disagrees with Kahneman about whether biases
in fact are rational heuristics (see Gigerenzer and Goldstein, 1996; Gigerenzer and Todd, 1999),
there are surprising commonalities between these two behavioral programs of research. Namely,
the emphasis in the heuristics program of research is also on perception and the nature of stimuli
themselves. The central construct of the heuristics program of research is an environment or
perceptual “cue.” Thus in their summary of the heuristics stream of research, Gigerenzer and
Gaissmaier (2011) emphasize varied factors related to environmental and perceptual cues. These
include “the number of cues,” “cue weighting,” “the correlation of cues,” “cue validity,” “cue
addition,” “the search through cues,” “positive cues,” “cue value,” “cue ordering,” “cue redun-
dancy,” “cue correlation,” “cue integration,” “cue combination,” “ cue favoring,” and so forth
(also see Gigerenzer and Goldstein, 1996).
Cues within the heuristics program of Gigerenzer then serve the equivalent function to
Kahneman’s stimulus characteristics and natural assessments. Thus in many ways, the ecological
rationality sub-area of behavioral economics—pioneered by Gigerenzer and others—links with
the areas of psychophysics and inverse optics which focus on perception as a function of stimulus
intensity, similarity, repetition, exposure, thresholds, and signal detection (see Kingdom and
Prins, 2016). This is also the basis of Bayesian views of perception (Knill and Richard, 1996),
which have also had a strong influence on the bounded rationality and cognition literatures
(e.g., Chater et al., 2010; Oaksford and Chater, 2007). And a similar type of emphasis on stimuli
and cues can also readily be found in other areas of psychology and cognitive science, such as
the situation construal and situation perception literatures (e.g., Rauthmann et al., 2014; also
see Chater et al., 2018; Funder, 2016). In sum, the behavioral program in psychology and

107
108

Teppo Felin and Mia Felin

economics implicitly makes the assumption that the nature of stimuli (Kahneman, 2003) or
the nature of cues (Gigerenzer and Gaissmaier, 2011) determines perception and obviousness.
In the case of Kahneman, what we see is a function of (for example) how large or proximate
something is. And in the case of Gigerenzer and others, cues are simply given and their varied
structure (ordering, aggregation, combination, etc.) are essential for salience and perceptual
awareness.
In an important sense these conceptions of rationality don’t—ironically, similar to the rational
expectation model of economics—meaningfully require any assumptions about the organism
itself. That is, the architecture of cognition is general (Anderson, 2013). The models assume an
objective conception of the environment (which has particular visual or statistical features), and
organisms then have some kind of delimited, bounded, or biased view of their environment, or
a view that is proximate and good enough to satisfice. This environmental focus is also captured
by Simon who argued that “an ant [or human being], viewed as a behaving system, is quite
simple. The apparent complexity of its behavior over time is largely a reflection of the com-
plexity of the environment in which it finds itself ” (1981: 63–65; also see Anderson, 2013). The
arguments, then, are species-independent and universal (Simon, 1980), as suggested by Simon’s
reference to organisms in general, whether we are talking of ants or human beings. This also can
be related to behavioral models in psychology that similarly encouraged scientists to move away
from studying organisms but environments instead. Schwartz, for example, argued that “if you
want to know why someone did something, do not ask. Analyze the person’s immediate envir-
onment until you find the reward” (1978: 6). Furthermore, the universality of these arguments
is reflected in how broadly these models of bounded rationality are applied, across different
species and even computers. Simon argued that “since Homo Sapiens shares some important psy-
chological invariants with certain nonbiological systems—the computers—I shall make frequent
reference to them also” (1990: 3, emphasis added). Thus the notion of bounded rationality is
more generally applied to any form of search and foraging (Abbott et al., 2015; Fawcett et al.,
2014; Gershman et al., 2015).
An important wrinkle with the behavioral program of research is the fact that humans and
other organisms seem to be visually blind to things that should be readily obvious—obvious
given the nature of the stimuli or cues. That is, something that is large, and right in front of
an organism, directly in its visual field, should be obvious (as predicted by the theory). But
somehow organisms, humans included, are seemingly blind to any number of large things. Our
visual scenes feature various obvious things—obvious in terms of their characteristics—but
somehow we nonetheless appear to miss them. The highly-cited, classic example of this is
Simons and Chabris’s (1999) study of inattentional blindness, where a person dressed in a gorilla
suit walks across a visual scene and many experimental subjects never see the gorilla. This is sur-
prising, because the gorilla has many of the hallmarks of a stimulus or cue that in fact ought to
make it highly salient, including the fact that it is large. Of course, in some sense these types of
findings question the very idea that perception is a function of stimulus characteristics (cf. Felin
et al., 2019). Next we provide an alternative way of viewing these types of findings, which we
think can account for them, and also shed light on the question of bounded rationality in the
context of judgment and decision making in economics.
To summarize and briefly return to the subheading of this section of our chapter—what
do we see and why?—the predominant emphasis in the behavioral program has been on the
nature of stimuli and cues in the immediate environment of organisms. This argument certainly
is, in some ways, an improvement over rational expectations, which, in effect, presume visual
omniscience. The bounded rationality literature seems to have provided a psychologically more
realistic conception of perception and rationality. But next we seek to suggest some extensions

108
109

Seeking rationality

and alternatives. In short, we argue that the question of “what do organisms see and why” can
be answered differently and that this has important implications for how we think about human
rationality and decision making in economic and social settings.

Perception and rationality

Insights from psychology and biology
In this section we consider some alternative ways to understand perception and rationality. We
specifically discuss two key points: (1) perception and the organism-environment relationship;
and (2) seeking or “looking for” rationality.
Before proceeding, it is worth noting that that we strongly concur with Gigerenzer and
Selten’s notion that “visions of rationality do not respect disciplinary boundaries” (2001: 1).
This interdisciplinarity certainly was evident in the pioneering work of Herbert Simon, whose
oeuvre included everything from psychology, management to computer science and beyond.
And, of course, behavioral economics is strongly anchored on Kahneman’s (2003) emphasis
on and contributions in psychophysics (cf. Kahneman, 1966)—along with providing extensive
links to psychology (Thaler, 2015)—and the attendant implications of all this for judgment and
decision making. Thus we also draw insights from other literatures, specifically from the psych-
ology of perception and biology.

Perception and the organism-environment relationship

In the literature on bounded rationality there is a strong emphasis on the nature of environments
and their role in shaping or determining behavior. The argument is that we can “discover, by a
careful examination of some of the fundamental structural characteristics of the environment...
the mechanisms used in decision making” (Simon, 1956: 130). This focus on the environment
is reflected in the aforementioned comment by Simon, namely, that the “apparent complexity
of [organism] behavior over time is largely a reflection of the complexity of the environment in
which it finds itself ” (1969: 65–66; also see Simon, 1990). This is also evident in the analogy
between visual scenes and environments, the idea that agent behavior is guided “by what they
happen to see at a given moment” (Kahneman, 2003: 1469). For example, Kahneman exten-
sively uses evidence from perceptual priming to make this point (2011: 52–68), and it is this
literature that argues that “the entire environment-perception-behavior sequence is automatic,
with no role played by conscious choice in producing the behavior” (Bargh and Chartrand,
1999: 466). The emphasis on the environment is equally explicit in the literature on ecological
rationality, though the emphasis is somewhat different. Gigerenzer and Gaissmaier (2011) argue
that environments (and their structures) are characterized by specific factors: the uncertainty,
redundancy, sample size and variability of cues. The upshot is that environments are treated as
objective and species-general (for a recent discussion, see Chater et al., 2018), thus leading to
universal models that are said to account for rationality across species (cf. Gershman et al., 2015;
Simon, 1990).
From a biological point of view, the notion that environments are general and can be object-
ively characterized is problematic. This is because cues and stimuli are specific to organisms
(Koenderink, 2014). That is, what an organism attends to (or as we’ll discuss, looks for) depends
on the nature of the organism itself. This creates a confound in the organism-environment
distinction. Certainly, scholars are likely to admit that organisms differ, as do their respective
environments. But the problem is that these differences have not been meaningfully articulated

109
110

Teppo Felin and Mia Felin

or discussed (Chater et al., 2018).6 And this species-specificity also fundamentally changes how
we think about perception, and by extension, rationality.
The fields of ethology and comparative biology offer the best evidence for the fact that
environments are species-specific rather than general (Tinbergen, 1963; for a recent review, see
Burkhardt, 2018). Each organism has its own, unique “Umwelt” (surrounding world), where
awareness and perception are a function of the nature of the organism itself (Uexküll, 2010). To
provide a specific example, frogs may not see a juicy cricket or locust (their food) even if it is
directly in front of them (Ewert, 1987). Thus an analysis of what ought to be salient or valuable
to an organism based on any general stimulus characteristics will not tell us what it is actually
aware of. We might of course label the frog blind or biased, but this merely creates a black box
rather than meaningfully explaining what the frog sees and why. Thus the a priori selection of
what should be obvious—in the presence of any number of other stimuli in a scene—cannot
generate a scientific explanation. It turns out that frogs are visually attuned to movement and
thus will snap as soon as the cricket or locust jumps. This highlights that what “stimulates”—or
which cues become salient and naturally assessed (cf. Kahneman, 2003)—has less to do with the
stimulus itself (or the objective or general “amount” of some cue (Gigerenzer and Gaissmaier,
2011). Rather, awareness of particular stimuli has more to do with the nature of the organism
(Ewert, 1987). Only those stimuli or cues that are species-specific “light up,” or are processed
and attended to, while many other things are ignored.7
The notion that stimuli are specific to organisms changes the way we need to study the
organism-environment relationship. It means that the mechanisms behind the organism-
environment interaction are unique to each species, thus raising questions about species-
general models (Chater et al., 2018), including extensions into the domain of bounded
rationality.8 This work was pioneered by ethologists, comparative biologists, like Niko
Tinbergen and Konrad Lorenz, whose work was linked to Jakob von Uexküll. Uexküll
metaphorically conceived of the species-specific environment as “a soap bubble around
each creature to represent its own world, filled with the perception which it alone knows”
(2010: 117). He argued that “every animal is surrounded with different things, the dog
is surrounded by dog things and the dragonfly is surrounded by dragonfly things”—that
is, “each environment forms a self-enclosed unit, which is governed in all its parts by its
meaning for the subject” (2010: 5). These biological models didn’t somehow disagree with
broader evolutionary processes that might impact species at the level of populations. But this
literature was highly attuned to the more proximate, immediate considerations that shape
species behavior and perception (Tinbergen, 1963), and thus also provide useful insights
for human settings. Thus the goal here isn’t to engage in any form of “species chauvinism”
(Winter, 2012)—in fact, quite the opposite—rather to merely point out species-specific
factors that impact perception.

Seeking or “looking for” rationality

As we’ve discussed above, perceptual relevance and meaning for the behavioral program are
driven by environmental characteristics, by the inherent and objective nature of objects and
stimuli. Our suggested alternative focuses on the relevance and meaning that organisms them-
selves bring to encounters with environments. Perception, in some sense, is driven by what an
organism has “in mind,” what it considers relevant and meaningful for specific purposes.9 Not
only are environments specific, as discussed above, but organisms have species-specific factors
that direct their perception and behavior toward certain ends and toward the selection of cer-
tain stimuli.

110
111

Seeking rationality

A powerful way to understand organism awareness and perception is by studying what the
organism is searching or looking for. This “looking for”-activity is guided by a species-specific
“Suchbild” (Uexküll, 2010; cf. Chater et al., 2018), a German word for search or seek image.
Search images are the equivalent of what an organism has in mind, in terms of an “answer” that
the organisms is looking for to satisfy its search. Frogs, for example, when hungry, are looking
for certain, line-like features that move in their environment. Their behavior is motivated and
directed by this search-image, and awareness is directed toward highly specific things (or what
might be called answers). This “looking for”-activity comes at the expense of any number of
other, obvious things that an organism might have right in front of it.
In the case of humans, the relevant search image is given by what is in the head or mind
of the human actor when encountering a visual scene or situation. If you are searching for
your keys, you are keenly attuned to key-like stimuli or features. Relevance and meaning are
provided by the question that you have in mind, and the task that you are engaged in. This
Suchbild or question might be seen as the most basic form of a heuristic. As noted by Polanyi,
“the simplest heuristic effort is to search for an object you have mislaid” (1957: 89). Note that
this “search for”-heuristic is fundamentally different from the heuristics that emphasize, say, the
nature of stimuli themselves or the number of cues (Gigerenzer and Gaissmaier, 2011).10 When
scanning a room for your keys, you are ignoring vast numbers of stimuli (including obvious
ones) and selectively directing awareness toward possibly relevant cues based on the “key search
image” you have in mind. Note that the keys are not somehow inherently visually salient, as
they are small and thus do not attract attention or awareness due to their size (as suggested by
psychophysics). In searching for keys—or anything else for that matter—we ignore an indef-
inite number of other, perhaps obvious and large, things around us. We also do not individually
“attend to” each item in our visual scene to decide whether it is the key or not. Rather, we
only focus on and select key-like visual stimuli. We concentrate on the problem, question and
task at hand: searching for keys. This type of motivated searching or “looking for”-activity is
impossible to account for from the perspective of psychophysics or a world-to-mind oriented
view of perception. In psychophysics there is no meaning or relevance (Felin, Koenderink,
and Krueger, 2017; Koenderink, 2012), there are only objective stimuli and cue characteristics
(Chater et al., 2018; Gigerenzer and Gaissmaier, 2011).
Any number of empirical findings from the cognitive sciences can readily be re-interpreted
with this Suchbild-oriented or “looking for”-lens, especially any work that purports to provide
evidence of human blindness (e.g., Simons, 1999); Simons and Rensink, 2005). That is, the
reason that humans miss large changes or objects in their visual scene is, simply, because they
have something else in mind. To put this colloquially: humans miss certain things because they
are looking for other things. The problem with the empirical studies of human blindness is that
they often deliberately distract experimental subjects (asking them to engage in some irrelevant
task or answer a particular question), and then point out something that is large and blatantly
obvious but missed, such as a gorilla (Felin, Felin, Krueger, and Koenderink, 2019). In other
words, in these perceptual experiments, humans are effectively given a question or Suchbild of
what to look for (whether counting basketball passes or engaging in other tasks), which distracts
them, and therefore they miss large changes or objects. Visual scenes, of course, are teeming
with many blatantly obvious things. But no progress will be made in understanding perception
and rationality without focusing on the Suchbilds that direct human perception, or without a
focus on the things that human actors are looking for.
What humans have “in mind” can be accounted for in many ways. Perception might be
directed by a problem or question, such as the aforementioned example of looking for one’s
keys. Or perception might be directed by a task, such as getting to some location (or counting

111
112

Teppo Felin and Mia Felin

basketball passes (Simons and Chabris, 1999). But again, it’s hard to argue that people are “blind
to the obvious” (Kahneman, 2011: 23–24), as visual scenes are teeming with obvious things.
And perhaps most relevant to the economic context (discussed further in the next section) is the
idea that perception might be directed by a hypothesis or theory that a human has in mind. This
is aptly captured by Popper who argued that “we learn only from our hypotheses what kind
of observations we ought to make: whereto we ought to direct our attention: wherein to take
interest” (1966: 346). This also goes for science: that is, salience is given by what we are the-
oretically looking for. As noted by Einstein, “whether you can observe a thing or not depends
on the theory which you use. It is the theory which decides what can be observed” (Polanyi,
1971: 604). Thus whether we are talking about the labs of behavioral economics, or encounters
with visual scenes in the real world, we have implicit questions or theories about what these
respective situations or scenes are about, and these then direct our perception and awareness.
In all, the upshot of this discussion is that organism- and theory-dependent factors structure
awareness and perception, with important implications for how we think about rationality,
judgment and decision making in economic and social settings as well.

Rationality and the perception of value

Opportunities and caveats
Next we discuss the economic implications of the arguments above. We also provide some
important caveats. Our overall goal with this concluding section is to link the aforementioned
discussion of perception and obviousness to central questions of rationality and belief hetero-
geneity in economic settings, as well as the idea of economic value. We make two points.
First, axiomatic assumptions about the human mind are important. As put by Simon,
“nothing is more fundamental in setting our research agenda and informing our research
methods than our view of the nature of the human beings whose behavior we are studying”
(1985: 303). Thus if our theoretical “priors” about human nature are largely focused on the
“prevalence of bias in human judgment” (Kahneman, 2011), then we are likely to simply
focus on this, at the expense of other things. But the cost of this is that we will likely miss
many other, more positive aspects of rationality and the human mind. Thus the psychology
that behavioral scholars have imported into economics has been relatively one-sided—perhaps
deliberately “putting in a stake” at the very opposite extreme from omniscience: rampant bias
and blindness. The concern is that these bias-oriented priors—a form of scientific confirm-
ation bias—will lead to the construction of studies that continue to show further instances of
bias and blindness, deviations from omniscience. But as mentioned at the outset, providing
evidence of shortfalls from omniscience is all too easy and convenient (cf. Thaler, 2015).
Much of the evidence for bias may reflect a bias on the part of scholars toward fun, surprising
and easy studies that prove irrationality (Krueger and Funder, 2004). This work has also led to
a widespread public perception that bias and blindness are rampant problems and key features
of the human mind, as suggested by popular books written by cognitive scientists (e.g., Chater,
2018; Chabris and Simons, 2010). Now, of course, humans make mistakes and errors. But
the obsession with those errors has led to a rather one-sided sampling of psychology in the
context of economics. Furthermore, the focus on biases and irrationality will not allow us to
account for the fact that we live in the best of times, at least when it comes to any number
of objective, measurable dimensions, including radical declines in poverty across the globe,
exponential growth in human knowledge and staggering technological and economic progress
(e.g., Pinker, 2018).

112
113

Seeking rationality

The behavioral program sees itself as an heir to Adam Smith’s program of research in eco-
nomics, in reintroducing a more realistic version of psychology to economics (Thaler, 2016;
also see Ashraf, Camerer, and Loewenstein, 2005). But it’s hard to see how the rich ways in
which Smith delved “into the sentiments and mind of actors” can be squared with the strong
focus on bias and irrationality. A careful re-reading of The Theory of Moral Sentiments or The
Wealth of Nations shows that Smith’s program of research was far more expansive and positive
(Rothschild, 2013). His assumptions about human nature and the mind certainly didn’t suggest
any form of omniscience on the part of economic agents, as behavioral scholars rightly have
pointed out. But Adam Smith certainly was not as error-oriented when discussing human
nature and the human mind. In fact, he sought to endow economic actors with the same theor-
etical and creative capacities that we as scientists have. Adam Smith’s program of research might
be summarized as an effort to develop—as put by economic historian Emma Rothschild—“a
theory of people with theories” (2013: 157). This suggests a radically different agenda from
one that focuses on the “prevalence of bias in human judgment” (Kahneman, 2011; Thaler,
2015). This type of sentiment was also channeled by Edith Penrose in American Economic Review
when she argued: “For the life of me I can’t see why it is reasonable (on grounds other than
professional pride) to endow the economist with this unreasonable degree of omniscience and
prescience and not entrepreneurs” (1952: 813). Of course, no one is arguing that omniscience
is psychologically or cognitively a defensible view. But allowing human agents a modicum of
rationality, of a very specific type, given the uncertain and fast-paced situations they find them-
selves in, provides a useful way forward.
Thus an important next step is to develop models that assume that the human agents we
study might in fact have some of the same theoretical capacities that we scientists do. Namely,
humans have the ability to theorize, to ask questions, to pose and solve problems. There are
major strands of psychology to support this view. For example, the work of William James
(1912) is filled with psychological insights about rationality and how individuals navigate
uncertain environments and the role that their beliefs play in doing so. More generally, there
is important research in psychology and the cognitive sciences on the generative capacities of
humans in the presence of impoverished stimuli and uncertain environments (Spelke et al.,
1992; for a review, see Felin, Koenderink, and Krueger, 2017). All of this psychological
and cognitive research deserves to be integrated into the context of judgment and decision
making in economic settings. Glimmers of this approach can in fact be found in some pockets
of economics and adjacent fields like management. For example, Karni and Vierø (2013)
discuss the role of “growing awareness” (what they call reverse Bayesianism) in uncertain,
economic environments. Eric Van den Steen (2016) discusses the role of beliefs and man-
agerial vision in markets and in the strategy of firms. And Felin and Zenger (2017) highlight
how heterogeneous managerial beliefs and theories shape markets, the emergence of firms
and the origins of economic value. Firms, from this perspective, can be seen as representing
a unique beliefs or “point of view” about what to do and how to structure production
(Coase, 1937; King et al., 2010). Others have looked at the role of entrepreneurial judgment
(Foss and Klein, 2015) or attention (Ocasio and Joseph, 2017) in the context of markets and
organizations. And scholars within the domain of bounded rationality have discussed such
concepts as creativity (Viale, 2016) and the wisdom of heuristics in guiding behavior in
uncertain settings (Grandori, 2010).
Of course, in some sense, our arguments might be seen as creating a caricature of behav-
ioral economics and the emphasis it places on the negative aspects of human nature and mind.
There are important exceptions to our characterization. Herbert Simon’s work in particular is
brimming with insights that deserve more careful attention, beyond what we have discussed

113
114

Teppo Felin and Mia Felin

here. For example, throughout much of Simon’s work there is an emphasis on problem-
solving and goals (Simon, 1964). The specification of problems and goals indeed seems to
provide a possible foundation for the type of organism-oriented conception of perception
and rationality that we have discussed in this chapter. However, unfortunately the literature
on problem-solving and goals has not been meaningfully incorporated into discussions of
bounded rationality in cognitive science or behavioral economics (Chater, 2018; Kahneman,
2003, 2011; Thaler, 2016).
A second implication of our arguments about perception relates to the informal $500 bill-
axiom we discussed at the beginning of this chapter. In some sense, the $500 provides an intri-
guing litmus test for explaining rational expectations and bounded rationality. The $500 bill,
of course, is a rather strange example, because it hardly provides a useful proxy for discussing
the vast set of possible objects and assets that might have value in economics settings. After all,
the $500 bill is clearly labeled: we know what it is worth. In most economic cases of interest,
value is scarcely obvious. Most things of (possible) value are rarely labeled, and there is likely
to be significant disagreement about their worth. The set of possible uses for objects and assets
is, quite simply, unlistable and unprestatable (Felin et al., 2016). And importantly, most of the
valuable assets and activities we observe in markets require (or were the result of) some form of
long-run organization and production. Thus casual references to market efficiency, using the
$500 bill as an example, vastly oversimplify the economic problem. And experiments that show
that humans miss obvious things, while interesting, miss the opportunity to develop theories
of how truly novel value is identified, produced and organized in the first place. Thus a focus
on heterogeneous perceptions, beliefs, hypotheses and theories can provide a powerful tool for
explaining value in firms and markets.
When economists do in fact focus on beliefs, the focus is usually on self-fulfilling proph-
ecies where irrational beliefs and delusions lead to bubbles or market collapse (e.g., Gennaioli
and Shleifer, 2018; Shiller, 2015). The human psychology that is applied to study markets is
about “animal spirits,” wrong-headed beliefs, and illusions, which are largely used to explain
negative outcomes (Akerlof and Shiller, 2010; Shiller, 2017). This type of work of course
has its place. But what about the obverse? That is, what about cases where beliefs appear to
be delusional (perhaps even to a large number of market constituents, including experts), or
where there is significant heterogeneity and ambiguity, but these beliefs and ideas nonetheless
become realities and generate value (cf. Soros, 2013; van den Steen, 2016)?11 These types of
heterogeneous beliefs and opinions ought to be at the very heart of understanding markets and
value creation.12 It seems that the fascination with delusion or bias has led us to lose track of the
type of heterogeneity that animates markets and provide the underlying engine of economic
growth. That is, despite biases and irrational beliefs, we nonetheless witness continued techno-
logical progress and ongoing economic growth. A central problem is that public markets and
many investors have a hard time assessing uniqueness and value (Benner and Zenger, 2016;
Zuckerman, 1999), thus raising questions about the obviousness of value and more general
market efficiency and rationality. Again, beliefs that appear to be delusional or wrong-headed
(to some, including experts and savvy market actors), somehow generate what later becomes
obvious economic value. Some beliefs, ones that objectively looked impossible, become real-
ities as new organizations are formed and certain market actors gravitate toward and invest in
what (to others) appear to be excessively risky and implausible investments. Of course, there
are no “rules for riches” here. But we think that our earlier discussion of heterogeneous per-
ception and beliefs might provide some useful, scientific and empirical starting points for this
type of analysis for understanding the origins and nature of economic value.

114
115

Seeking rationality

Conclusion
The aim of this chapter has been to discuss different forms of rationality—rational expectations
and bounded rationality—and to make explicit the underlying assumptions that are made
regarding perception, obviousness, and economic value. We discuss how insights from biology
and the psychology of perception might provide the preliminary foundations of an alternative.
We specifically emphasize the species-specific nature of both environments and what organisms
are “searching for,” suggesting central mechanisms for explaining judgment and decision
making in uncertain environments. We conclude by offering some high-level remarks and pos-
sible extensions to our argument, specifically by focusing on perception, belief heterogeneity
and the origins of value in markets.

Acknowledgments
An early version of this chapter was presented at the 2018 University of Oxford workshop
“Science, Humanity and Meaning.” The authors appreciate helpful conversations with and
feedback from George Ellis, Thomas Fink, Seth Jenson, Stuart Kauffman, Aaron Reeves,
Riccardo Viale, and Mark Wrathall. This chapter extends joint work published in Psychonomic
Bulletin and Review and Perception with Jan Koenderink and Joachim Krueger. Any errors are
our own.

Notes
1 Of course, value may not be obvious to everyone immediately, in the short run. But in the long-
run—as agents search, adapt, compete, and learn (e.g., Bray and Savin, 1986; Lucas, 1986)—even semi-
efficient markets will ensure that obvious value will be competed away as agents interact.
2 For more macroeconomic discussions of value and “bills left on the table,” see Clemens (2011) and
Olson (1996).
3 In his book on behavioral psychology and economics, Richard Thaler recounts what the psychologist
Thomas Gilovich said to him: “I never cease to be amazed by the number of convenient null hypotheses
economic theory has given you” (2015: 97). Of course, the notion that something is easy or convenient
should not be seen as a put-down, just as Thaler argues that his comment about neoclassical economics
isn’t a put-down (2015: 1579). Both omniscient and behavioral approaches to rationality have certainly
been useful and led to progress in our understanding of economic activity.
4 The research on bounded rationality has generated varied, proximate literatures and concepts including
“computational rationality” (Gershman et al., 2015), “algorithmic rationality” (Halpern and Pass, 2015),
“Bayesian rationality” (Oaksford and Chater, 2007), resource-rational analysis (Griffiths et al., 2015),
and bounded awareness (Chugh and Bazerman, 2007). These literatures are closely related and roughly
feature similar underlying assumptions about perception (for a recent discussion, see Chater et al., 2018;
Felin, Koenderink, and Krueger, 2017).
5 Depending on one’s theoretical sub-field, there are significant differences in how scholars treat situ-
ational perception versus scene statistics versus visual fields versus, and so forth (cf. Chater et al., 2018).
For the purposes of this chapter, we simply utilize the language and intuition developed by Simon
(1955) and Kahneman (2003), though we also build some links to existing vision research that provides
an alternative (Koenderink, 2012).
6 Intriguingly, there is a brief, early recognition of this by Herbert Simon (cf. Felin, Koenderink and
Krueger, 2017). Simon argues that “we are not interested in describing some physically objective world
in its totality, but only those aspects of the totality that have relevance as the ‘life space’ of the organisms
considered” (1956: 130). Our question, then, has to do with this “life space,” and specifically its unique
features relative to the organism in question. This organism-specific line of argument has subsequently
not received attention (see Chater et al., 2018).

115
116

Teppo Felin and Mia Felin

7 An additional bit of evidence for species-specificity comes from the idea of supernormal stimuli
(Barrett, 2010; Hoffman et al., 2015). Supernormal stimuli are stimuli that are disproportionately
attractive or salient to specific species. Thus certain birds may try to hatch a volleyball, because of its
large size, preferring it over their own eggs. Or, a specific beetle in Australia (Julodimorpha bakewelli)
nearly became extinct as discarded beer bottles became a supernormal stimulus for the male beetle,
who tried to copulate with the bottle (the nodules or glass beads on the bottle were highly attractive
to the beetle). Thus species have built-in instincts for specific perceptual aspects of their environments.
8 For related points, about embodied cognition, see Gallese et al. (Chapter 23 in this volume).
9 We are of course using “in mind” merely as informal shorthand for any number of endogenous,
species-specific factors behind perception and behavior. Our discussion of “Suchbild” provides an
overall way of capturing this notion. Though, as we discuss, this Suchbild (or what is “in mind,” what
an organization might be looking for) captures a number of disparate factors such as internal drives and
motivations, as well as the questions, problems or tasks that organisms impose on environments.
10 Perhaps the closest idea to this might be the so-called “one reason”-heuristic suggested by Gigerenzer
and colleagues (1999). However, the one reason heuristic is also explicitly tied to cue characteristics—
such as cue order, or other ideas like stopping rules—and thus is hard to directly square with the more
Suchbild-oriented approach suggested in this chapter.
11 Interestingly, some scholars have recently begun to recognize this issue under the guise of discussing
the role of “narratives,” which shape the attention, motivations, predictions and behavior of economic
actors (see Akerlof and Snower, 2016).
12 The idea of knowledge as “justified true belief ” and the social processes of belief justification (Goldman,
1999) create an interesting puzzle for value creation in the context of uncertain economic environ-
ments. That is, economic value is inherently created where there are disparate and conflicting beliefs
about what the relevant possibilities and facts are (for further discussion, see Felin and Zenger, 2017).
Thus the social mechanisms and “scaling” of beliefs (from some subjective state to a more objective
one)—and where justification happens on the basis of working toward belief realization and mobil-
izing others—provide an interesting opportunity for future work.

References
Abbott, J. T., Austerweil, J. L., and Griffiths, T.L. 2015. Random walks on semantic networks can resemble
optimal foraging. Psychological Review, 122(3), 558–569.
Akerlof, G. A., and Shiller, R. J. 2010. Animal spirits: How human psychology drives the economy, and why it
matters for global capitalism. Princeton, NJ: Princeton University Press.
Akerlof, G. A., and Snower, D. J. 2016. Bread and bullets. Journal of Economic Behavior & Organization,
126, 58–71.
Akerlof, G. A., and Yellen, J. L. 1985. Can small deviations from rationality make significant differences
to economic equilibria? American Economic Review, 75(4), 708–720.
Alchian, A. A. 1950. Uncertainty, evolution, and economic theory. Journal of Political Economy, 58(3),
211–221.
Anderson, J. R. 2013. The architecture of cognition. New York: Psychology Press.
Arrow, K. 1986. Rationality of self and others in an economic system. Journal of Business, 59, 385–399.
Arrow, K., and Debreu, G. 1954. Existence of an equilibrium for a competitive economy. Econometrica,
22, 265–290.
Ashraf, N., Camerer, C. F., and Loewenstein, G. 2005. Adam Smith, behavioral economist. Journal of
Economic Perspectives, 19(3), 131–145.
Bargh, J. A., and Chartrand, T. L. 1999. The unbearable automaticity of being. American Psychologist,
54(7), 462.
Barrett, D. 2010. Supernormal stimuli: How primal urges overran their evolutionary purpose. New York: WW
Norton & Company.
Benner, M. J., and Zenger, T. 2016. The lemons problem in markets for strategy. Strategy Science, 1: 71–89.
Braddick, O. 2018. Illusion research: An infantile disorder? Perception, 47(8), 805–806.
Bray, M. M., and Savin, N. E. 1986. Rational expectations equilibria, learning, and model specification.
Econometrica: Journal of the Econometric Society, 54(5), 1129–1160.
Buchanan, J. M. 1959. Positive economics, welfare economics, and political economy. The Journal of Law
and Economics, 2, 124–138.

116
117

Seeking rationality

Burkhardt, R. W. 2018. Patterns of behavior, Konrad Lorenz, Niko Tinbergen, and the founding of ethology.
Chicago: University of Chicago Press.
Chabris, C., and Simons, D. 2010. The invisible gorilla: And other ways our intuitions deceive us.
New York: Random House.
Chater, N., Felin, T., Funder, D. C., Gigerenzer, G., Koenderink, J. J., ... Stanovich, K. E. 2018. Mind,
rationality, and cognition: An interdisciplinary debate. Psychonomic Bulletin & Review, 25(2), 1–34.
Chater, N., Oaksford, M., Hahn, U., and Heit, E. 2010. Bayesian models of cognition. Wiley Interdisciplinary
Reviews: Cognitive Science, 1(6), 811–823.
Chugh, D., and Bazerman, M. H. 2007. Bounded awareness: What you fail to see can hurt you. Mind &
Society, 6(1),1–18.
Clark, J. M. 1918. Economics and modern psychology. Journal of Political Economy, 26(1), 1–30.
Clemens, M. A. 2011. Economics and emigration: Trillion-dollar bills on the sidewalk? Journal of Economic
Perspectives, 25(3), 83–106.
Coase, R. H. 1937. The nature of the firm. Economica, 4(16), 386–405.
Crandall, C. S., and Sherman, J. W. 2016. On the scientific superiority of conceptual replications for sci-
entific progress. Journal of Experimental Social Psychology, 66, 93–99.
Denrell, J., Fang, C., and Winter, S. G. 2003. The economics of strategic opportunity. Strategic Management
Journal, 24(10), 977–990.
Evans, G. W., and Honkapohja, S. 2005. An interview with Thomas J. Sargent. Macroeconomic Dynamics,
9(4), 561–583.
Ewert, J. P. 1987. Neuroethology of releasing mechanisms: Prey-catching in toads. Behavioral and Brain
Sciences, 10(3), 337–368.
Felin, T., Felin, M., Krueger, J. I., and Koenderink, J. 2019. On surprise-hacking. Perception, 48(2),
109–114.
Felin, T., Kauffman, S., Mastrogiorgio, A., and Mastrogiorgio, M. 2016. Factor markets, actors and
affordances. Industrial and Corporate Change, 25(1), 133–147.
Felin, T., Koenderink, J., and Krueger, J. I. 2017. Rationality, perception, and the all-seeing eye.
Psychonomic Bulletin & Review, 24(4), 1040–1059.
Felin, T. , and Zenger, T. R. 2017. The theory-based view: Economic actors as theorists. Strategy Science,
2(4), 258–271.
Foss, N. J., and Klein, P. G. 2015. Introduction to a forum on the judgment-based approach to entre-
preneurship: Accomplishments, challenges, new directions. Journal of Institutional Economics, 11(3),
585–599.
Frank, R. H., and Bernanke, B. 2007. Principles of microeconomics, 3rd ed. London: McGraw-Hill Irwin.
Frydman, R., and Phelps, E. S. (Eds.) 2013. Rethinking expectations: The way forward for macroeconomics.
Princeton, NJ: Princeton University Press.
Funder, D. C. 2016. Taking situations seriously: The situation construal model and the Riverside
Situational Q-Sort. Current Directions in Psychological Science, 25(3), 203–208.
Geisler, W. S. 2011. Contributions of ideal observer theory to vision research. Vision Research, 51(7),
771–781.
Gennaioli, N., and Shleifer, A. 2018. A crisis of beliefs: Investor psychology and financial fragility. Princeton,
NJ: Princeton University Press.
Gershman, S. J., Horvitz, E. J., and Tenenbaum, J. B. 2015. Computational rationality: A converging para-
digm for intelligence in brains, minds, and machines. Science, 349(6245), 273–278.
Gigerenzer, G., and Gaissmaier, W. 2011. Heuristic decision making. Annual Review of Psychology, 62,
451–482.
Gigerenzer, G., and Goldstein, D. G. 1996. Reasoning the fast and frugal way: Models of bounded ration-
ality. Psychological Review, 103(4), 650–669.
Gigerenzer, G., and Selten, R. 2001. Rethinking rationality. In Bounded rationality: The adaptive toolbox.
Cambridge, MA: MIT Press.
Gigerenzer, G., Todd, P. M., and the ABC Research Group (Eds.) 1999. Simple heuristics that make us smart.
Oxford: Oxford University Press.
Goldman, A. I. 1999. Knowledge in a social world. Oxford: Oxford University Press.
Grandori, A. 2010. A rational heuristic model of economic decision making. Rationality and Society, 22(4),
477–504.
Griffiths, T. L., Lieder, F., and Goodman, N. D. 2015. Rational use of cognitive resources: Levels of ana-
lysis between the computational and the algorithmic. Topics in Cognitive Science, 7(2), 217–229.

117
118

Teppo Felin and Mia Felin

Halpern, J. Y., and Pass, R. 2015. Algorithmic rationality: Game theory with costly computation. Journal
of Economic Theory, 156, 246–268.
Hoffman, D. D., Singh, M., and Prakash, C. 2015. The interface theory of perception. Psychonomic Bulletin
& Review, 22(6), 1480–1506.
James, W. 1912. The sentiment of rationality and other essays. London: Longmans, Green and Company.
Kahneman, D. 1966. Time-intensity reciprocity in acuity as a function of luminance and figure-ground
contrast. Vision Research, 6(3–4), 207–215.
Kahneman, D. 2003. Maps of bounded rationality: Psychology for behavioral economics. American
Economic Review, 93(5), 1449–1475.
Kahneman, D. 2011. Thinking, fast and slow. New York: Farrar, Straus and Giroux.
Kahneman, D., and Frederick, S. 2002. Representativeness revisited: Attribute substitution in intuitive
judgment. In T. Gilovich, D. Griffin, and D. Kahneman (Eds.), Heuristics and Biases: The Psychology of
Intuitive Judgment, Cambridge: Cambridge University Press.
Karni, E., and Vierø, M.L. 2013. “Reverse Bayesianism”: A choice-based theory of growing awareness.
American Economic Review, 103(7), 2790–2810.
Kersten, D., Mamassian, P., and Yuille, A. 2004. Object perception as Bayesian inference. Annual Review
of Psychology, 55, 271–304.
King, B. G., Felin, T., and Whetten, D. A., 2010. Perspective—Finding the organization in organizational
theory: A meta-theory of the organization as a social actor. Organization Science, 21(1), 290–305.
Kingdom, F. A. A., and Prins, N. 2016. Psychophysics: A practical introduction. New York: Academic Press.
Koenderink, J. 2012. Geometry of imaginary spaces. Journal of Physiology, 106, 173–182.
Koenderink, J. 2014. The all-seeing eye. Perception, 43, 1–6.
Knill, D. C. and Richards, W. 1996. Perception as Bayesian inference. Cambridge: Cambridge University Press.
Krueger, J. I., and Funder, D. C. 2004. Towards a balanced social psychology: Causes, consequences, and
cures for the problem-seeking approach to social behavior and cognition. Behavioral and Brain Sciences,
27(3), 313–327.
Lieder, F. and Griffiths, T. L. 2019. Resource-rational analysis: Understanding human cognition as the
optimal use of limited computational resources. The Behavioral and Brain Sciences, 43, 1–85.
Lieder, F., Griffiths, T. L., Huys, Q .J., and Goodman, N. D. 2018. The anchoring bias reflects rational use
of cognitive resources. Psychonomic Bulletin & Review, 25(1), 322–349.
Lucas Jr, R. E. 1986. Adaptive behavior and economic theory. Journal of Business, 59(4), S401–S426.
Marr, D. 1982. Vision: A computational investigation into the human representation and processing of visual infor-
mation. New York: Henry Holt and Co: Inc.
Melnikoff, D. E., and Bargh, J. A. 2018. The mythical number two. Trends in Cognitive Sciences, 22(4),
280–293.
Miller, J. B., and Sanjurjo, A. 2018. Surprised by the hot hand fallacy? A truth in the law of small numbers.
Econometrica, 86(6), 2019–2047.
Muth, J. F. 1961. Rational expectations and the theory of price movements. Econometrica: Journal of the
Econometric Society, 29(3), 315–335.
Oaksford, M., and Chater, N. 2007. Bayesian rationality: The probabilistic approach to approach to human
reasoning. Oxford: Oxford University Press.
Ocasio, W., and Joseph, J., 2017. The attention-based view of great strategies. Strategy Science, 3(1),
289–294.
Olson, M. 1996. Big bills left on the sidewalk: Why some nations are rich, and others poor. Journal of
Economic Perspectives, 10(2), 3–24.
Penrose, E. T. 1952. Biological analogies in the theory of the firm. American Economic Review, 42(5),
804–819.
Pinker, S. 2018. Enlightenment now: The case for reason, science, humanism, and progress. London: Penguin.
Polanyi, M. 1957. Problem solving. British Journal for the Philosophy of Science, 8, 89–103.
Polanyi, M. 1971. Genius in science. Archives de Philosophie, 34, 593–607.
Popper, K. 1966. Objective knowledge: An evolutionary approach. Oxford: Oxford University Press.
Ramscar, M. 2016. Learning and the replicability of priming effects. Current Opinion in Psychology,
12, 80–84.
Rauthmann, J. F., Gallardo-Pujol, D., Guillaume, E. M., Todd, E., Nave, C.S., ... Funder, D. C. 2014.
The situational eight DIAMONDS: A taxonomy of major dimensions of situation characteristics.
Journal of Personality and Social Psychology, 107(4), 677.
Rogers, B. 2014. Delusions about illusions. Perception, 43, 840–845.

118
119

Seeking rationality

Rogers, B. 2019. Where have all the illusions gone? Perception, 48(3), 193–196.
Rothschild, E. 2013. Economic sentiments. Cambridge, MA: Harvard University Press.
Schwartz, B. 1978. The psychology of learning and behavior. New York: W.W. Norton.
Shiller, R .J. 2015. Irrational exuberance. Princeton, NJ: Princeton University Press.
Shiller, R. J. 2017. Narrative economics. American Economic Review, 107(4), 967–1004.
Simon, H. A. 1955. A behavioral model of rational choice. The Quarterly Journal of Economics, 69(1),
99–118.
Simon, H. A. 1956. Rational choice and the structure of the environment. Psychological Review, 63(2), 129.
Simon, H. A. 1964. On the concept of organizational goal. Administrative Science Quarterly, 9(1), 1–22.
Simon, H. A. 1979. Models of thought. New Haven, CT: Yale University Press.
Simon, H. A. 1980. Cognitive science: The newest science of the artificial. Cognitive Science, 4(1), 33–46.
Simon, H. A. 1981. Sciences of the artificial. Cambridge, MA: MIT Press.
Simon, H. A. 1985. Human nature in politics: The dialogue of psychology with political science. American
Political Science Review, 79, 293–304.
Simon, H. A, 1990. Invariants of human behavior. Annual Review of Psychology, 41(1), 1–20.
Simon, H. A. 1991. Organizations and markets. Journal of Economic Perspectives, 5(2), 25–44.
Simons, D. J., and Chabris, C. F., 1999. Gorillas in our midst: Sustained inattentional blindness for dynamic
events. Perception, 28(9), 1059–1074.
Simons, D. J., and Rensink, R. A. 2005. Change blindness: Past, present, and future. Trends in Cognitive
Sciences, 9(1), 16–20.
Soros, G. 2013. Fallibility, reflexivity, and the human uncertainty principle. Journal of Economic Methodology,
20(4), 309–329.
Spelke, E. S., Breinlinger, K., Macomber, J., and Jacobson, K. 1992. Origins of knowledge. Psychological
Review, 99(4), 605–632.
Stanovich, K. E., West, R. F., and Toplak, M. E. 2016. The rationality quotient: Toward a test of rational
thinking. Cambridge, MA: MIT Press.
Thaler, R. H. 2015. Misbehaving: The making of behavioral economics. New York: WW Norton.
Thaler, R. H. 2016. Behavioral economics: Past, present, and future. American Economic Review, 106(7),
1577–1600.
Thaler, R. H. 2017. Behavioral economics. Journal of Political Economy, 125(6), 1799–1805.
Thaler, R., and Sunstein, C. 2008. Nudge: the gentle power of choice architecture. New Haven, CT: Yale
University Press.
Tinbergen, N. 1963. On aims and methods of ethology. Zeitschrift für Tierpsychologie, 20(4), 410–433.
Todd, P. M., and Gigerenzer, G. 2003. Bounding rationality to the world. Journal of Economic Psychology,
24(2), 143–165.
Tversky, A., and Kahneman, D. 1983. Extensional versus intuitive reasoning: the conjunction fallacy in
probability judgment. Psychological Review, 90(4), 293–315.
Uexküll, J. von 2010. A foray into the worlds of animals and humans, trans. J. D. O’Neil. Minneapolis,
MN: University of Minnesota Press.
van Buren, B., and Scholl, B. J. 2018. Visual illusions as a tool for dissociating seeing from thinking: a reply
to Braddick. Perception, 47(10–11), 999–1001.
van den Steen, E. 2016. A formal theory of strategy. Management Science, 63(8), pp.2616–2636.
Viale, R. 2016. Brain-based bounded creativity. In L. Macchi, M. Bagassi, and R. Viale, (Eds.), Cognitive
unconscious and human rationality. Cambridge, MA: MIT Press.
Winter, S. G. 2011. Problems at the foundation? Comments on Felin and Foss. Journal of Institutional
Economics, 7(2), 257–277.
Yuille, A., and Kersten, D. 2006. Vision as Bayesian inference: Analysis by synthesis? Trends in Cognitive
Sciences, 10(7), 301–308.
Zuckerman, E. 1999. The categorical imperative: Securities analysts and the illegitimacy discount.
American Journal of Sociology, 104(5), 1398–1438.

119
120

6
BOUNDED RATIONALITY,
DISTRIBUTED COGNITION,
AND THE COMPUTATIONAL
MODELING OF COMPLEX
SYSTEMS
Miles MacLeod and Nancy J. Nersessian

Introduction
Computational modeling and simulation of complex systems are playing a major role in twenty-
first-century scientific discovery. Given the rhetoric of “big data” and automated modeling
practices, one might expect that computer simulation is a resource for side-stepping human
cognitive capacities and constraints. Yet even these practices still require human cognitive
engagement with their products for the purposes of validation and interpretation. And, as we
have discovered in our investigations of computational systems biology, much problem-solving
still involves researchers building their own models in situations of scarce or inadequate data,
using laptop computers. Rather than freeing scientists from their cognitive limitations, using
computation to extend control and understanding over ever more complex systems requires
developing sophisticated problem-solving practices which push them toward these constraints.
Several philosophers of science have had the intuition that cognitive capacities and limitations
play a significant role in shaping methodological strategies and choices, especially in modeling
complex systems (see, for instance, Humphreys 2009). It would seem, then, that cognitive
considerations should be factored in with epistemic and practical considerations in advan-
cing justifications for specific methodological choices in such problem-solving processes, but
philosophers by and large have not examined these. Unlike most of the research in this area
of philosophy, which relies on published papers concerning finished models, our research is
based on the premise that it is necessary to examine the model-building practices to develop
an understanding of how human cognitive capacities enable and constrain these, as well as the
ways and extent to which modelers overcome cognitive limitations. Our analysis is based on
a 5-year ethnographic study of model-building practices in computational systems biology.
In our investigations of innovative modeling practices in systems biology, we have found the
notion of bounded rationality and its mechanisms of search and satisficing particularly useful for
explaining and justifying the development of modeling methods and strategies. The overarching
goal of contemporary systems biology is to create high-fidelity computational models and
simulations of complex biological systems. This goal largely eludes them. Instead modelers

120
121

Bounded rationality, distributed cognition

rely on strategies of building mid-level models that contain modest details of system com-
position and organization (“mesoscopic models”) and of “building-out” rough initial models
and incrementally exploring and modifying these. Thus, modelers build constrained problem
spaces, and solution search operates chiefly through generating inferences about erroneous
model structure, which requires both a set of search heuristics and constant feedback through
computational simulation. The use of rather simplified representations, which are often far
from high-fidelity, can be seen on cognitive and other grounds to be rational for the long-term
goals of systems biologists, given cognitive and other constraints. This is important, since most
models of biological systems currently produced in the field do not achieve the level of fidelity
required for prediction and control for medical or other purposes. Nonetheless we will show
that the modeling strategies currently employed provide a bounded rational solution to the
problem of building such models over the long term.
However, ours is a quite different analysis of scientific problem-solving than that advanced by
Simon. Specifically, we have been advancing the position that scientific problem-solving requires
a distributed cognition analysis. In the case at hand, accounting for scientific discoveries through
building computational representations requires a distributed cognition analysis of the processes of
building these representations, in which the model and modeler comprise a complex distributed
model-based reasoning system. Thus, we extend the notion of bounded rationality to distributed
cognitive systems. Although not entertained by Simon himself, we see the analytical framework
of distributed cognition as a means of further explicating his “parable of the ant,” where ant
and environment comprise a complex problem-solving system (Simon 1996, p. 63). In the par-
able, if one simply traces the track of the ant over a two-dimensional space, its problem-solving
behavior seems highly complicated. However, once the environment is factored in, each move
of the ant can be recognized as a relatively simple localized response to individual obstacles.
The ant’s behavior is largely a reflection of constraints in the environment and basic internal
representations. Simon suggested that “man” could be substituted for “ant.” However, while we
agree problem-solvers do make local responses based on limited information, we would add that
much human problem-solving behavior is not just a response to an environment but a more active
use of affordances in the environment to aid cognitive tasks. Further, humans create affordances
in the environments such as external representational artifacts that “off-load” cognitive tasks to
the environment, as in the case of the speed bug performing memory functions for the pilot in
the process of landing a plane (Hutchins 1995a). In this brief chapter: (1) we outline the mod-
eling tasks in systems biology; (2) we examine their modeling practices; (3) we present an over-
view of our cognitive analysis; and (4) make a case for their bounded rationality.

Modeling tasks in systems biology

Our ethnographic study of model-building practices tracked modeling in two computational
systems biology labs. In each lab the aim is to construct relatively detailed models of metabolic,
cell signaling, and gene-regulatory networks. These models are dynamic, tracking the changes
in concentration of each metabolite, or the activity of each gene, in a network over time.
Philosophically, computational systems biology takes the stance that only high-fidelity math-
ematical models can capture network behavior sufficiently to gain understanding and reliable
control of these networks for medical or technological purposes. However, although the basic
mathematical methods of systems biology have a long history, the potential to produce high-
fidelity models has only become possible over the last 20 years with the advent of readily avail-
able fast computation. It is now possible for a modeler using a laptop computer to create a model
which captures 50 or more interacting biochemical elements. For the most part, modelers aim

121
122

Miles MacLeod and Nancy J. Nersessian

to model these biochemical networks using coupled ordinary differential equations. These
networks, however, have a number of complexities which make the modeling task difficult.
Structurally, the biochemical networks are highly nonlinear, exhibiting feedback effects, and
having elements playing multiple roles at different points in a network (meaning that different
points in the network draw on the same chemical reservoir within a cell). Cognitively, non-
linearity makes it difficult for modelers to track causal relations in their model and antici-
pate what modifications will achieve. Additionally, modelers never have a complete biological
representation of their network (“pathway diagram”) at the beginning of their project and lack
sufficient data to build a highly detailed model. Indeed, since experimentalists generally do not
study all interactions within a network themselves, modelers (who know little biology) need to
search through the experimental literature and databases to build the pathways on their own.
With this information in hand, we can describe the modeling task. First, modelers need to
assemble a pathway representation of their network, which is a line diagram connecting elem-
ents in a chain according to the order in which they interact. The pathway representation allows
them to map variables to a mathematical formulation and choose interaction representations
to fill in that formulation. Finally, parameters to fit the model need to be estimated from the
available data. Any unknown parameters can then be determined through a parameter-fitting
algorithm which chooses a parameter set for the model, according to how well it can repro-
duce experimental data on the network behavior. Usually some data are held over for testing
the validity of the model that is produced. If all this works, then modelers can proceed to make
novel predictions about system behavior, such as how to intervene optimally on the systems for
medical purposes using target drug combinations.
Although this looks like a relatively straightforward set of procedures, in nearly all cases
we have examined, the starting biological pathway representations modelers assemble are
never adequate for the network behaviors they want to capture. Biochemical elements are
missing or interactions among elements in the network are missing. At the same time, initial
representations of those interactions might not be correct. The models they produce thus
invariably fail to test well against the available data first time around. The task, then, of the
modeler is to correct the models they have produced by deriving or inferring the location of
missing elements or poor interaction descriptions in their models and hypothesizing additions
or corrections to replace them. Usually a variety of corrections are required to build a model
that functions adequately.

Mesoscopic modeling and the building-out strategies

Given the complexity of the systems being modeled, modelers have developed strategies that
make the task cognitively tractable. In the first place, they build a lower-fidelity model that
bounds the problem space. Such models are called “mesoscopic” (Voit et al. 2012), meaning
they are ”in between” any one level of description or biological organization in two senses.
First, the systems-level is simplified such that only certain relationships or system-level behaviors
are represented, rather than the full range of potential behaviors. Second, the network itself
is simplified in a variety of ways, often relative to these system-level goals. Quite abstract
non-mechanistic representations of biochemical interactions are applied to simplify running
simulations, parameter-fitting, and mathematical analysis. Elements thought not to be rele-
vant to target behaviour are left out. Subsystems are typically blackboxed and treated as linear
or constant. Note that while such simplifications are typical of any modeling effort, they are
problematic to the extent that systems biology aims at high-fidelity models capable of robust
predictions.

122
123

Bounded rationality, distributed cognition

The mesoscopic model delimits the problem space for searching for solutions to correct the
model. Dimensions of the space represent different structural alternatives and biochemical inter-
action alternatives (in terms of their mathematical description). This space is also constrained by
what interactions and structures are biochemically possible, although a modeler (typically from
engineering) might not know all of these. The task of the modeler is to locate the alternatives
which best capture the functional performance of the target network they wish to capture
and also perform best in response to perturbations.1 It is impossible for modelers to simply
hypothesize the needed corrections in one hit. These more limited models still represent many
interacting elements, involving many non-linearities. Although it might be possible to predict
what one modification might do, understanding the implications of changing two elements is
usually impossible. There is thus an imperative on modelers to develop strategies for iteratively
localizing problems and fixing them. The strategy our modelers employ has been developed
around a central presupposition, namely, that the initial models they formulate are within a rela-
tively small set of changes from a good network representation. This is essential, since for the
most part their strategy is constructed almost entirely around making discrete modifications to
the original model and using the original model to guide their actions. We call this a “building-
out” strategy. In practice, the problem space modelers confront is constructed relative to this
initial model. In addition to this initial model representation, model simulation provides infor-
mation as to how the dynamics of the model differ in behavior from the target. Modelers are
thus, at the start, looking for modifications which bring their model into alignment with that
behavior. Of course, the nature of the divergences provides clues as to where to look and what
to change. With an initial model at hand, then modelers formulate their strategy around two
central tasks: explorations and modifications (MacLeod 2018). Explorations are steps modelers
employ to try to understand the causal relationships within the model. Explorations can be
done through pen and paper representations, but usually they also require running simulations
repeatedly and visualizing the results. To get the information they need, modelers will often
temporarily remove elements to try to isolate particular relations. Importantly, they can use the
affordances of computation to bracket nonlinear behaviors, through explorations of different
parameter and input variations. These repeated processes help modelers develop what they
often refer to as having a “feel for a model” (Voit et al. 2012). We have analyzed this notion in
detail (Chandrasekharan and Nersessian 2015; MacLeod and Nersessian 2018) and understand
it to refer to intuitions modelers build up on how particular variables and parameters in the
network causally affect specific others. In more cognitive terms we characterize these intuitions
as simulative mental models; internal representations which simulate causal relationships.
Explorations build up families of mental models which themselves represent at least part
of the structure of the overall model. These mental models help modelers predict potentially
fruitful modifications by allowing them to infer possible effects of modifications. Modifications
are hypotheses modelers make which should help move the model toward a sound represen-
tation. Working from the original model, modelers can entertain a sequence of increasingly
complex heuristics, including: (1) adjusting parameters; (2) adding or removing interactions
between elements already in the network; (3) changing interaction formulae; (4) “de-black-
boxing” a set of interactions; and (5) adding a new element to the model at different points in
the network. This list of options represents discrete, localized changes to models. Each can be
hypothesized and then tested according to how well the new model performs and, if heading in
the right direction, new modifications can then be hypothesized. In this respect these heuristics
follow a hill-climbing logic, with modelers hoping for improvements with each move but not
expecting to find any fully definite solution in one hit. Hopefully within a relatively contained
sequence of steps the model reaches the performance goals of the modeler. If the model fails

123
124

Miles MacLeod and Nancy J. Nersessian

to improve, then modelers will likely conclude that there is no cognitively accessible path from
their original model to an adequate representation. The model might be incorrect in too many
interrelated ways for them to untangle.
We view this building-out strategy as a means by which modelers of complex systems turn
problem-solving situations with ill-defined problems, unstructured task environments, and
quite general goals into a series of situations approximating those Simon envisaged. For each
iteration of the model-building process, there is a relatively well-defined problem space: the
modification process has a fixed representation, the goal of “fitting” that model, and a set of
heuristics for facilitating what is in essence a search process. However, to understand how such
computational model-building processes lead to scientific discoveries we need to embed the
notions of problem space, search, and bounded rationality a different kind of cognitive analysis
than that advanced by Simon.

Cognitive analysis: distributed model-based reasoning

Computational model-building provides an exemplar of what Hutchins calls creating cognitive
powers (1995b; also see Chandrasekharan and Nersessian 2015). Numerous iterations of model-
building (including visualizations of behavior), debugging, modifying, and simulation, build the
model and modeler into a “coupled” cognitive system. The highly integrated nature of this
distributed problem-solving system results in the modeler being able to manipulate and explore
a much more complex set of equations that would be possible “in the head” alone––even with
pen and paper resources. Scientific problem solving has always been dependent on creating and
using external representations, especially various kinds of diagrams. There is a vast literature
that examines the cognitive roles of these in both mundane and scientific reasoning. The roles
of computational representations, which dominate contemporary science, however, have
received scant attention, and what literature there is focuses on the function of model
visualizations in mental modeling (Trafton, Trickett, and Mintz 2005; Trickett and Trafton
2002, 2006, 2007) and distributed problem-solving processes (Alač and Hutchins 2004; Becvar
et al. 2008). The work of our research group has taken a different approach, focusing on the
processes of building computational representations and how discoveries emerge from these
processes (see, especially, Chandrasekharen and Nersessian 2015). Our analysis starts from
Nersessian’s account of scientific problem-solving processes in terms of “simulative model-
based reasoning” where inferences are drawn in distributed cognitive processes, comprising
building and manipulating mental models and external representations such as diagrams and
equations (Nersessian 1992, 2002, 2008) or physical models (Nersessian et al. 2003; Nersessian
2009). Inferences are drawn through processing information in the system comprising memory
(“internal”) and the environment (“external”)––what we call distributed model-based reasoning––
rather than as on the traditional account where all information is abstracted from the external
representations (environment) and represented and processed internally, as with Simon’s ant (see
Greeno 1989, for an earlier similar claim). We argue that the process of building models
integrates manipulations in the mental model and in the external representation, creating a
coupled cognitive system (Osbeck and Nersessian 2006; Nersessian 2008). Here we briefly discuss
the extensions required of the framework of distributed cognition to accommodate our account
of scientific problem solving, especially as it pertains to computational representations (a detailed
account is provided in Chandrasekharan and Nersessian 2015). Much of the research on the
function of external representations within the distributed cognition framework has focused on
highly defined task environments with well-defined problems and goals, such as piloting ships
and planes (Hutchins 1995b) or solving puzzles with fixed representations and rules (Zhang and

124
125

Bounded rationality, distributed cognition

Norman 1994). These analyses focus on the way external representations change the nature of
the cognitive tasks, especially by reducing cognitive load through off-loading memory to the
environment. These external representations are ready-to-hand. Scant attention has been
directed toward understanding the role of processes of building the external representations to
alter task environments in cognitive processes during problem solving (exceptions include Kirsh
1996 and Hall et al. 2010). Scientific practices provide a prime locus for studying processes of
building external representations since building problem-solving environments is a major com-
ponent of scientific research (Nersessian et al. 2003; Nersessian 2012). When studying science,
the focus needs to shift to the building of external structures consonant with how people
actively distribute cognition (Hall et al. 2010) through creating problem-solving environments
and in turn building cognitive powers (Hutchins 1995b). A research laboratory provides a good
example of a problem-solving environment. It contains artifacts salient to the research, people,
and socio-cultural practices. A computational model, itself, provides another example of a
problem-solving environment. Through building and running the model, the researcher creates
cognitive powers that extend her natural ones, such as the ability to synthesize a vast range of
data, to visualize complex dynamical processes, and to run through unlimited imaginative
(counterfactual) scenarios not possible in her mind alone. Indeed, we have witnessed how
building computational models enables systems biology modelers with scant biological know-
ledge to make fundamental biological discoveries (Chandrasekharan and Nersessian 2015). Our
extension of the framework to science has been based on ethnographic studies of the open,
ill-formed problem solving tasks of pioneering research and of the processes by which researchers
create their cognitive artifacts; in the case at hand, computational representations. The other
way is which the framework of distributed cognition is in need of extension is by providing an
account of the nature of the internal representations used by the human component of the
system. Most research on distributed cognitive processes is silent about mental representations.
Zhang and Norman (1994), which explicitly analyzes the interactions among external and
internal representations in problem solving, assume the internal representations to be mental
models. Nersessian (1992, 2002, 2008) has elaborated a “mental modeling framework,” which
she argues provides a cognitive basis for the range of model-based reasoning used in scientific
practice. This framework draws from the strand of research in the mental modeling literature
that examines the processes of constructing and manipulating a working memory model during
reasoning and problem solving and is agnostic about the nature of the long-term memory
representation. In thinking about scientific reasoning, Nersessian (2008) argues that we need to
move beyond the mental models literature per se and create a synthesis of an extensive range of
experimental literature spanning over 25 years on discourse and situation modeling (see, e.g.,
Johnson-Laird 1983; Perrrig and Kintsch 1985; Zwaan 1999), mental animation (see, e.g.,
Hegarty 1992; Schwartz 1995; Schwartz and Black 1996), mental spatial simulation (see, e.g.,
Shepard and Cooper 1982; Finke 1989; Kosslyn 1980), and embodied mental representation
and perceptual simulation (see, e.g., Glenberg 1997; Bryant and Tversky 1999; Barsalou 1999;
Brass, Bekkering, and Prinz 2001). She advances a “minimalist hypothesis” that “in certain
problem-solving situations humans reason by constructing a mental model of the situation,
events, and processes in working memory that in dynamic cases can be manipulated by simula-
tion” (2002, p. 143). Recent psychological research on scientists and engineers as they try to
solve research problems lends support to the hypothesis and further details the nature and role
of mental model (or “conceptual”) simulations (Trafton, Trickett, and Mintz 2005; Trickett and
Trafton 2007; Christensen and Schunn 2009). For instance, Trickett and Trafton (2007) estab-
lish the importance of conceptual simulation in the data analysis phase of several different kinds
of scientific fields. Most importantly, the use of such simulations increases in cases of inferential

125
126

Miles MacLeod and Nancy J. Nersessian

uncertainty where scientists appear to be trying to develop a general understanding of the

system under investigation. They see their findings as lending support to the extension of the
“minimalist hypothesis” to scientific problem-solving. Mental modeling is often carried out in
the presence of real-world resources, such as diagrams and the various objects being reasoned
about, as Simon has noted for design and other processes (Simon 1996). This is even more so
in the case of science, where visual representations, physical models, and computational models
are integral to the reasoning in problem-solving processes. Further, the representations and pro-
cessing required for such sophisticated reasoning are too detailed and complex to be “in the
head” alone. In cognitive psychology, the question of the interface between the mental capacity
and resources in the world has largely focused on diagrams and other visual representations.
Several of these analyses have proposed that the external representations are coupled with the
mental representations in inferential processing (see, e.g., Greeno 1989; Zhang and Norman
1994; Gorman 1997; Hegarty 2005; Nersessian 2008). Our research extends internal and
external representational coupling to dynamical representations: computational simulations and
mental simulations. By examining the process through which external and internal representa-
tional structures are built, we can begin to understand how these are incorporated into a
distributed system that performs cognitive functions such as memory and inference; in the case
at hand, distributed model-based reasoning through the coupling of mental modeling and com-
putational simulation processes. From our interview data and the pen and paper sketches the
computational modelers in our study make, we have been able to discern some features of the
mental representations they use in “debugging” computational models (MacLeod and Nersessian
2018). Broadly, they can be described as simulative models of causal relationships represented in
their mathematical networks, which operate in a manner similar to the kinds of mental models
captured in investigations of basic causal network reasoning (Hegarty 1992, 2004; Schwartz
1995). These models are qualitative (Roschelle and Greeno 1987), rather than quantitative, and
the inferences drawn from them are thus qualitative in nature. Further, as with everyday causal
network reasoning, these models tend to be selective and piecemeal representations of the
overall systems (Hergarty 1992). In particular, nonlinear relations can be identified and bracketed
with the aid of computation into different separate behaviors, allowing mental simulation of
each separately. For example, the range over which a feedback structure produces stable
oscillations can be separated from the range over which this structure produces more
equilibrium-type behavior, and these can handled separately. Incrementally building such
mental models in conjuction with processes of computational simulation and visualization
creates the coupled cognitive system through which novel inferences about both the computa-
tional and real-world biological systems can be made, often leading to important discoveries
(Chandrasekharan and Nersessian 2015; MacLeod and Nersessian 2018).

The bounded rationality of model-building practices in systems biology

We think it especially important to consider the rationality of modeling practices in systems
biology because the field currently is failing to reach its stated epistemological goals. The
practice of mesoscopic model-building and incremental modification through building-out
strategies provides an exemplar of what Simon, in a later analysis of bounded rationality, called
”procedural rationality” (Simon 1976), that is, the rationality of the processes through which
boundedly rational decisions or problem solutions (substantive rationality) are achieved.
To show procedural rationality for a particular strategy, two claims need to be made: first,
that a cognitive system is constrained (or bounded), and, second, that the strategy provides an
effective pathway given these constraints, toward reaching a set of goals. Building mesoscopic

126
127

Bounded rationality, distributed cognition

models bounds the problem space of the complex biological system. This facilitates the effect-
iveness of the building-out strategy as there is a limit to both the complexity of the model (in
terms, e.g., of the number of elements reponsible for any particular behavior) and the number
of faults which explain why a model departs from the data. Any model which breaches such
limits would not be tractable for mental modeling in the ways we have suggested, restricting
the possibility of employing the heuristic steps fruitfully. As we have seen, the nature of the
building-out strategy, to make discrete incremental modifications to a pre-existing model,
suggests the extent to which modelers work to keep their operations within their cognitive
limitations. Mental modeling research indicates that among other things there is a critical need
to keep modeling practices within the scope of working memory. Inferences are drawn based
on causal interactions among elements. As Hegarty has shown in the case of pulley systems, the
ability to reason about causal networks is constrained directly by working memory (Hegarty
1992, 2004). The building-out strategy depends on being able to draw inferences as to the
sources of errors and about modifications which might eliminate them, which requires starting
from a more limited computational model representation. Given these constraints, one can
make a good argument that the strategies of systems biology modelers provide a boundedly
rational option for their model-building tasks. Producing even a mesoscopic model which
captures the systems adequately at the outset is impossible due to incomplete information and
the complexity of the systems. The building-out process affords error-correcting inferences.
A step-wise approach takes advantage of this while keeping the modeling processes within cog-
nitive control. Modelers resolve the model not by planning several moves ahead, but through
selecting among a set of choices which, at any one time, amount to local fixes or improvements.
Overall, the rationality of the building-out strategy depends on both initial models being within
a relatively small number of steps from a good outcome and on the complexity of interactions
within the model being constrained, such that modelers can effectively draw inferences through
engagement with computational simulation. One cannot judge a priori how well initial models
meet the conditions they need to have in order for the building-out strategy to pay-off. But, in
our experience, modelers do manage to produce models through these techniques which meet
their limited data-matching and prediction goals. And even though the models are not high-
fidelity, some have produced significant discoveries, such as discovering unknown elements in
a well-established pathway (Chandraskeharan and Nersessian 2015; MacLeod and Nersessian
2018). Such achievements are in fact a powerful endorsement of the power of this approach to
modeling of complex biological systems, despite the ultimate goals of the field.
In general, we think our research shows that bounded rationality can provide not only a good
description of problem-solving practices, but also a justification for methodological decision-
making that, on the surface, might seem sub-optimal in fields that are attempting to understand,
predict, and control complex systems by relying heavily on computational modeling and simu-
lation. As noted, although the strategies considered enable the production of models which
meet limited representational goals, such models do not achieve the scale and fidelity at which
systems biology aims. A major part of the rhetoric of systems biology is that modern computa-
tion opens up the possibility of building high-fidelity models of large-scale systems, which are
required for robust predictions and to advance biological theory. However, as Voit et al. (2012,
p. 23) describe it, the majority of models in systems biology are not ”large enough to approach
the reality of cell or disease processes with high fidelity.” The strategies modelers use to build
these limited, mid-sized models, seem, on the face of it, to push modelers away from these
goals, especially since they are rarely capable of robust predictions (MacLeod and Nersessian,
2015, 2018). This has created a challenge for the field of justifying their current practices,
and systems biologists are remarkably reflective on the nature of the cogntive challenges they

127
128

Miles MacLeod and Nancy J. Nersessian

face and on the need to rationalize the current production of relatively small-scale models. As
pioneers in the field, Voit et al. (2012) and Voit (2014) argue that the value of such models to
the field is that, once built, they create tractable cognitive platforms for richer model devel-
opment and provide insight into the underlying systems, by virtue of their mesoscopic nature.
That is, they create potential pathways for scaling models up in detail and complexity, such that
higher fidelity models can be achieved sometime in the future. Interestingly, Voit et al. locate
the cognitive basis of

[their] strategy of locally increasing granularity ... in semantic networks of learning

and the way humans acquire complex knowledge ... hierarchical learning is very
effective, because we are able to start simple and add information as we are capable
of grasping it.
2012, p. 23

In other words, mesoscopic models, once constructed, serve as learning platforms, facilitating
a modeler’s ability to build richer, more detailed models that capture more of the biological
system’s complexity. Hence, although current systems biology might be failing in its aim of
constructing high fidelity models, the kinds of models being built are anticipated to facilitate
the ability of modelers to produce more realistic models without being overwhelmed by their
complexity. Detail can be added incrementally, in controlled steps using heuristics modelers
have been developing. For instance, once a mesoscopic model is in place, modelers can begin to
de-black-box systems which were initially black-boxed, so as to produce a model with greater
precision or a wider variety of accurate behaviors. A single term operating in the model can be
expanded as a sub-network. Importantly, the fact that the more expanded representation must
still reproduce the original behavior it produced as a single term is an important constraint that
modelers can use to help build this sub-network and pin down parameters without having to
reiterate the entire model. Similarly, interactions can be modeled using more complex relations,
and so on.
The analogy Voit draws with semantic learning reinforces the essential cognitive function of
mesoscopic models we gave above, which is to facilitate a process of cumulative controlled cog-
nitive development in complex contexts. This attempt by scientific practitioners to rationalize
their own activities, in effect, shares the main claims of the justification we have provided of
the strategy of building mesoscopic models, i.e., containing structure supports more complex
model development through building-out and other heuristic strategies. Modelers operate with
learning constraints, among others, and their methodological choice behavior is boundedly
rational. Although some kinds of purely computational systems might be capable of mod-
eling whole complex systems upfront, distributed human-computer problem-solving systems
are not.

Conclusion
In this chapter we have described the model-building processes of system biologists, arguing
that these practices should be understood as boundedly rational responses to the complexity of
the modeling problems they face. The concept of bounded rationality can thus play a useful
role in the field of systems biology, providing cognitive means to justify its present activities.
However, as we have seen, unpacking what precisely bounds problem-solving activity and why
a set of given responses might be rational in the case of systems biology, or indeed in the case of
any computational science, requires addressing the nature of the human-computational system

128
129

Bounded rationality, distributed cognition

interaction. Briefly, we have tried to argue that distributed cognition and simulative mental
modeling can play this role. Indeed, our case above illustrates how Simon’s notions of bounded
rationality and problem solving as search can be combined with a distributed cognitive frame-
work to help better articulate and study the complex nature of modern computational science.

Note
1 Given that one of the goals of systems biology is to be able manipulate biochemical systems, it thus
important that the models they build do not just capture natural equilibrium behavior, but also what
will happen in response to intervention on a system.

References
Alač, M., and Hutchins, E. (2004). I see what you are saying: Action as cognition in fMRI brain mapping
practice. Journal of Cognition and Culture, 4(3), 629–661.
Barsalou, L. W. (1999). Perceptions of perceptual symbols. Behavioral and Brain Sciences, 22(4), 637–660.
Becvar, A., Hollan, J., and Hutchins, E. (2008). Representing gestures as cognitive artifacts. In M. S.
Ackerman, C. Halverson, T. Erickson, and W. A. Kellog (Eds.), Resources, co-evolution, and artifacts: Theory
in CSCW (pp. 117–143). New York: Springer.
Brass, M., Bekkering, H., and Prinz, W. (2001). Movement observation affects movement execution in a
simple response task. Acta Psychologica, 106(1–2), 3–22.
Bryant, D. J., and Tversky, B. (1999). Mental representations of perspective and spatial relations from
diagrams and models. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25(1), 137.
Chandrasekharan, S., and Nersessian, N. J. (2015). Building cognition: The construction of computational
representations for scientific discovery. Cognitive Science, 39(8), 1727–1763.
Christensen, B. T., and Schunn, C. D. (2009). The role and impact of mental simulation in design. Applied
Cognitive Psychology, 23(3), 327–344.
Finke, R. A. (1989). Principles of mental imagery. Cambridge, MA: The MIT Press.
Glenberg, A. M. (1997). Mental models, space, and embodied cognition. In T. B. Ward, S. M. Smith,
and J. Vaid (Eds.), Creative thought: An investigation of conceptual structures and processes (pp. 495–522).
Washington, DC: American Psychological Association.
Gorman, M. E. (1997). Mind in the world: Cognition and practice in the invention of the telephone.
Social Studies of Science, 27(4), 583–624.
Greeno, J. G. (1989). Situations, mental models, and generative knowledge. In D. Klahr and K. Kotovsky
(Eds.), Complex information processing: The impact of Herbert A. Simon (pp. 285–318). Mahwah,
NJ: Lawrence Erlbaum.
Hall, R., Wieckert, K., and Wright. K. (2010). How does cognition get distributed? Case studies of
making concepts general in technical and scientific work. In M. Banich and D. Caccamise (Eds.),
Generalization of knowledge: Multidisciplinary perspectives, New York: Psychology Press.
Hegarty, M. (1992). Mental animation: Inferring motion from static displays of mechanical systems. Journal
of Experimental Psychology: Learning, Memory, and Cognition, 18(5), 1084.
Hegarty, M. (2004). Mechanical reasoning by mental simulation. Trends in Cognitive Sciences, 8(6),
280–285.
Hegarty, M. (2005). Multimedia learning about physical systems. In R. E. Mayer (Ed.), Cambridge hand-
book of multimedia learning (pp. 447–465). New York: Cambridge University Press.
Humphreys, P. (2009). The philosophical novelty of computer simulation methods. Synthese, 169,
615–626.
Hutchins, E. (1995a). How a cockpit remembers its speed. Cognitive Science, 19, 265–288.
Hutchins, E. (1995b). Cognition in the wild. Cambridge, MA: MIT Press.
Johnson-Laird, P. N. (1983). A computational analysis of consciousness. Cognition & Brain Theory, 6,
499–508.
Kirsh, D. (1996). Adapting the environment instead of oneself. Adaptive Behavior, 4, 415–452
Kosslyn, S. M. (1980). Image and mind. Cambridge, MA: Harvard University Press.
MacLeod, M. (2016). Heuristic approaches to models and modeling in systems biology. Biology &
Philosophy, 31(3), 353–372.

129
130

Miles MacLeod and Nancy J. Nersessian

MacLeod, M. (2018). Model-based inferences in modeling of complex systems. Topoi. https://doi.org/

10.1007/s11245–018–9569-x
MacLeod, M., and Nersessian, N. J. (2015). Modeling systems-level dynamics: understanding without
mechanistic explanation in integrative systems biology, Studies in History and Philosophy of Science Part
C – Biological and Biomedical Science, 49(1), 1–11.
MacLeod. M., and Nersessian, N. J. (2018). Cognitive constraints, complexity and model-building.
History and Philosophy of the Life Sciences, 40(17).
Nersessian, N. J. (1992). How do scientists think? Capturing the dynamics of conceptual change in science.
In R. Giere (Ed.), Minnesota studies in the philosophy of science (pp. 3–45). Minneapolis, MN: University
of Minnesota Press.
Nersessian, N. J. (2002). The cognitive basis of model-based reasoning in science. In P. Carruthers,
S. Stich, and M. Siegal (Eds.), The cognitive basis of science (pp. 133–153). Cambridge: Cambridge
University Press.
Nersessian, N. J. (2008). Creating scientific concepts. Cambridge, MA: MIT Press.
Nersessian, N. J. (2009). How do engineering scientists think? Model-based simulation in biomedical
engineering research laboratories. Topics in Cognitive Science, 1(4), 730–757.
Nersessian, N. J. (2012). Engineering concepts: The interplay between concept formation and modeling
practices in bioengineering sciences. Mind, Culture, and Activity, 19(3), 222–239.
Nersessian, N. J., Kurz-Milcke, E., Newstetter, W. C., and Davies, J. (2003). Research laboratories as
evolving distributed cognitive systems. In Proceedings of the Annual Meeting of the Cognitive Science Society.
Mahwah, NJ: Lawrence Erlbaum Associates.
Osbeck, L. M., and Nersessian, N. J. (2006). The distribution of representation. Journal for the Theory of
Social Behaviour, 36(2), 141–160.
Perrig, W., and Kintsch, W. (1985). Propositional and situational representations of text. Journal of Memory
and Language, 24, 503–518.
Roschelle, J., and Greeno, J. G. (1987). Mental models in expert physics reasoning (Tech. Report No.
GK-12). Berkeley, CA: University of California Press.
Schwartz, D. L. (1995). Reasoning about the referent of a picture versus reasoning about a picture as the
referent. Memory and Cognition, 23, 709–722.
Schwartz, D. L., and Black, J. B. (1996). Analog imagery in mental model reasoning: Depictive models.
Cognitive Psychology, 30(2), 154–219.
Shepard, R. N., and Cooper, L. A. (1982). Mental images and their transformations. Cambridge,
MA: MIT Press.
Simon, H. A. (1976). From substantive to procedural rationality. In S. J. Latsis (Ed.), Method and appraisal
in economics. Cambridge: Cambridge University Press.
Simon, H. A. (1996). The sciences of the artificial. Cambridge, MA; MIT Press.
Trafton, J. G., Trickett, S. B., and Mintz, F. E. (2005). Connecting internal and external
representations: Spatial transformations of scientific visualizations. Foundations of Science, 10(1), 89–106.
Trickett, S. B., and Trafton, J. G. (2002). The instantiation and use of conceptual simulations in evalu-
ating hypotheses: Movies-in-the-mind in scientific reasoning. In Proceedings of the Annual Meeting of the
Cognitive Science Society 24(24).
Trickett, S. B., and Trafton, J. G. (2006). Toward a comprehensive model of graph comprehension: Making
the case for spatial cognition. In Proceedings of International Conference on Theory and Application of
Diagrams (pp. 286–300). Heidelberg: Springer.
Trickett, S. B., and Trafton, J. G. (2007). “What if…”: The use of conceptual simulations in scientific
reasoning. Cognitive Science, 31(5), 843–875.
Voit, E. O. (2014). Mesoscopic modeling as a starting point for computational analyses of cystic fibrosis as
a systemic disease. Biochimica et Biophysica Acta (BBA)-Proteins and Proteomics, 1844(1), 258–270.
Voit, E. O., Newstetter, W. C., and Kemp, M. L. (2012). A feel for systems. Molecular Systems Biology, 8(1).
Voit, E. O., Qi, Z., and Kikuchi, S. (2012). Mesoscopic models of neurotransmission as intermediates
between disease simulators and tools for discovering design principles. Pharmacopsychiatry, 45(1): 22.
Zhang, J., and Norman, D. A. (1994). Representations in distributed cognitive tasks. Cognitive Science,
18(1), 87–122.
Zwaan, R. A. (1999). Embodied cognition, perceptual symbols, and situation models. Discourse Processes,
28, 81–88.

130
131

7
BOUNDED RATIONALITY
AND PROBLEM SOLVING
The interpretative function of thought

Laura Macchi and Maria Bagassi

Heuristics and insight problem solving

People whose reasoning is characterized by Wason’s confirmation bias and those who use
Kahneman and Tversky’s heuristics would hardly seem to belong to the same species as the
people Duncker and Simon write about, people who challenge with a certain degree of success
the limits of their cognitive system. We believe that the diversity of the outcomes of the research
on problem solving and reasoning cannot be satisfactorily explained by saying that, on the
one hand, human beings are reasonable problem solvers, and, on the other, they are obtuse
reasoners, predisposed to fallacies, particularly regarding principles and procedures, on which
the psychology of reasoning places particular importance.
We can try to explain this paradoxical enigma of reason. At the beginning of the twentieth
century, psychology was differentiating itself from philosophy and carving out a place among
the sciences; to this end, to comply with requirements of “scientificity,” it integrated the specu-
lative approach adopted by philosophy with the experimental approach, basing this on logic,
the discipline which constituted the foundation of the other sciences. However, it is worth
noting that when psychology of thought encountered logic in the 1920s, logic was configured
as mathematical, formal logic, characterized by a gradual process of “depsychologization” of
logical language and of disambiguant simplification compared to natural language, intentionally
pursued, and programmatically declared by modern logic. It is paradoxical that the psychology
of reasoning should have taken on a depsychologized discipline as a prescriptive and descriptive
model of the functioning of thought. As a consequence of logic being the normative schema in
research on reasoning, experimental tasks tend to be identified as exercises in logic presented in
a popularized version, dressed up as a “‘real-life’ situation” (Mosconi, 2016, p. 351).
We consider that the legacies of the extrapsychological disciplines, such as logic, the theory
of probability, and so on have conditioned the psychology of reasoning and have had a nega-
tive impact in terms of originality and adequacy on psychological research. The psychology
of reasoning, by adopting an ideological approach, superimposed on the psychological reality,
burdened itself with a great limitation: the assumption of a pre-established norm, borrowed
from long-standing tradition.
The development of problem solving followed a completely different path, phenomeno-
logical in nature, where the research was not affected by these legacies. Problem solving is

131
132

Laura Macchi and Maria Bagassi

an original contribution of psychology; it is a completely new field of research and therefore

unpolluted by any backdrop of speculation or theories that can affect other consolidated areas
of study. There has not been the temptation, or even the possibility, to appropriate approaches
or systems of knowledge created with objectives other than those of psychological research.
The most important contributions to problem solving were those of the Gestaltists in the
1920s and of Herbert Simon et al. 30 years later. Both of these contributions represent a
response to the associative and behaviorist approach, recognizing the contribution of the sub-
jective dimension in perceptual organization and, in more general terms, in thought itself.
Demonized as it had been by a materialist approach and eluding all attempts at measurement
and quantification, for the Gestaltists, the mind returned, enhanced with all the dignity of an
object worthy of scientific study; the essence of the activity of thought did not lie in the mere
registration of an aggregate of stimuli, but rather in the ability to understand relations.
In studies on visual perception, which was the Gestaltists’ original area of research, visual
input was identified on the basis of our knowledge and then enhanced by inferential operations
initiated by principles of relevance and coherency, or estimates of probability Cognitive
integrations and interpolations are not isolated cases, they are the rule in our daily interaction
with the world around us. Processes of amodal completion are those which best allow us to
highlight the analogies that exist between seeing and thinking.
Frequently, these are cases of genuine perceptual problem solving in which the available
cues are evaluated in the light of knowledge and context. Take, for example, the case when we
recognize a person at a considerable distance. Even a scene in which many details are lacking
receives a sort of interpretation (Figure 7.1).
At the time of the Gestaltists’ studies on visual perception, psychology was dominated by
elementaristic conceptions, inherited from philosophical associationism and the relative con-
viction that the study of the contents of conscience and behavior necessarily implied that they

Figure 7.1 The Kanizsa Triangle

132
133

Bounded rationality and problem solving

be broken down into their ultimate parts, pure sensations, acts of elementary volition, reflexes,
and so on.
This approach was, however, comprehensible, as it was natural that the founders of scientific
psychology would think that the procedures of the natural sciences would have endowed the
investigation of issues regarding the mind with the dignity of science. Unfortunately, the ana-
lytic method adopted rendered psychological investigation sterile; it fell victim to a futile hunt
for elements that did not permit the reconstruction of the real, living complexity of the psychic
functions in seeing, thinking, and problem solving.
Of the critical reactions to this situation, the Gestaltists were particularly outstanding for the
vivacity with which they attacked the doctrine of the elements.
For the Gestaltists, in every perception, there is a moment––the elementary and raw sensory
data––and, at the same time, another moment, realized by an act of production, with which
our conscience builds objects of a higher order, the structures. What determines, for example,
a certain melody are the relationships between the sounds. The phenomenal “melody” fact is
independent of its phenomenal constituents, sounds.1

The melodic phrase is fundamentally a relational structure: the melody remains a

melody even if all its sounds are substituted as long as all the temporal relations and
intervals that connect the notes are maintained, from the first note on.
Kanizsa, 1980, p. 11

Taking optical illusions as an example, Benussi showed the extent to which the productive act
identifies with psychic activity. He showed in fact that it is possible to counteract the illusionary
effect by forcedly focusing attention on the individual sensorial elements, thus impeding the
natural tendency to organize these elements into a totality (Figure 7.2).
We are particularly aware of this “act of production” when looking at ambiguous sketches
such as the candlesticks that can also be seen as two human profiles (Figure 7.3) or the face of
an old woman that can also be seen as a young girl (Figure 7.4), to cite two of the best known.
When we switch from one structure to the other, we have the perception of the ongoing “act
of production”.
When the interpretation process establishes new links between the elements in a scene
(restructuring), or allows the imagination to see the object that is either hidden or covered and
only partially visible (without anything truly perceptual being added), this is when the pro-
cess of integration or mental completion, which generally is part of an interpretative process,
takes over.
The Gestaltists considered that the concept of restructuring, to which they attributed a cen-
tral role in visual perception, was also central in problem solving. They did not use the term

Figure 7.2 The Müller-Lyer illusion

133
134

Laura Macchi and Maria Bagassi

Figure 7.3 The figure-ground organization

Figure 7.4 The young girl-old woman illusion

“problem solving,” however, preferring “productive thinking,” as they were convinced that
thinking is not merely a reproduction of the past, a reemergence of ideas, images, behaviors
that already existed; in addition to the reproductive activities of thought, there are processes
that create and produce that which does not yet exist.
The Gestalt psychologists focused their research on establishing the phenomenology of
these productive processes and the characteristics that distinguish them from the reproductive
processes; they studied the conditions that give rise to these processes or allow them to arise,
and those which, on the contrary, impede them, and the decisive moments in the process
when light dawns and understanding is reached. The famous “Aha!” experience of genuine
insight accompanies this change in representation, the restructuring. This is characterized by a
switch in direction that occurs together with the transformation of the problem or a change
in our understanding of an essential relationship. The decisive points in thought processes, the

134
135

Bounded rationality and problem solving

moments of sudden comprehension, of the “Aha!,” of the new, are always at the same time
moments in which such a sudden restructuring of the thought-material takes place, in which
something “tips over” (Duncker, 1945, pp. 29–30).
Thirty years later, the reaction of cognitivism to behaviorism had many analogies with the
Gestaltists’ reaction to associationism, to the point where Herbert Simon could, in a certain
sense, be considered the heir of the second generation of the Gestalt theory, of which Karl
Duncker was the most eminent scholar. Simon himself admitted this, sharing the key concepts
of Duncker and Selz’s theory, and expressing them by a Human Information Processing
approach (HIP), which is characterized by the observability principle of the cognitive process.
“we can see a number of quite specific connections between the formulations of Duncker and
Selz, on the one hand, and our own, on the other …” (Simon 1984, p. 149).
Prior to the rise of information-processing theories, perhaps only Duncker fully shared with
Selz this notion of the proper goal for problem-solving research, although some of the same
flavor comes through in the writings of Koehler and the other Gestaltists. Certainly, in these
two writers, the orientation to process comes through far more clearly than in other writings
on this topic prior to World War II (Simon, 1984, p. 153).

The simulation models of the 1950s were offspring of the marriage between that had
emerged from symbolic logic and cybernetics, on the one side, and Würzburg and
Gestalt psychology, on the other … From Würzburg and Gestalt psychology were
inherited the ideas that long-term memory is an organization of directed associations,
and that problem solving is a process of selective goal-oriented search.
Simon, 1979, pp. 364–365

The legacy of Gestaltism can be considered a fruitful vantage point from which to view
current research in the field of problem solving. This entails acknowledging the affinities,
and, to a certain extent, the continuity between the approaches of Duncker and HIP theory,
and the connections between the formulations of Duncker and Simon and colleagues, as they
have exerted a decisive influence on the field of problem solving for more than half a century.
Let us now consider crucial concepts of the two theories, underlining both similarities and
differences.
Duncker was not only interested in the moment when the solution was found, but also in
the thought process, the phases by which the restructuring is reached:

But while, with undoubted justification, great stress was laid on the significance of
restructuration or reorganization in thinking, another side of the problem was almost
completely lost from view. In what way do these restructurations, and with them the
solution, arise?
Duncker, 1945, pp. 29–30

The final form of a solution is typically attained by way of mediating phases of the process, of
which each one, in retrospect, possesses the character of a solution, and, in prospect, that of
a problem; on the contrary, the principle, the functional value of the solution, typically arises
first, and the final form of the solution in question develops only as this principle becomes
successively more and more concrete. In other words, the general or “essential” properties of a
solution genetically precede the specific properties (Duncker, 1945, pp. 7–8).
The solution process involves a selective research in phases, over a delimited area; previously
explored paths that turned out to be dead ends must be discarded and new paths adopted. The

135
136

Laura Macchi and Maria Bagassi

new path can even have been evaluated in a previous phase. In fact, as Duncker stressed from
the beginning of his work:

In short, from the very first, the deliberating and searching [are] always confined to
a province, which is relatively narrow as to space and content. Thus preparation is
made for the more discrete phases of a solution by certain approximate regional
demarcations, i.e., by phases in which necessary but not yet sufficient properties of the
solution are demanded.
Duncker, 1945, p. 10

Duncker’s genealogical tree of the solution is made up of the aggregate of ways in which the solu-
tion can be reached, starting from the problem situation. It represents an ideal solution process
similar to the HIP typical protocol, and, in a certain sense, anticipates the concept of “program.”

As yet we have dealt only with the progress from the superordinate to the subordinate
phases (or vice versa), in other words, with progress along a given genealogical line …
Here the line itself is continually changed, and one way of approach gives way to
another. Such a transition to phases in another line takes place typically when some
tentative solution does not satisfy, or when one makes no further progress in a given
direction. Another solution, more or less clearly defined, is then looked for/
Duncker, 1945, p. 12

Duncker’s description of how a problem solver deals with the search for the solution to a
problem prefigures the labyrinth model of information processing:

The problem solver's search for the solution, is an “odyssey” through problem space
from one state of knowledge to another, until his/her knowledge state includes the
solution. A person faced with a problem moves within the problem space just as in a
labyrinth; he searches for the right path, retraces his steps when he comes up against
a dead end, and sometimes even returns to the starting point; he will form and apply
a strategy of sorts, making a selective search in the problem space.
Simon, Newell, & Shaw, 1979, p. 148

Duncker and Simon used real protocols, registering attempts to solve the problem, to rebuild
what is assumed to be a solution process but which will only correspond fortuitously to the
succession of actual attempts made to solve the problem.
The technique used here is thinking aloud, which is completely different, for example, from
Wundt’s introspection:

The subjects (Ss), who were mostly students of universities or of colleges, were given
various thinking problems, with the request that they think aloud … This instruc-
tion, "Think aloud", is not identical with the instruction to introspect which has
been common in experiments on thought-processes. While the introspecter makes
himself as thinking the object of his attention [and so differentiates himself intention-
ally from himself as a thinking subject],2 the subject who is thinking aloud remains
immediately directed to the problem, so to speak allowing his activity to become
verbal.
Duncker, 1945, p. 2

136
137

Bounded rationality and problem solving

They both adopted a descriptive-phenomenological approach to problem solving. Just as the

Gestaltists studied the principles of organization of reality in perception, which sometimes
are present when a problem comes into being, so, in Simon’s theory, the basic characteristics
of the human cognitive system are those which generally lead to the solution of a problem.
The boundaries are indeed characteristics, but characteristics without negative connotations.
This phenomenological approach, which is in no way conditioned by an ideal functioning
model borrowed from extra-psychological normative theories, allows us to think of heuristics
as expressions of our intelligence.
According to Duncker, the problem solver never blunders along blindly, seeking the solution
by trial and error; he proceeds by applying heuristic strategies and methods:

the whole process, from the original setting of the problem to the final solution,
appears as a series of more or less concrete proposals … But however primitive they
may be, this one thing is certain, that they cannot be discussed in terms of meaningless,
blind, trial-and-error reactions.
Duncker, 1945, p. 3

What is even more relevant in Duncker’s concept is that the heuristic captures the intrinsic
necessity of the solution itself, guiding the problem solver to incessantly analyze the conflict
and to eliminate his fixation. In this sense, it constitutes the main, if not the only3 method of
thought in problem solving:

We can therefore say that “insistent” analyses of the situation, especially the endeavor to vary
appropriate elements meaningfully sub specie of the goal, must belong to the essential nature of
a solution through thinking. We may call such relatively general procedures, “heuristic
methods of thinking.”
Duncker, 1945, pp. 20–21

According to Simon, on the other hand, the term “heuristics” indicates any principle and
expedient that contributes to reducing the time that would normally be necessary to reach a
solution. Given that often an exhaustive search of all the possible ways to solve a problem is
just not viable for a human being, the use of heuristics reduces the problem-solving process to
manageable proportions.
But the economy of heuristics is not only an advantage, it is also an essential condition for
the possibility of dealing with and solving problems, albeit at a price––the uncertainty of the
outcome. Efficacious thought is only possible through the heuristic reduction of the search.

Heuristic methods [allow us] to carry out highly selective searches, hence to cut down
enormous problem spaces to sizes that a slow, serial processor could handle. Selectivity
of search, not speed, was taken as the key organizing principle … Heuristic methods
that make this selectivity possible have turned out to be the central magic in all human
problem solving that has been studied to date.
Simon & Newell, 1971, p. 147

Simon and Newell add:

The power of heuristics resides in their capability for examining small, promising
regions of the entire space and simply ignoring the rest. We need not be concerned

137
138

Laura Macchi and Maria Bagassi

with how large the haystack is, if we can identify a small part of it in which we are
quite sure to find a needle.
1971, p. 151

This changes radically when considering the psychology of reasoning, and in particular Kahneman
and Tversky’s research program “Heuristics and Biases” on probability judgment. In their work,
they attribute a quite different definition to heuristics, in spite of the fact that the authors
explicitly state that they have taken the concept from Simon. Not only are their heuristics sim-
plified or abbreviated procedures, but they are subject to their own criteria and mechanisms,
completely foreign to the logic of the problem being dealt with and the decision to be taken.
It is no longer a case of procedures that “do not guarantee finding the solution” depending on
their economy, but of procedures that will lead to the correct solution of the problem purely
by chance.
The uncertainty of the outcome that derives from the adoption of heuristic methods has
been radicalized by the acceptance of heuristics by Kahneman and Tversky, which leads to an
irrational deviation, which had nothing to do with Simon’s view. According to Simon, simplifi-
cation is the identification of what is salient and crucial, while Kahneman and Tversky consider
it to be “neglect” of crucial information for probability judgment.
A direct consequence of this theoretical framework of the psychology of reasoning is the
introduction of the theoretical construct of the bias, i.e., an intrinsic tendency to error of the
human cognitive system. According to Kahneman and Tversky, it is the use of heuristics that
produces biases.

There are situations in which people assess the frequency of a class or the probability
of an event by the ease with which instances or occurrences can be brought to mind
… However, availability is affected by factors other than frequency and probability.
Consequently, the reliance on availability leads to predictable biases …
Kahneman, Slovic, & Tversky, 1982, p. 11

The challenging issue of insight problem solving

The affinities and a certain continuity between Duncker’s formulation and that of the Human
Information Processing Theory are quite surprising, to the extent that it is possible to consider
Simon, with all due distinctions, as the heir of the Gestalt psychology. That said, the differences
are crucial and given the antithetical theoretical framework of the two authors, it would not be
reasonable to expect anything different. For Duncker, the theoretical framework remains that of
the early Gestaltists and therefore is restructuralist. The study of the process of thought provides
the means of studying “how restructuring comes about.”
According to Duncker: “The finding of a functional value of a solution means each time a
reformulation of the original problem” (1945, p. 8):

Every solution consists in some alteration of the given situation. But not only this or
that in the situation is changed …; over and beyond this the psychological structure of
the situation as a whole or of certain significant parts is changed. Such alterations are
called restructurations.
Duncker, 1945, p. 29

138
139

Bounded rationality and problem solving

This discontinuity, this alteration is not present in problem-solving process as studied by Simon
(take the Crypto-arithmetic task, for example, or the Cannibals and Missionaries problem).
The problem solver may make mistakes (illegal moves), following paths that turn out to be
dead ends and so has to retrace his steps, and slowly but surely (step-by-step) makes his way to
the solution: it is only a question of time. The difficulty lies in the calculation, the number of
operations to be performed, and the quantity of data to be processed and remembered.
When the Weltanschauung of cognitivism informed psychology, and confidence in the possi-
bility of making testable models for the functioning of the mind was at its highest, Simon wrote:

The only solution to this problem is the hard solution. Psychology is now taking the
road taken earlier by other sciences: it is introducing essential formalisms to describe
and explain its phenomena. Natural language formulations of the phenomena of
human thinking did not yield explanations of what was going on; formulations in
information-processing languages appear to be yielding such explanations.
Simon & Newell, 1971, p. 148

In that particular historical phase, to keep faith with this “hard” methodology, the aim of which
is to measure––observe––every step of the problem-solving process, Simon inevitably had to
exclude insight problems from his Human Information Processing research as he considered
them 'ill defined' for the task environment (Hayes & Simon, 1974; Newell & Simon, 1972;
Simon & Hayes, 1976). He continued to hold that “the programmability of the theories is the
guarantor of their operationality, an iron-clad insurance against admitting magical entities into the
head” (Simon, 1979).
Some years later, in response to an article Michael Wertheimer wrote in 1985, Simon
expressed the same absolute faith in the possibility of translating the problem solving process for
insight problems into computer programming:

One serious difficulty in testing Gestalt theories is that the concepts that play central
roles in these theories are not often defined in terms of operations or measurements,
but are usually assumed to be directly understood and observable … If we put aside
our nostalgia for a rich, but largely non-operational, vocabulary, and when we focus
instead on the observable processes of thought, we see that we are much further
advanced in understanding these processes … the phenomena of insight are fully vis-
ible in the behavior of today's computers when they are programmed appropriately to
simulate human thinking.
Simon, 1986, pp. 242, 254

The explanation for these very fundamental differences in the two approaches lies in the diver-
sity of the problems they were studying. The focus on the restructuration privileged insight
problems as an object of study, while interest for observability privileged the procedural, non-
insight problems. The choice of the object of study was not accidental, but congenial to the
different intent pursued. Futile taxonomies apart, the only distinction worthy of note is that
proposed by Mosconi (2016), focusing squarely on the different kind of difficulty, i.e., between
tasks and problems.
In the problems studied by Simon (the tasks) the difficulty lies in their intractability, given
the basic characteristics of the human information-processing system (alias, the human problem
solver): “serial processing, small short-term memory, infinite long-term memory with fast

139
140

Laura Macchi and Maria Bagassi

retrieval but slow storage – [which] impose strong constraints on the ways in which the system
can seek solutions to problems in larger problem spaces” (Simon & Newell, 1971, p. 149).
In the problems studied by the Gestaltists (the problems), on the other hand, the difficulty
does not lie in the complexity of the calculations and the burdensome process necessary to
reach the solution, but in one or more critical points, that create a situation of conflict and,
as a consequence, an unsurmountable impasse. None of this happens in Simon’s research on
problem solving.
In fact,

If a situation is introduced in a certain perceptual structuration, and if this structure is

still “real” or “alive”, thinking achieves a contrary structuration only against the resist-
ance of the former structure. The degree of this difficulty varies among individuals.
Duncker 1945, p. 108

This, then, is the question at hand. When, in 1990, Simon, in collaboration with Kaplan,
explored in depth the peculiarities of the search and the steps of the solution process for insight
problems, they dealt with the issue of a change in representation, alias restructuration, in their sem-
inal article “In search of insight.” They provided

a major extension of the standard information processing theory of problem solving to

handle insight problems … the same processes that are ordinarily used to search within
problem space can be used to search for a problem space (problem representation).
Kaplan & Simon, 1990, p. 376

According to the authors, the problem solver who finds himself at an impasse must first of all
make a conscious decision to change his way of looking at the problem by adopting the “Try a
Switch” meta-heuristic, given that the initial representation of the problem did not provide him
with the means of reaching the solution, “but the space of possible problem spaces is exceed-
ingly ill-defined, in fact, infinite.” Once this conscious decision has been made, “the subject has
to have or obtain strong constraints that guide [the] search and make it highly selective,” i.e.,
he must be on the lookout for the important cues that will help him select the representation
that will lead to the solution, from the infinite number available. In his search for these decisive
cues to switch the representation in the representation metaspace, he is assisted by adopting a
heuristic initiated by boundaries of salience,4 that fits the problem.

The simulation demonstrates that once the decision has been reached to search for a new
representation, [in the Mutilated Checkerboard insight problem] attention to the parity
cue is sufficient to construct the new representation by modifying the initial one and
to find the solution by means of a reasoning process. What leads subjects to consider
the critical cues that are prerequisite to such a switch?
The Notice Invariants heuristic is of great importance in permitting subjects, par-
ticularly those not given strong perceptual or verbal hints, to narrow their search for
new representations.
Kaplan & Simon, 1990, p. 403

Yet again, it is worth noticing the connection between Simon and Duncker, who defines
invariants as the functional principles of a solution:

140
141

Bounded rationality and problem solving

To the same degree to which a solution is understood, it can be transposed, which

means that under altered conditions it may be changed correspondingly in such a way
as to preserve its functional value. For, one can transpose a solution only when one
has grasped its functional value, its general principle, i.e., the invariants from which,
by introduction of changed conditions, the corresponding variations of the solution
follow each time.
Duncker, 1945, p. 5

With insight problems too, the way Simon explains the difficulty revolves around a sort of
“quantitative unmanageability,” in this case constituted by the excessive, “infinite” number of
possible representations of the problem.
This way of dealing with the switch of problem space––the change in representation––
describes what happens when the person solves an insight problem, rather than actually
explaining the solution process. However, what makes the problem solver focus on certain
cues, rather than others, which are identified as crucial for a switch that will lead him out of
the impasse, remains obscure.
Keeping the search on a level of consciousness in order to simulate the solution process,
Simon conserves the key theoretical constructs of his theory such as the concepts of search and
process in which the steps can always be reconstructed, and in which self-conscious and delib-
erative connotation is the heart.
Although he did pose the question as to what determines the switch, Simon was not able
to find answers to the crucial interrogatives and so ended up with a sort of explanatory circu-
larity. In point of fact, deciding to change the representation alone is not sufficient to solve the
problem, if we do not change our understanding of an essential, new relationship.5
The emphasis on the observability and registrability of each step in the solution process,
generating a typical protocol that can be simulated in a calculator, translates into a sort of bias,
when he assumes insight problems as the object of his research, as the solution to these problems
is characterized by discontinuity and the transition from the impasse to the solution remains a
mystery to the problem solver himself.
Moreover, the solution can appear after a period of incubation that may vary in time, during
which the problem solver withdraws into himself.6
The difficulty continues to be quantitative even with insight problems such as the Mutilated
Checkerboard (MC) problem (see Appendix A), to the point at which, “the task would be
hopeless without some source of search constraint.” In the domain of the MC problem, the
authors identified four potential sources of constraint: “perceptual cues in the problem, hints
provided by the experimenter, prior knowledge that the subject might bring to the problem,
and heuristics – in particular the Notice Invariants heuristic” (Kaplan & Simon 1990, p. 413).
There is still something magical in how problem solvers adopt the heuristic that is applic-
able to the solution and which is on the same level as the other constraints, only, in the light of
the results, considered to be more efficient in identifying the representation or problem space
that is most suited to the problem, as “focusing attention on invariant features of the problem
situation allows subjects to convert a search in an enormous and unmanageable space (in which
they have no relevant generators) to a search in a smaller space (with generators available)”
(Kaplan & Simon 1990).
The Gestaltists would have said that the problem solvers discover the solution, when they
“see” that in the Mutilated Checkerboard problem there is no longer reciprocal parity between
the black and the white squares and so they restructure.

141
142

Laura Macchi and Maria Bagassi

Simon himself was fully aware that the difficulty lies in the fact that “you do not know
beforehand where the solution may lie” and the representation switch has still to take place.
His question: “But how is this change of representation brought about?” remains without an
answer; it continues to be a mysterious event, if we do not venture beyond the conscious layer
of solution processing.

The role of unconscious analytic thought in insight problem solving:

the emergence of the shadow area
The difficulty in changing the representation does not appear to involve the working memory
capacity, nor the conscious retrieval from memory of solutions or crucial parts of solutions to
reproduce. It requires of necessity not conscious, implicit processing––incubation––and for
this reason it is a challenge to the current theories of thought, regarding the central role of
consciousness, given that, in these cases, restructuring (or change in representation) is not
performed consciously.
Incubation, which remains the core issue in the renewed interest in insight problems is still a
crucial question to be solved (Bagassi & Macchi 2016; Ball, Marsh, Litchfield, Cook, & Booth,
2015; Fleck & Weisberg, 2013; Gilhooly, Ball, & Macchi, 2018; Gilhooly, Georgiou, & Devery,
2013; Gilhooly, Georgiou, Sirota, & Paphiti-Galeano, 2015; Macchi & Bagassi, 2012, 2015; Sio
& Ormerod, 2009; Threadgold, Marsh, & Ball, 2018).
A heterogeneous complex of unresolved critical issues underlies the research on this subject
and still revolves around the controversy of the relationship between the conscious and uncon-
scious layers of thought in the solution of insight problems.
However, the various mechanisms that have been proposed only describe the characteristics
of the solution but do not explain the processes of reasoning that have made the solution
possible (these include, for example, eliciting new information, selective forgetting, strategy
switching, and relaxing self-imposed inappropriate constraints).
Moreover, a general characteristic that is common to the literature on insight problems in
general, and in particular on the incubation-solution relationship, is the total absence of an ana-
lysis of the types of difficulty found in individual insight problems. In other words, what makes
a difficult insight problem difficult? What kinds of difficulty are we facing? If it were possible
to lay them out as a continuum in an ascending order of difficulty, we would see that, in fact,
the difficulty correlates with the incubation, and this, in turn, with the possible discovery of the
solution, thus allowing the restructuring process to occur. Therefore, incubation may offer a
measure of the degree and type of difficulty of the problem, as it may vary in length, depending
on the degree of gravity of the state of impasse (Macchi & Bagassi, 2012, 2015; Segal, 2004).
We obtained experimental evidence in support of our hypothesis, using the horse-trading
problem, whose answer is given here.

The horse-trading problem

A man bought a horse for £70 and sold it for £80, then he bought it back for £90, and sold it for
£100. How much did he make?
In this problem the correct response (£20) is found by approximately half of those who attempt
to solve it, by restructuring in two separate financial transactions7 (see Appendix B).

142
143

Bounded rationality and problem solving

We have also examined a problem that is one of the most intriguing insight problems, a genuine
challenge, the study window problem, which at first sight appears impossible to solve (Macchi
& Bagassi, 2014, 2015; Mosconi & D’Urso, 1974). This is an example of a problem which
cannot be solved analytically, in which the impasse seems to be unsurmountable, an incubation
period is necessary and the solution is the outcome of a deep restructuring (see Appendix C).

The study window problem

The study window measures 1 meter in height and 1 meter wide (Figure 7.5). The owner decides
to enlarge it and calls in a workman. He instructs the man to double the area of the window
without changing its shape and so that it still measures 1 meter by 1 meter. The workman carried
out the commission. How did he do it?

At this point, the question is, what kind of unconscious intelligent thought operates during
the incubation to solve these special problems? Through experiments in brain imaging, it
is now possible to identify certain regions of the brain that contribute both to unconscious
intuition and to the processing that follows. Jung-Beeman, Bowden, Haberman, Frymiare,
Arambel-Liu, Greenblatt, et al. (2004) found that creative intuition is the culmination of a
series of transitional cerebral states that operate in different sites and for different lengths of
time. According to these authors, creative intuition is a delicate mental balancing act that
requires periods of concentration from the brain but also moments in which the mind wanders
and retraces its steps, in particular during the incubation period or when it comes up against
a dead end.

Incubation is actually the necessary but not sufficient condition to reach the solution.
It allows the process but does not guarantee success; however, if it is inhibited, for
example, by compelling participants to verbalize, the solution process will be impeded.

B D

Figure 7.5 The study window problem solution

143
144

Laura Macchi and Maria Bagassi

It has been observed that under these circumstances, subjects giving thinking aloud
protocols often fall silent. When asked what they are thinking about, they may say
“I’m not thinking about anything,” an indication that they have no available strategy
for the meta-level search for the new representation.
Kaplan & Simon, 1990

When the participants in these studies (Macchi & Bagassi, 2012; Schooler et al., 1993) were still
exploring ways of discovering a new problem representation, they were not able to express con-
sciously and therefore to verbalize their attempts to find the solution. Indeed, our data showed
that serial “on-line” verbalization, compelling participants to “restless” verbalization, impairs
reasoning in insight problem solving; this provides support for the hypothesis of the necessity
of an incubation period during which the thinking processes involved are mainly unconscious.
During this stage of wide-range searching, the solution still has to be found, and verbaliza-
tion acts as a constraint, continuously forcing thought back to a conscious, explicit level and
maintaining it in the impasse of the default representation. Conscious explicit reasoning elicited
by verbalization clings to the default interpretation, thus impeding the search process, which is
mainly unconscious and unreportable.
Hence, we speculate that the creative act of restructuring implies a high-level implicit
thought, a sort of unconscious analytic thought; where, however, analytic thought is not to
be understood in the sense of a gradual, step-by-step simplification of the difficulties in the
given problem. It would rather be a gentle, unconscious analysis of the conflict between the
initial problematic representation and the goal of the problem, until the same data are seen in
a different light and new relations are found by exploring different interpretations, neither by
exhaustive searches nor by abstractions but by involving a relationship between the data that is
most pertinent to the solution of the problem. In this way, each individual stimulus takes on a
different meaning with respect to the other elements and to the whole, contributing to a new
representation of the problem, to the understanding of which it holistically concurs. Indeed,
the original representation of the data changes when a new relation is discovered, giving rise to
a gestalt, a different vision of the whole, which has a new meaning.
In other words, solving an insight problem––restructuring––means discovering a new per-
spective, a different sense to the existing relations. The interrelations between the elements of
the default, usual interpretation have to be loosened in order to perceive new possibilities, to
grasp from among many salient cues what is the most pertinent with the aim of the task, or
in other words, to reach understanding. This type of process cuts across both conscious and
unconscious thought.
The default representation (a preferred organization of stimulus and a generalized interpret-
ation of the text of the problem) does not allow participants to pursue the solution.
Consider, for instance, the above-mentioned study window problem. This problem was
investigated in a previous study (Macchi & Bagassi, 2014). For all the participants the problem
appeared impossible to solve, and nobody actually solved it. In our view, the solution is blocked
by the problem solver’s mental representation of the window described in the statement, which
she/he will “visualise” in the usual, orthogonal orientation (with the sides parallel to the
vertical-horizontal coordinates). The information provided regarding the dimensions brings a
square form to mind. The problem solver interprets the window to be a square 1 meter high
by 1 meter wide, resting on one side. Furthermore, the problem states “without changing its
shape,” intending geometric shape of the two windows (square, independently of the orientation
of the window), while the problem solver interprets this as meaning the phenomenic shape of
the two windows (two squares with the same orthogonal orientation). And this is where the

144
145

Bounded rationality and problem solving

difficulty of the problem lies, in the mental representation of the window and the concur-
rent interpretation of the text of the problem (see Appendix C). Actually, spatial orientation
is a decisive factor in the perception of forms (Mach, 1914). Two identical shapes seen from
different orientations take on a different phenomenic identity.
The “inverted” version of the problem gave less trouble:

The owner decides to make it smaller and calls in a workman. He instructs the man
to halve the area of the window ...

With this version, 30 percent of the participants solved the problem. They started from the
representation of the orthogonal square (ABCD) (see Figure 7.5) and looked for the solution
within the square, trying to respect the required height and width of the window, and inev-
itably changing the orientation of the internal square. This time the height and width are the
diagonals, rather than the side (base and height) of the square.
In another version (the “orientation” version) (Figure 7.6), it was explicit that orientation
was not a mandatory attribute of the shape, and this time 66 percent of the participants found
the solution immediately (see Appendix C). This confirms the hypothesis that an inappropriate
representation of the relation between the orthogonal orientation of the square and its geo-
metric shape is the origin of the misunderstanding.
In our view, overcoming a fixation requires elaboration and comprehension of the context
that is more relevant to the aim. The interrelations between the parts of the default interpret-
ation have to be loosened in order to perceive the new, in other words, to restructure. The
impasse, the difficulty of reaching a solution, should activate an increasingly focussed and rele-
vant search and a reconsideration of the available data in the light of the goal. It is analytical
research “that recognizes the primacy of the whole, and its dynamic nature, and its decisive role
in articulating its parts” (Wertheimer 1985, pp. 24, 25).
The relevance of text and the reinterpretation of perceptual stimuli, goal-oriented to the
task, were worked out in unison in an interrelated interpretative “game.” Over and above
the macro-distinction between insight and non-insight problems, the crucial characteristic of
insight problems which distinguishes them from non-insight problems is that they originate

D C

A B

Figure 7.6 The Wertheimer’s Parallelogram-Square problem

145
146

Laura Macchi and Maria Bagassi

from a misunderstanding, of varying complexity and difficulty, but which always requires a
restructuring process to overcome.
The Gestalt vision has been invoked, with different inflections, by the special process views
of insight problem solving, investigating the qualitatively diverse processes that elude the con-
trol of consciousness through spreading activation of unconscious knowledge (Ohlsson, 2012;
Ollinger, Jones, & Knoblich, 2008; Schooler, Ohlsson, & Brooks, 1993). This search goes
beyond the boundaries of the working memory. The process that leads to the discovery of the
solution through restructuring is mainly unconscious, characterized by a period of incubation,
and can only be described a posteriori (Gilhooly, Fioratou, & Henretty, 2010). The characteristic
unconsciousness with which these processes are performed has led to them being defined as
automatic, spontaneous associations (Ash & Wiley, 2006; Fleck, 2008; Fleck & Weisberg, 2013;
Schooler et al., 1993).
On the other hand, the cognitivist approach, known as business as usual, requires a greater
working memory capacity for explaining insight problems: intelligence fades without the
“light” of consciousness and therefore identifies with conscious reflective thought.
Both these approaches have critical aspects that reveal the complexity of the issue to
hand: the former grasps the specificity of the phenomenon of discovery, which characterizes
insight problems and creative thought, but is not in a position to identify the explicative
processes because it does not attribute a selective quality to unconscious processes, as they
continue to be merely associative, automatic, and capable of producing associations that
will contribute to finding the solution almost by chance. The limit of the business as usual
approach, on the other hand, is that it levels off the specificity of genuine insight problems,
lumping them together with problems that can be solved with explicit analytic thought
processes.
Recently, Weisberg (2015) has proposed an integrated theory of insight in problem solving.
According to Weisberg, analytic thinking, given its dynamic nature, can produce a novel out-
come; in problem solving in particular, it can generate a complex interaction between the
possible solutions and the situation, such that new information constantly emerges, resulting
in novelty. When we change strategy, we select a better route; this change is known as
restructuration without insight and remains on the conscious, explicit plane.
Weisberg rather optimistically claims that when the restructuring does not occur and the
subject is at an impasse, which

may result in coming to mind of a new representation of the problem … That new
representation may bring to mind at the very least a new way of approaching the
problem and, if the problem is relatively simple, may bring with it a quick and smooth
solution.
Weisberg 2015, p. 34

However, this concept of intelligence does not explain how we suddenly see the solution to
a problem after a period of fumbling in the dark (the impasse), during which our conscious
analytical thought has stalled and does not extract any useful information from the failure.
Thought is incapable of throwing light on the situation, and if the solution is hit upon, it has
the characteristics of a “mysterious event.”
Finally, also Weisberg’s approach makes little progress in explaining the so-called “mysterious
event” in the solution of insight problems, relegating them to situations of normal adminis-
tration that can be dealt with by conscious analytical thought. However, when the solution is
mainly the result of a covert and unconscious mind-wandering process, it cannot be attributed,

146
147

Bounded rationality and problem solving

even exclusively, to reflective, conscious thinking (Baird, Smallwood, Mrazek, Kam, Franklin,
et al., 2012; Macchi & Bagassi, 2012; Smallwood & Schooler, 2006).
These problems may seem bizarre or constructed to provide an intellectual divertissement,
but they are a paradigmatic case of human creativity in which intelligence is at its acme. Their
study provides a privileged route to understanding the processes underlying creative thought,
scientific discovery, and innovation, and all situations in which the mind has to face something
in a new way.

Restructuring as reinterpreting: the interpretative heuristic

The search for meaning, in view of an objective, characterizes every activity of the human
cognitive system at all levels, from perception to language and reasoning. For instance, syntax
in itself does not give direct access to meaning, only on the basis of the rules that discipline it.
Similarly, perceiving external reality is not just a question of registering stimuli; sensorial data
have to be interpreted and organized, and relations created.
When analyzing sensory experience, it is important to realize that our conscious sensations
differ qualitatively from the physical properties of stimuli because, as Kant and the idealists
predicted, the nervous system extracts only certain pieces of information from each stimulus
while ignoring others. It then interprets this information within the constraints of the brain’s
intrinsic structure and previous experience. Thus sounds, words, music, color, tones are mental
creations constructed by the brain out of sensory experience. They do not exist as such outside
the brain (Gardner & Johnson, 2013, p. 455).
Our view, therefore, is that this interpretative function is a characteristic inherent to all reasoning
processes and is an adaptive characteristic of the human cognitive system in general. Rather
than abstracting from contextual elements, this function exploits their potential informativeness
(Levinson, 1995, 2013; Mercier & Sperber, 2011; Sperber & Wilson, 1986; Tomasello, 2009).
It guarantees cognitive economy when meanings and relations are familiar, permitting recog-
nition in a “blink of an eye.”
This same process becomes much more arduous when meanings and relations are unfamiliar,
obliging us to face the novel. When this happens, we have to come to terms with the fact that
the usual, default interpretation will not work, and this is a necessary condition for exploring
other ways of interpreting the situation. A restless, conscious, and unconscious search for other
possible relations between the parts and the whole ensues until everything falls into place and
nothing is left unexplained, with an interpretative heuristic-type process.
This ability always implies an analytic, multilayered reasoning, which works on a conscious
and unconscious layer by processing explicit and implicit contents, presuppositions, and beliefs,
but is always informed by relevance.8
Sometimes, however, our initial interpretation does not support understanding, and
misunderstanding is inevitable; as a result, sooner or later we come up against an impasse.
We are able to get out of this impasse by neglecting the default interpretation and looking for
another one that is more pertinent to the situation and which helps us grasp the meaning that
matches both the context and the speaker’s intention; this requires continuous adjustments until
all makes sense.
For instance, grasping a witty comment and understanding rhetorical figures of speech such
as irony and metaphor are the result of an interpretative process adhering to or detracting from
the explicit meaning as the case requires.
This analysis allows us to consider insight problems as paradigmatic examples of
misunderstanding, in that they arise from a quid pro quo, a glitch in communications. When

147
148

Laura Macchi and Maria Bagassi

insight problems are used in research, it could be said that the researcher sets a trap, more or
less intentionally, inducing an interpretation that appears to be pertinent to the data and to the
text; this interpretation is adopted more or less automatically because it has been validated by
use. But in point of fact, the solution presupposes a different, unusual interpretation that is only
theoretically possible, just as when someone makes a witty comment, playing on an unusual
interpretation of a sentence.
Indeed, the solution—restructuring––is a re-interpretation of the relationship between the
data and the aim of the task, a search for the appropriate meaning carried out at a deeper level,
not by automaticity. If this is true, then a disambiguant reformulation of the problem that
eliminates the trap into which the subject has fallen, should produce restructuring and the way
to the solution.
Our view, therefore, is that the interpretative heuristic is a characteristic inherent to all
insight problem-solving processes. This in itself is not mysterious or inexplicable, it is the
result of an incessant search for possible relations between the parts and the whole until every-
thing falls into place and nothing is left unexplained, but makes sense using an interpretative
heuristic-type process.
The original representation of the data changes when a new relation is discovered, giving
rise to a gestalt, a different vision of the whole, something which has a new sense. The same
data, seen in a different light, provide a view that is different to the default, the original starting
point. As we have seen, new relations are found by exploring different interpretations, and not
by exhaustive searches or abstractions.
Gestalt and Human Information Processing have really grasped the core of human
thinking, describing its functioning and identifying the beauty as well as the simplicity of
human intelligence, which is able to deal with both areas of tasks that are too large, reducing
them to a manageable representation, but also capable of changing representation, discovering
misunderstandings, restructuring. In both cases, thought always proceeds by adopting a heur-
istic interpretative approach, essentially aimed at identifying relationships. We think, we per-
ceive in a context.
In the case of the insight problems, the solution requires a selective processing of the con-
text at a deeper level than that of the default, by re-examining the relations between the data,
adopting an interpretative heuristic-type process. In this case, our mind fluctuates between con-
scious and unconscious levels in a wide-ranging analytical search.
This perspective allows us to reconsider the question of bounded rationality. The study of
how we reason highlights the principles that describe the functioning of the human thought;
these are not limits, biases, which in themselves lead to error––cognitive illusions––but
characteristics by which we perceive, think, and communicate. Take optical illusions, for
instance. It is the same perceptive mechanism of distortion that creates three-dimensionality
from the two-dimensional painting Images of art, as well as all images, do not represent reality
but rather perceptions, imagination, and expectations of the viewer, and his knowledge of
other images recalled from memory. Art is, in part, a creation of the mind; the sensory infor-
mation, organized in accordance with Gestaltist principles, allows reality to be created by
the mind.
Our claim is that the cognitive illusions as well as the perceptual illusions, are the side
effect, the epiphenomenon of the functioning of the cognitive system, which is realized in
the absence of context, with the occurrence of a misunderstanding. In the optical illusion of
Mueller-Lyer, for example, the experimenter measures the objective two-dimensional length

148
149

Bounded rationality and problem solving

Figure 7.7 The Lamentation of Christ by Andrea Mantegna

of the two segments, neglecting the three-dimensional effects produced by extensions to the
apexes (see Figure 7.2). Art, instead, cooperatively exploits the three-dimensional interpretation
of two-dimensional data.
By dizzyingly foreshortening the perspective, Andrea Mantegna inserted the body of his
dead Christ into a dilated space that has the depth of a room, in a painting of only 68 cm. by
81 cm (Figure 7.7). If the viewer moves, he preserves the perspective as in reality. In this case,
the tacit cooperation between viewer and artist, on which the fruition of the artwork is based,
gives life to a powerful representation of reality.

Appendix A

The classic mutilated checkerboard (MC) problem

Cover the 62 remaining squares in Figure A7.1 with 31 dominoes.

Each domino covers two adjacent squares or:
Prove logically why such a covering is impossible

149
150

Laura Macchi and Maria Bagassi

Figure A7.1

Appendix B

Horse-trading problem

A man bought a horse for £70 and sold it for 80, then he bought it back for 90, and sold it for
100. How much did he make?
The majority of problem solvers come up with one of two answers: £20 and £10 (Macchi
and Bagassi, 2015).
The typical explanation for the first answer, which is the correct one, is, “the dealer makes
£10 on the first transaction (70 - 80) and 10 on the second (90 - 100),” while for the second
answer, the usual explanation given is “the dealer makes £10 on the first transaction, but loses
this gain when he buys the horse back for 90. He then makes £10 on the last transaction, when
he sells the horse again for 100.”
According to Maier and Burke (1967), the difficulty of the horse-trading problem is not
due to the answer being hard to find, but due to a misrepresentation of the problem. The “10
pound” answer appears to be due to the perception of the problem as being constituted of three
separate financial transactions, while in fact it becomes much clearer when it is seen as two, and
not three, financial operations.

Appendix C

The study window problem

The study window measures 1 meter in height and 1 meter wide. The owner decides to enlarge
it and calls in a workman. He instructs the man to double the area of the window without
changing its shape and so that it still measures 1 meter by 1 meter. The workman carried out
the commission. How did he do it?

150
151

Bounded rationality and problem solving

Solution
The solution is to be found in a square (geometric form) that “rests” on one of its angles, thus
becoming a rhombus (phenomenic form). Now the dimensions given are those of the two
diagonals of the represented rhombus (ABCD) (Figure A7.2).

B D

Figure A7.2

The “inverted” version

A study window measures 1 meter in height and 1 meter wide. The owner decides to make it
smaller and calls in a workman. He instructs the man to halve the area of the window without
changing its shape and in such a way that it still measures 1 meter by 1 meter. The workman
carries out the commission. How did he do it?
Once again, 30 problem solvers attempted to solve the enigma. Nine started from the
representation of the orthogonal square (ABCD) and looked for the solution within the square,
trying to respect the required height and width of the window, and inevitably changing the
orientation of the internal square. This time the height and width are the diagonals, rather than
the sides (base and height) of the square.
We investigated this problem further by submitting a new experimental version to 30 under-
graduate students from humanity faculties of the University of Milano-Bicocca, of both sexes
and aged mostly between 19 and 25 years. Each participant was presented with the problem
individually, with no time limit. In this version it was explicit that orientation was not a man-
datory attribute of the shape.

The “orientation” version

A study window measures 1 meter in height and 1 meter wide. The owner decides to make
it smaller and calls in a workman. He instructs the man to halve the area of the window: the

151
152

Laura Macchi and Maria Bagassi

workman can change the orientation of the window, but not its shape and in such a way that it
still measures 1 meter by 1 meter. The workman carries out the commission. How did he do it?
In this version, in comparison with the control condition, 66 percent of the participants (21
out of 30; chi-square D 32.308; p < 0.001) found the solution immediately, as they were able
to loosen the problem knot.

Notes
1 In this sense, in Meinong’s terminology, the superiora are independent from the inferiora; on the other
hand, since a melody cannot be made without notes, this independence is not absolute. It is an inde-
pendence constrained by rules. The relationship systems between the pieces of the basic material con-
stitute the mediation between the different orders. The melody remains.
2 Phrase inserted into the text by the authors of the chapter.
3 In the horse-trading problem, for instance, the correct response can be found by applying two methods,
either by restructuration, or by a routine analytical process; however, only when the former method is
applied can we talk of an insight problem. The study window problem, on the contrary, originates from
an insurmountable fixation. This time the analytical process is not an option and the problem is solvable
only by restructuration. (See Appendix B and Appendix C.)
4 “The salience, defined as strikingness or unusualness, would be responsible for the solution of insight
problems” (Kaplan & Simon, 1990, p. 385).
5 This can be seen from the literature on the 9-dot problem, where the hint given to the problem solvers
regarding the virtual square did not determine any change (Weisberg & Alba, 1981).
6 Even the Gestaltists didn’t describe the covert work that goes on during this incubation period (and
how could they, given that they were most interested in the restructuration) and this point was strongly
criticized by the cognitivists and by Simon in particular.
7 However, even reasoning on the basis of three transactions in sequence, a very limited number of the
participants (8 out of 37) gave the correct answer. In this case, in their mind, the loss of the £10 gained
in the previous transaction translated into the reduction of the actual cost of the horse to £80. When
they look at the final transaction (the sale of the horse for £100), they calculate correctly that the dealer
has made £20.
8 The hypothesis that interpretation uses the conscious-unconscious level of thought is supported by
the studies on selective spatial attention in the cluttered displays of Dehaene and Changeux (2011).
According to Dehaene and Changeux, the gate to the conscious access (working memory) is always the
outcome of a nonconscious selection, “separation of relevant versus irrelevant information, based on its
saliency or relevance to current goals … that regulates which information reaches conscious processing”
(pp. 201, 202).

References
Ash, I. K., & Wiley, J. (2006). The nature of restructuring in insight: An individual-differences approach.
Psychonomic Bulletin & Review, 13, 66–73.
Bagassi, M., & Macchi, L. (2016). The interpretative function and the emergence of unconscious ana-
lytic thought. In L. Macchi, M. Bagassi, & R. Viale (Eds.), Cognitive unconscious and human rationality
(pp. 43–76). Cambridge, MA: MIT Press.
Baird, B., Smallwood, J., Mrazek, M. D., Kam, J. W. Y, Franklin, M. S., & Schooler, J. W. (2012). Inspired
by distraction: Mind wandering facilitates creative incubation. Psychological Science, 23, 1117–1122.
Ball, L. J., Marsh, J. E., Litchfield, D., Cook, R. L., & Booth, N. (2015). When distraction helps: Evidence
that concurrent articulation and irrelevant speech can facilitate insight problem solving, Thinking &
Reasoning, 21, 1, 76–96.
Dehaene, S., & Changeux, J. P. (2011). Experimental and theoretical approaches to conscious processing.
Neuron, 70, 200–227.
Duncker, K. (1945). On problem solving. Psychological Monographs, 58 (270), i–113. (Original in German,
Psychologie des produktiven Denkens, Berlin: Springer, 1935).
Evans, J. S. B. T. (2010). Dual-process theories of reasoning: facts and fallacies. In K. J. Holyoak & R.
G. Morrison (Eds.), The Oxford handbook of thinking and reasoning. Oxford: Oxford University Press.

152
153

Bounded rationality and problem solving

Evans, J. St. B. T., Barston, J. L., & Pollard, P. (1983). On the conflict between logic and belief in syllo-
gistic reasoning. Memory & Cognition, 11(3), 285–306.
Fleck, J. I. (2008). Working memory demands in insight versus analytic problem solving. European Journal
of Cognitive Psychology, 20, 139–176.
Fleck, J. I., & Weisberg, R. W. (2013). Insight versus analysis: Evidence for diverse methods in problem
solving. Journal of Cognitive Psychology, 25(4), 436–463.
Frankish, K., & Evans, J. St. B. T. (Eds.) (2009). In two minds. Oxford: Oxford University Press.
Gardner, E. P., & Johnson, K. O. (2013). Sensory coding. In E. R. Kandel, J. Schwartz, T. J. Jessell, S. A.
St. Siegelbaum, & A. J. Hudspeth (Eds.), Principles of neural science. New York: McGraw-Hill.
Gilhooly, K. J., Ball, L. J., & Macchi, L. (2018). Insight and creativity in problem soling. London: Routledge.
Gilhooly, K. J., Fioratou, E., & Henretty, N. (2010). Verbalization and problem solving: Insight and spatial
factors. British Journal of Psychology, 101, 81–93.
Gilhooly, K. J., Georgiou, G. J., & Devery, U. (2013). Incubation and creativity: Do something different.
Thinking and Reasoning, 19, 137–149.
Gilhooly, K. J., Georgiou, G. J., Sirota, M., & Paphiti-Galeano, A. (2015). Incubation and suppression
processes in creative problem solving. Thinking and Reasoning, 21, 1, 130–146.
Jung-Beeman, M., Bowden, E. M., Haberman, J., Frymiare, J. L., Arambel-Liu, S. ... Kounios, J. (2004).
Neural activity when people solve verbal problems with insight. PloS Biology, 2(4), 1–11.
Hayes, J. R., & Simon, H. A. (1974). Understanding written problem instructions. In L. Gregg (Ed.), Knowledge
and cognition. Potomac, MD: Lawrence Erlbaum Associates.
Henle, M. (1962). On the relation between logic and thinking. Psychological Review, 69, 366–378.
Kahneman, D., & Tversky, A. (1972). Subjective probability: A judgement of representativeness. Cognitive
Psychology, 3, 430–454.
Kahneman, D., Slovic, P., & Tversky, A. (1982). Judgement under uncertainty: Heuristics and biases.
Cambridge: Cambridge University Press.
Kandel, E. R. (2012). The age of insight. New York: Random House.
Kanizsa, G. (1980). Grammatica del vedere. Bologna: Il Mulino.
Kaplan, C. A., & Simon, H. A. (1990). In search of insight. Cognitive Psychology, 22, 374–419.
Levinson, S. C. (1995). Interactional biases in human thinking. In E. N. Goody (Ed.) Social intelligence and
interaction (pp. 221–261). Cambridge: Cambridge University Press.
Levinson, S. C. (2000). Presumptive meanings: The theory of generalized conversational implicature. Cambridge,
MA: MIT Press.
Levinson, S. C. (2013). Cross-cultural universals and communication structures. In M. A. Arbib (Ed.),
Language, music, and the brain: A mysterious relationship (pp. 67–80). Cambridge, MA: MIT Press.
Macchi, L., & Bagassi, M. (2012). Intuitive and analytical processes in insight problem solving: A psycho-
rhetorical approach to the study of reasoning. Mind & Society, 11(1), 53–67.
Macchi, L., & Bagassi, M. (2014). The interpretative heuristic in insight problem solving. Mind & Society,
13(1), 97–108.
Macchi, L., & Bagassi, M. (2015). When analytic thought is challenged by a misunderstanding. Thinking
& Reasoning, 21(1), 147–164.
Macchi, L., Bagassi, M., & Viale, R. (Eds.) (2016). Cognitive unconscious and human rationality. Cambridge,
MA: MIT Press.
Mach, E. (1914). The analysis of sensations. Chicago: Open Court.
Maier, N. R. F., & Burke, R. J. (1967). Response availability as a factor in the problem-solving perform-
ance of males and females. Journal of Personality and Social Psychology, 5(3), 304–310.
Mercier, H., & Sperber, D. (2011). Why do human reason? Arguments for an argumentative theory.
Behavioral and Brain Sciences, 34(2), 57–74.
Mercier, H., & Sperber, D. (2017). The enigma of reason. Cambridge, MA: Harvard University Press).
Metcalfe, J. (1986). Feeling of knowing in memory and problem solving. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 12(2), 288–294.
Mosconi, G. (1990). Discorso e pensiero. Bologna: Il Mulino.
Mosconi, G. (2016). Closing thoughts. In L. Macchi, M. Bagassi, & R. Viale (Eds.), Cognitive unconscious
and human rationality (pp. 355–363). Cambridge, MA: MIT Press.
Mosconi, G., & D’Urso, V. (1974). Il farsi e il disfarsi del problema. Firenze: Giunti-Barbera.
Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall.
Ohlsson, S. (2012). The problems with problem solving; Reflections on the rise, current status, and pos-
sible future of a cognitive research paradigm. Journal of Problem Solving, 5(1), 101–128.

153
154

Laura Macchi and Maria Bagassi

Ollinger, M., Jones, G., & Knoblich, G. K. (2008). Investigating the effect of mental set on insight
problem solving. Experimental Psychology, 55(4), 270–282.
Perkins, D. (1981). The mind’s best work. Cambridge, MA: Harvard University Press.
Rizzolatti, G., & Strick, P.L. ( 2013). Cognitive functions of the premotor systems. In E. R. Kandel,
J. Schwartz, T. Jessell, S. A. Siegelbaum, & A. J. Hudspeth (Eds.), Principles of neural science.
New York: McGraw-Hill..
Schooler, J. W., Ohlsson, S., & Brooks, K. (1993). Thoughts beyond words: When language overshadows
insight. Journal of Experimental Psychology: General, 122(2), 166–183.
Schooler, J. W., Smallwood, J., Christoff, K., Handy, T. C., Reichle, E. D., & Sayette, M. A. (2011) Meta-
awareness, perceptual decoupling and the wandering mind. Trends in Cognitive Science, 15, 319–326.
Segal, E. (2004). Incubation in insight problem solving. Creativity Research Journal, 16, 141–148.
Simon, H. A. (1979). Information processing models of cognition. Annual Review of Psychology, 30,
363–396.
Simon, H. A. (1984). Otto Selz and information processing psychology. In N. H. Frijda & A. de Groot
(Eds.), Otto Selz: His contribution to psychology. The Hague: Mouton..
Simon, H. A. (1986). The information processing explanation of gestalt phenomena. Computers in Human
Behavior, 2, 241–255.
Simon, H. A., & Hayes, J. R. (1976). The understanding process: Problem isomorphs. Cognitive Psychology,
8, 165–190.
Simon, H. A., & Newell, A. (1971). Human problem solving: The state of theory. American Psychologist,
21(2), 145–159.
Simon, H. A., Newell, A., & Shaw, J. C. (1979). The processes of creative thinking. In H. A. Simon
(Ed.), Models of thought (pp. 144–174). New Haven, CT: Yale University Press. (Original work
published 1962.)
Sio, U. N., & Ormerod, T. C. (2009). Does incubation enhance problem solving? A meta-analytic review.
Psychological Bulletin, 135(1), 94–120.
Smallwood, J., & Schooler, J.W. (2006). The restless mind. Psychological Bulletin, 132, 946–958.
Sperber, D., & Wilson, D. (1986). Relevance: Communication and cognition. Oxford: Blackwell.
Sternberg, R. J., & Davidson, J. E. (Eds.) (1986). Conceptions of giftedness. New York: Cambridge
University Press.
Stupple, E. J. N., Ball, L. J., Evans, J. S. B. T., & Kamal-Smith, E. (2011). When logic and belief
collide: Individual differences in reasoning times support a selective processing model. Journal of
Cognitive Psychology, 23(8), 931–941.
Threadgold, E., Marsh, J. E., & Ball, L. J. (2018). Normative data for 84 UK English Rebus puzzles.
Frontiers in Psychology, 13.
Tomasello, M. (2009). Why we cooperate. Cambridge, MA: MIT Press.
Tversky, A., & Kahneman, D. (1983). Extension versus intuitive reasoning: The conjunction fallacy in
probability judgment. Psychological Review, 90(4), 293–315.
Weisberg, R. W. (2015). Toward an integrated theory of insight in problem solving. Thinking & Reasoning,
21, 5–39.
Weisberg, R. W., & Alba, J. W. (1981). An examination of the alleged role of “fixation” in the solution of
several “insight” problems. Journal of Experimental Psychology: General, 110(2), 169–192.
Wertheimer, M. (1925). Drei Abhandlungen zur Gestalttheorie. Erlangen: Verlag der Philosophischen
Akademie.
Wertheimer, M. (1985). A Gestalt perspective on computer simulations of cognitive processes. Computers
on Human Behavior, 1, 19–33.

154
155

8
SIMON’S LEGACIES
FOR MATHEMATICS
EDUCATORS
Laura Martignon, Kathryn Laskey, and Keith Stenning

Introduction
Herbert Simon is famous for having said: “The goal of science is to make the wonderful and
the complex understandable and simple – but not less wonderful” (1996). The fundamental
truth underlying Simon’s words has an important parallel in mathematics education, where one
might say that the goal is to devise ways to make wonderful, sometimes complex mathematical
results understandable and simple––but not less wonderful. Explaining our ideas means decom-
posing them into understandable pieces and gluing them together with accessible arguments.
In a proper “waterproof ” explanation, these arguments must be backed by logical rigor. Logical
rigor is sometimes an enemy of beauty and even, at times, of intuition. Mathematics, no less
than the natural sciences, is in itself a collection of wonders. It has grown at the interface
between the human mind/brain and nature, enabling the mind both to model natural phe-
nomena and to tame uncertainty. In science in general and mathematics in particular, there is a
certain risk in decomposing concepts into their components and explaining processes formally
because fascination may fade through the exercise of formal decomposition and explanation.
According to our version of Simon’s statement, the good mathematics instructor should teach
and explain mathematical facts while keeping both motivation and wonder alive in his/her
students’ minds. We cite a well-known mathematical example: It is a wonderful result of math-
ematics that there exist exactly five Platonic solids. Explaining this wonder requires a minimum
of detailed formalism and going through this formalism demands special skills of a mathematics
educator, if she wants to make the result clear and understandable without tarnishing its beauty.
This nugget of wisdom is just one of many that are found in Simon’s writings. Simon’s
work on the sciences of the artificial yields innumerable bits of inspiration for the education of
young students in mathematical thinking. For him, the artificial includes models and structures
expressed in symbols developed by the mind! In other words, for Simon, mathematics and
computer science are among the sciences of the artificial, as is the subject of how the human
mind pursues these topics.
Many of the legacies left by Simon for the mathematics educator appear in his book, The
Sciences of the Artificial, which first appeared in 1969 (Simon, 1996). Some of the chapters in
that revolutionary book were specifically devoted to four topics of intense current interest in
mathematics and computer science education: learning, representing information, bounded

155
156

Laura Martignon et al.

rationality, and problem solving. Our aim in this chapter is to analyze some of these legacies
and discuss their relevance to education today.

Simon’s predictions
Simon and his colleague Allen Newell were among the handful of founders of the field called
Artificial Intelligence. In his Preface to the second edition of The Sciences of the Artificial (1996),
Simon explained the difference between “natural” and “artificial” intelligence, contrasting the
goal-oriented essence of the artificial with the evolutionary fit of natural systems to their envir-
onments. One naturally wonders how Herbert Simon, one of the fathers of artificial intelli-
gence, would view the accomplishments of the field, were he alive today. In a 1992 interview
for UBS Nobel Perspectives,1 Simon mentioned his four predictions on the future of computers
and computer programs:

1 By the mid-1960s, computers would play chess so well that they would beat the greatest
chess masters.
2 Also by the mid-1960s, computers would compose aesthetically pleasing music.
3 Computers would prove at least one new theorem.
4 Most theories of the mind’s cognitive processes would be expressed as computer programs.

Simon was prescient in that all his predictions have eventually come to pass, at least to some
degree. In the case of chess, his timing was off by a few decades, but he was correct that
computers would eventually play at the highest levels. Computers could play quite well by the
1980s, but it was not until 1997 that a computer beat World Chess Champion Garry Kasparov.2
Simon was also early but correct regarding musical composition. In 2017, a program from Aiva
Technologies became the first musical composition program to be given official status, when
the French Société des Auteurs, Compositeurs et Éditeurs de musique (SACEM) conferred
upon it the official status of Composer.3 In 1976, the Four Color Theorem, a longstanding
mathematical conjecture that had frustrated mathematicians since it had been proposed in
1852, became the first major theorem to be proven by computers and human minds working
together. While controversial at first, because it was too complex to be checked by hand, the
proof is widely accepted today. As for the fourth prediction, Herbert Simon himself saw to it
that computer theories of the mind were developed, and he became a driving force behind
their acceptance by the cognitive science community. It is difficult to estimate the ratio of full
computational theories to less formally expressed theories of the mind, and thus to evaluate his
assessment that most theories would be computational. Nevertheless, computer modeling of
cognitive processes has become a vibrant, prolific, and highly successful sub-field of psychology.
Simon’s insight that many cognitive processes can be explained in terms of computer models
was revolutionary. The question today is rather, which are those cognitive processes, which
cannot be explained in terms of computer models?
One reason Simon, along with many early AI researchers, was far too optimistic about the
time frame during which critical milestones would be achieved, was the contrast between
how computers and humans process information. Humans are excellent at pattern recognition
and analogy, still exceeding the capabilities of computers on many tasks requiring these skills.
Computers are designed to perform the kind of exacting, step-by-step, logical reasoning tasks
humans find especially difficult. Because computers were so good at tasks humans find so chal-
lenging, it was easy to overestimate the difficulty of making computers perform well on tasks
humans find routine and even less intelligent animals can perform well.

156
157

Simon’s legacies for mathematics educators

Simon and the early AI researchers also underestimated the difficulty of developing AI
systems to tackle ill-structured problems. While acknowledging that there can be no crisp,
formal boundary between well- and ill-structured problems, Simon (1973) set out the
characteristics of a problem that can be regarded as well-structured, i.e., one that is amenable
to solution by a computer problem solver. An ill-structured problem is then a residual category,
i.e., an ill-structured problem is simply one that is not well-structured. Simon argued that
whether a problem is well- or ill-structured can be a matter of context, and that “any problem
with a sufficiently large base of relevant knowledge can appear to be an ill-structured problem”
(Simon, 1973, p. 197). He concluded: “There appears to be no reason to suppose that concepts
as yet un-invented and unknown would be needed to automate the solution of ill-structured
problems.” Whether Simon was correct in this conclusion is still a matter of debate among AI
researchers. It is unquestionable that it has taken longer than Simon anticipated for some of the
more challenging problems to succumb to automation.
In 1956, von Neumann was invited to give a set of lectures comparing the ways in which
computers and brains process information. He never gave the lectures, but worked on the
manuscript until his death, and it was published posthumously (von Neumann, 1958).
Acknowledging that he was a mathematician and not a neuroscientist, he identified a number
of ways in which computers and brains operate differently:

1 Brains are analog and computers are digital.

2 Brains process in parallel and computers process serially.
3 Unlike brains, computers have the ability to use instructions stored in memory to modify
control of processing.
4 Individual brain component processes are too slow to perform the serial processes at which
computers excel.

Von Neumann concluded that the language and logic used by the brain must be radically
different from that of computers.
Despite these fundamental differences, the project of computational modeling of cogni-
tive phenomena has been extremely successful, and has led to many important insights. These
insights have sparked advances in decision support, automation, and also education. We are here
mainly concerned with Simon’s seminal insights: that cognitive processes can be described by
means of computer models and that the representation of information in a given problem is a
key factor for finding its solution. These insights have had far-reaching consequences for math-
ematics and computer science education.

Ecologically rational representations

One of Simon’s most famous quotes concerns the fundamental importance of representation
for solving problem: “Solving a problem simply means representing it so as to make the solu-
tion transparent” (Simon, 1996, p. 132). A solution becomes transparent if it emerges from the
structure in terms of which the problem has been modeled. The structure of the problem is the
result of an attempt at adaptation between the problem itself and the conceptual constructions
of the mind. An adaptation is successful if it is ecologically rational (cf., Gigerener et al., 1999).
Ecological rationality is a fundamental characteristic of successful representations. It refers to
behaviors and thought processes that are adaptive and goal-oriented in the context of the
environment in which an organism is situated, thus giving survival advantage to the organism.
Discovering a representation that makes a problem easily solvable gives evolutionary advantage

157
158

Laura Martignon et al.

in terms of time resources to adopters of the new representation, because they are able to solve
the problem more quickly and easily than those holding on to the old representation.
Modern mathematics education treats representations of mathematical entities as a funda-
mental aspect of didactics in the classroom. Special attention has been devoted, for instance,
to the issue of multiple representations of numerical entities and the advantages of switching
between them as a means of achieving mathematical competencies for dealing with them
(Dreher & Kuntze, 2014).
For many types of problems, such as fractions and probabilities, we argue that different
mathematically equivalent representations are far from cognitively equivalent. That is, some
representations are more adaptive and advantageous than others, because they seem to be better
aligned to the cognitive systems of human problem solvers.
We begin by describing the relevance of information formats for inference, a major com-
ponent of human reasoning. Both problem solving and decision-making require inference.
From old knowledge, we create and acquire new knowledge by means of chains of inferences.
From partial solutions, we advance to more complete solutions by means of inferences. The
introduction of so-called inference machines, that is well-described systems of rules for infer-
ence, has had immense implications for science. The inferences of these inference machines are,
more often than not, produced under uncertainty. Classical logic and probability provide the
two most salient, standard mechanisms invented in human history for inference machines. We
will discuss the ecological rationality of representations in the context of these two inference
approaches, inverting the historical order of their inception. We begin by approaching tasks of
probabilistic inference.

The ecological rationality of natural frequencies for probabilistic inference

One of the problems that makes mathematics so difficult is the already mentioned tension
between intuition and rigor. In a variety of areas, mathematicians have developed elegant,
abstract, logically coherent theories that encompass and generalize intuitively natural concepts.
Examples include numbers, probabilities, and algebraic structures. These elegant abstract the-
ories have a pristine beauty that attracts mathematicians, and have led to powerful innovations,
but can be daunting to beginning students and to the lay public. Mathematics educators there-
fore face the challenge of helping students to build on their natural intuition to bridge the gap
between human intuition and mathematical theory.

Natural frequency and probability

An important instance of the tension between intuition and rigor is probability theory. Our
intuitive, innocent, and natural understanding of probabilities fits uneasily with the formal
definition as countably additive measures on sigma-algebras of sets. Formalism and rigor, espe-
cially introduced too early, may kill intuition and with it motivation. We aim at scrutinizing
the natural Bayesian who reasons, as in Gerd Gigerenzer’s recommendation, by forming simple
proportions of outcomes starting from well-defined populations of cases. The natural Bayesian
goes through a sequence of stages: from multiple narratives stemming from observations that are
ultimately represented as a frequency tree, to expected outcomes in future phases of updating.
The natural frequency representation is adapted to our cognitive processes in a way that
makes the solution to evidential reasoning problems transparent. In other words, it is ecologic-
ally rational.

158
159

Simon’s legacies for mathematics educators

Consider a physician who must reason from evidence, such as symptoms and test results, to
a hypothesis, such as whether or not a patient has a disease. To approach this problem formally,
as presented in textbooks, we begin with two ingredients: (1) a prior probability that the dis-
ease is present; and (2) likelihoods, or probabilities that the evidence would be observed if the
disease was present or absent. For our example, we imagine that the evidence is a test for the
disease, and it has come out positive. We use the symbols D+ and D- to denote the presence
and absence of the disease, and T+ to denote a positive test result. We use our prior probability
and test likelihoods to calculate the posterior probability that the disease is present given the
positive test result:

P (T + | D + )P ( D + )
(D +|T + ) = (8.1)
P (T +|D + ) P ( D + ) + P (T +|D − ) P ( D − )

The formula (8.1) is called Bayes rule after its inventor, the eighteenth-century philosopher and
minister Thomas Bayes.
Formula (8.1) gives correct answers to the evidential reasoning problem, but it requires
calculations of which only a relatively few trained individuals are capable. There is an immense
literature on people’s difficulties with this kind of reasoning based on conditional probabilities,
and some scientists have been convinced that these difficulties point to some form of human
irrationality.
There is, on the other hand, a way of representing such evidential reasoning tasks that
makes the solution transparent. To use this method, called natural frequencies (e.g., Gigerenzer &
Hoffrage, 1995), the doctor imagines a population of fictitious people, say, 1,000 of them. She
divides them into those who do and do not have the disease. For example, if the disease is pre-
sent in only 1 percent of the population, she would partition her imaginary 1,000 patients into
10 who have the disease and 990 who do not. Of those who have the disease, let us imagine
that 80 percent will test positive. Therefore, the doctor imagines that 8 of the 10 ill patients will
test positive and 2 will test negative. Now, suppose that 90 percent of those who do not have
the disease will have a negative test result. In our doctor’s imaginary population, this works out
to 99 well patients who test positive and 891 who test negative.
This thought experiment uncovers a peculiar, and very important, characteristic of problems
characterized by uncertainty: the fundamental role played by the base rate. Our doctor, like
many untrained people, may be surprised to learn that, although the vast majority of ill patients
have tested positive and the vast majority of well patients have tested negative, only 7 percent
of the patients who test positive actually have the disease. This occurs because of the low base
rate––that is, a very small proportion of our original population was ill.
Figure 8.1 compares the probability and the natural frequency representation of this eviden-
tial reasoning problem. As many an educator has discovered, students taught to apply formula
(8.1) often find the result P(D|T+) = 7% surprising, sometimes even refusing to believe that
the disease remains unlikely after a positive test result. By contrast, most students who work
with the figure on the right immediately grasp why most patients who test positive are actu-
ally disease-free. It is readily apparent that although 107 positives is a small proportion of the
990 well patients, it is still nevertheless many more than the 8 true positives out of only 10 ill
patients.
There is a wealth of literature demonstrating that people tend to neglect base rates when
reasoning about this kind of problem. Even many people who have been taught Bayes rule
often fall prey to this fallacy. However, performance on such tasks improves dramatically when

159
160

Laura Martignon et al.

1000

0.01 0.99 C ¬C

No 10 990
Breast cancer
breast cancer

0.8 0.2 0.1 0.9 M+ M– M+ M–

M+ M– M+ M– 8 2 99 891

P(C|M+) = 7% P(C|M+) = 8 of 107

Figure 8.1 Probability and natural frequency representation of evidential reasoning

people are taught the natural frequency interpretation (Gigerenzer & Hoffrage, 1995). This
improved performance has been observed in a variety of application domains (Hoffrage &
Gigerenzer, 1998; Hoffrage, Lindsey, Hertwig, & Gigerenzer, 2000; Gigerenzer, 2002) and
in more complex problems than reasoning with a single cue (Hoffrage, Krauss, Martignon, &
Gigerenzer, 2015).
To summarize, an ecologically rational strategy is one our cognitive apparatus can per-
form naturally and easily, and that has adaptive value in a given environment. In other
words, ecological rationality involves discovery of a representation that makes an optimal
or nearly optimal solution transparent. For reasoning from evidence to hypothesis, because
the natural frequency representation is ecologically rational, it allows humans to perform
an otherwise difficult task quickly and easily. Because the natural frequency representation
is ecologically rational, its use in educational settings facilitates understanding and improves
performance.

Icon arrays and Venn diagrams

Back in the 1960s and 1970s, Venn diagrams were introduced as representations of sets in pri-
mary and secondary schools in most European countries and several other countries around
the world. There were protests everywhere. In Germany, for instance, the protests both of
schoolteachers and parents were so strong, that set theory was banned from primary school.
This had as a consequence a reluctance to introduce probabilities earlier than in ninth grade.
Also for the communication of probabilistic information in the medical and pharmaceutical
domain, Venn diagrams did not enjoy acceptance.
Here again, an ecologically rational format has made its way into many sectors of commu-
nication. Otto Neurath introduced such a format during the first half of the twentieth century
(he actually proposed what he called isotypes and arrays of such isotypes).
Icon arrays are displays of icons, which represent individuals or animals or items. The
icons can exhibit special features, thus allowing for classifications. In the example in
Figure 8.2a, the depicted icon array represents dogs and cats, with or without the feature
“bell.” One more structural way of representing the information about these house pets is

160
161

Simon’s legacies for mathematics educators

(b)
10
Pets

6 4
Cats Dogs

3 3 2 2
Cats with bell Cats no bell Dogs with bell Dogs no bell

5 5
With bell No bell

10
All together

Figure 8.2 (a) An icon array representing dogs and cats, some with a bell and some without; (b) the
corresponding double tree for transparent Bayesian reasoning
Source: (a) www.eeps.com/projects/wwg/ a free webpage for Risk literacy by Tim Erickson and Laura
Martignon.

161
162

Laura Martignon et al.

to organize them in a double tree: from top to bottom we first partition our house pets into
cats and dogs and continue by looking in each subgroup whether the pet has a bell or not
(Figure 8.2b). From bottom to top, we first look at bells, partitioning our pets into those
with a bell and those without. School kids can use this representation to learn an elementary
Bayesian reasoning. They learn to assess the probability––expressed in natural frequencies––
that a pet with a bell is a cat, or that a dog has a bell. Thus, icon arrays combined with trees
and double trees communicate information on base rates, sensitivities, specificities and pre-
dictive values in a straightforward, transparent way. They have been used in primary school
with great success (Martignon & Hoffrage, 2019) coupled with corresponding natural fre-
quencies doubletrees.

From inference based on one cue to inference based on many cues:

the ecological rationality of lexicographic strategies
We commonly model situations by characterizing them in terms of a set of cues or features
we can extract from them. We make inferences on these situations based on those cues. We
have treated above the case of judgments or classifications based on just one cue––in that case
a test––characterizing a disease. Actually, a physician usually needs more than just one test
result to diagnose a disease. For instance, in the case of breast cancer, a doctor will systemat-
ically look at two cues, namely, mammography and an ultrasound test. We illustrate a possible
Natural Frequency tree drawn in the diagnostic direction, i.e., with the end nodes corresponding
to “having breast cancer” or “not having breast cancer” (Figure 8.3), in contrast with the tree
of Figure 8.1, which follows the causal direction.
The facilitating effect of natural frequencies disappears when more cues are considered: As
the number of cues grows, the size of the natural frequency tree explodes and ecological ration-
ality is lost. The huge tree representing 6 cues, for instance, suffers from “brittleness.” As it
grows, the number of end nodes may become very large, while the number of cases per end
node may become very small. These small numbers of cases per end node will not support
reliable predictions.

10 000
M+ M–

1030 8970
U+ U– U+ U–

114 916 377 8593

76 38 4 912 19 358 1 8592

Figure 8.3 A natural frequency tree for classifying patients as having or not having breast cancer, by
looking at the values of two cues, namely, a mammography and an ultrasound test

162
163

Simon’s legacies for mathematics educators

Is there an alternative ecologically rational strategy that maintains a tree structure and
eliminates non-relevant information? There is, in fact, a heuristic strategy that fits the bill.
Inspired by Simon, the heuristic consists of radically pruning the full tree into a minimal one
based on the same set of cues. The resulting tree becomes fast-and-frugal, according to the def-
inition of Martignon et al. (2003). The fast-and-frugal tree has a single exit at every level before
the last one, where it has two. The cues are ordered by means of very simple ranking criteria.
Fast-and-frugal trees are implemented step by step, reduce memory load, and can be set up and
executed by the unaided mind, requiring, at most, paper and pencil.
Step-by-step procedures, that follow well-specified sequences, are a typical aspect of human
behavior. We pronounce one word at a time, for instance. Time organizes our actions and
thoughts in one-dimensional arrays. We are tuned for doing “first things first” and then
proceeding to second and third things. This innate tendency for sequencing makes so-called lex-
icographic strategies for classification and decision ecologically rational. This section illustrates
how sequential treatment of cues can extend and expand classifications based on one cue.
A fast-and-frugal tree does not classify optimally. Rather, to use a term coined by Simon,
it satisfices, producing good enough solutions with reasonable cognitive effort. The predictive
accuracy and the robustness of the fast-and-frugal tree have been amply demonstrated (e.g.,
Laskey & Martignon, 2014; Woike, Hoffrage, & Martignon, 2017; Luan, Gigerenzer, &
Schooler, 2011).
Martignon et al. (2003) provided a characterization of these trees as lexicographic, supporting
an ecologically rational step-by-step execution of the classification process, which can be stated
as the following theorem:

Theorem: For binary cues with values 0 or 1, a fast-and-frugal tree is characterized

by the existence of a unique cue profile of 0’s and 1’s that operates as a splitting profile
of the tree; this means that any item with a profile lexicographically lower than the
splitting profile will be classified in one of the two categories, while the rest of items
will be classified in the other one.

This theorem is illustrated in Figure 8.4.

111
D
110

101
100

011

010

001
–D
000

Figure 8.4 An example of a splitting profile characterizing a fast-and-frugal tree for classifying items
into two complementary categories based on three cues
Source: Martignon et al. (2003).

163
164

Laura Martignon et al.

From classification to comparison and back

The trees for classification described above belong to the class of simple heuristics inspired by
Simon and analyzed for a variety of tasks by the ABC Research Group, led by Gerd Gigerenzer.
Another of these heuristics tackles comparison: When two items have to be compared as to
which has a higher value on a given criterion, the fast-and-frugal heuristic at hand, called Take-
the-Best, proceeds as follows: It ranks the binary cues according to a simple criterion (e.g.,
validity) and looks at the cue profiles, comparing them lexicographically. Imagine that there
were just three binary cues and the corresponding profiles were 110 and 101, as depicted in
Figure 8.4. When they are compared lexicographically, 110 is larger than 101 because the first
digit on the left-hand side coincides in both, but the second differs. Thus, the larger second
digit corresponds to the larger number.
This simple principle is the essence of Take-the-Best. Assume, for instance, one wants to
compare German cities, say, Duisburg and Ludwigsburg, as to their population. The com-
parison is to be based on binary cues, just as in the original example treated by Gigerenzer
and Goldstein (1996). Here more than three binary cues may be necessary, if the pairs of
cities to be compared can be any of all pairs in the reference of class of cities with more
than 100,000 inhabitants. In fact, Gigerenzer and Goldstein considered nine binary cues,
of which the first one, was simple recognition: “Do I recognize the name of the city?”. If
this is true for only one of the cities in the pair, then the recognized one is to be considered
the larger one. If both are recognized, then one can compare the remaining “cue profiles”
lexicographically. Other cues are “Is the city a state capital?,” “Does the city have a soccer
team playing in the National League?,” or “Does it have a station where ICE trains stop?”
In summary, Take-the-Best analyzes each cue, one after the other, according to the ranking
by validity (i.e., its overall predictive value, or the probability of making a correct prediction
if it discriminates between the two cities) and stopping the first time a cue discriminates
between the items, concluding that the item with the larger value has also a larger value on
the criterion.
Both the fast-and-frugal tree for classification and Take-the-Best are simple heuristics which
belong in more than one of the great mathematical classes of models. On the one hand, they
can be represented by trees, and, on the other hand, they are lexicographic strategies. An add-
itional property that may be useful for computer implementation and use in applications, as well
as for the comparisons in performance with models like regression (see Şimşek’s Chapter 21
in this volume) is that these methods can be described in terms of linear models, which are
those that “weigh and sum” cue values, in the most elementary way. In a more formal context,
Martignon and Hoffrage (2002) showed that such lexicographic rules are always equivalent to
weighted linear classification models with non-compensatory coefficients. A list of weights are
non-compensatory if each of them is larger than the sum of all weights that follow in the list
(see Figure 8.5 for an example).
On the left-hand side of Figure 8.5, we illustrate non-compensatory weights, which make
a linear weighted model behave like a lexicographic strategy. On the right-hand side, we illus-
trate equal weights, which correspond to the tallying heuristic, which simply “counts” 1s in
profiles of 1s and 0s.
In the next section, we illustrate the connection between the fast-and-frugal Take-the-Best
with comparisons of natural numbers. The striking fact is that Hindu-Arabic numerals are a
representation of numbers that allows the magnitude of numbers to be compared by following
the Take-the-Best strategy.

164
165

Simon’s legacies for mathematics educators

Figure 8.5 On the left side non-compensatory weights, like 2, 1, ½, ¼, 1/8. On the right side 5
weights all equal to 1.

Fast-and-frugal number comparison with the Hindu-Arabic system

Herbert Simon wrote, in Chapter 5 of The Sciences of the Artificial: “We all believe that arith-
metic has become easier since Arabic numerals and place notation replaced Roman numerals,
although I know of no theoretic treatment that explains why” (Simon, 1996).
Our aim is to use the tenets of ecological rationality to explain why arithmetic became so
much easier with this new representation. We claim that “the theoretic treatment of why the
decimal positional system made arithmetic calculations and comparisons so much easier than
the Roman numerals” is in no small part due to the ecological rationality of the lexicographic
heuristic for comparison it allows: Obviously, if a natural number has more digits than another
natural number, then it is larger. What happens when they have the same number of digits?
If one has to compare, say, 2056 and 2049, then one proceeds lexicographically stopping at
the third place, where digits differ and declaring 2056 to be the larger one because 5 > 4.
The same procedure is also valid if numbers are written in base 2, like 100000000100 and
10000000001.
Observe that these numbers written in Roman numerals are MMLVI and MMXLIX,
allowing no natural, organic way for comparing them: in fact, the number with a longer
representation is actually smaller, which makes little sense from the perspective of ecological
rationality. More importantly, arithmetic operations become simple with place notation, being
far more cumbersome with the Roman system.
For mathematics education, the connections between ecological rationality and the number
system are relevant. School students learn a minimum of mathematical history. They learn
about the Roman system and they learn why even Europeans finally replaced it by a system
that––from the structural point of view––is far more ecologically rational. It remains a mys-
tery why the European continent maintained the Roman numeral system for such a long time.
Roman numerals were merely used to describe quantities and data, but not for comparisons
nor for calculations. They mimicked, at least for numerals between 1 and 50, fingers and
hands: 10 corresponded to two hands, 20 to 4 hands and 33 to 3 hands and 3 single fingers.
The letter L for 50 meant “half of C,” and C was the first letter of Centum, which meant 100
in Latin. The system was additive and cumulative, requiring ever more letters for representing
larger numbers. Although its basic symbols may seem natural, because they depict fingers,

165
166

Laura Martignon et al.

their representation is not based on an ecologically rational structuring. Not just comparisons
but also computations were cumbersome. Until the twelfth century, Italian merchants had to
outsource computations on the prices of their merchandise by communicating their numbers
to “abacists” who translated Roman numbers into groups of marbles and operated with these
groups in a cumbersome way.
In India, and later in Persia and the Islamic countries, the positional number system had
been used widely for computations since the seventh century. It was Leonardo da Pisa, also
known as Fibonacci, who made the first attempts to import the Hindu-Arabic positional system
to Europe, well into the twelfth century (Figure 8.6). The Church opposed this novelty and

Figure 8.6 An allegory of Arithmetic from Margarita Philosophica: a young woman, smiling benevolently
at the young man who uses the Hindu-Arabic positional system, while his fellow at the other table uses
marbles
Source: Gregor Reisch, twelfth century.

166
167

Simon’s legacies for mathematics educators

even declared the number “0,” which the Arabs had called “Zifer” (from which the term
“cipher” was derived), as the number of the devil. Yet Fibonacci’s dream came true in the end
because its ecological rationality provided such great practical utility.

The ecological rationality of defeasible logic and its role in

mathematics education
In Chapter 3 of The Sciences of the Artificial, Simon recommends the use of multiple logics and in
Chapter 5, he explains that “Multiple logics may become necessary when approaching heuristic
decision making” (Simon, 1996).This is particularly true in the mathematics classroom today
when instructors have to understand students’ reasoning without declaring it as “wrong” simply
because it does not follow the laws of classical logic. One of the responsibilities of mathematics
educators today is to instruct students in identifying different forms of inferences––not just for
decision making but also as tools for mathematical thinking in general. The main point we
want to make here is that classical logic and probabilities are not sufficient as instruments for
understanding and describing human reasoning.
Remember how, in previous sections, we treated a conditional sentence of the kind:

If she tests positive, she has the disease

We imagined a physician reasoning from evidence to hypothesis, based on knowledge of ana-

tomical facts but also on experience. How high is the physician’s confidence in this conditional?
In the first part of this chapter, we have extensively discussed how the physician can treat this
conditional probabilistically by estimating the probability that the patient has the disease, given
that she tests positive. It is well established that physicians do not necessarily treat this impli-
cation, basing their assessment of frequencies. A fundamental treatise on physicians’ decision
making in inferences is David Eddy’s book (Eddy, 1996), in which he presents conclusions
gained from his decades of work examining his colleagues’ reasoning. In fact, Eddy became
known for propagating the necessity of physicians acquiring statistical literacy. One observation
he made seems particularly relevant to our discussion of multiple logics: He was surprised by
the fact that excellent doctors could be so bad at correctly estimating predictive values (Eddy,
1982). What they apparently had was a profound knowledge of the reasons or factors that could
enable a positive test. Basing inferences on such “reasons” instead of frequencies is not prob-
abilistic and it also does not follow the rules of classical logic. Nevertheless, it can be extremely
successful. For example, a physician with extensive knowledge of the reasons a test could be
positive without the presence of the illness in question is in a good position to order follow-up
tests and to observe the patient carefully to rule out these other reasons before making a final
diagnosis. Precisely here we see other logics at work, namely, defeasible logics, which take into
account so-called enablers and defeaters of causal conditionals.
Psychologists have examined these different ways of reasoning and making inferences suc-
cessfully. Back in 1995, Denise Cummins performed a seminal experiment. She let participants
generate “defeaters” for causal conditionals, such as, “If John studies, he passes the test.” She
then discovered that the number of defeaters generated by a group of participants inversely
predicted the confidence of another group of participants in the statements of the conditionals.
In terms of heuristics, one can state that simply “tallying” defeaters inversely predicts confi-
dence in the inference expressed by a conditional. They rely on their tallying abnormalities and
alternative causes, that is “reasons” that cause conditional inferences of the Modus Ponens type
or of the Affirmation of the Consequent type to be defeated on occasions. Stenning, Martignon,

167
168

Laura Martignon et al.

and Varga (2017) replicated this experiment in a within-subject design, and obtained basically
the same results. They further discovered that people are ecologically rational, in that they
switch from tallying defeaters or enablers to using just one of them if it is particularly strong,
or sometimes two of them, if together they seem to back up their judgment strongly enough.
The relevant conclusion from Cummins’ experiment and our replication is that representations
for inference in daily life are not just numerical or strictly logical in a classical sense but can be
expressed by narratives, which obey defeasible logic.
In fact, strict classical logic is the instrument of adversarial communication, while defeasible
logic is often the instrument of cooperative communication (Stenning, Martignon, & Varga,
2017). Adversarial communication is the mode adopted by someone who knows that peers
and colleagues will be checking each step in a dialectical mode, whereas cooperative com-
munication is the mode adopted by someone arguing to achieve progress toward a common
goal. Adversarial communication can place students on the defensive and inhibit learning.
Cooperative communication, on the other hand, places teachers and students in a collaboration
to support learning. When a mathematics educator analyzes her students’ answers, she should
be aware of students’ different logics before declaring that her students are committing serious
mistakes. Mathematics educators are likely to be more successful if they engage their students
in a cooperative pursuit of knowledge than if they act as adversaries pouncing on their students’
every perceived mistake.

Conclusion
Simon devoted much of his career to promulgating the importance of what we have called eco-
logically rational representations and procedures. These adapt well to our cognitive apparatus
and produce solutions that are good enough for the purpose. Thus, as Simon argued, solving a
solution means finding a representation that makes a satisficing solution transparent. In a class
attended by one of the authors, Simon argued for the necessity of satisficing with characteristic
wry humor, announcing that he was insured to precisely the point that his family was indif-
ferent as to whether he lived or died. As a brilliant pedagogical strategy, this little joke cemented
in his students’ mind both the impossibility of finding a genuinely optimal insurance strategy,
and the necessity of satisficing to protect one’s family.
This work has explored the role of ecologically rational representations and procedures for
several common mathematical reasoning tasks: number comparison, classification, and causal
inference. In each class of problems, we illustrated how ecologically rational representations
and procedures make satisficing solutions transparent. As tools for education, such ecologic-
ally rational strategies support understanding, and maintain the wonder of mathematics in the
minds of students.

Notes
1 The relevant excerpt can be found at this link: www.youtube.com/watch?v=ABucG05nurs.
2 See https://en.wikipedia.org/wiki/Human–computer_chess_matches.
3 See https://aibusiness.com/aiva-is-the-first-ai-to-officially-be-recognised-as-a-composer/.

References
Cummins, D. (1995) Naïve theories and causal cognition. Memory and Cognition, 23(5), 646–659.
Dreher, A., & Kuntze, S. (2014) Teachers’ professional knowledge and noticing: The case of multiple
representations in the mathematics classroom. Educational Studies in Mathematics, 88(1): 1–26.

168
169

Simon’s legacies for mathematics educators

Eddy, D. M. (1982) Probabilistic reasoning in clinical medicine: Problems and opportunities. In D.

Kahneman, P. Slovic, & A. Tversky (Eds.), Judgment under uncertainty: Heuristics and biases (pp. 249–
267). Cambridge: Cambridge University Press.
Eddy, D. M. (1996). Clinical decision making: From theory to practice: a collection of essays published by the
American Medical Association. New York: Jones and Bartlett Publishers.
Gigerenzer, G., & Goldstein, D. G. (1996). Reasoning the fast and frugal way: Models of bounded ration-
ality. Psychological Review, 103, 650–669.
Gigerenzer, G., & Hoffrage, U. (1995). How to improve Bayesian reasoning without instruction: Frequency
formats. Psychological Review, 102, 684–704.
Gigerenzer, G., Todd, P. M., & the ABC Research Group (Eds.) (1999). Simple heuristics that make us smart.
New York: Oxford University Press.
Green, L., & Mehr, D. R. (1997). What alters physicians’ decisions to admit to the coronary care unit? The
Journal of Family Practice, 45(3), 219–226.
Hoffrage, U., & Gigerenzer, G. (1998). Using natural frequencies to improve diagnostic inferences.
Academic Medicine, 73(5), 538–540.
Hoffrage, U., Gigerenzer, G., Krauss, S., & Martignon, L. (2002). Representation facilitates reasoning: What
natural frequencies are and what they are not. Cognition, 84, 343–352.
Hoffrage, U., Krauss, S., Martignon, L., & Gigerenzer, G. (2015). Natural frequencies improve Bayesian
reasoning in simple and complex tasks. Frontiers in Psychology, 6(1473), 1–14.
Hoffrage, U., Lindsey, S., Hertwig, R., & Gigerenzer, G. (2000). Communicating statistical information.
Science, 290, 2261–2262.
Laskey, K. B., & Martignon, L. (2014). Comparing fast and frugal trees and Bayesian networks for risk
assessment. In K. Makar, B. de Sousa, & R. Gould (Eds.), Sustainability in statistics education. Proceedings
of the Ninth International Conference on Teaching Statistics (ICOTS9, July 2014), Flagstaff, Arizona, USA .
Voorburg, The Netherlands: International Statistical Institute.
Luan, S., Schooler, L. J., & Gigerenzer, G. (2011). A signal detection analysis of fast-and-frugal trees.
Psychological Review, 118, 316–338.
Martignon, L., & Hoffrage, U. (2002). Fast, frugal, and fit: Simple heuristics for paired comparison. Theory
and Decision, 52(1): 29–71.
Martignon, L., & Hoffrage, U. (2019). Wer wagt, gewinnt? Göttingen: Hogrefe.
Martignon, L., Vitouch, O., Takezawa, M., & Forster, M. (2003). Naive and yet enlightened: from nat-
ural frequencies to fast and frugal trees. In D. Hardman & L. Macchi (Eds.), Thinking: Psychological
perspectives on reasoning, judgment, and decision making, (pp. 189–211). Chichester: Wiley.
Simon, H. (1973). The structure of ill-structured problems. Artificial Intelligence, 4, 181–201.
Simon, H. (1996). The sciences of the artificial (3rd edn). Cambridge, MA: MIT Press.
Stenning, K., Martignon, L., & Varga, A. (2017). Probability-free judgment: Integrating fast and frugal
heuristics with a logic of interpretation. Decision, 4(3), 136–158.
Von Neumann, J. (1958). The computer and the brain. New Haven, CT: Yale University Press.
Woike, J. K., Hoffrage, U., & Martignon, L. (2017). Integrating and testing natural frequencies, naïve
Bayes, and fast-and-frugal trees. Decision, 4(4), 234–260.

169
170

9
BOUNDED KNOWLEDGE
Cristina Bicchieri and Giacomo Sillari

Introduction
As is well known, there exists a chasm between the picture of perfect rationality on which
modern mainstream economics is largely built, and the limited abilities in reasoning and deci-
sion making displayed by real agents. Herbert Simon’s groundbreaking work (e.g., Simon,
1947; 1955; 1957) pioneered research on bounded rationality. At its outset, bounded rationality
was a fundamental theme in the study of administrative and organizational behavior (Simon,
1947). After its inception, the study of boundedly rational agency has flourished in a variety
of areas. In decision theory, the momentous conceptual shift is from focusing on a normative
theory of optimal behavior to focusing on a descriptive theory of satisficing behavior (Simon,
1955). In satisficing, instead of maximizing her expected utility, a boundedly rational agent
chooses by using a heuristic process: (1) fix aspiration levels for relevant variables; (2) scan pos-
sible outcomes until she identifies one that goes over her aspiration level; (3) choose accord-
ingly.1 Simon (1967) claims that satisficing is closer to the way in which actual human agents
make their decisions, and his descriptive approach to theorizing decision making has become
central to the work of behavioral economics. Instead, in neoclassical economics, the theory
assumes, as Simon puts it, an “Olympian,” perfect rationality of agents. Neoclassical agents
are perfect reasoners who invariably come up with an ideal, canonical, objective representation
of the decision space, and are able to perform the computationally demanding calculations
leading to a choice of action that maximizes their expected utility. In order to do so, agents are
required to be able to perform sophisticated logical, mathematical, and probabilistic reasoning.
While common sense indicates that human agents are not able to perform such calculations,
experimental evidence shows moreover that human agents (1) are persistently biased in their
reasoning abilities, and (2) use, instead of probability calculus, a variety of heuristics. Heuristics
may allow agents to deviate from the rationality canon and yet achieve satisficing, if suboptimal
outcomes (see, e.g., Simon, 1972).
Ideal agents, however, are not only assumed to be perfect in their treatment of probabil-
istic reasoning. Their logical thinking is also supposed to be carried out flawlessly, which, of
course, is descriptively untenable. Famously, the Wason Selection Task indicates that human
logical abilities are highly context-dependent and their results often questionable (Wason,

170
171

Bounded knowledge

1966). In this chapter, we look at such limitations with regards to the agents’ logical abilities or,
more generally speaking, limitations to their epistemic rationality.2 There are, of course, many
different ways to model such limitations, even though, with few exceptions, logicians have
not been particularly interested in providing descriptively adequate representations of human
logical thinking.3 Here we will explore one possible avenue to represent bounded epistemic
rationality. The idea, in a nutshell, is that subjects display different levels of access to items of
knowledge and belief. Beliefs that are accessible can be explicitly held and possibly used in the
agent’s reasoning. Beliefs to which an agent has no access are only implicitly held and are not
used in her explicit reasoning.4

Awareness and accessibility

In the remainder of this chapter we will look at a formal proposal to characterize differential
access to information but, before we offer a formalization, it should be clear that differen-
tial accessibility is a descriptively relevant notion for behavioral economics and the charac-
terization of bounded rationality: Bounded knowledge, as based on accessibility, is a crucial
component of bounded rationality. Indeed, in his Nobel lecture, Kahneman (2003) depicts
accessibility as a crucial notion for behavioral economics. The dual process theory defended
by Kahneman rests on accounts of accessibility: “The core concept of the present analysis
of intuitive judgments and preferences is accessibility – the ease with which particular mental
contents come to mind” (Kahneman, 2003, p. 452). In particular, our System 1, intuitive,
“fast” thinking and reasoning are characterized by accessibility. Various items affect accessi-
bility and hence influence our intuitive reasoning: among them stimulus salience, selective
attention, and response activation. The list is not exhaustive and, as Kahneman acknowledges,
we do not have a full-fledged theory of accessibility. That does not mean that the notion of
accessibility is not a powerful one for explanatory work. For instance, prospect theory––in
which agents evaluate their possible decisions on the basis of prospects of gains and losses
rather than on the expected utility associated with outcomes––is based on the observation
that changes and differences are psychologically more salient, and hence more accessible, than
absolute levels of stimulation. Combined with the dual process reasoning theory, the notion of
accessibility explains phenomena, such as heuristics, framing effects, and risk attitudes in deci-
sion making. For example, when it comes to framing––the violation of rational choice theory
such that decision makers tend to make different choices in non-transparently equivalent
decision problems—accessibility, linked to the higher salience of differential characteristics
in the description of a decision problem, influences preference, leading to different choices
in equivalent settings. A cognitive effort could be made (and indeed, when it is done, it
tends to succeed in reducing framing effects) to provide a canonical representation of deci-
sion problems that would transparently reveal equivalences and hence lead to same choices
in equivalent decision problems. In its absence, however, decision makers are left with the
vagaries of the influence exerted by contextual cues.
Even in the realm of logical reasoning, one can see accessibility at work. Take, for instance,
the well-known asymmetry in providing a successful solution to the Wason’s Selection Task in
its abstract formulation and in a concrete formulation in which participants are asked to iden-
tify a cheater (Cox & Griggs, 1982). The latter version has a much higher rate of successful
solutions, which can be readily explained by observing that formal logical rules have, in general,
less accessibility than concrete “cheating-discovery module” rules (Cosmides, 1989; Cosmides
& Tooby, 1989).

171
172

Cristina Bicchieri and Giacomo Sillari

Representing knowledge and belief

To precisely characterize the link between accessibility and reasoning, we now confine our
attention to epistemic rationality, rather than full-fledged logical deductions. The formal frame-
work in which we propose to implement the distinction between accessible and inaccessible
items of knowledge is epistemic logic. Epistemic logic encodes the mathematical representation
of knowledge and belief. It was pioneered by Jaakko Hintikka in the 1960s, then revived by
the interest of both economists and computer scientists in the 1980s and today is thriving at
the intersection of disciplines like philosophy, economics, informatics, and cognitive science.
The basic idea behind formal representations of knowledge and belief is to give a semantics
of such notions in terms of possible worlds that may or may not be epistemically possible for the
agents. Each world validates a certain set of atomic propositions. An agent is said to know a
given proposition at world w1 if it holds true in all possible worlds that are epistemically possible
for that agent from w1.
Let us illustrate this key notion with a simple example (van Benthem, 2008): Ann, Bob, and
Carol receive one card each from a set comprising one red, one white, and one blue card. They
can see what card they have been dealt, but they cannot tell which card the other kids hold. We
can represent the relevant worlds by the initials of the cards distributed to Ann, Bob, and Carol,
respectively. For instance, BWR is the world in which Ann has the blue card, Bob the white
card, and Carol the red card. Atoms have the form bAnn, meaning that Ann holds the blue card.
bAnn is true in both BWR and BRW and only in those two possible worlds. Thus, for instance,
Ann cannot distinguish between BWR and BRW. We can schematically represent the model M
as follows, representing Ann’s accessibility relation with solid lines, Bob’s with dotted lines and
Carol’s with dashed lines (and omitting reflexive arrows) (Figure 9.1).
It is now easy to answer questions about what players know about each other. For instance: at
BWR, does Ann know that Bob has the white card? She does not, since both BWR and
BRW are epistemically accessible for her, and Bob has white in one state but red in the other.
However, she knows that Carol does not have blue, since both in BWR and BRW, bAnn holds.
More complex propositions can be tested: does Bob know that Ann does not know that Bob
has white? From BWR, Bob accesses BWR and RWB. At BWR Ann doesn’t know bBob. From
RWB, Ann accesses RBW and RWB, and she does not know that bBob either. So at all worlds
accessible from BWR for Bob, Ann doesn’t know bBob. Hence Bob knows that Ann doesn’t
know that Bob holds blue.
Knowledge and belief are distinguished in the language by imposing different axiomatic
constraints. For instance, it is customary to accept the veridicality axiom5 for knowledge (where
Kφ stands for “the agent knows φ”):6

BRW WRB

BWR RWB

WBR RBW

Figure 9.1 A representation of the knowledge states of three agents in terms of possible worlds and
epistemic accessibility relations

172
173

Bounded knowledge

(T) Kφ → φ.
In English, the axiom ensures that if an agent knows φ, then φ must be true. Instead of assuming
veridicality, accounts of belief impose a weaker axiomatic constraint (where Bφ stands for “the
agent believes φ”):
(B) Bφ → ¬B¬φ.
Axiom (B) makes sure that beliefs are consistent, that is if an agent believes φ, then the agent
does not believe its negation. These kinds of axiomatic constraints translate into structural prop-
erties of possible worlds semantics. For instance, veridicality entails that the accessibility relation
be reflexive, as the agent must be able to see that φ holds at the world she is at. Veridicality is
not the only property of knowledge and belief posited by epistemic logicians. Both positive and
negative introspective axioms such as

(4) Kφ → KKφ
and

(5) ¬Kφ → K¬Kφ

are required. Axiom (4) says that agents know what they know, while the Socratic agent of
axiom (5), if she does not know something, knows that she does not know it. These axioms,
along with a few others, ensure that the accessibility relation on the set of possible worlds is an
equivalence relation. This in turn entails that the accessibility relation induces a partition of the
space of possible worlds, in fact, it establishes the equivalence between possible worlds models
favored by philosophers and partitional models used in economics to model agents’ knowledge
and information. This approach to modeling knowledge and belief was proven to be extremely
fruitful in economics and computer science alike, generating relevant philosophical debates
as well.
While partitional models of this kind are widely used in economics or computer science
applications, they present a problem, as they entail several forms of so-called logical omnisci-
ence: Agents are, by construction, supposed to have unrealistic logical, epistemic, and doxastic
abilities.7 In a certain sense, logical omniscience reveals that modeling epistemic agents in this
formal framework is akin to endowing them with “Olympian epistemic rationality.” Indeed, a
structural axiom characterizing both knowledge and belief is

(K) K(φ → φ') → Kφ & Kφ'.

Combining (K) and the axioms of propositional logic, epistemic systems entail that agents’
knowledge is closed under logical consequence. Hence, for instance, given basic knowledge of
the axioms of arithmetic, an agent is supposed to grasp all theorems logically deducible from
them, and know whether an old unsolved problem such as Goldbach’s conjecture8 is true. But
this particular epistemic property (closure under logical implication) represents only one facet
of the so-called problem of logical omniscience. That is to say, epistemic logic is an adequate
representation of agent’s epistemic reasoning only if we accept unrealistic idealizations but
not so if we are trying to model realistic agents. The problem was already apparent to Jaakko
Hintikka in his seminal 1962 monograph Knowledge and Belief (Hintikka, 1962). His clever yet
partial way out was to distinguish two senses of “knowledge”: a weak sense, in which know-
ledge stands simply for information, and a strong sense, in which knowledge is information held
by the agent by virtue of some kind of justification.9 Thus, the semantic characterization of
knowledge described above would in fact be a characterization of information, while knowledge

173
174

Cristina Bicchieri and Giacomo Sillari

proper would consist of items of information that are entertained (that is, accessed) by the agent
in some relevant (epistemological or perhaps psychological) sense.10 Thus, epistemic logic offers
a powerful tool to formally and mathematically (set-theoretically) characterize the epistemic
characteristics and reasoning of agents, yet the characterization is flawed in the same sense in
which the characterization of rational agents in economics is flawed. The agents populating
epistemic logic models, just like the agents populating microeconomic models, are perfect
(logical) reasoners: They know all tautologies, they know all the consequences of what they
know, and so on. In a word, their knowledge is unbounded. Can we nevertheless provide a char-
acterization of bounded knowledge retaining the useful framework of epistemic logic?

Newell on knowledge representation

A suggestion on the direction where to look for an answer to the question above was offered
by Allen Newell, one of Herbert Simon’s most important co-authors in the field of Artificial
Intelligence. In the “cognitive revolution” that took place in the 1950s and 1960s, the attention
of psychologists moved from a basic, neural level of analysis based on stimulus-response, to a
higher, symbolic level of analysis of cognitive architectures. In the first presidential address to
the American Association for Artificial Intelligence, Newell (1982) points out that there is
a further, higher level of analysis for cognitive systems. He calls it the knowledge level and in
describing its characteristics points at the answer we are looking for.
An intelligent system, in the functional view of agency endorsed by Newell, is embedded
in an action-oriented environment. The system’s activity consists in the process from a per-
ceptual component (that inputs task statements and information), through a representa-
tion module (that represents tasks and information as data structures), to a goal structure
(the solution to the given task statement). In this picture, knowledge is perceived from the
external world, and stored as it is represented in data structures.11 Newell claims that there
is a distinction between knowledge and its representation, in a way that is similar to the
difference that there is between the symbolic level of a computer system and the phys-
ical implementation supporting the symbolic manipulation. More precisely, in a computer
system, Newell defines a level as consisting of four general elements: (1) a medium (which
is to be processed); (2) components together with laws of composition; (3) a system; and
(4) determining the behavior of the system, laws of behavior. Different levels are determined
by different elements. For instance, at the symbolic level, we can describe a computer system
through the following elements: The medium is given by the content of computer mem-
ories, the components and their laws of composition are symbols and their syntax, while the
laws of behavior are given by the semantics of the logical operators. Below the symbolic level
we can describe a computer system at the logical level of its circuits and devices. Here the
medium is bit vectors, components and composition are given by functional units (logical
gates, etc.) in the processing chip, and laws of behavior are given by logical operations. We
could descend to describing a system at lower level, reducing the description all the way to
the physical level. Or, we could ascend, from the symbolic level, to describing the system at
a higher level, what Newell calls the knowledge level.
At the knowledge level, the system is the agent; the components are goals, actions, and bodies
of knowledge; the medium is knowledge, and the behavioral rule is rationality. Notice that the
symbolic level constitutes the level of representation. Hence, since every level is describable in
terms of the next lower level, knowledge can be represented through symbolic systems. But
can we provide a description of the knowledge level without resorting to the level of represen-
tation? It turns out that we can, although we only can if we do not decouple knowledge and

174
175

Bounded knowledge

action. In particular, says Newell, “it is unclear in what sense [systems lacking rationality] can
be said to have knowledge” (Newell, 1982, p. 100), where “rationality” stands for “principles
of action.” Indeed, an agent, at the knowledge level, is but a set of actions, bodies of knowledge,
and a set of goals, rather independently of whether the agent has any physical implementation.
What, then, is knowledge? Knowledge, according to Newell, is whatever can be ascribed to
an agent, such that the observed behavior of the agent gets to be explained (that is, computed)
according to the laws of behavior encoded in the principle of rationality. The principle of
rationality appears to be unqualified: “If an agent has knowledge that one of its actions will lead
to one of his goals, then the agent will select that action” (Newell, 1982, p. 102). Thus, the
definition of knowledge is a procedural one: an observer notices the action undertaken by the
agent; given that the observer is familiar with the agent’s goals and its rationality, the observer
can therefore infer what knowledge the agent must possess. Knowledge is not defined structur-
ally, for example, as physical objects, symbols standing for them and their specific properties
and relations. Knowledge is rather defined functionally as what mediates the behavior of the
agent and the principle of rationality governing the agent’s actions. Can we not sever the bond
between knowledge and action by providing, for example, a characterization of knowledge in
terms of a physical structure corresponding to it? As Newell explains,

The answer in a nutshell is that knowledge of the world cannot be captured in a finite
structure ... Knowledge as a structure must contain at least as much variety as the set
of all truths (i.e. propositions) that the agent can respond to.
Newell, 1982, p. 107

Hence, knowledge cannot be captured in a finite physical structure, and can only be considered
in its functional relation with action.
Thus, (a version of) the problem of logical omniscience presents itself when it comes to
describing the epistemic aspect of an intelligent system. Ideally (at the knowledge level), the
body of knowledge an agent is equipped with is unbounded, hence knowledge cannot be
represented in a physical system. However, recall from above how a level of interpretation of
the intelligent system is reducible to the next lower level. Knowledge should therefore be redu-
cible to the level of symbols. This implies that the symbolic level necessarily encompasses only
a portion of the unbounded body of knowledge that the agent possesses.
The interesting question is then: In what way does an agent extract representation from
knowledge? Or, in other terms: Given the definition of representation above, what can its theory
be? Building a theory of representation involves building a theory of access, to explain how
agents manage to extract limited, explicit knowledge (working knowledge, representation)
from their unbounded implicit knowledge. The crucial idea is that agents do so “intelligently,”
i.e., by judging what is relevant to the task at hand. Such a judgment, in turn, depends on
the principle of rationality. Hence, knowledge and action cannot be decoupled and know-
ledge cannot be entirely represented at the symbolic level, since it involves both structures
and processes. Logics, as they are “one class of representations ... uniquely fitted to the analysis
of knowledge and representation” seem to be suitable for such an endeavor. In particular,
epistemic logics, in which the notion of access can be built in, in order to achieve the distinc-
tion between (unbounded) knowledge and its (limited) representation, are natural candidates
for axiomatizing theories of explicit knowledge representation. If we understand that human
agents may fail to display unbounded rationality, then the bounds to an agent's knowledge
depend essentially on the way accessibility displays itself in behavior, that is through phenomena
like selective attention, priming, salience, and so on, resulting in biased judgments and choices.

175
176

Cristina Bicchieri and Giacomo Sillari

Models of (un)awareness
In the case of Allen Newell's analysis of agency at the knowledge level, knowledge representation
(readily available to the agent) is distinguished by knowledge tout court in that it consists of the
latter plus some form of access to it. The slogan, by Newell, is thus:

Representation = Knowledge + Access.

How can we incorporate the issue of accessibility into the fabric of epistemic logic? An answer
comes again from computer science. Fagin and Halpern (1988) proposed a logic of awareness in
which the picture of knowledge as based on access finds a precise formalization. In their logic,
there are three operators pertaining to issues epistemic: (1) an implicit knowledge operator,
which behaves in the standard way illustrated above; (2) an awareness operator, which has in its
scope formulas to which the agent has access to; and (3) an explicit knowledge operator, which
represents the realistic notion of knowledge we are interested in and that, intuitively, is used
for those items of implicit knowledge to which the agent has access or, in Fagin and Halpern’s
language, of which she has awareness.
The idea of introducing a function that filters an agent’s unbounded knowledge to distill
those items of knowledge she is in fact actively entertaining represents a first step toward the
explicit introduction of context representation in logical formalisms. Technically, the intuition
is that the juncture where issues of accessibility become prominent is the linguistic representa-
tion of propositions; hence the language distinguishes formulas an agent is aware of (i.e., those
she has access to) and those she is not aware of. At the semantic level, each world is associated
with an “awareness set,” representing the propositions to which the agent has access. Implicit
knowledge is defined as usual, while explicit knowledge is yielded, syntactically, by the con-
junction of implicit knowledge and awareness while, semantically, the truth conditions for an
agent to explicitly know φ at w hold if and only if she implicitly knows φ and φ belongs to the
agent’s awareness set at w.
Awareness logics are versatile, in that various properties of awareness can be axiomatically
included in the system. A kind of awareness (particularly favored by economists) is the so-
called awareness as generated by primitive propositions. In these systems, a subset of the set
of atoms in the language is specified, and agents are aware of all (and only) those formulas that
mention atoms from the relevant subset.12 In the economic literature, attempts to model agent’s
awareness information go back to Modica and Rustichini (1994, 1999). At first, following the
lead of Geanakoplos (1989, 1992), they attempted to model awareness in standard structures by
dropping the negative introspection axiom and positing that Aφ ↔ (Kφ ∨ ¬K¬Kφ), or that an
agent is aware of φ iff they know φ or, if they do not, when they ignore that they do not know
φ. However, Dekel, Lipman, and Rustichini (1998) show that the standard approach trivializes
awareness (either if an agent has access to one formula, she has access to all formulas or she has
access to no formulas at all), and Modica and Rustichini (1999) came up with a non-standard
approach introducing the notion that agents have access to a (possibly proper) subset of the set
of all atoms in the language. Halpern (2001) shows that their approach, although lacking an
implicit knowledge operator, is equivalent to a special case of the epistemic logic of awareness
introduced by Fagin and Halpern (1988).13

An application to strategic reasoning

Limits to epistemic rationality are readily understood as deviations from the standards of ration-
ality and, as such, taken to be undesirable or in violation of the normative requirements of such

176
177

Bounded knowledge

standards. In fact, one can question (Mercier & Sperber, 2011) the normativity of the standards,
one can find (Rips, 2011) discrepancies between the standards and actual behavior, or one can
find examples of theories in which limited knowledge provides a more realistic behavioral
model. One such example comes from game theory and has to do with the problem of back-
ward induction (Bicchieri & Antonelli, 1995, for a different yet related “logical” approach to
the problem, see Bonanno, 1991).
Any extensive game of perfect information can be solved by backwards induction (BI).
However, this solution concepts runs into difficulties were a deviation from the equilibrium
path to occur. For instance, let us say that the backward-induction equilibrium involves a move
by player 1 ending the game at the first node, and yet player 2 finds herself to play at the second
node. That means that player 1 has deviated from the equilibrium path, and therefore that he
has acted not rationally. What should player 2 do now? She cannot choose an action under the
assumption that player 1 is rational, since, were player 1 in fact rational, player 2 would not have
had to choose an action in the first place! Hence the assumption of rationality (and of common
knowledge thereof) has to be revised, since it leads to the paradoxical conclusion that player
2 should not be called into play at all (Bicchieri, 1989). This means that when a player finds
herself off the equilibrium path, she cannot assume the rationality of the other player (other-
wise she would not have found herself in a deviation in the first place) and hence she lacks the
ability to choose an action at an off-equilibrium node. Bicchieri and Antonelli (1995) point out
that this difficulty lies at the meta-language level. The analysis of BI deviations requires: (1) the
ability to reason counterfactually (since rational players do not deviate and rationality is com-
monly known); and (2) a theory of belief revision (since the fact that a deviation has occurred
requires a reassessment of the rationality assumptions made so far). The difficulties do not
appear, however, at the level of local decision making, as agents can be endowed with a minimal
theory of the game that is sufficient to compute the BI solution starting from any node and does
not come with global rationality assumptions.
Such a “modular” or “sequential” theory of the game belongs to agents whose epistemic
access is limited. Epistemic access is bounded to the analysis of future nodes, while previous nodes
(and possible deviations occurred thereof) remain inaccessible to players' theories. Bicchieri and
Antonelli, thus, propose to adopt a dual point of view in the analysis of backward induction: the
theorist’s and the players’. The game theorist’s view is a bird’s-eye perspective on the totality of
the game that does not involve sequential or modular rationality. Since the analysis of backwards
induction involves paradoxical reasoning about nodes that are not reached in equilibrium (see,
e.g., Bicchieri, 1989, 1997), a theory that involves global rationality has to be equipped with
logical tools to cope with counterfactual reasoning and belief revision (Bicchieri, 1988). The
players’ theories, on the other hand, make assumptions about rationality at a node and at sub-
sequent nodes, while history of previous play (and hence possible deviations off equilibrium,
and the need for counterfactual analysis and belief revision stemming from those deviations) are
not part of the theory. Bicchieri and Antonelli suggest that the logical formalisms of the two
theories be distinct. A way to understand the distinction is to put to work the notion of bounded
knowledge analyzed in this chapter and presume that players only have epistemic access, and thus
full belief, to a subset of rationality ascriptions, the accessible subset being maximal at the root
of the tree while shrinking as nodes get closer to the leaves of the game tree.

Conclusion
In this chapter we have looked at a specific aspect of bounded rationality: limits to human
logical abilities and, in particular, limits to the way real agents reason about knowledge. In

177
178

Cristina Bicchieri and Giacomo Sillari

principle, infinite, human knowledge is necessarily bounded and we argued here that a useful
way to understand its limits is differential accessibility. We applied the notion to the case of the
backwards induction solution concept, yet once we accept the link between bounded know-
ledge and accessibility, many more applications become relevant. Besides the rather abstract
ones mentioned above, bounded knowledge becomes relevant, for instance, when it comes
to explaining the great variability of behavior observed in many experiments in response to
slight situational variations (Bicchieri, 2006:46). Differential awareness elicits different behav-
ioral norms, and hence considerations about bounded knowledge, once the static framework
offered here is supplemented with suitable dynamics, can be extremely helpful in order to pro-
mote behavior change through social norm change (Bicchieri, 2016).

Notes
1 Satisficing, introduced conceptually by Simon (1947) and developed more formally in Simon (1955)
has been studied subsequently, in particular, by Sauermann and Selten (1962) and more recently by
Güth and collaborators (see, for a general account, Güth, 2010, or for a specific application, Güth
et al., 2008).
2 For a discussion highlighting the relevance of epistemic rationality in game theory and in economics,
see Bicchieri (1997), Chapter 3. In game theory, epistemic rationality deals with ways to achieve con-
cordant beliefs leading to choosing a Nash solution (e.g., tracing procedure; Harsanyi & Selten, 1988, or
rational deliberation; Skyrms, 1990). In this sense, epistemic rationality involves procedural or substantive
instances of belief change, be it in terms of Bayesian upgrading, or in terms of belief revision. More
generally, epistemic rationality is ascribable to an agent that entertains and believes propositions for
which she has good evidence, and avoids believing those for which she does not, changing her belief
in light of new evidence. For a philosophical analysis and argument in favor of the distinction between
practical and epistemic rationality, see Kelly (2003).
3 The road to a deeper exploration of the psychological foundations of human thought has also been
followed in the fields of logic and probability, whereas very few attempts have been made to modify
the axioms of these two theories in order to predict the real behavior of individuals. For instance,
Pei Wang (2001) gives an explanation of the phenomenon in terms of his non-Aristotelian NARS
inference system. Another important program is based on the work of Lance Rips (1995; 2011),
who developed cognitive models for deductive reasoning, such as the so-called PSYCOP, short for
“Psychology of Proof ” (Rips, 2008) based on assumptions about reasoners' memory, mastery of infer-
ence rules and control on specific proof processes. The model is validated empirically by comparing
its “deductive abilities” with those of human, boundedly rational, laboratory experiment participants,
showing that there is a significant match in the results of the deductions carried over by PSYCOP
and by human reasoners. Or, for a view from the field of cognitive psychology, see Stenning and
Van Lambalgen (2008). For descriptively adequate logical systems that try to account for observable
behavior but are not necessarily based on psychological processes, see Balbiani et al. (2016) and Solaki
et al. (2019). This contribution is intended as a step in such a direction.
4 While there is some overlap between the notion of implicit and tacit knowledge, it is important to stress
that the two notions are quite different. For instance, explicit and implicit knowledge, in the literature
we consider here, differ as a matter of an agent’s accessibility (the former is accessible, the latter is not),
while the definition tacit knowledge, since Polanyi’s fundamental contribution (Polanyi, 1967), does
not hinge on accessibility. In fact, tacit knowledge can be perfectly accessible but not readily codifiable,
or transmissible, or learnable (see e.g., Howells, 1996).
5 Customarily, in epistemic logic the veridicality axiom is axiom T (for truth). Axiom B characterizes
belief, while axioms 4 and 5 characterize the introspective abilities of agents.
6 We confine our argument to the single agent case. Formally, it can be generalized straightforwardly, although
adding interactions and group epistemic operators need not be neutral in more substantive implications.
Considering such implications rests beyond the scope of this chapter. In general, for an introduction to epi-
stemic logic and application, the reader is referred to Fagin et al. (2004) or van Benthem (2011).
7 Notice that the problem of logical omniscience applies to both epistemic (knowledge-representing) and
doxastic (belief-representing) systems. To see this, notice that some of the issues pertaining to logical
omniscience depend on the fact that the semantic structures validating the axioms are partitional. But

178
179

Bounded knowledge

both epistemic and doxastic semantics are based on partitional structure, one main difference being
that in epistemic structures agents are never mistaken about the position of the actual world, while in
doxastic structures, validating axiom (B) but not axiom (T), we allow agents to be “mistaken” about
which one is the actual state of affairs. Besides the possibility of mistakes, agents are still fully omnis-
cient, e.g., they believe in all the logical consequences of their, possibly mistaken, beliefs.
8 The claim that any integer greater than 2 is the sum of two prime numbers.
9 For a philosophical discussion of the distinction, see Sillari (2006). Artemov develops (in Artemov and
Kuznets, 2009, and elsewhere) a logic of justified knowledge.
10 It is interesting to note that, though not at all based on psychological evidence, Hintikka’s argument
in several places appeals to intuitions recoverable among ordinary language users (e.g., validity was
rethought of as “immunity from certain standards of criticism”).
11 While we follow here Newell (1982), where the subject is artificial agents, this account of agency is
neutral regarding its object: it can be artificial agents, as it can be human agents. As Newell (1994)
points out:
to claim that humans can be described at the knowledge level is to claim there is a way of for-
mulating them as agents that have knowledge and goals, such that their behavior is successfully
predicted by the law that says: all the person's knowledge is always used to attain the goals of the
person.
Notice that this entails that knowledge is always functional. One can think of seemingly non-
instrumental knowledge that, however, does play (tacitly, maybe) an instrumental role, for instance, a
signaling one.
12 Clearly, these systems are still a heroic idealization of agents’ deductive capabilities, since the agent is
fully logical omniscient relative to the subset of the language of which the agent is aware. Thus, such
models are perhaps seen as models of unawareness rather than awareness.
13 There are many other attempts to characterize (un)awareness. Heifetz et al. (2006) characterize it in
set-theoretic terms, Sillari (2008a), in first-order logic, Halpern and Rêgo (2009) in second-order
epistemic logic and several applications indicate the relevance of the notion, for instance, when it
comes to Lewis’s theory of convention (Sillari, 2005; 2008b), no-trade theorems (Heifetz et al.,
2013), dynamic models (van Benthem and Velàzquez-Quesada, 2010; Hill, 2010; Karni and Vierø,
2013), its relation with inductive reasoning (Grant and Quiggin, 2013). A utility-based approach cog-
nate to the idea of awareness as capturing differential access is developed by Golman and Loewenstein
(2015).

References
Artemov, S., & Kuznets, R. (2009). Logical omniscience as a computational complexity problem. In
Proceedings of the 12th Conference on Theoretical Aspects of Rationality and Knowledge (pp. 14–23). ACM.
Balbiani, P., Fernández-Duque, D., & Lorini, E. (2016). A logical theory of belief dynamics for resource-
bounded agents. In Proceedings of the 2016 International Conference on Autonomous Agents & Multiagent
Systems (pp. 644–652). International Foundation for Autonomous Agents and Multiagent Systems.
Bicchieri, C. (1988). Strategic behavior and counterfactuals. Synthese, 76, 135–169.
Bicchieri, C. (1989). Self-refuting theories of strategic interaction: A paradox of common knowledge.
Erkenntnis, 30, 69–85.
Bicchieri, C. (1997). Rationality and coordination. Cambridge: Cambridge University Press.
Bicchieri, C. (2006). The grammar of society: The nature and dynamics of social norms. Cambridge: Cambridge
University Press.
Bicchieri, C. (2016). Norms in the wild: How to diagnose, measure, and change social norms. Oxford: Oxford
University Press.
Bicchieri, C., & Antonelli, G. A. (1995). Game-theoretic axioms for local rationality and bounded know-
ledge. Journal of Logic, Language and Information, 4(2), 145–167.
Bonanno, G. (1991). Extensive forms and set-theoretic forms. Economics Letters, 37(4), 363–370.
Cosmides, L. (1989). The logic of social exchange: Has natural selection shaped how humans reason?
Studies with the Wason selection task. Cognition, 31(3), 187–276.
Cosmides, L., & Tooby, J. (1989). Evolutionary psychology and the generation of culture, Part II: Case
study: A computational theory of social exchange. Ethology and Sociobiology, 10(1–3), 51–97.

179
180

Cristina Bicchieri and Giacomo Sillari

Cox, J. R., & Griggs, R. A. (1982). The effects of experience on performance in Wason’s selection task.
Memory & Cognition, 10(5), 496–502.
Dekel, E., Lipman B. L., & Rustichini A. (1998). Standard state-space models preclude unawareness.
Econometrica, 66(1), 159–173.
Fagin, R., & Halpern J. (1988). Belief, awareness and limited reasoning. Artificial Intelligence, 34(1), 39–76.
Fagin, R., Halpern, J. Y., Moses, Y., & Vardi, M. (2004). Reasoning about knowledge. Cambridge,
MA: MIT Press.
Geanakoplos, J. (1989). Game theory without partitions, and applications to speculation and consensus.
Cowles Foundation Discussion Paper.
Geanakoplos, J. (1992). Common knowledge, Bayesian learning and market speculation with bounded
rationality. Journal of Economic Perspectives, 6, 58–82.
Golman, R., & Loewenstein, G. (2015). An information-gap framework for capturing preferences about
uncertainty. In Proceedings of the 15th conference on theoretical aspects of rationality and knowledge. Available
at: www.cmu.edu/dietrich/sds/docs/golman/Information-Gap Framework Golman...
Grant, S., & Quiggin, J. (2013). Inductive reasoning about unawareness. Economic Theory, 54(3), 717–755.
Güth, W. (2010). Satisficing and (un)bounded rationality: A formal definition and its experimental val-
idity. Journal of Economic Behavior & Organization, 73(3), 308–316.
Güth, W., Levati, M. V., & Ploner, M. (2008). Is satisficing absorbable? An experimental study. The Journal
of Behavioral Finance, 9(2), 95–105.
Halpern, J. (2001). Alternative semantics for unawareness. Games and Economic Behavior, 37(2), 321–339.
Halpern, J. Y., & Rêgo, L. C. (2009). Reasoning about knowledge of unawareness. Games and Economic
Behavior, 67(2), 503–525.
Harsanyi, J. C., & Selten, R. (1988). A general theory of equilibrium selection in games. Cambridge,
MA: MIT Press.
Heifetz, A., Meier, M., & Schipper, B. C. (2006). Interactive unawareness. Journal of Economic Theory,
130(1), 78–94.
Heifetz, A., Meier, M., & Schipper, B. C. (2013). Unawareness, beliefs, and speculative trade. Games and
Economic Behavior, 77(1), 100–121.
Hill, B. (2010). Awareness dynamics. Journal of Philosophical Logic, 39(2), 113–137.
Hintikka, J. (1962). Knowledge and belief. Ithaca, NY: Cornell University Press
Howells, J. (1996). Tacit knowledge. Technology Analysis & Strategic Management, 8(2), 91–106.
Kahneman, D. (2003). Maps of bounded rationality: Psychology for behavioral economics. American
Economic Review, 93(5), 1449–1475.
Karni, E., & Vierø, M. L. (2013). “Reverse Bayesianism”: A choice-based theory of growing awareness.
American Economic Review, 103(7), 2790–2810.
Kelly, T. (2003). Epistemic rationality as instrumental rationality: A critique. Philosophy and Phenomenological
Research, 66(3), 612–640.
Mercier, H., & Sperber, D. (2011). Why do humans reason? Arguments for an argumentative theory.
Behavioral and Brain Sciences, 34(2), 57–74.
Modica, S., & Rustichini, A. (1994). Awareness and partitional information structures. Theory and
Decision, 37(1), 107–124.
Modica, S., & Rustichini, A. (1999). Unawareness and partitional information structures. Games and
Economic Behavior, 27(2), 265–298.
Newell, A. (1982). The knowledge level. Artificial Intelligence, 18(1).
Newell, A. (1994). Unified theories of cognition. Cambridge, MA: Harvard University Press.
Polanyi, M. (1967/2009). The tacit dimension. Chicago: University of Chicago Press.
Rips, L. J. (1995). The psychology of proof: Deductive reasoning in human thinking. Cambridge, MA: MIT Press.
Rips, L. J. (2008). Logical approaches to human deductive reasoning. In J. E. Adler & L. J. Rips
(Eds.), Reasoning: Studies of human inference and its foundation (pp. 187–205). Cambridge: Cambridge
University Press.
Rips, L. J. (2011). Lines of thought: Central concepts in cognitive psychology. Oxford: Oxford University Press.
Sauermann, H., & Selten, R. (1962). Anspruchsanpassungstheorie der Unternehmung. Zeitschrift für die
gesamte Staatswissenschaft/Journal of Institutional and Theoretical Economics, 4, 577–597.
Sillari, G. (2005). A logical framework for convention. Synthese, 147(2), 379–400.
Sillari, G. (2006). Models of awareness. In G. Bonanno, W. van der Hoek, & M. Wooldridge (Eds.),
Logic and the foundations of game and decision theory (pp. 209–240). Amsterdam: Amsterdam University
Press,.

180
181

Bounded knowledge

Sillari, G. (2008a). Quantified logic of awareness and impossible possible worlds. Review of Symbolic Logic,
1(4), 1–16.
Sillari, G. (2008b). Common knowledge and convention. Topoi, 27(1), 29–40.
Simon, H. A. (1947). Administrative behavior. New York: The Macmillan Company.
Simon, H. A. (1955). A behavioral model of rational choice. Quarterly Journal of Economics, 69, 99–118.
Simon, H. A. (1957). Models of man. New York: Wiley.
Simon, H. A. (1967). Motivational and emotional controls of cognition. Psychological Review, 74(1), 29–39.
Simon, H. A. (1972). Theories of bounded rationality. Decision and Organization, 1(1), 161–176.
Skyrms, B. (1990). The dynamics of rational deliberation. Cambridge, MA: Harvard University Press.
Solaki, A., Berto, F., & Smets, S. (2019). The logic of fast and slow thinking. Erkenntnis. June.
Stenning, K., & van Lambalgen, M. (2008). Human reasoning and cognitive science. Cambridge,
MA: MIT Press.
van Benthem, J. (2008). Logic and reasoning: Do the facts matter? Studia Logica, 88, 67–84, special issue,
“Psychologism in Logic?,” edited by H. Leitgeb.
van Benthem, J. (2011). Logical dynamics of information and interaction. Cambridge: Cambridge
University Press.
van Benthem, J., & Velázquez-Quesada, F. R. (2010). The dynamics of awareness. Synthese, 177(1), 5–27.
Wang, P. (2001). Wason’s cards: What is wrong. In Proceedings of the Third International Conference on
Cognitive Science (pp. 371–375).
Wason, P. C. (1966). Reasoning. In B. M. Foss (Ed.), New horizons in psychology. Harmondsworth: Penguin.

181
182
183

PART II

Cognitive misery and mental

dualism
184
185

10
BOUNDED RATIONALITY,
REASONING AND DUAL
PROCESSING
Jonathan St. B. T. Evans

Introduction
Traditionally, philosophers distinguish between instrumental and epistemic rationality which
may be defined roughly as follows:

Instrumental rationality
Acting in such a way as to achieve one’s goals.
Epistemic rationality
Attaining and maintaining true beliefs.

Both concepts are slippery. Take, for example, my goals of being an Open Golf champion and
the world’s most famous cognitive psychologist. As I approach 70, the chance of the former
had vanished more than 50 years ago, as several years of play and practice had resulted in only a
middling 15 handicap. As to the latter, any realistic hope was extinguished perhaps 40 years ago.
But I am surely not irrational for failing to achieve these goals, even if I was wildly optimistic in
even contemplating them. This, of course, is what Herb Simon’s bounded rationality is all about.
We can only achieve goals that are within our capability, either as a species or as individuals. It
does make the whole idea of instrumental rationality a bit messy though. What goals should
we strive to attain? Only those that are fully realistic? Or should we set our sights high to reach
the limits we can?
Let us take the game of chess, whose study fascinated Simon. I also love the game and in
semi-retirement play a lot online and study the game, striving to improve my standard. The
game is, of course intractably difficult due to the vast search space. No human can ever master
it, nor will any machine we can imagine, even though they have now overtaken the best human
players. Except in simplified endgame positions, or those allowing a forced win, no human or
machine can be sure that it is picking the best move. None of us can optimise; so as Simon
put it, in chess as in all complex decision making, we can only satisfice – choose moves which
appear to be good enough. Unfortunately, the heuristic search methods promoted by Simon
(e.g., Newell & Simon, 1972) fell by the wayside in computer AI which came to be dominated
by brute force as computers acquired ever bigger memories and faster processors. Eventually,
a program of this type, Deep Blue, came to beat the world champion Gary Kasparov in 1997,

185
186

Jonathan St. B. T. Evans

even if there were deep suspicions about human intervention by the IBM team at the time.
Writing recently about this, Kasparov (2017) dismisses these suspicions on the grounds that the
‘human-like’ moves played by Deep Blue on an IBM supercomputer were replicated within
a few years by chess engines available cheaply on personal computers. He had long under-
stood that computers would eventually prevail – it was just a matter of speed. What is remark-
able, however, is how well the best human players can compete with the current generation
of programs that can search billions of chess positions before selecting a move. Significantly,
Kasparov points out that this is because computers are weak in two areas where humans are
strong: pattern recognition and strategic thinking.
The striking thing about the cognitive science of chess is just how differently computers
had to be designed to play the game in order to match human ability. This shows us that the
human brain, with its massively parallel architecture, has adapted to the solution of complex
problems by means that do not require brute force computation. And of course, Simon was
one of the first to understand the role of perception and pattern recognition in human expertise
(Chase & Simon, 1973). Like machines, human players do calculate continuations and move
sequences, but only a tiny fraction of what the machines do. The recognition of which moves
deserve attention, cued by patterns learnt by long study, is what gives humans playing strength.
This, of course, is an example of dual processing (Evans, 2008; Evans & Stanovich, 2013). These
rapid recognition processes (Type 1, autonomous) greatly reduce the load of explicit calculation
(Type 2, engaging working memory). If we think of rationality as depending simply on explicit
calculation and reasoning, then we do our species a great disservice. Our working memories
are small and slow but our capacity for implicit learning of patterns is vast. Thus, the bounds on
our rationality are a lot less severe than one might imagine. Unfortunately for the human race,
however, recent developments in artificial intelligence have shown that chess programs based on
machine learning and pattern recognition are even better than the best of brute force engines.
Reportedly, AlphaZero, originally developed to play Go but adapted for chess, taught itself to
beat the world’s best brute force program in just four hours (Klein, 2017)!
If instrumental rationality is a slippery concept, then so too is epistemic rationality. In
fact, the psychology of reasoning has been radically reconstructed over the past 20 years, as
the model handed by philosophers proved inadequate. The traditional emphasis was on the
importance of logic as the basis of rational thinking (Henle, 1962; Inhelder & Piaget, 1958),
which led to the foundation of the traditional study of deductive reasoning by psychologists.
If you think epistemic rationality is about truth, then deductive reasoning is very important
because it is truth-preserving. That is to say, any deductively valid conclusion is true, provided
that its premises are true. How disappointing, then, when hundreds of laboratory studies
showed that human adults, typically of undergraduate student level intelligence, made
numerous logical errors and showed systematic cognitive biases (Evans, Newstead, & Byrne,
1993). Moreover, they were heavily influenced by the content or meaning of logical problems,
which was irrelevant to their formal structure. Such findings led to a sharp rationality debate,
when the philosopher Jonathan Cohen claimed that such experimental findings could not
demonstrate that people were irrational (Cohen, 1981). Eventually, psychologists working this
field confronted the dilemma before them: either people are inherently irrational or else there
was something wrong with the paradigm they were using to study reasoning (Evans, 2002;
Oaksford & Chater, 1998).
One problem is that nothing new can be learnt by deductive reasoning, which makes it a
poor standard for rationality. We can form new beliefs by induction, but we cannot be sure
that they are true. Strangely, induction has been the poor relation, with far fewer psychological
studies of inductive inference carried out (Feeney & Heit, 2007). Reasoning from the particular

186
187

BR, reasoning and dual processing

to the general is immensely beneficial but not logically valid. Such inductions are not truth-
preserving and their conclusions can only be held provisionally, or with a degree of probability
pending further evidence. But how could we cope in the world without the ability to learn?
Even if one takes a modular view of the human mind, dividing it into many specialised infor-
mation processing systems, one must assume that many of these have the ability to learn and
modify themselves with experience (Carruthers, 2006).
Those working within the ‘new paradigm’ in the psychology of reasoning (Elqayam & Over,
2012; Evans, 2012; Oaksford & Chater, 2013; Over, 2009) accept that people routinely use
their beliefs when reasoning and allow them to draw conclusions with degrees of probability.
Some authors have suggested that Bayesian theory should replace the role once played by logic
(Oaksford & Chater, 2007) although this is not a necessary constraint. It is important to rec-
ognise, however, that the new paradigm does not mean that psychologists have switched from
studying deduction to induction, which is still a separate paradigm. Because the new paradigm
allows people to express degrees of belief in their conclusions, it would be easy to think that it
was inductive inferences that were being studied. However, the field has rather moved towards
the study of uncertain deduction, where uncertainty in premises is reflected by uncertainty in
conclusions drawn from them (Evans & Over, 2012; Evans, Thompson, & Over, 2015). At one
conference, I put up a slide with one of Sherlock Holmes’ famous ‘deductions’ (see Box 9.1)
and asked the audience of reasoning researchers whether the inference he drew was actually
inductive, deductive, or abductive. The great majority voted for ‘inductive’ but I was able to
show that Holmes’ argument was in fact deductively valid. Their intuition, correctly, was that
Holmes’ inference was uncertain. However, the uncertainty lay entirely in the premises from
which his deduction was made. Watson may have picked the mud up somewhere else, visited
the Post Office to meet a friend, and so on. But if Holmes’ premises were true, then so was his
conclusion.

Box 9.1 Extract from Conan-Doyle’s The Sign of Four (1890)

(HOLMES TO WATSON) “Observation shows me that you have been to the Wigmore Street
Post-Office this morning, but deduction lets me know that when there you dispatched a telegram.”
“Right!” said I. “Right on both points! But I confess that I don’t see how you arrived at it. It was
a sudden impulse upon my part, and I have mentioned it to no one.”
“It is simplicity itself,” he remarked, chuckling at my surprise, – “so absurdly simple that an
explanation is superfluous; and yet it may serve to define the limits of observation and of deduction.
Observation tells me that you have a little reddish mould adhering to your instep. Just opposite the
Wigmore Street Office they have taken up the pavement and thrown up some earth which lies in
such a way that it is difficult to avoid treading in it in entering. The earth is of this peculiar reddish
tint which is found, as far as I know, nowhere else in the neighbourhood. So much is observation.
The rest is deduction.”
“How, then, did you deduce the telegram?”
“Why, of course I knew that you had not written a letter, since I sat opposite to you all morning.
I see also in your open desk there that you have a sheet of stamps and a thick bundle of post-cards.
What could you go into the post-office for, then, but to send a wire? Eliminate all other factors,
and the one which remains must be the truth.”

187
188

Jonathan St. B. T. Evans

The one domain in which truth is valued above all else is science. Science strives for true
explanations and the avoidance of contradictions. If two theories conflict in their proposals
then we take it for granted that at least one will be false, even if there is no immediate means
to prove which. The history of science teaches us that false beliefs, for example, about the
motion of the planets, or the evolution of life will be eventually abandoned, no matter how
much prior belief is operating in their favour. Important though science is, however, it is only
one domain in which (some) people need to operate. Many scientists also hold religious and
political beliefs, for example, which may be directly inconsistent with their scientific beliefs, or
at least not provable by the scientific method. As another example, my scientific view of psych-
ology is inconsistent with many of the tenets of folk psychology which I resist in my scientific
thinking. In everyday life, however, I apply the native folk psychology which enables all of us
to function socially. Our built-in folk psychology and theory of mind are good examples of fit-
for-purpose beliefs. They do not stand up to scientifically rigorous examination, but they don’t
need to. They are fit for the purpose for which they evolved, which is communication, social
interaction and the formation of social structures.
In discussing chess earlier, I pointed out that our pattern recognition and implicit learning
abilities compensate to a great extent for the limited computational capacity of executive pro-
cessing, requiring working memory. So one approach to bounded rationality is the idea that we
can often rely heavily on gut feelings and quick and simple heuristics for our decision making
(Dijksterhuis, Bos, Nordgren, & von Baaren, 2006; Gigerenzer, 2007; Gigerenzer & Todd,
1999; Gladwell, 2005). However, there are also vast amounts of research showing the import-
ance of general intelligence, working memory and explicit forms of reasoning in the solution
of a wide variety of cognitive tasks. This supports the proposals of dual-process theory (Evans
& Stanovich, 2013), in which rapid and intuitive forms of processing (Type 1) are said to com-
bine with slower and more effortful reasoning, the latter engaging working memory (Type 2).
One of the dual-process theorists who most strongly advocates the importance of Type 2 pro-
cessing in rational decision making is Keith Stanovich, who together with colleagues applied
the theory to individual differences in reasoning and decision-making ability (Stanovich, 1999,
2011; Stanovich, West, & Toplak, 2016). The argument here is that while people often rely on
heuristics and intuitions, this can lead them into serious errors and biases. Such errors can be
avoided by successful intervention with high effort reasoning, provided that the individual has
the cognitive capability and the relevant knowledge. So, at this point, I need to examine the
dual-process case in more detail.

Default-interventionist dual-process theory and bounded rationality

Dual-process theory is something of an intellectual minefield for several reasons. One problem
is that there are many authors in cognitive and social psychology advocating various form of
dual processing which appear similar but may differ in detail. All broadly identify two forms
of processing: Type 1 which is intuitive and typically quick, and Type 2 which is reflective
and typically slow. One important structural difference is between theories that have a parallel-
competitive approach and those which are default-interventionist (terms originally introduced by
Evans (2007b). Parallel-competitive theories envisage that associative (Type 1) and rule-based
(Type 2) processes proceed in parallel and compete for control of our behaviour (Sloman,
1996; Smith & DeCoster, 2000). The more popular form, however, is what I termed ‘default-
interventionist’ (DI). Theorists of this persuasion include myself, Stanovich and Kahneman
(Evans, 2007a; Evans & Stanovich, 2013; Kahneman, 2011; Stanovich, 2011). The common
framework is the assumption that Type 1 processes, being relatively quick and automatic, provide

188
189

BR, reasoning and dual processing

Intui ve
processing
(Type 1)

A1
COGNITIVE
MOTIVATIONAL Reflec ve RESOURCES
FACTORS processing
(Type 2) Time available
Instruc onal set
Compe ng tasks
Thinking disposi on
Mindware
FOR Set degree of WMC
cri cal effort

no Resources
A1 available for
yes jus fied?
rethinking?

yes
no
No answer/
Answer A1 Answer A2
guess / A1

Figure 10.1 The default-interventionist process

default intuitions which are subject to scrutiny and possible intervention by Type 2 processing,
which are slower and effortful. A great deal of research has been conducted by now showing
that the answers people give to typical laboratory problems are affected by both experimental
manipulations (e.g., time available, concurrent working memory load, instructions) and by
individual differences in both cognitive ability and cognitive style. This work is reflected in the
DI theory I proposed a few years ago (Evans, 2011) and shown in Figure 10.1.
The model assumes that Type 1 processing leads to a default intuitive answer A1 which
appears in consciousness without effort or reflection. A1 is then subject to scrutiny by Type
2 processes which are effortful and require working memory. However, such scrutiny can be
shallow or deep, which depends on a number of factors. Motivational factors include how
people are instructed, whether they have a rational thinking disposition (which inclines people
to check intuitions more carefully) and the feeling of rightness (FOR) or subjective confidence
that they have in A1 (Thompson, Prowse Turner, & Pennycook, 2011). Cognitive resources
include time available (speeded tasks will reduce Type 2 checking), individual differences in
working memory capacity (highly correlated with IQ) and ‘mindware’, of which more shortly.
When A1 is scrutinised, it may or may not be deemed satisfactory. Whether the intuition can
successfully be replaced by an answer based on explicit reasoning and knowledge, however,
depends on the cognitive resources available as Stanovich (2011) has argued. He has provided
much empirical evidence (see also Stanovich et al., 2016) on how decision making is affected
by individual differences in both cognitive ability and rational thinking dispositions. The latter
measure the extent to which people will rely on intuitions or check them out by reasoning. On
the basis of this work, Stanovich argues that a number of conditions must be met for someone
to successfully engage in Type 2 reasoning and solve a demanding and novel experimental

189
190

Jonathan St. B. T. Evans

problem. First, people must be aware of the need for intervention by Type 2 reasoning, which
is where rational thinking dispositions and feelings of rightness come into play. However, if they
do intervene, then this will only be successful if both (1) they have sufficient cognitive cap-
acity for the hypothetical reasoning involved; and (2) they have the relevant mindware – that is,
formal or procedural knowledge required by the context.
The DI dual-process framework is clearly one of bounded rationality. Stanovich (2011)
writes extensively about the cognitive miser, which is the idea that we evolved to expend as
little mental effort as we can get away with. Most of our behaviour, most of the time is con-
trolled by autonomous, Type 1 processes. This makes sense when you consider that Type 2 or
executive processing engaged in working memory is a limited and essentially singular resource
which can only perform one task at a time with any efficiency. Hence, this valuable resource,
which corresponds with the focus of our conscious attention, must be used sparingly and for
the most important tasks only. However, in laboratory tasks at least, the cognitive miser default
can lead to cognitive biases and normative errors. Stanovich suggests this happens in three ways
(see Stanovich, 2011, pp. 98–104). First, the system may detect no need for Type 2 interven-
tion and default to autonomous processing. Second, a simplified and low effort form of Type
2 processing may be used to solve a problem and is equally likely to lead to a biased response.
Finally, a genuine effort may be made at intervention but fail to change the default answer. As
Elqayam (2012) points out, only those equipped with sufficient ability and mindware to solve
the problems can fairly be accused of cognitive miserliness should they fail to do so. Stanovich
(e.g., 2009) has a particular focus on the case of people with high IQs who nevertheless fail to
make what he considers to be rational decisions.
There are some correspondences with Stanovich’s arguments in my own writing and theor-
etical framework. For example, I have suggested (Evans, 2007a) that there is both a fundamental
heuristic bias (Type 1) and a fundamental analytic bias (Type 2) and I apportioned approximately
equal ‘blame’ to each in explaining the various biases in the literature. The Type 1 bias is to
focus only on information which is preconsciously cued as relevant and the Type 2 bias is
to focus on singular hypotheses and fail to consider alternatives. The latter corresponds to
Stanovich’s lazy and ineffective Type 2 processing. I also agree with Stanovich that the inter-
vention with Type 2 reasoning does not necessarily lead to an improved response, as indicated
in Figure 10.1. However, there are a couple of differences in our approach, also. First, in all my
models, Type 2 processing is always involved before a decision is made, even though the default
answer may well continue unchanged. Second, a major reason why people persist with A1 in
Figure 10.1 is that they seek justification for an initial intuition and often end up rationalising
this response. In fact, my very earliest papers on dual processing, in collaboration with Pete
Wason, were concerned entirely with this rationalising function (Evans & Wason, 1976; Wason
& Evans, 1975). On the Wason selection task, with which these papers were concerned, it was
many years before I could find clear evidence that Type 2 processing might lead to a change of
the intuitive response at all, although I eventually did (Evans & Ball, 2010).
This rationalisation aspect of Type 2 processing does not fit as neatly with the cognitive
miser hypothesis because it does require mental effort and working memory resources to carry
out, even though the default intuition is maintained. However, it is supported by evidence that
we spend most time thinking about responses that are intuitively cued (Ball, Lucas, Miles, &
Gale, 2003; Evans, 1996). It also fits with evolutionary accounts of explicit reasoning that link
this facility to argumentation and social discourse (Mercier & Sperber, 2011). Entertaining this
argument for a moment, one way in which an evolved argumentation function could become
adapted to the assistance of problem solving would be precisely in the kind of model I am
proposing here. Instead of simply making an intuitive response, we first seek arguments and

190
191

BR, reasoning and dual processing

justifications, by an internal mental process, to support them. While this may bias the system
towards rationalisation and confabulation, it does at least allow the possibility that error will be
detected and corrected. This is more likely to happen when more time and effort are expended
on examining the intuitive response, which reflects the individual differences in rational thinking
dispositions studied by Stanovich and others. These measures, even when based on self-report,
do predict accuracy on cognitive tasks when effects of cognitive ability are statistically removed,
as many studies in cognitive and social psychology have shown.

Metacognitive feelings
Metacognition is the knowledge or feeling that we have about our own cognitive processes.
When we make judgements or predictions, for example, these may come with feelings of
confidence which we can report upon if asked. A well-established literature in the study of
judgement and decision-making reports a general phenomenon of overconfidence. When we
are more confident, we are in fact more likely to be right, but not as likely as we think we are
(Koehler, Brenner, & Griffin, 2002). There has been a lot of research into feeling of knowing
(FOK) in memory research, which again shows imperfect correlation with actual accuracy of
memories (Koriat, 1993). Of more relevance to this chapter and the bounded rationality argu-
ment is the suggestion that these metacognitive feelings have a regulatory purpose, controlling
the amount of effort that is expended (Ackerman, 2014).
Valerie Thompson (2009) was the first to introduce the issue of metacognitive feelings to the
psychology of reasoning. It was an attempt to solve a fundamental issue for dual process theories
at the time of writing, namely, what determines whether an intervention with Type 2 processing
will occur. She suggested that the default intuitive answer that comes to mind carries with it
a feeling of rightness (FOR) which determines the amount of effort that will be made. When
FOR is high, we will tend to accept our intuitions with little effort at reasoning and when it
is low, we will intervene more. A number of empirical studies followed which supported these
claims. Thompson invented a two-response paradigm, in which people are asked to give a
quick intuitive answer together with the FOR rating. They are then allowed to think further
for as long as they like and change the answer if they wish. On a number of different cognitive
tasks, Thompson and colleagues showed that low FOR leads to greater rethinking time and
more likelihood that initial answers will be changed (Thompson et al., 2011).
On the face of it, the FOR research seems to fit a neat story of evolution and bounded
rationality. Our intuitions come helpfully packaged with feelings of confidence which tell us
whether we need to expend effort or not in checking them out. However, there is a snag. None
of this research – on reasoning and decision making – has provided evidence for even a partial
accuracy of the FOR judgement, as observed in other fields where metacognition is studied.
The answers which people rethink and change are as likely to be right as wrong and low FOR
can lead to abandonment of an initially correct judgement. Some cues which lead to errors are
particularly convincing, such as the matching bias which dominates choices on the abstract selec-
tion task (Thompson, Evans, & Campbell, 2013) and the compelling but wrong intuitions on
the cognitive reflection task (Frederick, 2005). Belief bias is a major cause of error in syllogistic
reasoning (Evans, Barston, & Pollard, 1983; Klauer, Musch, & Naumer, 2000) and yet believ-
able conclusions inspire positive feelings of liking and confidence (Morsanyi & Handley, 2012;
Shynkarkuk & Thompson, 2006). None of which evidence supports the claim that we evolved
these metacognitive feelings because they are of assistance to the cognitive miser.
There are two problems with all of these metacognitive measures. First, they are intro-
spective reports whose validity can only be established by relating them to other behavioural

191
192

Jonathan St. B. T. Evans

measures. Second, these relations are correlational rather than causal. What we know about
FOR is that strong feelings are associated with answers that are given quickly and generally
held on to with little effort to rethink them. Putting these two problems together, we can
see that there is no clear reason to think that FOR corresponds to a stage of cognitive pro-
cessing in the manner that researchers in the field assume. They could simply be conscious
feelings that are by-products of the processes that determine the behaviour measures. In
other words, a brain process that leads to a quick and firmly held decision, also generates
conscious feelings of rightness and confidence, which in themselves are of no functional
significance.
Recent studies with the Thompson two-response paradigm have suggested than on some
simpler forms of syllogistic reasoning, most of the people who get it right after reflection,
also got it right with their initial intuitive response (Bago & De Neys, 2017). So, it appears
that no slow reflective Type 2 process was apparently involved. Connected to this is another
recent finding that people of higher cognitive ability or with better mindware for the numer-
ical problems studied have stronger intuitions leading them to the right answers, whereas
those of lower ability are more likely to be influenced intuitively by prior belief (Thompson,
Pennycook, Trippas, & Evans, 2018). There are actually a number of recent studies showing
that people may have ‘logical intuitions’ (De Neys, 2012, 2017). While some authors have taken
such findings as evidence against the traditional default-interventionist account, the case is not
that clear-cut (Evans, 2018). First, most of these recent studies use relatively simple tasks where
cues for logically correct response might well be present without need for working memory
engagement. Second, those of higher cognitive ability may have acquired mindware that has
become automated with practice. Although laboratory tasks that are both novel and requiring
a degree of complexity for their solution typically require Type 2 processing, it is a fallacy to
assume that Type 1 processes are necessarily biased or that Type 2 processing always leads to
correct answers (Evans & Stanovich, 2013).

Conclusion
In the early part of this chapter, I showed that Simon’s notion of bounded rationality applies
both to how we can achieve our goals and to how we can organise and validate our knowledge
about the world. I also argued, using the domain of chess, that while our capabilities of explicit
calculation are sharply limited compared with modern computers, the brain has evolved rapid,
pattern recognition systems which compensate to a large extent for this limitation. When we
examine the field of dual processing in the study of reasoning and decision making, however,
the dice have been loaded heavily in favour of tasks that require Type 2 processing (explicit
calculation and reasoning loading on working memory) for their solution. This is because the
tasks presented tend to be novel and difficult to solve on the basis of prior experience. For
the same reason, this literature has given relatively bad press to intuitive or Type 1 processes
which operate quickly and autonomously. In fact, the same kinds of process that enable the
best human chess players to approach the standard of brute force computer programs! And yet
researchers are somehow surprised when they simplify the standard laboratory tasks and find a
lot of accurate Type 1 processing is involved in their solution.
I have shown that the default-interventionist dual process theories that have dominated in
these fields are essentially within the bounded rationality tradition. That is to say, it is assumed
that Type 1 processing will dominate by default and high effort Type 2 processing will be select-
ively applied. There is much accumulated evidence that this is the case (Evans, 2007a; Evans &
Stanovich, 2013; Stanovich, 2011) although recent studies have somewhat muddied the waters

192
193

BR, reasoning and dual processing

(De Neys, 2017). The studies by Thompson and colleagues of feelings of rightness are of much
interest but produce a mixed picture. On the one hand, FOR most definitely correlates with
the amount of time and effort people expend on reasoning, as opposed to reliance on initial
intuitions. On the other hand, there is no evidence that these feelings are related to actual
accuracy, and some major cognitive biases (e.g., matching bias, belief bias) are supported by
strong and false feelings of confidence. It is at least possible that these feelings are no more than
a by-product of the brain processes that lead to strongly or weakly held decisions.
While DI dual-process theory is under pressure from those inside as well as outside of the
paradigm, we should not lose sight of the broader picture. First, there is no doubt that the great
bulk of our behaviour is controlled by fast and automatic processing. Outside of the laboratory,
such processing is often very effective and compensates for the limitations of our explicit com-
putational capacity. There is also no doubt that we do have a central working memory system
which is implicated in the successful execution of a very large range of cognitive functions
(Baddeley, 2007). So the way in which our brains provide our bounded rationality simply has
to be some kind of dual process story, whether we call these Type 1 and 2 or just autonomous
and executive processes. Although some evolutionary psychologists have argued for a massively
modular brain that does not depend on any general-purpose reasoning system (Cosmides &
Tooby, 1992) and other prominent researchers state that we can safely rely on intuitions and gut
feelings for complex decisions (Gigerenzer, 2007), these argument simply do not hold up. More
than 100 years of research on general intelligence and 50 plus years of research on working
memory attest to the fact that there is a central cognitive system allowing behaviour to be based
on reasoning and hypothetical thought and which operates distinctly from autonomous pro-
cessing. The intellectual success of our own species quite clearly depends upon it.

Acknowledgements
The author is grateful to Shira Elqayam for a critical reading of an earlier draft of this chapter.

References
Ackerman, R. (2014). The diminishing criterion role for metacognitive regulation of time investment.
Journal of Experimental Psychology General, 143, 1349–1368.
Baddeley, A. (2007). Working memory, thought and action. Oxford: Oxford University Press.
Bago, B., & De Neys, W. (2017). Fast logic?: Examining the time course assumption of dual process
theory. Cognition, 158, 90–109. doi:10.1016/j.cognition.2016.10.014
Ball, L. J., Lucas, E. J., Miles, J. N. V., & Gale, A. G. (2003). Inspection times and the selection task: What
do eye-movements reveal about relevance effects? Quarterly Journal of Experimental Psychology, 56(6),
1053–1077.
Carruthers, P. (2006). The architecture of the mind. Oxford: Oxford University Press.
Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4, 55–81.
Cohen, L. J. (1981). Can human irrationality be experimentally demonstrated? Behavioral and Brain
Sciences, 4, 317–370.
Cosmides, L., & Tooby, J. (1992). Cognitive adaptations for social exchange. In J. H. Barkow, L. Cosmides
& J. Tooby (Eds.), The adapted mind: Evolutionary psychology and the generation of culture (pp. 163–228).
New York: Oxford University Press.
De Neys, W. (2012). Bias and conflict: A case for logical intuitions. Perspectives on Psychological Science,
7, 28–38.
De Neys, W. (Ed.) (2017). Dual process theory 2.0. London: Routledge.
Dijksterhuis, A., Bos, M. W., Nordgren, L. F., & von Baaren, R. B. (2006). On making the right
choice: The deliberation-without-attention effect. Science, 311, 1005–1007.
Elqayam, S. (2012). Grounded rationality: Descriptivism in epistemic context. Synthese, 189, 39–49.

193
194

Jonathan St. B. T. Evans

Elqayam, S. & Over, D. E. (2012). New paradigm in psychology of reasoning: Probabilities, beliefs and
dual processing. Mind & Society, 11, 27–40.
Evans, J. St. B. T. (1996). Deciding before you think: Relevance and reasoning in the selection task. British
Journal of Psychology, 87, 223–240.
Evans, J. St. B. T. (2002). Logic and human reasoning: An assessment of the deduction paradigm.
Psychological Bulletin, 128, 978–996.
Evans, J. St. B. T. (2007a). Hypothetical thinking: Dual processes in reasoning and judgement.
Hove: Psychology Press.
Evans, J. St. B. T. (2007b). On the resolution of conflict in dual-process theories of reasoning. Thinking
& Reasoning, 13, 321–329.
Evans, J. St. B. T. (2008). Dual-processing accounts of reasoning, judgment and social cognition. Annual
Review of Psychology, 59, 255–278.
Evans, J. St. B. T. (2010). Thinking twice: Two minds in one brain. Oxford: Oxford University Press.
Evans, J. St. B. T. (2011). Dual-process theories of reasoning: Contemporary issues and developmental
applications. Developmental Review, 31, 86–102.
Evans, J. St. B. T. (2012). Questions and challenges for the new psychology of reasoning. Thinking &
Reasoning, 18, 5–31.
Evans, J. St. B. T. (2018). Dual-process theory: Perspectives and problems. In W. De Neys (Ed.), Dual
process theory 2.0 (pp. 137–155). London: Routledge.
Evans, J. St. B. T., & Ball, L. J. (2010). Do people reason on the Wason selection task?: A new look at the
data of Ball et al. (2003). Quarterly Journal of Experimental Psychology, 63(3), 434–441.
Evans, J. St. B. T., Barston, J. L., & Pollard, P. (1983). On the conflict between logic and belief in syllo-
gistic reasoning. Memory & Cognition, 11, 295–306.
Evans, J. St. B. T., Newstead, S. E., & Byrne, R. M. J. (1993). Human reasoning: The psychology of deduction.
Hove: Lawrence Erlbaum.
Evans, J. St. B. T., & Over, D. E. (2012). Reasoning to and from belief: Deduction and induction are still
distinct. Thinking & Reasoning, 19(3–4), 267–283. doi:10.1080/13546783.2012.745450.
Evans, J. St. B. T., & Stanovich, K. E. (2013). Dual process theories of higher cognition: Advancing the
debate. Perspectives on Psychological Science, 8, 223–241.
Evans, J. St. B. T., Thompson, V. A., & Over, D. E. (2015). Uncertain deduction and conditional reasoning.
Frontiers in Psychology, 6. doi:10.3389/fpsyg.2015.00398.
Evans, J. St. B. T., & Wason, P.C. (1976). Rationalisation in a reasoning task. British Journal of Psychology,
63, 205–212.
Feeney, A., & Heit, E. (2007). Inductive reasoning. Cambridge: Cambridge University Press.
Frederick, S. (2005). Cognitive reflection and decision making. Journal of Economic Perspectives, 19(4), 25–42.
Gigerenzer, G. (2007). Gut feelings: The intelligence of the unconscious. London: Penguin.
Gigerenzer, G., & Todd, P. M. (1999). Simple heuristics that make us smart. Oxford: Oxford University Press.
Gladwell, M. (2005). Blink. London: Penguin.
Henle, M. (1962). On the relation between logic and thinking. Psychological Review, 69, 366–378.
Inhelder, B., & Piaget, J. (1958). The growth of logical thinking. New York: Basic Books.
Kahneman, D. (2011). Thinking, fast and slow. New York: Farrar, Straus and Giroux.
Kasparov, G. (2017). Thinking deeply: Where machine intelligence ends. London: John Murray.
Klauer, K. C., Musch, J., & Naumer, B. (2000). On belief bias in syllogistic reasoning. Psychological Review,
107, 852–884.
Klein, M. (2017). Google’s AlphaZero destroys Stockfish in 100 game match. Available at: www.chess.
com/news/view/google-s-alphazero-destroys-stockfish-in-100-game-match.
Koehler, D. J., Brenner, L., & Griffin, D. (2002). The calibration of expert judgment: heuristics and biases
beyond the laboratory. In T. Gilovich, D. Griffin, & A. Kahneman (Eds.), Heuristics and biases: The
psychology of intuitive judgment. Cambridge: Cambridge University Press.
Koriat, A. (1993). How do we know what we know? The accessibility model of the feeling of knowing.
Psychological Review, 100, 609–639.
Mercier, H., & Sperber, D. (2011). Why do humans reason? Arguments for an argumentative theory.
Behavioral and Brain Sciences, 34, 57–111.
Morsanyi, K., & Handley, S. (2012). Logic feels so good – I like it. Evidence for an implicit detection of
logicality in syllogistic reasoning. Journal of Experimental Psychology-Learning Memory and Cognition, 38,
596–616.
Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall.

194
195

BR, reasoning and dual processing

Oaksford, M., & Chater, N. (1998). Rationality in an uncertain world. Hove: Psychology Press.
Oaksford, M., & Chater, N. (2007). Bayesian rationality: The probabilistic approach to human reasoning.
Oxford: Oxford University Press.
Oaksford, M., & Chater, N. (2013). Dynamic inference and everyday conditional reasoning in the new
paradigm. Thinking & Reasoning, 19(3–4), 346–379. doi:10.1080/13546783.2013.808163.
Over, D. E. (2009). New paradigm psychology of reasoning. Thinking & Reasoning, 15, 431–438.
Shynkarkuk, J. M., & Thompson, V. A. (2006). Confidence and accuracy in deductive reasoning. Memory
& Cognition, 34, 619–632.
Sloman, S. A. (1996). The empirical case for two systems of reasoning. Psychological Bulletin, 119, 3–22.
Smith, E. R., & DeCoster, J. (2000). Dual-process models in social and cognitive psychology: Conceptual
integration and links to underlying memory systems. Personality and Social Psychology Review, 4(2),
108–131.
Stanovich, K. E. (1999). Who is rational? Studies of individual differences in reasoning. Mahwah, NJ: Lawrence
Erlbaum Associates.
Stanovich, K. E. (2009). The thinking that IQ tests miss. Scientific American, November/December, 33–39.
Stanovich, K. E. (2011). Rationality and the reflective mind. New York: Oxford University Press.
Stanovich, K. E., West, C., & Toplak, M. E. (2016). The rationality quotient: Towards a test of rational thinking.
Cambridge, MA: MIT Press.
Thompson, V. A. (2009). Dual-process theories: A metacognitive perspective. In J. S. B. T. Evans & K.
Frankish (Eds.), In two minds: Dual processes and beyond (pp. 171–196). Oxford: Oxford University Press.
Thompson, V. A., Evans, J. S. T., & Campbell, J. I. D. (2013). Matching bias on the selection task: It’s fast
and feels good. Thinking & Reasoning, 19(3–4), 431–452. doi:10.1080/13546783.2013.820220.
Thompson, V. A., Pennycook, G., Trippas, D., & Evans, J. St. B. T. (2018). Do smart people have better
intuitions? Journal of Experimental Psychology General. 147(7), 945–961.
Thompson, V. A., Prowse Turner, J. A., & Pennycook, G. (2011). Intuition, reason, and metacognition.
Cognitive Psychology, 63(3), 107–140.
Wason, P. C., & Evans, J. St. B. T. (1975). Dual processes in reasoning? Cognition, 3, 141–154.

195
196

11
WHY HUMANS ARE
COGNITIVE MISERS AND
WHAT IT MEANS FOR THE
GREAT RATIONALITY DEBATE
Keith E. Stanovich

Introduction
That humans are cognitive misers has been a major theme throughout the past 50 years of
research in psychology and cognitive science (see Dawes, 1976; Kahneman, 2011; Simon, 1955;
1956; Shah & Oppenheimer, 2008; Taylor, 1981; Tversky & Kahneman, 1974). Humans are
cognitive misers because their basic tendency is to default to processing mechanisms of low
computational expense. Humorously, Hull (2001) has said that “the rule that human beings
seem to follow is to engage the brain only when all else fails—and usually not even then” (p. 37).
More seriously, Richerson and Boyd (2005) have put the same point in terms of its origins in
evolution: “In effect, all animals are under stringent selection pressure to be as stupid as they can
get away with” (p. 135). Miserly cognitive tendencies have evolved for reasons of computational
efficiency. But that same computational efficiency simultaneously guarantees that humans will
be less than perfectly rational—that they will display, instead, bounded rationality.

Miserly processing and human evolution

Of course, evolution guarantees human rationality in the dictionary sense of “the quality or state
of being able to reason” because evolution built the human brain. But evolution does not guar-
antee perfect rationality in a different sense—the sense used throughout cognitive science: as
maximizing subjective expected utility. In contrast to maximization, natural selection works
on a “better than” principle. The variation and selective retention logic of evolution “design”
for the reproductive advantage of one organism over the next, not for the optimality of any
one characteristic (including rationality). Natural selection is geared to immediate advantage
rather than long-term strategy. Human rationality, in contrast, must incorporate the long-term
interests of the individual and thus it can diverge from the short-term strategies of evolutionary
adaptation.
Organisms have evolved to increase the reproductive fitness of genes, not to increase the
rationality of humans, and increases in fitness do not always entail increases in rationality. For
example, beliefs need not always track the world with maximum accuracy in order for fitness
to increase (Mercier & Sperber, 2017; Stanovich, 2004). Evolution might fail to select out

196
197

Why humans are cognitive misers

epistemic mechanisms of high accuracy when they are costly in terms of organismic resources
(for example, in terms of memory, energy, or attention). Unreliable, error-prone, risk-aversive
strategies may well be favored by natural selection (Stich, 1990).
It is likewise in the domain of goals and desires. The purpose of evolution was not to maxi-
mize the happiness of human beings. As has become clear from research on affective forecasting
(Gilbert, 2006; Kahneman, 2011), people are remarkably bad at making choices that make
them happy. This should be no surprise. The reason we have pleasure circuits in our brains is
to encourage us to do things (survive and reproduce, help kin) that propagate our genes. The
pleasure centers were not designed to maximize the amount of time we are happy.
The instrumental rationality1 of humans is not guaranteed by evolution for two further
reasons. First, many genetic goals may no longer serve our ends because the environment
has changed. The goals underlying these mechanisms have become detached from their evo-
lutionary context (Li, van Vugt & Colarelli, 2018). Finally, the cultural evolution of rational
standards is apt to occur at a pace markedly faster than that of human evolution (Richerson &
Boyd, 2005; Stanovich, 2004)—thus providing ample opportunity for mental mechanisms of
utility maximization to dissociate from local genetic fitness maximization.
That evolution does not guarantee perfect rationality in humans is the first fundamental
concept that we need in order to resolve the Great Rationality Debate in cognitive science—
the debate about how much rationality to ascribe to people (Cohen, 1981; Gigerenzer, 1996;
Kahneman & Tversky, 1996; Kelman, 2011; Lee, 2006; Polonioli, 2015; Samuels & Stich,
2004; Stanovich, 1999, 2004; Stanovich & West, 2000; Stein, 1996; Tetlock & Mellers,
2002). The other two concepts that are needed are dual-process cognitive theory and an
understanding of the logic of goals within the human organism. The first is well-known and
has been exhaustively discussed, so I turn first to the latter (see Stanovich, 2004, for a fuller
discussion).

The logic of goals in organisms of differing complexity

I will rely here on Dawkins’ (1976, 1982) discussion of replicators and vehicles: replicators as
entities (e.g., genes) that copy themselves and vehicles as the containers (e.g., organisms) in
which replicators house themselves. It is vehicles that interact with the environment, and the
differential success of the vehicles in interacting with the environment determines the success
of the replicators that they house. Humans have proven to be good vehicles for genes, as have
bees. But the goal structures of bees and humans are very different.
As a creature characterized primarily by a so-called Darwinian mind (see Dennett, 1996,
2017), a bee has a goal structure as indicated in Figure 11.1. The area labeled A indicates the
majority of cases where the replicator and vehicle goals coincide. Not flying into a brick wall
serves both the interests of the genes and of the bee itself as a coherent organism. Of course,
the exact area represented by A is nothing more than a guess. The important point is that there
exists a nonzero area B—a set of goals that serve only the interests of the replicators and that
are antithetical to the interests of the vehicle itself.2 A given bee will sacrifice itself as a vehicle
if there is greater benefit to the same genes by helping other bees (e.g., causing its own death
when it loses its stinger while protecting its genetically-related hive-Queen).
All of the goals in a bee are genetic goals pure and simple. Some of these goals overlap with
the interests of the bee as a vehicle and some do not, but the bee does not know enough to
care. Of course, the case of humans is radically different. The possibility of genetic interests
and vehicle interests dissociating has profound implications for humans as self-contemplating
vehicles.

197
198

Keith E. Stanovich

Goal structure:
Darwinian creature

A
Goals serving both
vehicle and genes’
interests

B
Goals serving only the
genes’ interests

Figure 11.1 Goal structure of a Darwinian creature. The areas indicate overlap and nonoverlap of
vehicle and genetic “interests.”

Humans were the first organisms capable of recognizing that there may be goals embedded
in their brains that serve the interests of their genes rather than their own interests and the first
organisms capable of choosing not to pursue those goals. An organism with a flexible intel-
ligence and long-leash goals can, unlike the situation displayed in Figure 11.1, develop goals
that are completely dissociated from genetic optimization. For the first time in evolutionary
history, we have the possibility of a goal structure like that displayed in Figure 11.2 (again, the
sizes of these areas are pure conjecture). Here, although we have area A as before (where gene
and vehicle goals coincide) and area B as before (goals serving the genes’ interests but not the
vehicle’s), we have a new area, C, which shows that, in humans, we have the possibility of goals
that serve only the vehicle’s interests and not those of the genes.
Why does area C come to exist in humans? When the limits of coding the moment-by-
moment responses of their vehicles were reached, the genes began adding long-leash strategies to
the brain (Dennett, 1996, 2017; Stanovich, 2004). At some point in evolutionary development,
these long-leash strategies increased in flexibility to the point that—to anthropomorphize—the
genes said the equivalent of: “Things will be changing too fast out there, brain, for us to tell
you exactly what to do—you just go ahead and do what you think is best given the general
goals (survival, sexual reproduction) that we (the genes) have inserted.” But once the goal has
become this general, a potential gap has been created whereby behaviors that might serve the

198
199

Why humans are cognitive misers

Goal structure:
humans

A
Goals serving both
vehicle and genes’
interests

B
Goals serving only the
genes’ interests

C
Goals serving only the
vehicle’s interests

Figure 11.2 The logic of the goal structure in a human

vehicle’s goal might not serve that of the genes. We need not go beyond the obvious example
of sex with contraception—an act which serves the vehicle’s goal of pleasure without serving
the genes’ goal of reproduction. The logic of the situation here is that the goals of the vehicle—
being general instantiations of things that probabilistically tend to reproduce genes—can diverge
from the specific reproductive goal itself.

Genetic and vehicle goals in a dual-process organism

The last global concept that is needed to contextualize the Great Rationality Debate is that of
dual-process theory. To simplify the discussion, we need only the most basic assumptions of
such a theory (Stanovich & Toplak, 2012) along with the subsequent clarifications and caveats
that have been much discussed in the literature (Evans, 2008, 2014, 2018; Evans & Stanovich,
2013; Stanovich, 2011). In many such theories, the defining feature of System 1 processing
is its autonomy. Execution of these processes is mandatory when their triggering stimuli are
encountered, and they are not dependent on input from high-level control systems. The cat-
egory of autonomous, System 1 processes would include: processes of emotional regulation;
the encapsulated modules for solving specific adaptive problems that have been posited by evo-
lutionary psychologists; processes of implicit learning; and the automatic firing of overlearned
associations.

199
200

Keith E. Stanovich

In contrast to System 1 processing, System 2 processing is nonautonomous and computa-

tionally expensive. Many System 1 processes can operate in parallel, but System 2 processing is
largely serial. One of the most critical functions of System 2 processing is to override nonoptimal
System 1 processing (for extensive discussion of the details of these broad generalizations, see
De Neys, 2018; Evans & Stanovich, 2013; Pennycook et al., 2015; Stanovich, 2004, 2011,
2018; Thompson, 2009). There are individual differences in the System 2 tendency to override,
and thus there are individual differences in how miserly people are (Stanovich, West, &
Toplak, 2016).
System 1 is partially composed of older evolutionary structures (Amati & Shallice, 2007;
Mithen, 1996, 2002; Reber, 1993) that more directly code the goals of the genes (repro-
ductive success), whereas the goal structure of System 2—a more recently evolved brain cap-
ability (Evans, 2010; Mithen, 1996, 2002; Stanovich, 2004, 2011)—is more flexible and on an
ongoing basis attempts to coordinate the goals of the broader social environment with the more
domain-specific short-leash goals of System 1. System 2 is primarily a control system focused
on the interests of the whole person. It is the primary maximizer of an individual's personal goal
satisfaction.
Because System 2 is more attuned to the person’s needs as a coherent organism than is
System 1 (which is more directly tuned to the ancient reproductive goals of the subpersonal
replicators), in the minority of cases where the outputs of the two systems conflict, people will
often be better off if they can accomplish an override of the System 1-triggered output. Such
a system conflict could be signaling a vehicle/replicator goal mismatch and, statistically, such a
mismatch is more likely to be resolved in favor of the vehicle (which all of us should want) if
the System 1 output is overridden.
Figure 11.3 displays a graphic representation of the logic of the situation (of course, the exact
size of the areas of overlap are mere guesses; it is only the relative proportions that are necessary
to sustain the argument here). It illustrates that override is a statistically good bet in cases of
conflict because System 1 contains a disproportionate share of the goals serving only the genes’
interests and not the vehicle’s (area A) and System 2 contains a disproportionate share of the
goals serving only the vehicle’s interests and not the genes’ (area F). An assumption reflected in
Figure 11.3 is that vehicle and gene goals coincide in the vast majority of real-life situations (the
areas labeled B and E). For example, accurately navigating around objects in the natural world
fostered evolutionary adaptation—and it likewise serves our personal goals as we carry out our
lives in the modern world. But the most important feature of Figure 11.3 is that it illustrates the
asymmetries in the interests served by the goal distributions of the two systems.
Many of the goals instantiated in System 1 were acquired nonreflectively—they have not
undergone an evaluation in terms of whether they served the person’s interests (area A in
Figure 11.3). They have in fact been evaluated, but by a different set of criteria entirely: whether
they enhanced the longevity and fecundity of the replicators in the evolutionary past. From the
standpoint of the individual person (the vehicle), these can become dangerous goals because
they reflect genetic goals only.3 They are the goals that sacrifice the vehicle to the interests of
replicators—the ones that lead the bee to sacrifice itself for its genetically-related hive-Queen.
These are the goals that should be strong candidates for override.
The right side of Figure 11.3 indicates the goal structure of System 2. Through its exer-
cise of a reflective intelligence, this system derives flexible long-leash goals that often serve
the overall ends of the organism but thwart the goals of the genes (area F in Figure 11.3—for
example, sex with contraception; resource use after the reproductive years have ended, etc.). Of
course, a reflectively acquired goal can, if habitually invoked, become part of System 1 as well
(Bago & De Neys, 2017; Stanovich, 2018). This fact explains why there is a small4 section (area

200
201

Why humans are cognitive misers

Goal structure
System 1 System 2
Goals reflecting Goals reflecting

Genes’ interests D

Genes’ interests
A

Genes’ and vehicle’s

interests coincide

Genes’ and vehicle’s E

interests coincide

B Vehicle’s interests

Vehicle’s interests C F

Figure 11.3 Genetic and vehicle goal overlap in System 1 and System 2

C) in System 1 representing goals that serve the vehicle’s interests only. Reflectively acquired
goal-states might be taken on for their unique advantages to the vehicle (advantages that might
accrue because they trump contrary gene-installed goals—“don’t flirt with your boss’s wife”)
and then may become instantiated in System 1 through practice. We might say that in situations
such as this, System 1 in humans reflects the consequences of residing in a brain along with a
reflective System 2. This is why the goal-structure of System 1in humans does not simply recap-
itulate the structure of a Darwinian creature depicted in Figure 11.1.
Nevertheless, with the small but important exception of area C, System 1 can be under-
stood, roughly, as the part of the brain on a short genetic leash. In contrast, most of the goals
that the System 2 is trying to coordinate are derived goals. When humans live in complex
societies, basic goals and primary drives (bodily pleasure, safety, sustenance) are satisfied indir-
ectly by maximizing secondary symbolic goals such as prestige, status, employment, and remu-
neration. In order to achieve many of these secondary goals, the more directly-coded System
1 responses must be suppressed—at least temporarily. Long-leashed derived goals create the
conditions for a separation between the goals of evolutionary adaptation and the interests of
the vehicle.
Because of its properties of autonomy, System 1 will often provide an output relevant to a
problem in which System 2 is engaged. Such a system conflict could be signaling a vehicle/
replicator goal mismatch and, statistically, such a mismatch is more likely to be resolved in
favor of the vehicle (which all of us should want) if the System 1 output is overridden (area
E + F exceeds area B + C). This is why, in cases of response conflict, override is a statistically
good bet.

201
202

Keith E. Stanovich

Reconciling the opposing positions in the Great Rationality Debate

Researchers working in the heuristics and biases tradition tend to see a large gap between
normative models of rational responding and descriptive models of what people actually do.
These researchers have been termed Meliorists (Stanovich, 1999, 2004, 2010) because they
assume that human reasoning is not as good as it could be, and that thinking could be improved
(Stanovich et al., 2016).
However, over the last several decades, an alternative interpretation of the findings from the
heuristics and biases research program has been championed. Contributing to this alternative
interpretation have been philosophers, evolutionary psychologists, adaptationist modelers, and
ecological theorists (Cohen, 1981; Gigerenzer, 2007; Oaksford & Chater, 2007, 2012; Todd &
Gigerenzer, 2000). They have reinterpreted the modal response in most of the classic heuristics
and biases experiments as indicating an optimal information processing adaptation on the part
of the subjects. This group of theorists—who argue that an assumption of maximal human
rationality is the proper default position to take—have been termed the Panglossians.
The Panglossian theorists often argue either that the normative model being applied is not
the appropriate one because the subject’s interpretation of the task is different from what the
researcher assumes it is, or that the modal response in the task makes perfect sense from an evo-
lutionary perspective. The contrasting positions of the Panglossians and Meliorists define the
differing poles in what has been termed the Great Rationality Debate in cognitive science—the
debate about whether humans can be systematically irrational.
A reconciliation of the views of the Panglossians and Meliorists is possible, however. I argued
above that the statistical distributions of the types of goals being pursued by System 1 and
System 2 processing are different. Because System 2 processing is more attuned to the person’s
needs as a coherent organism, in the minority of cases where the outputs of the two systems
conflict, people will often be better off if they can accomplish a System 1 override (the full
argument5 is contained in Stanovich, 2004). Instances where there is a conflict between the
responses primed by System 1 and System 2 processing are interpreted as reflecting conflicts
between two different types of optimization—fitness maximization at the subpersonal genetic
level and utility maximization at the personal level.
A failure to differentiate these interests is at the heart of the disputes between researchers
working in the heuristics and biases tradition and their critics in the evolutionary psychology
camp. First, it certainly must be said that the evolutionary psychologists are on to something
with respect to the tasks they have analyzed, because in most cases the adaptive response is the
modal response in the task—the one most subjects give. Nevertheless, this must be reconciled
with a triangulating data pattern relevant to this discussion—an analysis of patterns of covari-
ation and individual differences across these tasks. Specifically, we have found that cognitive
ability often (but not always) dissociates from the response deemed adaptive from an evolu-
tionary analysis (Stanovich & West, 1998, 1999, 2000).
These two data patterns can be reconciled, however. The evolutionary psychologists are
probably correct that most System 1 processing is evolutionarily adaptive. Nevertheless, their
evolutionary interpretations do not impeach the position of the heuristics and biases researchers
that the alternative response given by the minority of subjects is rational at the level of the
individual. Subjects of higher analytic intelligence are simply more prone to override System 1
processing in order to produce responses that are epistemically and instrumentally rational. This
rapprochement between the two camps was introduced by Stanovich (1999) and subsequent
research has only reinforced it (see Kahneman and Frederick, 2002; Kelman, 2011; Samuels &
Stich, 2004; Stanovich, 2004, 2011).

202
203

Why humans are cognitive misers

It is possible to continue to resist this rapprochement, of course, but only at the expense
of taking rather extreme positions. A Meliorist could resist the rapprochement by continuing
to deny the efficacy of much of our cognition from the standpoint of evolution—a position
that denies much of evolutionary cognitive science. A Panglossian might decide to reject the
rapprochement by siding with the goals of the genes over the goals of the vehicle when the
two conflict. But most people find this choice unpalatable, and few of those who claim they
do not have considered exactly what they are endorsing when they do. For example, Cooper
(1989), in an essay describing how some nonoptimal behavioral tendencies could be genetic-
ally optimal, admits that such behaviors are indeed detrimental to the reasoner's own welfare.
Nonetheless, he goes on to counter that the behaviors are still justified because: “What if the
individual identifies its own welfare with that of its genotype?” (p. 477).
But who are these people with such loyalty to the random shuffle of genes that is their geno-
type? I really doubt that there are such people. To be precise, I am doubting whether there are
people who say they value their genome and have an accurate view of what they are valuing when
they say this. For example, in such a case, the person would have to be absolutely clear that
valuing your own genome is not some proxy for valuing your children; be clear that having chil-
dren does not replicate one's genome; and be clear about the fact that the genome is a subpersonal
entity. Most people, I think, would eschew this Panglossian path if it were properly understood
and adopt the view of philosopher Alan Gibbard, who offers the more reasoned view that

a person’s evolutionary telos explains his having the propensities in virtue of which
he develops the goals he does, but his goals are distinct from this surrogate purpose.
My evolutionary telos, the reproduction of my genes, has no straightforward bearing
on what it makes sense for me to want or act to attain ... A like conclusion would
hold if I knew that I was created by a deity for some purpose of his: his goal need not
be mine.
1990, pp. 28–29

Gibbard’s view is shared by distinguished biologist George Williams (1988), who feels that

There is no conceivable justification for any personal concern with the interests (long-
term average proliferation) of the genes we received in the lottery of meiosis and fer-
tilization. As Huxley was the first to recognize, there is every reason to rebel against
any tendency to serve such interest.
p. 403

Hence the title of an earlier book of mine, The Robot’s Rebellion (Stanovich, 2004). The oppor-
tunity exists for a remarkable cultural project that involves advancing human rationality by
honoring human interests over genetic interests when the two do not coincide. Its emancipa-
tory potential is lost if we fail to see the critical divergence of interests that creates the distinc-
tion between genetic fitness and maximizing human satisfaction.

Notes
1 I define instrumental rationality standardly here as: Behaving in the world so that you get exactly what
you most want, given the resources (physical and mental) available to you. More formally, economists
and cognitive scientists define maximizing instrumental rationality as choosing among options based on
which option has the largest expected utility.

203
204

Keith E. Stanovich

2 Strictly speaking, there are two conceptually different subspaces within area B. There are goals that are
currently serving genetic fitness that are antithetical to the vehicle’s interests, and there are goals within
this area that serve neither genetic nor vehicle interests. The reason there are the latter is because genetic
goals arose in the ancient environment in which our brains evolved (the environment of evolutionary
adaptation, EEA). Environments can change faster than evolutionary adaptations, so that some genetic
goals may not always be perfectly adapted to the current environment. Whether these goals currently
facilitate genetic fitness—or only facilitated reproductive fitness in the past—is irrelevant for the pre-
sent argument. In either case, goals which diverge from vehicle goals reside in the brain because of the
genes. For example, whether the consumption of excess fat serves current reproductive fitness or not, it
is a vehicle-thwarting tendency (for most of us!), and it is there because it served reproductive fitness at
some earlier point in time.
3 The caveat in note 2 is relevant here as well. When something is labeled a genetic goal, it does not
necessarily mean that the goal is currently serving the interests of reproductive fitness—only that it did so
sometime in the past in the EEA.
4 Of course, the absolute sizes of the areas in Figure 11.3 are a matter of conjecture. The argument here
depends only on the assumption that area A is larger than area D.
5 The full treatment in Stanovich (2004) also discusses the case of goals serving the interests of memes
(and not genes or vehicles).

References
Amati, D., & Shallice, T. (2007). On the emergence of modern humans. Cognition, 103, 358–385.
Bago, B., & De Neys, W. (2017). Fast logic? Examining the time course assumption of dual process theory.
Cognition, 158, 90–109.
Cohen, L. J. (1981). Can human irrationality be experimentally demonstrated? Behavioral and Brain
Sciences, 4, 317–370.
Cooper, W. S. (1989). How evolutionary biology challenges the classical theory of rational choice. Biology
and Philosophy, 4, 457–481.
Dawes, R. M. (1976). Shallow psychology. In J. S. Carroll & J. W. Payne (Eds.), Cognition and social behavior
(pp. 3–11). Hillsdale, NJ: Erlbaum.
Dawkins, R. (1976). The selfish gene. New York: Oxford University Press.
Dawkins, R. (1982). The extended phenotype. New York: Oxford University Press.
De Neys, W. (Ed.) (2018). Dual process theory 2.0. London: Routledge.
Dennett, D. C. (1996). Kinds of minds: Toward an understanding of consciousness. New York: Basic Books.
Dennett, D. C. (2017). From bacteria to Bach and back. New York: Norton.
Evans, J. St. B. T. (2008). Dual-processing accounts of reasoning, judgment and social cognition. Annual
Review of Psychology, 59, 255–278.
Evans, J. St. B. T. (2010). Thinking twice: Two minds in one brain. Oxford: Oxford University Press.
Evans, J. St. B. T. (2014). Reasoning, rationality and dual processes. London: Psychology Press.
Evans, J. St. B. T. (2018). Dual process theory: Perspectives and problems. In W. De Neys (Ed.), Dual pro-
cess theory 2.0. London: Routledge.
Evans, J. St. B. T., & Stanovich, K. E. (2013). Dual-process theories of higher cognition: Advancing the
debate. Perspectives on Psychological Science, 8, 223–241.
Gibbard, A. (1990). Wise choices, apt feelings: A theory of normative judgment. Cambridge, MA: Harvard
University Press.
Gigerenzer, G. (1996). On narrow norms and vague heuristics: A reply to Kahneman and Tversky (1996).
Psychological Review, 103, 592–596.
Gigerenzer, G. (2007). Gut feelings: The intelligence of the unconscious. New York: Viking Penguin.
Gilbert, D. T. (2006). Stumbling on happiness. New York: Alfred A. Knopf.
Hull, D. L. (2001). Science and selection: Essays on biological evolution and the philosophy of science.
Cambridge: Cambridge University Press.
Kahneman, D. (2011). Thinking, fast and slow. New York: Farrar, Straus & Giroux.
Kahneman, D., & Frederick, S. (2002). Representativeness revisited: Attribute substitution in intuitive
judgment. In T. Gilovich, D. Griffin, & D. Kahneman (Eds.), Heuristics and biases: The psychology of
intuitive judgment (pp. 49–81). New York: Cambridge University Press.

204
205

Why humans are cognitive misers

Kahneman, D., & Tversky, A. (1996). On the reality of cognitive illusions. Psychological Review, 103,
582–591.
Kelman, M. (2011). The heuristics debate. New York: Oxford University Press.
Lee, C. J. (2006). Gricean charity: The Gricean turn in psychology. Philosophy of the Social Sciences, 36,
193–218.
Li, N., van Vugt, M., & Colarelli, S. (2018). The evolutionary mismatch hypothesis: Implications for psy-
chological science. Current Directions in Psychological Science, 27, 38–44.
Mercier, H., & Sperber, D. (2017). The enigma of reason. Cambridge, MA: Harvard University Press.
Mithen, S. (1996). The prehistory of mind: The cognitive origins of art and science. London: Thames and
Hudson.
Mithen, S. (2002). Human evolution and the cognitive basis of science. In P. Carruthers, S. Stich, & M.
Siegel (Eds.), The cognitive basis of science (pp. 23–40). Cambridge: Cambridge University Press.
Oaksford, M., & Chater, N. (2007). Bayesian rationality: The probabilistic approach to human reasoning.
Oxford: Oxford University Press.
Oaksford, M., & Chater, N. (2012). Dual processes, probabilities, and cognitive architecture. Mind &
Society, 11, 15–26.
Pennycook, G., Fugelsang, J. A., & Koehler, D. J. (2015). What makes us think? A three-stage dual-process
model of analytic engagement. Cognitive Psychology, 80, 34–72.
Polonioli, A. (2015). Stanovich’s arguments against the “Adaptive rationality” Project: An assessment.
Studies in History and Philosophy of Biological and Biomedical Sciences, 40, 55–62.
Reber, A. S. (1993). Implicit learning and tacit knowledge. New York: Oxford University Press.
Richerson, P. J., & Boyd, R. (2005). Not by genes alone: How culture transformed human evolution.
Chicago: University of Chicago Press.
Samuels, R., & Stich, S. P. (2004). Rationality and psychology. In A. R. Mele & P. Rawling (Eds.), The
Oxford handbook of rationality (pp. 279–300). Oxford: Oxford University Press.
Shah, A. K., & Oppenheimer, D. M. (2008). Heuristics made easy: An effort-reduction framework.
Psychological Bulletin, 134, 207–222.
Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics, 69,
99–118.
Simon, H. A. (1956). Rational choice and the structure of the environment. Psychological Review, 63,
129–138.
Stanovich, K. E. (1999). Who is rational? Studies of individual differences in reasoning. Mahwah,
NJ: Erlbaum.
Stanovich, K. E. (2004). The robot's rebellion: Finding meaning in the age of Darwin. Chicago: University of
Chicago Press.
Stanovich, K. E. (2010). Decision making and rationality in the modern world. New York: Oxford
University Press.
Stanovich, K. E. (2011). Rationality and the reflective mind. New York: Oxford University Press.
Stanovich, K. E. (2018). Miserliness in human cognition: The interaction of detection, override, and
mindware. Thinking & Reasoning, 24, 423–444.
Stanovich, K. E., & Toplak, M. E. (2012). Defining features versus incidental correlates of Type 1 and
Type 2 processing. Mind & Society, 11, 3–13.
Stanovich, K. E., & West, R. F. (1998). Individual differences in rational thought. Journal of Experimental
Psychology: General, 127, 161–188.
Stanovich, K. E., & West, R. F. (1999). Discrepancies between normative and descriptive models of deci-
sion making and the understanding/acceptance principle. Cognitive Psychology, 38, 349–385.
Stanovich, K. E., & West, R. F. (2000). Individual differences in reasoning: Implications for the rationality
debate? Behavioral and Brain Sciences, 23, 645–726.
Stanovich, K. E., West, R. F., & Toplak, M. E. (2016). The rationality quotient: Toward a test of rational
thinking. Cambridge, MA: MIT Press.
Stein, E. (1996). Without good reason: The rationality debate in philosophy and cognitive science. Oxford: Oxford
University Press.
Stich, S. P. (1990). The fragmentation of reason. Cambridge, MA: MIT Press.
Taylor, S. E. (1981). The interface of cognitive and social psychology. In J. H. Harvey (Ed.), Cognition,
social behavior, and the environment (pp. 189–211). Hillsdale, NJ: Erlbaum.
Tetlock, P. E., & Mellers, B. A. (2002). The great rationality debate. Psychological Science, 13, 94–99.

205
206

Keith E. Stanovich

Thompson, V. A. (2009). Dual-process theories: A metacognitive perspective. In J. Evans & K. Frankish

(Eds.), In two minds: Dual processes and beyond (pp. 171–195). Oxford: Oxford University Press.
Todd, P. M., & Gigerenzer, G. (2000). Precis of simple heuristics that make us smart. Behavioral and Brain
Sciences, 23, 727–780.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185,
1124–1131.
Williams, G. C. (1988). Huxley’s Evolution and Ethics in sociobiological perspective. Zygon, 23, 383–407.

206
207

12
BOUNDED RATIONALITY
AND DUAL SYSTEMS
Samuel C. Bellini-Leite and Keith Frankish

Introduction
The notion of bounded rationality was initially developed by Herbert Simon as a cor-
rective to the ideal models of rationality used in economics, psychology, and philosophy
(Simon, 1957). Bounded rationality moderates the requirements on rational agents to reflect
the limitations of human cognitive capacity and the conditions of the task environment
(Simon, 1955, 1989). In a similar spirit, Christopher Cherniak has argued that traditional
conceptions of rationality have no relevance for cognitive science. To adopt ideal general
rationality conditions, Cherniak claims, is to deny or ignore our ‘finitary predicament’ –
the basic fact that human reasoning is limited by constraints of time and cognitive capacity
(Cherniak, 1986).
The history of the cognitive psychology of reasoning is to a large extent the story of accu-
mulating evidence for human failure to meet standards of ideal rationality. Peter Wason started
the modern study of reasoning with tasks, such as the Thog problem and the Wason selection
task, which challenged the notion that human cognitive development culminates in a formal-
logical stage (e.g., Wason, 1966). In the study of judgement and decision making, Daniel
Kahneman and collaborators showed that humans routinely violate the axioms of probability
theory and rational choice theory (Kahneman et al., 1982; Tversky, 1969). These pioneering
advances led to the identification of various inherent biases in human cognition, and it is
now common for researchers on reasoning and decision-making to adopt limited standards of
rationality, though there is continued dispute over the precise nature of the limitations dictated
(Gigerenzer, 1996; Kahneman, 2003).
A popular way to explain cognitive biases is to appeal to some form of dual-process theory
(DP theory or DPT), developed by cognitive and social psychologists since the late 1970s (for
reviews, see Frankish and Evans, 2009; Evans and Stanovich, 2013). Such theories hold that
human reasoning and decision making are supported by two different processing systems: System
1, which is fast, autonomous, effortless, and nonconscious, and System 2, which is slow, con-
trolled, effortful, and conscious.1 It is common to ascribe biases to System 1 processing (which
is seen as geared to norms of evolutionary rationality), and normative responses to System 2
processing. Individual differences in the activation and use of System 2 are often cited to explain

207
208

Samuel C. Bellini-Leite and Keith Frankish

why some individuals produce the modal, biased response in reasoning tasks, whereas others
respond in line with standards of normative rationality (Stanovich, 1999).
But is the DP approach compatible with the tenets of bounded rationality? Does DPT really
abandon the notion of ideal rationality, or does it merely treat it as a feature of System 2 rather
than of cognition as a whole? Certainly, some writers in the bounded rationality tradition are
suspicious of DPT. They deny that there is any fundamental division of mental systems and
account for ‘higher cognitive judgements’ by positing a range of heuristics, from which a selec-
tion is made as needed (Gigerenzer and Regier, 1996; Gigerenzer and Selten, 2001; Kruglanski
and Gigerenzer, 2011). In part at least, this negative attitude towards DPT may reflect a view of
System 2 as some sort of ideal central system. For example, Gigerenzer writes:

Heuristics are sometimes subsumed into a ‘System 1’ that is supposedly responsible

for associations and making errors and is contrasted with a ‘System 2’ that embodies
the laws of logic and probability, again without specifying models of the processes in
either system.
2008, p. 21

Now, one thing that is meant by saying that System 2 is normatively rational is that by engaging
in effortful, conscious, System 2 thinking we can apply learned rules of inference, follow task
instructions, and correct biased intuitive responses.2 This has always been a central claim of DPT.
It does not follow, however, that System 2 processes must be normatively rational in this sense. For
we can also go wrong in our effortful, conscious thinking – applying unsound rules of inference,
misunderstanding task instructions, and overriding intuitive responses that were in fact normatively
correct. We may also consciously employ learned heuristics and rules of thumb, which give us quick
and dirty solutions that may be inaccurate or biased. What System 2 gives us is a new level of control,
which can be used to arrive at normatively correct responses but can also be misused to produce
errors. It is not definitional of System 2 that it follows normatively correct principles. This point was
perhaps not sufficiently stressed by early DP theorists, but it is now clearly acknowledged in the DP
literature (see, e.g., Evans and Stanovich, 2013). To this extent, DPT contains no commitment to
ideal rationality and there is no conflict with bounded rationality.
There is another worry about DPT, however, which concerns the role of System 2 in abductive
inference and belief fixation. The worry has its roots in an earlier form of DPT proposed by
Jerry Fodor, which divides the mind into encapsulated input systems and unencapsulated cen-
tral systems (Fodor, 1983, 2001). This early form of DPT does involve a commitment to
unbounded central processing, which it represents as sensitive to global assessments of context
and relevance, and Fodor himself drew a bleak moral for cognitive science. Since later forms of
DPT have been heavily influenced by Fodor, it is appropriate to ask whether this commitment
carries over to them. If it does, then DPT not only conflicts with bounded rationality but also
threatens to render part of the mind inaccessible to cognitive science.
We shall set out this worry in some detail and then go on to outline a reply to it. We
shall show that there are various ways in which System 2 processing is limited and that con-
textual processing can be done by System 1. We shall conclude that DPT need not include a
commitment to ideal rationality and can incorporate the insights of bounded rationality.

Fodor’s DPT and the limits of cognitive science

The mental division proposed by Fodor is between input systems, which interpret perceptual
stimuli, and central systems, which are involved in belief fixation and general problem-solving

208
209

Bounded rationality and dual systems

(Fodor, 1983). Fodor argues that input systems are modular, and he develops a detailed account
of the features of modular systems. One key property of modules is that they are domain-specific.
They are dedicated to processing stimuli of a specific type, such as colours, faces, voices, or
uttered sentences, and each is structured to deal with its own specific domain, perhaps incorp-
orating information about it. A second core property of modular systems is that they are infor-
mationally encapsulated – insensitive to information stored elsewhere in the cognitive system.
Thus, a module for colour perception would not draw on general knowledge about the colours
of particular things. Fodor does concede that contextual information can bias perception, but
he argues that this is because perception is not limited to input analysis, so the bias can be
applied later by central systems (Fodor, 1983, p. 73). Similarly, he argues that priming effects
in language (where experience of one word facilitates the recognition of another, semantically
related one) may be due to superficial associations between lexical items inside the language
module, rather than to the top-down influence of background knowledge about the world.
Informational encapsulation of this kind is the core feature of modules, as Fodor conceives of
them, and he suggests that their other features can be explained as resulting from it (Fodor,
1983, pp. 79–82).
Central systems, in Fodor’s view, have contrasting features to input systems. They are domain
neutral (not tailored to any specific type of task), and they are unencapsulated (and therefore
nonmodular). The latter feature, Fodor argues, is a consequence of the fact that belief fixation
typically involves abductive inference – finding the best explanation of the information available –
and such inference is a global process, which is isotropic and Quineian (Fodor, 1983, 2001). To
say that it is isotropic is to say that it is open-ended: any item of knowledge could in principle
be relevant to the confirmation of any belief (knowledge about astronomy could be relevant
to problems in subatomic physics; knowledge of economics could be relevant to evolutionary
theory, and so on). To say that abductive inference is Quineian is to say that it is a holistic pro-
cess: the degree of confirmation a belief receives depends on considerations such as simplicity
and conservatism, which are determined globally, by the belief ’s relations to the rest of the
belief system.
Fodor argues that because central processes have this isotropic and Quineian character,
cognitive science cannot get any explanatory purchase on them. The argument (developed
at length in Fodor, 2001) is at heart simple. Fodor argues that the only serious approach
in contemporary cognitive science is the computational theory of mind, which identifies
mental processes with formal operations upon mental representations. But such operations
are sensitive only to local properties of representations, not to global properties of the
belief system. Hence, the more global a process is, the less contemporary cognitive science
understands it. Fodor calls this ‘Fodor’s First Law of the Nonexistence of Cognitive Science’
(1983, p. 107).
This problem for computationalism, Fodor notes, manifests itself in the field of artificial
intelligence as the notorious frame problem (McCarthy and Hayes, 1969; Pylyshyn, 1987). In
traditional computational AI (‘Good-Old-Fashioned AI’ or GOFAI), a robot guides its behav-
iour by reference to an internal model of the world, and the programmer must provide it with a
procedure for updating this model each time it acts, determining what will and will not change
as a result of each action. The problem is to find a tractable way of doing this, since, given the
right context, any action could change anything. What will change depends on the background
state of the world, and working out what revisions to make thus involves holistic, abductive
inference. Fodor concludes that cognitive science can deal only with modular, information-
ally encapsulated systems, the rest being inexplicable by contemporary cognitive science and
resistant to artificial modelling.

209
210

Samuel C. Bellini-Leite and Keith Frankish

Now, if we look at Simon’s description of the ideally rational ‘economic man’ posited by
traditional economic theory, we find a similarity to Fodor’s conception of central processing.
Like Fodorian central systems, the economic man is able to make rational choices because he has

knowledge of the relevant aspects of his environment which, if not absolutely com-
plete, is at least impressively clear and voluminous ... a well-organized and stable
system of preferences, and a skill in computation that enables him to calculate, for the
alternative courses of actions that are available to him, which will permit him to reach
the highest attainable point on his preference scale.
Simon, 1955, p. 99

By analogy, Fodorian central processing can be thought of as a ‘cognitive’ economic man – a

powerful homunculus with a voluminous knowledge base, a coherent system of preferences,
and the computational skill to determine the rational response in the light of both. Simon notes
that this idealized conception provides a poor foundation for economic theory, and we think
that Fodor’s conception of central systems provides a poor foundation for cognitive theory.
If DPT is committed to a similarly idealized conception of System 2, then DP theorists have
reason to be worried.

From Fodor to contemporary DPT

How does Fodor’s division between input systems and central systems relate to the two systems
posited by contemporary DPT? There are similarities. In DPT, System 1 includes input systems
of the sort Fodor discusses, while System 2 processes are all central ones. However, the corres-
pondence is not straightforward. System 1 also includes many processes of reasoning, problem
solving, and belief fixation, which are all central ones in Fodor’s scheme. In fact, in DPT, the
two sorts of processing are distinguished not by their functions, but by their computational
characteristics (speed, effort, seriality, consciousness, and so on).
Another similarity with Fodor’s account is that many DP theorists regard System 1 processes
as modular. Stanovich, for example, conceives of System 1 as a set of adaptative problem-solving
sub-systems, mostly modular in character (he refers to it as The Autonomous Set of Systems,
or TASS) (Stanovich, 2004). However, the notion of modularity in play here is weaker than
Fodor’s. Stanovich does not treat encapsulation or domain-specificity as defining features of
TASS. He notes that encapsulation is a problematic feature, which may be a matter of degree,
and he holds that TASS includes domain-general processes of learning and emotion-mediated
behavioural regulation. In addition, he does not assume that TASS systems are innately speci-
fied with fixed neural structures, and he allows that modules can be developed through practice,
as a controlled process becomes autonomous (Stanovich, 2004, pp. 38–40).
For Stanovich, the defining feature of TASS processes is that they are autonomous or manda-
tory – that is, they are not subject to higher-level control and, once triggered by their propri-
etary stimuli, they run automatically to completion. This is why they are fast and do not load on
central resources, such as working memory. This does not mean that TASS/System 1 processes
cannot be overridden, since higher-level processes may intervene to prevent their outputs from
influencing behaviour.3
Despite these differences with Fodor, Stanovich’s model (and modern DPT generally) still
retain a sharp division between autonomous, System 1 processes and central, System 2 ones.
Stanovich suggests that System 2 (he also refers to it as the analytic system) can be roughly
characterized as exhibiting ‘serial processing, central executive control, conscious awareness,

210
211

Bounded rationality and dual systems

capacity-demanding operations, and domain generality in the information recruited to aid

computation’ (Stanovich, 2004, p. 45).
Thus, though DP theorists do not treat System 2 as comprising all central processes, they
agree with Fodor that there are some genuinely nonmodular, domain-general central processes,
namely, the conscious, controlled ones. The challenge, then, for DP theories is to explain how
System 2 can be flexible and domain-general while still being a limited, bounded system. We
shall not attempt to provide an answer here. (For a detailed proposal in the spirit of DPT, see
Carruthers, 2006.) Instead, we shall review some evidence relevant to a solution, showing
that the division between autonomous, nonconscious, System 1 processes and controlled,
conscious, System 2 processes should not be understood in Fodorian fashion as a contrast
between rigidly encapsulated systems and unbounded, ideally rational ones. We shall highlight
evidence that conscious, controlled cognition is computationally limited, and thus certainly
not ideally rational, and then indicate how contextual processing may be done by autono-
mous, nonconscious processes. This points to an explanation of how the two systems together
can approximate to flexible, general-purpose cognition by balancing the computational load
between them.

The limitations of System 2

Cognitive psychology shows decisively that our conscious cognitive capacities are limited.
This is evidenced by work on short-term and working memory (e.g., Baddeley, 1992; Miller,
1956), attention (e.g., Simon and Chabris, 1999), competing tasks with divided attention (e.g.,
Schneider and Shiffrin, 1977), and cognitive theories of consciousness (e.g., Baars, 1988).
When anything remotely like conscious central systems are involved, what we observe is a
competition for limited resources. Although we can perform several automatic tasks simultan-
eously (for example, singing while driving), when conscious, controlled cognition is required,
we have a hard time multitasking. We can see, feel, hear, and move at the same time, but we
cannot attend to a lecture, do maths problems, and talk about politics at the same time.
These limitations show up in various ways. Selective attention experiments reveal that people
can be conscious of only a few items in a rich perceptual scene. Subjects asked to monitor the
flow of certain information in a perceptual scene, such as details in a basketball game, become
unconscious of other, unrelated information in the scene.4 Other stimuli are certainly being
processed, since we react to meaningful stimuli outside a monitored flow of events, such as
someone calling our name. But if we switch focus to the new stimulus, we miss aspects of the
scene being monitored, revealing competition for limited capacity.
In dual-task experiments, subjects are asked to complete two tasks at the same time. Initially,
task interference reduces performance, but if one of the tasks becomes predictable, performance
increases. This is attributed to a process of automatization – a transfer from controlled, System 2
processing to autonomous, System 1 processing, with the consequent freeing up of the limited
resources available for the former (Schneider and Shiffrin, 1977).
The limits of conscious processing are closely bound up with those of working memory (see
Baars and Franklin, 2003). (Indeed, some DP theorists define System 2 processes as the ones
that load on working memory; Evans and Stanovich, 2013.) Working memory is believed to
comprise three short-term storage capacities, or loops, for visuo-spatial, phonological, and epi-
sodic information, each of which is limited in the number of items it can retain and the time
for which it can retain them (e.g., Baddeley, 1992). For example, in the Corsi block-tapping
test, a visuo-spatial task that requires participants to track the order of cubes the experimenter
points to, people can only keep up with a sequence of about five. Likewise, the phonological

211
212

Samuel C. Bellini-Leite and Keith Frankish

loop, which can be used to mentally rehearse words and numbers for a few minutes, has a limit
of about seven items (Miller, 1956).
Could limited-capacity systems like these be isotropic and Quineian? We shall return to this
later, but it is obvious at the outset that they could not be Quineian, since that would require
that the confirmation of a single belief be influenced by many others (in principle, all of them),
and certainly by many more than the seven-item limit of phonological working memory. As
Carruthers notes: ‘it has traditionally been assumed by philosophers that any candidate new
belief should be checked for consistency with existing beliefs before being accepted. But in fact
consistency-checking is demonstrably intractable, if attempted on an exhaustive basis’ (2006,
p. 52). Of course, a rational agent must have some sort of coherence in their belief set, but they
cannot be conscious of all the important confirmation relations in their whole web of beliefs.
We suspect that the problem here stems from the fact that Fodor explicitly models the
psychological processes of abductive inference and belief fixation on the processes of theory
construction and confirmation in science (Fodor, 1983, pp. 104–105). There are certainly
parallels between them, but there are also big differences – not least the fact that science is a
social enterprise, which is carried out over an extended period of time and with large amounts
of external scaffolding in the form of records and notes. At any rate, while the truth-value of a
belief or the consistency of a preference may depend on its links to many other mental states,
no more than a few of these links can be checked consciously. This is one thing that condemns
us to bounded rationality.
If the conscious mind is so limited, why should we be tempted to assign ideal rationality to
it? Perhaps the answer is that we identify our conscious processes with us – with the self that is
supposed to be responsive to norms of rationality. From a phenomenological perspective, this
is tempting, and it may lead us to think of System 2 as a powerful homunculus – an executive
system or ‘central meaner’ (Dennett, 1991), which accesses all our knowledge and is the locus
for rational decision making. This would be a serious mistake, however. Our conscious, con-
trolled processes are part of our minds, just as our automatic, nonconscious ones are, and they
must depend heavily on nonconscious, autonomous processing.5
Where does this leave us? Fodor supposed that central processes had to be highly sensi-
tive to context and background information precisely because autonomous, peripheral systems
were not (Fodor, 2001). Within the context of DPT, we can reverse this line of reasoning.
Since System 2 processes are in fact severely limited, it follows that System 1 must do the work
of supplying relevant, contextualized information for conscious central systems to work with
(Evans, 2009). As we shall see in the next section, this conclusion accords well with new models
of brain functioning, which locate contextual processing in nonconscious, probabilistic systems
that are not limited by the capacity of working memory or other central resources.

The flexibility of System 1

If System 1 processing is flexible and context-sensitive, then System 2 can be freed from the
burden of ideal rationality. It need not work on complex webs of content but merely on those
items that have been pre-selected for relevance by System 1. And there are, in fact, good reasons
for thinking that this is indeed the case.
First, as we have already noted, DP theories generally employ a weaker notion of modularity
than Fodor’s, placing less stress on features such as encapsulation. Carruthers, for example, notes
that although modules must be frugal, in the sense that they draw on only a limited amount of
information in their processing, it does not follow that they each have access only to a specific
subset of information. A module might have access in principle to all the information in the

212
213

Bounded rationality and dual systems

system but draw on only a tiny fraction of it on each occasion, perhaps using simple search
heuristics to make the selection (Carruthers, 2006, pp. 58–59).
Second, it is now widely accepted that modules needn’t be persisting, innately specified
structures. Rather, they may be ‘soft modules’ – temporary constructions, assembled from pre-
existing components to deal with specific environmental problems. This point is stressed by
Michael Anderson (Anderson, 2014). Although a critic of rigid modularity, Anderson does not
deny that the cortex contains functionally differentiated regions, which are biased towards different
response profiles (Anderson, 2014, p. 52). However, he argues that these regions can interact with
each other and form coalitions to perform other tasks. When existing functionally differentiated
regions are unable to solve a task as they stand, coalitions between them can be formed ‘on the
fly’ to attempt a local resolution. Anderson dubs such constructions, composed of coalitions of
different networks, ‘transiently assembled local neural subsystems’ or TALoNS. These coalitions
generate new possibilities of interaction but are still constrained by the available abilities of the
participating regions. Being formed for specific tasks, these coalitions can generate responses in a
rapid, effortless manner, without engaging resource-hungry, conscious control processes.
Third, recent models of brain functioning undermine the rigid Fodorian distinction
between strictly encapsulated systems and free unencapsulated systems, even within perceptual
processing itself. According to currently popular predictive processing (PP) theories, the brain’s
goal is to predict proximal stimuli rather than simply to process past input (e.g., Friston, 2005;
Clark, 2013, 2016). These theories see the brain as having a hierarchical structure, with lower
levels being close to perceptual input mechanisms and higher levels receiving information from
diverse multimodal regions. Signals flow both up and down the hierarchy, with top-down
signals predicting lower-level activity and bottom-up signals flagging errors in those predictions.
Predictions are based on probabilistic generative models distributed through the hierarchy. These
models monitor statistical patterns in the layer below, generating nonconscious hypotheses in
an attempt to accommodate incoming data.6
Within this architecture, the influence of different neural regions is modulated to reflect
their success in prediction, through a mechanism known as precision weighting. Neural regions
that do badly must accept input from other regions, whereas ones that constantly do well are
granted more autonomy, allowing a form of encapsulation. Thus, the informational closure
of a region reflects its success in prediction, generating a soft module, like one of Anderson’s
TALoNS. As Clark puts it: ‘Distinctive, objectively identifiable, local processing organizations
… emerge and operate within a larger, more integrative, framework in which functionally
differentiated populations and sub-populations are engaged and nuanced in different ways so
as to serve different tasks …’ (2016, p. 150). Coalition formation and shifts in control are
determined contextually, by success in accommodating new stimuli.
If this approach is on the right track, then Fodor is quite wrong to characterize input systems
as strongly encapsulated. For, as Clark emphasizes, context sensitivity is pervasive in the PP
account of perception, with top-down predictions guiding how stimuli are interpreted and
signal distinguished from noise. For example, if someone says to you, ‘Oedipus married his
own nother’, the currently dominant top-down prediction might constrain interpretation so
that the /n/ sound is nonconsciously interpreted as /m/, and the word is consciously heard as
‘mother’ (Clark, 2013).
Such contextually sensitive processing can be done rapidly since stimuli that can be well
accommodated by current predictions are ignored. Instead of letting all worldly informa-
tion move up to higher layers, lower layers pass on only stimuli that generate high prediction
error. This selectional effect is further enhanced by precision weighting. As Clark puts it, ‘very

213
214

Samuel C. Bellini-Leite and Keith Frankish

low-precision prediction errors will have little or no influence upon ongoing processing and
will fail to recruit or nuance higher level representations’ (2016, p. 148). By these means, rele-
vance is selected for right from the start, facilitating rapid responding.
A central feature of the PP framework is that it draws no sharp boundary between per-
ception and cognition (or, indeed, between perception, cognition, and action). Judgement-
like top-down predictions are continually modifying and being modified by perception-like
bottom-up error signals. Thus, what DPT theorists would think of System 1 judgements may
be more akin to perceptual processes than to explicit System 2 judgements, sharing in the
context-sensitivity of perception. If one’s previously calm companion starts to show signs of
anger, bottom-up error signals will cause one’s generative models to adapt, yielding a new pre-
diction with a judgement-like form (‘He is angry and may become violent’).
Although few DP theorists have explicitly adopted PP approaches, some have stressed the
perception-like character of System 1 judgements. For example, Kahneman and Frederick
observe that the perception/judgement boundary ‘is fuzzy and permeable: the perception of a
stranger as menacing is inseparable from a prediction of future harm’ (Kahneman and Frederick,
2002, p. 50). They note, moreover, that the heuristics and biases research programme (which
was a major inspiration for dual-process theories) was from the start ‘guided by the idea that
intuitive judgments occupy a position ... between the automatic parallel operations of percep-
tion and the controlled serial operations of reasoning’ (Kahneman and Frederick, 2002, p. 50).
There are good reasons, then, for thinking that the dual-process conception of System 1 and
PP make a natural partnership.
The moral of this is that the representations that are made available for explicit System 2
processing will have already gone through considerable prior processing and filtering. Only
high error (unexpected and thus informative) stimuli reach higher levels of processing and
become available for explicit System 2 reasoning. Thus, on the view we have outlined, System
2 reasoning would be neither isotropic nor Quineian. It would not be isotropic because it
would not be true that any element could influence it (it receives only a small subset of the
data, selected as relevant) and it would not be Quineian since it would not have access to all
the information required to check for holistic consistency. In fact, it would be limited in just
the ways the evidence suggests it in fact is. Thus, it could meet only bounded standards of
rationality.

Conclusion
It is tempting to think of conscious thought as open-ended and unconstrained, and of
nonconscious processes as inflexible and encapsulated. Cognitive psychology and computa-
tional neuroscience show that this is wrong. Conscious cognition is in fact severely limited
in capacity, while nonconscious processes are tuned to the heavy demands of contextual and
relevance processing. It is thus vital that modern dual-process theories do not carry over
the conception of rationality implicit in Fodor’s precursor theory. In thinking about the
functions and capacities of the posited dual systems, the perspective of bounded rationality is
essential. System 2, if it exists, is not an idealized reasoner, a cognitive analogue of Simon’s
‘economic man’.

Acknowledgements
The authors thank Riccardo Viale for his detailed and helpful comments on an earlier draft of
this chapter.

214
215

Bounded rationality and dual systems

Notes
1 More recently, there has been a tendency to speak of two types of processing rather than two systems
(see e.g., Evans and Stanovich, 2013; Bellini-Leite, 2018). For present purposes, however, we shall use
‘systems’ terminology, since it better expresses the concerns we wish to address about architectural inter-
pretations of dual-process theory.
2 Of course, the fact that System 2 can be used to override intuitive responses does not mean that it will
be. System 2 thinking is effortful and demanding of cognitive resources, and people engage in it only
when they have sufficient time, motivation, and cognitive capacity. And even when they do engage in
it, they may still fail to override an intuitive System 1 response. For discussion, see Viale (2018).
3 Another similarity between Stanovich’s DPT and Fodor’s architecture is that Stanovich sees System 2
representations as decoupled from online perceptual processing, so that they can be used for offline,
hypothetical thinking (e.g., Stanovich and Toplak, 2012). However, this decoupling need not bring
with it the problematic features of Fodorian central processing. Decoupled representations are detached
from the world, but this does not mean that they are integrated with the rest of the system’s knowledge
in reasoning processes that are Quineian and isotropic. Hypothetical thinking may involve only a limited
number of representations held in working memory.
4 The most effective illustration is Simons and Chabris’ (1999) experiment, in which participants who are
required to the count passes in a video of a basketball game completely fail to notice when a person in
a gorilla costume appears.
5 In fact, those who see System 2 as a personal-level system – in the sense that its processes involve inten-
tional actions – are also those who argue against a ‘central meaner’ and for an illusionist view of con-
sciousness (Dennett, 1991; Frankish 2015, 2016).
6 The claim that System 1 is supported by probabilistic calculations does not entail that the system’s overt
responses will respect probabilistic principles. The calculations govern the system’s internal workings,
and its overt responses could conform to different principles, such as those of classical logic. This is not
to say that we cannot plausibly infer internal computational principles from behaviour, but we must
consider a wider range of evidence, including the speed, accuracy, and range of overt responses.

References
Anderson, M. (2014). After phrenology: Neural reuse and the interactive brain. Cambridge, MA: MIT Press.
Baddeley, A. (1992). Working memory. Science, 255(5044): 556–559.
Baars, B. J. (1988). A cognitive theory of consciousness. Cambridge: Cambridge University Press.
Baars, B., and Franklin, S. (2003). How conscious experience and working memory interact. Trends in
Cognitive Science, 7(4): 166–172.
Bellini-Leite, S. (2018). Dual process theory: Systems, types, minds, modes, kinds or metaphors? A critical
review. Review of Philosophy and Psychology, 9(2): 213–225.
Carruthers, P. (2006). The architecture of the mind. Oxford: Oxford University Press.
Cherniak, C. (1986). Minimal rationality. Cambridge, MA: MIT Press.
Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science.
Behavioral and Brain Sciences, 36(3): 181–204.
Clark, A. (2016). Surfing uncertainty: Prediction, action, and the embodied mind. Oxford: Oxford University Press.
Dennett, D. (1991). Consciousness explained. New York: Little, Brown.
Evans, J. (2009). How many dual-process theories do we need? One, two, or many? In J. Evans and K.
Frankish (Eds.), In two minds: Dual processes and beyond (pp. 33–54). Oxford: Oxford University Press.
Evans, J., and Stanovich, K. (2013). Dual-process theories of higher cognition: Advancing the debate.
Perspectives on Psychological Science, 8(3): 223–241.
Fodor, J. (1983). The modularity of mind. Cambridge, MA: MIT Press.
Fodor, J. (2001). The mind doesn’t work that way. Cambridge, MA: MIT Press.
Frankish, K. (2015). Dennett’s dual-process theory of reasoning. In C. Muñoz-Suárez and F. De Brigard
(Eds.), Content and consciousness revisited (pp. 73–92). Cham: Springer.
Frankish, K. (2016). Illusionism as a theory of consciousness. Journal of Consciousness Studies, 23(11–12): 11–39.
Frankish, K., and Evans, J. (2009). The duality of mind: An historical perspective. In J. Evans and K.
Frankish (Eds.), In two minds: Dual processes and beyond (pp. 1–29). Oxford: Oxford University Press.
Friston, K. (2005). A theory of cortical responses. Philosophical Transactions of the Royal Society of London
B: Biological Sciences, 360(1456): 815–836.

215
216

Samuel C. Bellini-Leite and Keith Frankish

Gigerenzer, G. (1996). On narrow norms and vague heuristics: A reply to Kahneman and Tversky.
Psychological Review, 103(3): 592–596.
Gigerenzer, G. (2008). Why heuristics work. Perspectives on Psychological Science, 3(1): 20–29.
Gigerenzer, G., and Regier, T. (1996). How do we tell an association from a rule? Comment on Sloman
(1996). Psychological Bulletin, 119(1): 23–26.
Gigerenzer, G., and Selten, R. (Eds.) (2001). Bounded rationality: The adaptive toolbox. Cambridge,
MA: MIT Press.
Kahneman, D. (2003). Maps of bounded rationality: Psychology for behavioral economics. The American
Economic Review, 93(5): 1449–1475.
Kahneman, D., and Frederick, S. (2002). Representativeness revisited: Attribute substitution in intuitive
judgment. In T. Gilovich, D. Griffin, and D. Kahneman (Eds.), Heuristics and biases: The psychology of
intuitive judgment (pp. 49–81). Cambridge: Cambridge University Press.
Kahneman, D., Slovic, P., and Tversky, A. (Eds.) (1982). Judgment under uncertainty: Heuristics and biases.
Cambridge: Cambridge University Press.
Kruglanski, A. W., and Gigerenzer, G. (2011). Intuitive and deliberative judgments are based on common
principles. Psychological Review, 118(1): 97–109.
Miller, G. (1956). The magical number seven, plus or minus two: Some limits on our capacity for pro-
cessing information. Psychological Review, 63(2): 81–97.
McCarthy, J., and Hayes P. (1969). Some philosophical problems from the standpoint of artificial intelli-
gence. In B. Meltzer and D. Michie (Eds.), Machine intelligence 4 (pp. 463–502). Edinburgh: Edinburgh
University Press.
Pylyshyn, Z. (Ed.) (1987). The robot’s dilemma: The frame problem in artificial intelligence. Norwood, NJ: Ablex.
Schneider, W., and Shiffrin, R. (1977). Controlled and automatic human information processing
I: Detection, search and attention. Psychological Review, 84(1): 1–66.
Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics,
69(1): 99–118.
Simon, H. A. (1957). Models of man: Social and rational – Mathematical essays on rational human behavior in a
social setting. New York: Wiley.
Simon, H. A. (1989). Cognitive architectures and rational analysis: Comments. Paper presented at 21st
Annual Symposium on Cognition, Carnegie-Mellon University.
Simons, D., and Chabris, C. (1999). Gorillas in our midst: Sustained inattentional blindness for dynamic
events. Perception, 28(9): 1059–1074.
Stanovich, K. (1999). Who is rational? Studies of individual differences in reasoning. Mahwah, NJ: Lawrence
Erlbaum.
Stanovich, K. (2004). The robot’s rebellion: Finding meaning in the age of Darwin. Chicago, IL: University of
Chicago Press.
Stanovich, K. E., and Toplak, M. E. (2012). Defining features versus incidental correlates of Type 1 and
Type 2 processing. Mind & Society, 11(1): 3–13.
Tverksy, A. (1969). Intransitivity of preferences. Psychological Review, 76(1): 31–48.
Viale, R. (2018). The normative and descriptive weaknesses of behavioral economics-informed
nudge: Depowered paternalism and unjustified libertarianism. Mind & Society, 17(1–2): 53–69.
Wason, P. (1966). Reasoning, in B. Foss (Ed.), New horizons in psychology (pp. 135–151). Harmondsworth:
Penguin.

216
217

13
MODELS AND RATIONAL
DEDUCTIONS
Phil N. Johnson-Laird

Introduction
In daily life, we reason to try to reach conclusions that are true. But, since our premises are
often uncertain, a sensible goal is at least to make what logicians refer to as valid deductions.
These are inferences whose conclusions hold in every case in which their premises hold (see
Jeffrey, 1981, p. 1), and so if the premises are true their conclusions are true too, e.g.:

If I’m in Britain then I drive on the left.

I’m in Britain.
Therefore, I drive on the left.

Whatever rational might mean – and it is open to many interpretations, rational deductions
should at least be valid. They have no counterexamples, which are cases in which their premises
are true but their conclusions false.
The inference above is so simple that you might think that all valid deductions in daily
life can follow the formal rules of inference for a logical calculus. For instance, one such rule,
appropriate for the inference above, is:

If A then B; A; Therefore, B

where A and B are variables that range over propositions. This rule is from the calculus that concerns
the analogs in logic of if, or, and, and not (Jeffrey, 1981). Everyday reasoning would be straightforward
if it could follow rules that were complete in that they captured all and only valid deductions, that
yielded a decision about validity or invalidity in a finite number of steps, and that did so in a tractable
way, that is, depending on only reasonable amounts of time and memory. Alas, some inferences in
life violate these desirable properties: their logic is incomplete, and they lack a decision procedure or
a tractable one. An instructive example (from Johnson-Laird, 1983, p. 140) is:

More than half the musicians were classically trained.

More than half the musicians were in rock groups.
So, at least one of the musicians was both classically trained and in a rock group.

217
218

Phil N. Johnson-Laird

The concept of “more than half ” calls for a logic in which variables range over sets, and this logic
is not complete, and has no decision procedure (Jeffrey, 1981, Chapter 7). Likewise, reasoning
about the domain of two-dimensional spatial relations, such as “The cup is on the left of the
saucer”, to which we return later, is intractable for all but the simplest deductions (Johnson-Laird,
1983, p. 409; Ragni, 2003). So, even before we consider the limitations of human reasoning,
deduction is bounded for any finite device (cf. Simon 2000). And yet it is to a finite device, the
human brain, that we owe the proofs of incompleteness, undecidability, and intractability.
Validity is a notion that can be applied to any system of inference provided that we know
the conditions in which its assertions are true. What is invalid in one logic of possibilities, for
instance, is valid in another logic of possibilities, because the truth conditions for “possible”
differ between the two logics (e.g., Girle, 2009). You might therefore think that, given the truth
conditions for everyday assertions, rational deduction is nothing more than valid deduction.
Whatever you understand rationality to mean, this idea is wrong. The reason why is illustrated
in these three deductions, which are each valid and each silly:

1. I’m in Britain.
Therefore, it is raining or it isn’t raining.
2. I’m in Britain, and if so then I drive on the left.
Therefore, if I’m in Britain, which I am, then I drive on the left.
3. I’m in Britain, and if so then I drive on the right.
Therefore, I’m in Britain or I drive on the right, or both.

In (1), the conclusion is bound to be true given the premises, because it is a tautology. But it is
silly, because its conclusion has no relation to the premises. A sensible deduction needs to depend
on the premises. In (2), the conclusion is bound to be true given the premises, and it does
depend on them. But it is silly, because it merely repeats their contents. In (3), the conclusion
is bound to be true given the premises. But it is silly, because it throws information away: the
premise implies the conjunction of the two propositions. A sensible deduction is in contrast:

I’m in Britain, and if so then I drive on the right.

Therefore, I drive on the right.

Sensible deductions are accordingly inferences that maintain the information in the prem-
ises (and are thereby valid), that are parsimonious, and that yield a conclusion that was not
explicit in the premises (see Johnson-Laird, 1983, p. 37; Johnson-Laird & Byrne, 1991, p. 22).
If no conclusion meets these requirements then people tend to say that nothing follows from
the premises – a judgement that stands in stark contrast to logic, in which infinitely many
conclusions follow from any premises whatsoever. As the three examples above illustrate, most
of them are silly. So, let us say: To make a rational deduction is at least to maintain the information in
the premises, to simplify, and to reach a new conclusion.
Can naïve individuals – those who are ignorant of logic or its cognate disciplines – make
rational deductions? Yes, of course. Without this ability, Aristotle and his intellectual descendants
would have been unable to develop logic. You might suppose, as some theorists argue, that
human beings would not have evolved or survived as a species had not at least some of them
been capable of some rational deductions. This way of framing the matter recognizes two robust
phenomena: people differ in deductive ability (Stanovich, 1999), and some rational deductions
pertinent to daily life defeat almost everyone. Perhaps that is why logic exists.

218
219

Models and rational deductions

How do individuals – naïve ones should henceforth be taken for granted – make rational
deductions? The question has been under investigation for over a century, but only in recent
years have cognitive scientists begun to answer it. This chapter outlines one such answer – the
theory of mental models. It rests on five principles corresponding to sections in the chapter.
Mental models are not uncontroversial, but the final section of the chapter shows that they
explain phenomena beyond other theories of human deduction.

Possibilities underlie reasoning

The theory of mental models postulates that all inferences – even inductions, which are outside
the scope of the present chapter – depend on envisaging what is possible, given premises such
as assertions, diagrams, or direct observations (Johnson-Laird, 2006; Johnson-Laird & Ragni,
2019). A possibility is a simple sort of uncertainty, but a conclusion that holds in all the possi-
bilities to which the premises refer must be valid. The major principle of human reasoning is
accordingly that inferences are good only if they have no counterexamples, that is, possibilities in
which the premises hold, but the conclusion does not.
Suppose you know about a particular electrical circuit:
If there was a short in the circuit then the fuse blew.
The salient possibility to which this assertion refers is:
Short in circuit fuse blew

But, other cases are possible too, namely:

Not a short fuse blew
Not a short fuse did not blow

The first of these latter two possibilities occurs when the circuit is overloaded, and the second
of them is the norm. It is hard to keep three distinct possibilities in mind, and so people tend to
focus on the first one, and to make a mental note that the other cases (in which the if-clause of the
conditional assertion does not hold) are possible. Of course, people don’t use words and phrases
to represent possibilities, but actual models of the world akin to those that the perceptual system
constructs.
The mental models of the premise above are represented in the following diagram:

Short in circuit fuse blew

...

The first model is the salient one in which the if-clause holds. The second model (shown as an
ellipsis) is an implicit one representing the other possibilities in which there wasn’t a short in
the circuit. Mental models underlie intuitive reasoning. You learn that there was a short in the
circuit. It picks out only the first model, from which it follows at once:

The fuse blew.

Possibilities and necessities can be defined in terms of one another: if an event is possible then
it is not necessarily not the case, and if an event is necessary, then it is not possibly not the case.
So, what evidence shows that people model possibilities rather than necessities? If possibilities

219
220

Phil N. Johnson-Laird

are fundamental, then reasoners should be more accurate in inferring what’s possible than in
inferring what’s necessary. Indeed, they are faster and more accurate (Bell & Johnson-Laird,
1998). A typical experiment concerned games of one-on-one basketball, in which only two
can play. So, how would you answer the question about this game:

If Alan is in the game then Betty is in the game.

If Cheryl is in the game then David is not in the game.
Can Betty be in the game?

The correct answer is “yes”, because both premises hold in case Alan and Betty are in the game.
But, now suppose that the question is instead about a necessity:

Must Betty be in the game?

You should grasp at once that this question is harder. In fact, if you deliberate about the matter, you
will discover that there are only three possible games: Alan plays Betty, Betty plays Cheryl, or Betty
plays David. So, Betty must be in any game. When the correct answer is “no”, the question about
whether a player must be in the game is easier to answer than the question about whether a player
can be in a game. A single model that is a counterexample suffices to justify the negative answer
to the first question, whereas all models of possibilities are needed to justify the negative answer to
the second question. These differences in difficulty between the two “yes” answers, and the switch
between the two “no” answers, show that models represent possibilities rather than necessities.

Deductions are in default of information to the contrary

An assertion, such as:

The fault is in the software or in the cable, or both

implies three possibilities:

It is possible that the fault is in the software.

It is possible that the fault is in the cable.
It is possible that the fault is in the software and in the cable.

People make these three deductions, and reject only the deduction that it is possible that the
fault is in neither the software nor the cable (Hinterecker, Knauff, & Johnson-Laird, 2016). Yet,
none of these inferences is valid in modal logics, which concern possibilities and necessities. To
see why, consider the case in which it is impossible for the fault to be in the software but true
that it is in the cable. The premise is true, but the first conclusion above is false. So, the infer-
ence is invalid. Analogous cases show that the other two inferences are also invalid. Why, then,
do individuals make these inferences? The answer is that they treat the disjunction as referring
to the set of possibilities in default of information to the contrary. Conversely, if they know that
it is possible that the fault is in the software, possible that it is in the cable, and possible that it is
in both, they can use the disjunction above to summarize their knowledge. One reason for the
assumption of defaults is their ubiquity in daily life.
Facts often reveal that a valid conclusion that you drew is false. It may be a surprise, but you
cope. You no longer believe your conclusion. Yet, in logic, no need exists for you to withdraw it.

220
221

Models and rational deductions

Logic means never having to be sorry about any valid conclusion that you’ve drawn. The reason is
that a self-contradiction, such as one between a conclusion and a fact, implies that any conclusion
whatsoever follows validly in logic. The jargon is that logic is “monotonic”, whereas everyday
reasoning is “non-monotonic”. The divergence is just one of many between logic and life.
When the facts contradict a valid deduction, you should change your mind, and one view of a
rational change is that it should be minimal. As William James (1907, p. 59) wrote, “[The new fact]
preserves the older stock of truths with a minimum of modification, stretching them just enough
to make them admit the novelty.” This parsimony seems sensible, and many cognitive scientists
advocate minimalism of this sort (e.g., Harman, 1986; Elio & Pelletier, 1997). It implies, of course,
that the rational step in dealing with a contradiction to a valid inference is to make a minimal
amendment of the premises. That’s not what happens in daily life. Consider this example:

If a person is bitten by a cobra then the person dies.

Viv was bitten by a cobra. But, Viv did not die.
What would you infer?

Minimalism predicts that you should make a minimal change to the premises, e.g.:

Not everyone dies if bitten by a cobra.

In fact, most individuals respond in a different way. They create an explanation that resolves the
inconsistency, e.g.:

Someone sucked the venom from the bite so it did not get into Viv’s bloodstream.

Hardly a minimal change. Yet, people tend to create such causal explanations, and to judge them
to be more probable than minimal amendments to the premises (Johnson-Laird, Girotto, &
Legrenzi, 2004). After they have created such explanations, they even find inconsistencies harder
to detect (Khemlani & Johnson-Laird, 2012). Just as individuals assume possibilities by default,
so they draw conclusions by default. They give them up in the light of evidence to the contrary.

Knowledge modulates the meanings of logical terms

In logic, the analogs of such words as if, or, and and, have constant meanings. Earlier we saw
that conditionals – assertions based on “if ” – usually refer to three distinct possibilities. But,
consider the conditional:

If God exists then atheism is wrong.

It ought to refer to the possibility in which God does not exist and atheism is wrong. But it
doesn’t. Knowledge of the meaning of “atheism”, as disbelief in God, blocks the construction
of the corresponding model. So, the preceding assertion is in reality a biconditional equivalent to:

If, and only if, God exists then atheism is wrong.

Our knowledge modulates our interpretations of words (e.g., Quelhas & Johnson-Laird, 2017).
Yet, we are unaware for the most part of these modulations. In one experiment (Juhos, Quelhas,
& Johnson-Laird, 2012), the participants drew their own conclusion from premises containing
a biconditional:

221
222

Phil N. Johnson-Laird

If the client makes an order then the goods are shipped. The goods are shipped.
What follows?

They tended to infer:

The client made an order.

They also drew their own conclusion from the similar premises:

If the client makes an order then the goods are shipped. The client makes the order.
What follows?

They tended to infer:

The goods are shipped.

They were unaware of a subtle difference between the two sorts of conclusion: their first con-
clusion was in the past tense, whereas their second conclusion was in the present tense (or in
the future tense in Portuguese, which was the participants’ native tongue). The model theory
predicts this difference. The participants use their knowledge to infer the sequence in which
the two events occur, i.e., the ordering the goods comes before their shipping. The conclusion
to their first inference refers to the ordering, which comes before the shipping. So, the conclu-
sion calls for the past tense. In contrast, the conclusion to their second inference refers to the
shipping, which comes after the ordering. So, the conclusion calls for the present tense, which
can refer to future events in both English and Portuguese, or to the Portuguese future tense. (As
its native speakers often don’t realize, English has no future tense.)
The implementation of modulation in a computer program revealed that some assertions
should be judged to be true a priori, i.e., without the need for evidence, and that other assertions
should be judged to be false a priori. An experiment corroborated the predictions (Quelhas,
Rasga, & Johnson-Laird, 2017). Individuals judged an assertion, such as:

If Mary has flu then she is ill

to be true. And they judged an assertion such as:

If Mary has flu then she is healthy

to be false. Such judgements run contrary to an influential view in philosophy that the diffe-
rence between a priori assertions and those contingent on evidence is “an unempirical dogma
of empiricism” (Quine, 1953, p. 23). Not any more.

Reasoning depends on intuitions or on deliberations, or both

Human reasoners are equipped with two systems for reasoning, which interact with one
another. Intuitions depend on mental models, which represent only what is true. As we saw
earlier, the assertion:

If there was a short in the circuit then the fuse blew

222
223

Models and rational deductions

has these mental models:

a short in circuit fuse blew

...

Suppose you learn:

The fuse didn’t blow.

It eliminates the first of the two mental models above, and so only the implicit model remains.
It has no explicit content, and so it seems that nothing follows from the two premises. Many
people have this intuition. If they think harder, however, their deliberations may lead them to
fully explicit models of the possibilities:

a short in circuit fuse blew

not a short in circuit fuse blew
not a short in circuit fuse did not blow

The premise that the fuse did not blow rules out the first two of these models. So, only the third
model remains, and it yields the conclusion:

There was not a short in the circuit.

Some people are able to make this valid deduction, but, as the theory predicts, it is harder and
takes longer than the inference described earlier. It depends on deliberation and fully explicit
models whereas the earlier inference depends on intuition and mental models.
Consider the following inference, where each premise refers to two alternative possibilities:

Either the pie is on the table or else the cake is on the table.
Either the pie isn’t on the table or else the cake is on the table.
Could both of these assertions be true at the same time?

Most people say, “yes”. Their inference depends on mental models. Each premise refers to the
possibility that the cake is on the table, and so it seems that both assertions could be true. But,
let’s spell out the fully explicit possibilities. The first premise refers to these two possibilities for
what’s on the table:

the pie not the cake

not the pie the cake

The second premise refers to these two possibilities:

not the pie not the cake

the pie the cake

A careful examination reveals that the two assertions have no possibility in common. So, the
correct conclusion is that they cannot both be true at the same time. The intuitive inference
is an illusion, just one of many different sorts (Khemlani & Johnson-Laird, 2017). What they

223
224

Phil N. Johnson-Laird

have in common is that they follow from mental models of premises, but not from their fully
explicit models. The systematic occurrence of illusory inferences is by far the most unexpected
prediction of the model theory. It is a litmus test to show that reasoning is relying on mental
models rather than on fully explicit models.

Simulations based on spatial, temporal, and kinematic models

So far, you may have the impression that models are just words and phrases rather than actual
models of the world. In fact, models can be three-dimensional structures underlying your grasp,
say, of how to get from an office on one floor of a building to its main entrance. If someone
asks you to draw a path in the air of your route you can do so: the ability is a prerequisite for
navigation, especially the sort that Micronesian islanders carry out without instruments, not
even a compass (see, e.g., Gladwin, 1970).
When individuals make two-dimensional spatial inferences, they rely on models. Here is
one such problem:

The cup is on the right of the plate.

The spoon is on the left of the plate.
The knife is in front of the spoon.
The saucer is in front of the cup.
What is the relation between the knife and the saucer?

The premises describe a layout of the sort in this diagram of a table-top:

spoon plate cup

knife saucer

where the items at the bottom of the diagram are in front of those at the top. It is quite easy to infer:

The knife is on the left of the saucer.

If instead you used logical rules to derive your answer, your first step would be to deduce the
relation between the spoon and cup (using the logical properties of on the left and on the right):

The spoon is on the left of the cup.

You would then use axioms for the two-dimensional relations in the remaining premises to
deduce from this intermediate conclusion that:

The knife is on the left of the saucer.

A similar problem is most revealing:

The plate is on the right of the cup.

The spoon is on the left of the plate.
The knife is in front of the spoon.
The saucer is in front of the plate.
What is the relation between the knife and the saucer?

224
225

Models and rational deductions

The premises are consistent with two distinct layouts, because of the uncertainty of the relation
between the spoon and the cup:

spoon cup plate cup spoon plate

knife saucer knife saucer

Yet, the relation between the two items in the question is the same in both layouts, and so it
follows that:

The knife is on the left of the saucer.

If you used models to infer the conclusion, the deduction should be harder than the previous
one, because you have to construct two models. But, if you used logical rules to infer the
conclusion, then the deduction should be easier than the previous one. You no longer have
to deduce the relation between the spoon and the plate, because it is stated in the second
premise. Hence, the experiment is crucial in that mental models and logical rules make
opposite predictions. The results of experiments corroborated the model theory, and they
did so even when a later study corrected for the fact that the first premise is irrelevant in
the second inference (see Byrne & Johnson-Laird, 1989; Schaeken, Girotto, & Johnson-Laird,
1998). Analogous results occur when inferences concern temporal rather than spatial relations
(Schaeken, Johnson-Laird, & d’Ydewalle, 1996). And a corollary is that diagrams that make it
easy to envisage alternative possibilities both speed up and enhance the accuracy of deductions
(Bauer & Johnson-Laird, 1993).
The description of a sequence of events can elicit a kinematic model that itself unfolds
in time. Imagine a railway track that runs from left to right. It has a siding onto which cars
enter from the left, and exit to the left. Five cars are at the left end of the track: A B C D
E. Figure 13.1 is a diagram of the situation. Here is a problem for you to solve:

All but one of the cars enter the siding. The remaining car at the left end of the track
moves over to the right end of the track. Now each car on the siding, one at a time,
moves back to the left end of the track and immediately over to the right end. What
is the resulting order of the cars at the right end of the track?

Adults, and even 10-year-old children, can deduce the final order of the cars: E D C B A. They
do so by envisaging the effect of each move until the sequence ends. In other words, they simu-
late the sequence in a kinematic mental model (Khemlani et al., 2013). Children accompany
their simulations with gestures indicating the moves they would make if they were allowed to
move the cars – an outward sign of inward simulation. And they are less accurate if they are
prevented from gesturing (Bucciarelli et al., 2016).

Figure 13.1 A diagram of a small railway track with five cars at the left-hand side of the track

225
226

Phil N. Johnson-Laird

Conclusion
Rational deductions depend on mental models. The theory is controversial in psychology,
but it has led to evidence that challenges its rivals. The earliest theories of human reasoning
proposed that the mind is equipped with formal rules of inference akin to those of logic (e.g.,
Rips, 1994). These theories cannot explain the illusory inferences described earlier, and, as we
saw, when models were pitted against logic, number of models rather than number of logical
steps predicted the difficulty of deductions. Because models represent possibilities, modal logic
might be relevant because it deals with possibilities and necessities. But, even though there are
many different modal logics (see, e.g., Girle, 2009), human reasoning diverges from all of them.
One such divergence is in inferences such as:

The fault is in the software or in the cable, or both.

Therefore, it is possible that the fault is in the software.

It is invalid in all modal logics (for the reasons described on p. 220). Yet, as we saw, most people
make such inferences (Hinterecker et al., 2016).
A radical alternative is that reasoning relies, not on logic, but on probabilities (e.g., Oaksford
& Chater, 2007). This proposal has a more limited purview than the model theory. It offers
no explanation of illusory or kinematic inferences. It applies to Aristotelian syllogisms, but
the model theory fits the results of experiments better (Khemlani & Johnson-Laird, 2013).
Possibilities with appropriate numbers are probabilities. So, the model theory offers an account
of probabilistic reasoning (Johnson-Laird et al., 1999; Khemlani, Lotstein, & Johnson-Laird,
2015). In contrast, theories based on probabilities alone cannot distinguish between certainty
and necessity, so they are unable to explain inferences about who must play in one-on-one
games of basketball, as discussed above.
Our reasoning is bounded. We make deductions in domains that are undecidable or intract-
able in their demands on time and memory. Yet, we are rational in principle. We grasp the force
of counterexamples, and spontaneously use them to refute invalid inferences. Our deductions
from fully explicit models yield valid conclusions based on all the information in the premises.
Indeed, models of possibilities are enough to establish principles of rationality, and the model
theory itself.

References
Bauer, M. I., and Johnson-Laird, P. N. 1993. How diagrams can improve reasoning. Psychological Science
4, 372–378.
Bell, V., and Johnson-Laird, P. N. 1998. A model theory of modal reasoning. Cognitive Science
22, 25–51.
Boolos, G. S., and Jeffrey, R. C. 1980. Computability and Logic (3rd ed.). Cambridge: Cambridge
University Press.
Bucciarelli, M., Mackiewicz, R., Khemlani, S. S., and Johnson-Laird, P. N. 2016. Children’s creation of
algorithms: Simulations and gestures. Journal of Cognitive Psychology 28, 297–318.
Byrne, R. M. J., and Johnson-Laird, P. N. 1989. Spatial reasoning. Journal of Memory and Language 28,
564–575.
Elio, R., and Pelletier, F.J. 1997. Belief change as prepositional update. Cognitive Science 21, 419–460.
Girle, R. 2009. Modal Logics and Philosophy (2nd ed.). London: Routledge.
Gladwin, T. 1970. East Is a Big Bird. Cambridge, MA: Harvard University Press.
Harman, G. 1986. Change in View: Principles of Reasoning. Cambridge, MA: MIT Press.
Hinterecker, T., Knauff, M., and Johnson-Laird, P. N. 2016. Modality, probability, and mental models.
Journal of Experimental Psychology: Learning, Memory, and Cognition 42, 1606–1620.

226
227

Models and rational deductions

James, W. 1907. Pragmatism. New York: Longmans, Green.

Jeffrey, R. 1981. Formal Logic: Its Scope and Limits (2nd ed.). New York: McGraw-Hill.
Johnson-Laird, P. N. 1983. Mental Models. Cambridge, MA: Harvard University Press.
Johnson-Laird, P. N. 2006. How We Reason. New York: Oxford University Press.
Johnson-Laird, P. N., and Byrne, R. M. J. 1991. Deduction. Hillsdale, NJ: Erlbaum.
Johnson-Laird, P. N., Girotto, V., and Legrenzi, P. 2004. Reasoning from inconsistency to consistency.
Psychological Review 111, 640–661.
Johnson-Laird, P. N., Legrenzi, P., Girotto, V., Legrenzi, M., and Caverni, J-P. 1999. Naive probability: A
mental model theory of extensional reasoning. Psychological Review 106, 62–88.
Johnson-Laird, P. N., and Ragni, M. 2019. Possibilities as the foundation of reasoning. Cognition 193,
130950.
Juhos, C. A., Quelhas, C., and Johnson-Laird, P. N. 2012. Temporal and spatial relations in sentential
reasoning. Cognition 122, 393–404.
Khemlani, S. S., and Johnson-Laird, P. N. 2012. Hidden conflicts: Explanations make inconsistencies
harder to detect. Acta Psychologica 139, 486–491.
Khemlani, S. S., and Johnson-Laird, P. N. 2013. The processes of inference. Argument and Computation
4, 4–20.
Khemlani, S. S., and Johnson-Laird, P. N. 2017. Illusions in reasoning. Minds and Machines 27, 11–35.
Khemlani, S. S., Lotstein, M., and Johnson-Laird, P. N. 2015. Naive probability: Model-based estimates
of unique events. Cognitive Science 39, 1216–1258.
Khemlani, S. S., Mackiewicz, R., Bucciarelli, M., and Johnson-Laird, P. N. 2013. Kinematic mental
simulations in abduction and deduction. Proceedings of the National Academy of Sciences 110, 16766–16771.
Oaksford, M., and Chater, N. 2007. Bayesian Rationality: The Probabilistic Approach to Human Reasoning.
Oxford: Oxford University Press.
Quelhas, A. C., and Johnson-Laird, P. N. 2017. The modulation of disjunctive assertions. Quarterly Journal
of Experimental Psychology 70, 703–717.
Quelhas, A. C., Rasga, C., and Johnson-Laird, P. N. 2017. A priori true and false conditionals. Cognitive
Science 41, 1003–1030.
Quine, W. V. O. 1953. From a Logical Point of View. Cambridge, MA: Harvard University, Press.
Ragni, M. 2003. An arrangement calculus, its complexity and algorithmic properties. In A. Günter,
R. Kruse, and B. Neumann (Eds.), Advances in Artificial Intelligence: Proceedings of 26th Annual German
Conference on AI (pp. 580–590). Berlin: Springer.
Rips, L. 1994. The Psychology of Proof. Cambridge, MA: MIT Press.
Schaeken, W. S., Girotto, V., and Johnson-Laird, P. N. 1998. The effect of an irrelevant remise on temporal
and spatial reasoning. Kognitionswisschenschaft 7, 27–32.
Schaeken, W. S., Johnson-Laird, P. N., and d’Ydewalle, G. 1996. Mental models and temporal reasoning.
Cognition 60, 205–234.
Simon, H. A. 2000. Bounded rationality in social science: Today and tomorrow. Mind & Society 1, 25–39.
Stanovich, K. E. 1999. Who Is Rational? Studies of Individual Differences in Reasoning. Hillsdale, NJ: Erlbaum,

227
228

14
PATTERNS OF DEFEASIBLE
INFERENCE IN CAUSAL
DIAGNOSTIC JUDGMENT
Jean Baratgin and Jean-Louis Stilgenbauer

Introduction: what is diagnostic reasoning?

Diagnostic reasoning consists in going back to the causes that triggered one or multiple effects. For
example, we reason from effect to cause when we notice the existence of certain symptoms (skin
eruptions, sneezing, conjunctivitis, eczema, …) and try to identify the agent that triggered them
(pollen allergy, food allergy, mite allergy, others …). The most basic form of diagnostic inference,
which will be our main interest in this chapter, consists in estimating the probability of the cause
from the knowledge of the effect. Formally this probability is written Pr(cause|effect). When
physicians notice that one of their patients has conjunctivitis, they might, for example, try to find
out if this symptom is linked to a pollen allergy. For this, they would need to calculate the diagnostic
probability associated with the cause of interest by estimating Pr(Pollen-Allergy|Conjunctivitis).
The rational norm of causal diagnostic reasoning has evolved a lot. It went from a purely
probabilistic (statistic) norm centered around the Bayes’s rule to a norm centered around causal
Bayesian networks. However, regardless of the rationality framework considered, people’s
performances systematically deviate from the predictions of these rationality norms. The under-
estimation of the base rate and the neglect of alternative causes are the main violations of ration-
ality traditionally reported by psychologists. These results, which are incidentally very robust,
suggest that the estimation of Pr(cause|effect) is calculated in a limited context of rationality,
or of degraded rationality, and they also suggest that this estimating process relies on the use of
heuristic strategies. In this chapter, we will focus on two verbal strategies of diagnostic reasoning
taking the shape of probable conditional inferences (the defeasible Modus Ponens and the defeasible
affirming the consequent). Before detailing these heuristics, we will first re-contextualize the basic
diagnostic activity of interest here in regard to associated reasoning. We will then clarify the
level on which our work is based in comparison to research in the field of causality.
Estimating Pr(cause|effect) reflects, as we mentioned, an elementary form of diagnostic
reasoning (Tversky & Kahneman, 1974). It is a basic reasoning triggered when the diagnostic
process starts and when we try to account for a new phenomenon (effect). Calculating the diag-
nostic probability helps people evaluate the intensity of the (statistical) link between the variable
considered to be the potential cause (or cause of interest), and the effect to explain. However,
this initial step is, insufficient to access the complete characterization of the causal role of the
cause of interest and two other types of reasoning will allow the access to a deeper level of

228
229

Defeasible inference in causal judgment

Symbol:
Type of inference Target Example
Quantity to estimate
3 - How can a pollen allergy cause
What is the mechanism by conjuncvis?
Explanatory which the cause triggers the - OR -
Pr(c → m → e|c → e)
mechanism effect? - Why does pollen allergy cause
conjuncvis?

2
What proporon of the
Causal Pr(c → e|e) Is conjuncvis triggered by
occurrence of effect is due to
a ribuon cause? pollen allergy?

1
What is the strength of
Stascal Is conjuncvis related to the
associaon between cause Pr(c|e)
associaon presence of pollen allergy?
and effect?

Figure 14.1 The three types of inferences involved in the production of a diagnosis. Reasoning
is initiated by the observation of an effect (noted e), step (1) consists in detecting a potential cause
(noted c) and in measuring the force of association between e and c. This first step corresponds to
the elementary diagnostic reasoning with the estimation of Pr(c|e). Step (2) consists in estimating the
contribution of c in the causal influence triggering e. This inference is usually called causal attribution
and is represented by the expression Pr(c → e|e). Step (3) is the most complex of all three as it is at this
level that an explanatory mechanism (noted m) will be selected\conceived. This mechanism is a process
(containing at least one variable) through which the influence of the cause c will be transmitted to the
effect e. We can summarize this final step of the diagnostic reasoning with the expression Pr(c → m →
e | c → e). This step corresponds to an inference toward the best explanation.

understanding (see Figure 14.1). To show the effective causal role of the cause of interest, we
need to extract the statistical data by producing an inference called causal attribution (Cheng &
Novick, 2005; Hilton & Slugoski, 1986). This reasoning consists in estimating the probability
of the causal link between the cause of interest and the effect. A more elaborate form of infer-
ence will then complete the diagnostic by encouraging the selection\conception of an explana-
tory mechanism which will explain the mode of action of the cause. This type of reasoning is
usually called inference to the best explanation (Douven, 2016; Lipton, 2004).
Finally, since this chapter focuses on a form of reasoning relying on causal links, it seems
legitimate to put our argument in perspective in the field of research on causality. According
to Williamson (2007), three great axes of research on causality exist. The first focuses on
the problem of the nature of causality, the second on learning causal links, and the third on
causality-based reasoning. Our work on diagnostic inference strategies is a natural extension of
research in the third axis, but also in the second one since in the field of diagnosis, learning and
reasoning are often quite entangled. The first axis is in contrast to the other two, completely
orthogonal to the question asked in this chapter, and as a consequence we will not comment
further on contemporary theories on the nature of causality.1

Rationality framework: from a statistical norm to causal models

Research led by the heuristics and bias program (Tversky & Kahneman, 1974) yet shows
that the estimation of the diagnostic probability of individuals deviates from the diagnostic
probability calculated by Bayes’s rule.2 Different tasks like the “Lawyer-Engineer problem”

229
230

Jean Baratgin and Jean-Louis Stilgenbauer

(Kahneman & Tversky, 1973), the “Taxis problem” (Bar-Hillel, 1980) and the “Mammography
problem” (Eddy, 1982), show the systematic neglect of the base rate of the cause of interest
when estimating the diagnostic probability (for a review, see Baratgin & Politzer, 2006; Barbey
& Sloman, 2007; Koehler, 1996). In the previous example of the allergy to pollens, the base
rate was then defined by the prior probability (or prevalence) of this allergy within the refer-
ence population.
Yet the pure statistical information is not enough to establish an accurate diagnosis as indi-
viduals actually infer this kind of information through a causal model of the situation (or causal
structure).3 Figure 14.1 shows an elementary causal model with only one cause of interest
(pollen allergy) and only one effect (conjunctivitis). This kind of object was introduced in the
field of cognitive psychology in the 1990s (Waldmann & Holyoak, 1992; Waldmann, Holyoak,
& Fratianne, 1995), it is a compact representation of the cause-effect relationships connecting
the variables of interest to each other. Causal models have later been developed and updated
to provide heuristics models to psychology allowing the testing of new hypotheses, and, on
the other hand, these models provide a new framework of rational analysis of human causal
inferences (Griffiths & Tenenbaum, 2005, 2009).
A causal model contains information that goes beyond the simple pattern of co-occurrences
of the variables of interest, meaning the statistical data of the studied system. Individuals
possessing a causal model will be able to understand the process that generated this data. This
higher level of understanding implies attributing a new state to the variables of the system: each
variable will be interpreted as either the cause or the effect. One other refinement consists in
enriching a causal model through the introduction of mechanisms which will help accurately
define the relationships between variables. Figure 14.2 represents an elementary causal model

Alternaves Mechanisms M alt_1 ; M alt_2 ; … M alt_n

Pollen Allergy Conjuncvis Causal Model

Cause Effect

Mechanism M 1 … Mechanism M n

Mechanism M 2
Time

Figure 14.2 Causal model containing a single cause of interest (pollen allergy) causally linked to a single
effect (conjunctivitis). The model includes a supplementary variable representing all the alternative
causes potentially also triggering the effect. The elementary diagnostic reasoning shown here allows
going back to the cause from the knowledge of the effect. This process consists in the estimation of the
probability Pr(cause|effect) and is symbolized by the curved double arrow pointing to the cause. The
causal model can also be improved through the introduction of a mechanism explaining why the cause
triggers the effect (see Protzko, 2018 for a recent discussion on the concepts of cause and mechanism).
In general, multiple mechanisms M1, M2, … Mn (dashed arrows) compete to characterize the causal link
of interest. Alternative causes also possess their own explanatory mechanisms Malt_1, Malt_2, … Malt_n.

230
231

Defeasible inference in causal judgment

based on a structure containing a single cause and a single effect. The arrow connecting those
two variables implies a link of causality. The model also contains a supplementary variable
representing all the different alternative causes able to trigger the same effect regrouped into
one entity.
Today, representing causal knowledge of people through the use of causal Bayes nets (CBN) is
quite common. The CBN are formal tools that were initially developed in the field of Artificial
Intelligence (Pearl, 1988, 2000; Spirtes, Glymour, & Scheines, 2000). This formalism is com-
monly used to study causal cognition in psychology. For a detailed review of its uses in this
field, see Rottman and Hastie (2014) and Rottman (2017). The great modularity of the CBN
offers the possibility to represent causal structures like the one in Figure 14.2 and also allow
the expression of the great variety of reasoning processes related to it. For example, Meder,
Mayrhofer, and Waldmann (2009, 2014) recently suggested a model of diagnostic reasoning
(structure induction model) establishing a new rational framework for elementary inferences
produced from a causal structure similar to the one shown in Figure 14.2.4 Despite being useful,
not only to establish a standard of rationality but also for their heuristic qualities, we believe that
this model remains insufficient in order to explain human diagnostic inferences. People have
indeed limited cognitive resources (memory, attention, computational power) and of course,
cannot process the complex computations required by CBN. In order to avoid this obstacle,
people must build strategies to estimate the diagnostic probability. In the following section of
this chapter, we will briefly describe some of these well-known strategies linked to a misuse of
the Bayes’s rule and will then detail a new type of strategies relying on conditional reasoning
schemas.

Strategies related to the sub-optimal use of the Bayes’s rule in the

estimation of the diagnostic probability
The heuristics and bias program of the 1970s revealed that people’s estimations of the diag-
nostic probability deviate from the Bayesian model. Yet, those first researchers did not take
into account the influence of causal knowledge in the probabilistic reasoning. Krinski and
Tenenbaum (2007) have shown that estimations of the diagnostic probability in the mammog-
raphy problem were close to the Bayesian norm when the statistical information (pattern of co-
occurrence of the input variables) was associated with an explicit causal model of the situation,
see also Fernbach and Rehder (2012) for results of the same nature. Recent work nuances this
observation (Hayes, Ngo, Hawkins, & Newell, 2018) for despite the introduction of the causal
model, some people keep using non-Bayesian estimation strategies. Some participants indeed
estimate the diagnostic probability Pr(cause|effect) from the rate of false positives, by taking
into account only the probability Pr(effect|¬cause). Other non-Bayesian estimation strategies
have been described by Cohen and Staub (2015), such as those consisting in calculating a
pondered sum of the probabilities mentioned in the Bayes’s rule. Some participants try to cal-
culate the diagnostic probability from the addition of the false positives rate Pr(effect|¬cause)
to the likelihood (rate of true positives) Pr(effect|cause).
The various strategies used by people in order to estimate the diagnostic probability suggest
that the underlying cognitive processes can vary from one person to another, added to the fact
that nothing prevents the observation of variations within the persons themselves depending on
the time or context. The systematic study of estimation strategies appears to be crucial in order
to improve the understanding of human diagnostic inferences. A limiting aspect of previous
studies is the nature of the strategies identified. Indeed, these studies have specifically focused
on strategies consisting in combining (in a sub-optimal fashion) quantities of the Bayes’s rule (in

231
232

Jean Baratgin and Jean-Louis Stilgenbauer

particular the rate of false positives and true positives). Yet, other strategies of diagnostic estima-
tion can be considered like those formed from schemas of conditional inferences.

New diagnostic estimation strategies: defeasible Modus Ponens and

defeasible Affirming the Consequent
In general, diagnostic probability estimation strategies are distinguished from rational models,
either purely statistical like the Bayes’s rule or causal like CBN framework. If we use Marr’s
terminology (1982), these models are at the computational level of the general cognitive archi-
tecture. They are functional models that define a standard of rationality to calculate the diag-
nostic probability. Strategies used by people are instead at the algorithmic level. They represent
the effective cognitive processes followed by people to estimate this probability. This level being
purely descriptive, we believe the influence of language and of mechanisms related to linguistic
communication in the estimation process cannot easily be ignored. It is indeed natural for
people to be communicating and to be using their causal beliefs in reasoning, forming prop-
ositions and organizing them in the shape of arguments. According to Mercier and Sperber
(2011), it is even the essential function of reasoning. An entire literature on causal arguments
exists establishing a strong link between causal cognition and argumentation (see Hahn, Bluhm,
& Zenker, 2017 for a recent review). Among the causal arguments, conditional arguments are
central to the human cognition. Two patterns have been of special interest to us here because
of their high psychological plausibility. The first is based on the affirming the consequent schema
(AC), the second based on the Modus Ponens (MP). These two schemas will be considered in
particular in defeasible forms, as a consequence, we will be using the terms defeasible affirming
the consequent (DAC) and defeasible Modus Ponens (DMP). These reasonings are represented in
Figure 14.3.
Initially, AC and MP were studied in their causal form by Cummins, Lubart, Alksnis, and
Rist (1991), Cummins (1995) and Politzer and Bonnefon (2006). In these research studies,
AC was built on a conditional in which the antecedent and the consequent respectively
corresponded to a cause and to an effect. The first premise of the argument consists in declaring
the effect (we will use here the term “effect!” to mark the initial stage of reasoning that begins
with the arrival of new information), the second premise corresponds to the conditional
statement and the conclusion is formed by the cause. AC coincides with a form of reasoning
called abduction by the philosopher C. S. Peirce (for recent reviews, see Douven, 2011; Park,
2017). Cummins (1995) and Politzer and Bonnefon (2006) also studied a specific form of MP
built on an inverted conditional in which the antecedent was an effect and the consequent
was the cause. The first premise and the conclusion of this form of MP remained untouched
compared to AC. They were respectively instantiated by the effect on one hand and the cause
on the other. We represent below the structure of AC and of MP such as they appear in the
aforementioned studies:

AC: MP:
Effect! Effect!
If <cause> then <effect> If <effect> then <cause>
Cause Cause
Those two arguments constitute, in our point of view, good candidates in order to define diag-
nostic probability estimation strategies, as these forms of reasoning both consist in going back
to the cause (conclusion) from the observation of the effect (first premise). The conditional
(the second premise) defines the type of strategy. For AC, the inference is based on the trust

232
233

Defeasible inference in causal judgment

Defeasible Affirming the Consequent (DAC):

Effect ! Conjuncvis !
If cause, then Pr(effect) If a paent is allergic to pollen, then he will probably have conjuncvis
example
Pr(cause) Probably pollen allergy

Defeasible Modus Ponens (DMP):

Effect ! Conjuncvis !
If effect, then Pr(cause) example If a paent has conjuncvis, then he is probably allergic to pollen
Pr(cause) Probably pollen allergy

Figure 14.3 These two reasoning patterns represent two formally different strategies for estimating
the diagnostic probability Pr(cause|effect). They correspond respectively in the A.I. and philosophical
literature to two modes of inference: abduction for DAC and deduction defeasible for DMP. First, the
defeasible deduction is a weakened form of deduction that permits the production of provisionally
true and/or probable conclusions. Second, abduction is a heuristic reasoning that serves to identify
explanatory causes or hypotheses.

put in the causality link between the antecedent and the consequent. This trust depends on the
existence of alternative causes that are, just like the cause of interest, capable of triggering the
effect. For MP, the mechanism which the inference relies on is, in our point of view, different,
for it is rather based on the explanatory qualities of the cause of interest. Here, it’s the strength
of the mechanism by which the effect is produced by the cause of interest that will guide the
inference.
Despite AC and MP being a priori interesting in order to define diagnostic reasoning strat-
egies, those arguments are not, as they are, completely satisfying. Indeed, as we’ve explained
above, diagnostic reasoning is an uncertain reasoning allowing the estimation of the prob-
ability of the cause of interest, given the knowledge of the effect, or formally the estimation
of the probability Pr(cause|effect). AC and MP schemas must be generalized and considered
in a probable form as is shown in Figure 14.3. As is shown in Figure 14.3, the probabilistic
aspect is defined by the degree of probability placed, on one hand, on the consequent of the
conditionals, and on the other, on the reasoning conclusions. A good diagnostic strategy should
also possess the quality of being defeasible, for it is natural for people to draw a conclusion or to
update its probability in the case of the arrival of new information (taking an additional premise
into account for example). The defeasible aspect is shown through the dashed inference line in
Figure 14.3.
In order to clearly differentiate the standard schemas AC and MP from their probable and
defeasible generalization, we will from now on be talking respectively about DAC and DMP.
Those two schemas, as causal diagnostic reasoning strategies, have recently been studied by
Stilgenbauer and Baratgin (2018, 2019) who have shown their psychological relevance in esti-
mating Pr(cause|effect). The authors submitted participants to an experimental paradigm of
rule production, where people had to rebuild the conditional rule of DAC and DMP schemas.

233
234

Jean Baratgin and Jean-Louis Stilgenbauer

Participants were only able to use two pieces of information. They received the first premise
of the schema (which was certain) corresponding to the phase of the declaration of the effect.
This step is symbolized in Figure 14.3 by the term “effect!.” Participants then were informed of
the (probable) conclusion of the reasoning, represented in Figure 14.3 by the term “Pr(cause).”
The task consisted in the production of a conditional rule allowing the inference of the con-
clusion from the first premise. To answer, participants had at their disposal jumbled words from
which they could form the rule of their choice: [1] if cause then Pr(effect) or the reversed rule
[2] if effect then Pr(cause). The kind of rule produced revealed the diagnostic estimation strategy
the participants preferred. The production of a rule similar to [1] signaled a preference for the
DAC estimation strategy. Similarly, the production of a rule similar to [2] revealed a preference
for the DMP strategy.
The results of Stilgenbauer and Baratgin (2018, 2019) show that the participants preferred
DMP, as they built rules similar to [2] to infer the conclusion “Pr(cause)” from the premise
“effect!.” The use of a rule like if effect then Pr(cause) is indeed quite natural in this situation
since a direct correspondence exists between that conditional and the diagnostic probability
Pr(cause|effect).5 This is not the case with DAC inferences though, which rely on rules similar
to [1] and which instead make the predictive probability Pr(effect|cause) stand out, since in this
case the antecedent of the rule is made of the cause, and the consequent is made of the effect.
The results of Stilgenbauer and Baratgin (2018) also show that the participants’ preferred
strategy was not always DMP. Indeed, in certain situations, the DAC strategy can strongly
compete with the DMP strategy. This change of preference is linked to the perceived value
of the predictive probability Pr(effect|cause).6 This result has been obtained while control-
ling the relative levels of the diagnostic and predictive probabilities. In a first probabilistic con-
text, participants received information indicating that the diagnostic probability was higher
than the predictive probability: Pr(cause|effect) > Pr(effect|cause). In a second condition, it
was the opposite: the information given indicated that the diagnostic probability was lower than
the predictive probability Pr(cause|effect) < Pr(effect|cause). The results show that conditional
rules built by the participants depended on the probabilistic context (meaning the relative levels
of the diagnostic and predictive probabilities). People, in the main, build rules similar to [2] if
effect then Pr(cause) (DMP strategy) in the context Pr(cause|effect) > Pr(effect|cause). Yet, in the
context of Pr(cause|effect) < Pr(effect|cause), the proportion of rules similar to [1] if cause then
Pr(effect) (DAC strategy) significantly increases and does not differ from the proportion of rules
similar to [2]. These results confirm the idea that people can estimate the diagnostic probability
following strategies taking the form of defeasible schemas of inferences. In the following section
we will report new experimental results confirming the robustness of the data we just exposed.

New experimental data: a test of diagnostic strategies through

a rule evaluation paradigm
We propose here a new experiment expanding on the logic of Stilgenbauer and Baratgin
(2018). The task no longer consists in producing a conditional rule, but to evaluate (and com-
pare) the two rules that DAC and DMP are made of. The experiment procedure was the
following: participants begin by receiving the first premise of the diagnostic strategy “effect!” as
well as the conclusion “Pr(cause);” we then explicitly and simultaneously show them the con-
ditional rules [1] If cause then Pr(effect) and [2] If effect then Pr(cause). The task of the participants
consists in judging if those rules are judicious to infer the conclusion “Pr(cause)” from the
premise “effect!.” Each rule is evaluated on a seven points scale (1 = not very judicious; 7 = very
judicious). We assume that the rule judged to be the most judicious will indicate the preferred

234
235

Defeasible inference in causal judgment

diagnostic estimation strategy. If the rule [1] is judged to be more judicious than rule [2], this
will show that DAC is the preferred strategy to estimate the diagnostic probability. But if [2] is
judged more judicious than [1], this will show a preference for DMP. In this experiment, we also
recreated two probabilistic contexts identical to those in the study of Stilgenbauer and Baratgin
(2018). The judgment of the participants was recorded in the context of Pr(cause|effect) >
Pr(effect|cause) and also in the context Pr(cause|effect) < Pr(effect|cause).
A scenario of fault detection has been used to cover the situation of diagnostic reasoning. This
scenario involved the operation of an industrial machine producing auto parts. The probabilistic
context was introduced in a between-subject design, using pictures as shown in Figure 14.4. No
matter the probabilistic context, a machine composed of multiple components was introduced
to the participants. Components of this machine can malfunction, occasioning the production
of faulty parts. The components of the machine represent the causes and the defects on the parts
represent the effects. In Figure 14.4, the picture on the left refers to the context Pr(cause|effect) >
Pr(effect|cause) and the picture on the right to the context Pr(cause|effect) < Pr(effect|cause).
For example, for the picture on the left, the machine has two components that can malfunc-
tion resulting in parts produced with four defects. The malfunction of components X1 or X2
can cause the defects Xa, Xb, Xc, or Xd. For the two probabilistic contexts, it is impossible to
have multiple components malfunctioning at the same time, neither is it possible to have mul-
tiple defects at the same time. For the left picture, the diagnostic and predictive probabilities are
respectively equal to Pr(cause|effect) = ½ and Pr(effect|cause) = ¼.
Participants of the experiment were then told of the malfunction of the machine, and we
used the conversation between the repairmen attempting to solve the issue to introduce sim-
ultaneously the conditional rules [1] and [2]. Before taking apart the machine, the repairmen

Probabilisc context: Probabilisc context:

Pr(Component|Defect) > Pr(Defect|Component) Pr(Component|Defect) < Pr(Defect|Component)

Defect Xa
Component X1

Defect Xb Component X2 Defect Xa

Component X1

Component X3 Defect Xb
Component X2 Defect Xc

Component X4
Machine Defect Xd

Machine

Figure 14.4 Experimental set-up introducing the situation of causal diagnostic reasoning. In each
probabilistic context, an industrial machine is symbolized by dashed rectangles. The machines are made
up of a certain number of components and their malfunction (causes) can trigger the production of
faulty parts with defects (effects). The figure on the left defines the probabilistic context Pr(cause|effect)
> Pr(effect|cause). For example, if we’re interested in the defect Xa (effect) and to the component X1
(cause), we have Pr(X1|Xa) = 1/2 and Pr(Xa|X1) = 1/4. The figure on the right defines the opposite
probabilistic context Pr(cause|effect) < Pr(effect|cause) with Pr(X1|Xa) = 1/4 and Pr(Xa|X1) = 1/2.

235
236

Jean Baratgin and Jean-Louis Stilgenbauer

Preferenal rules:
Effec ves

If effect then Pr(cause)

DMP strategy
If cause then Pr(effect)
DAC strategy
No preference
DMP = DAC

Pr(cause|effect) < Pr(effect|cause) Pr(cause|effect) > Pr(effect|cause)

Probabilis c Context
Figure 14.5 Experimental results obtained in rule evaluating. Distribution of the preferred rules of
participants depending on the probabilistic context defined by the relative levels of the diagnostic and
predictive probabilities.

thought out loud and one of them suggested rule [1] If component X1 then Pr(defect Xa).
Another repairman then contradicted him and suggested the reversed rule [2] If defect Xa then
Pr(component X1).
We recruited 101 participants for the experiment and the results are shown in Figure 14.5.
We categorized participants depending on the grades they attributed to each rule. When the
grade of the rule [1] was higher than the grade of the rule [2], participants were categorized
in DAC. They were categorized in DMP if they graded [2] higher than [1]. In the case where
participants graded the two rules the same, they were categorized in “No preference.”
Calculating a χ² revealed a significant link between the preferred rule and the probabilistic
context. In the situation Pr(cause|effect) > Pr(effect|cause), participants believe rules similar
to [2] to be more judicious than rules similar to [1]. Yet, in the situation of Pr(cause|effect)
< Pr(effect|cause), there was no particular preference visible between the two kinds of
rules: 20 participants believed rule [1] to be more judicious than rule [2] and 19 participants
believed the opposite. In the context Pr(cause|effect) > Pr(effect|cause), there are also 5
participants grading both rules as equally judicious, and 13 in the context Pr(cause|effect) <
Pr(effect|cause).
These results obtained with a paradigm of rule evaluation are interesting for they corrob-
orate those of Stilgenbauer and Baratgin (2018) who used a paradigm of rule production. This
new data seems to validate the idea that it is natural for people to use defeasible strategies like
DMP and DAC in order to estimate the diagnostic probability. This experiment also confirms
that these two strategies are not equally preferred by individuals. Our data indicate that DMP

236
237

Defeasible inference in causal judgment

seems to constitute the strategy by default, but in situations where the perceived value of the
predictive probability increases, DAC strongly competes with this strategy.

Conclusion: the diversity of diagnostic reasoning shapes

In this chapter, we have been interested in an elementary form of diagnostic reasoning consisting
in estimating the probability of a cause from the knowledge of its effect: Pr(cause|effect). We
have limited ourselves to the study of inferences produced from the causal structure known a
priori and given to participants. This structure was composed of a single cause of interest and a
single effect (see Figure 14.2), we will refer to it by using the expression cause → effect.7 In the
literature, the majority of studies consist in the evaluation of the performance of participants
compared to the rational norm that today is made up of causal Bayesian networks. The aim of this
work was different. It intended to remain at the level of psychological processes (in other words,
processes that are at the algorithmic level according to the terminology of D. Marr) to test the
plausibility of two strategies of diagnostic inferences. The first was based on a defeasible affirming
the consequent and the second was based on a defeasible Modus Ponens. Our results show that
participants understand and can follow these patterns of inference to estimate the probability of
a cause from the knowledge of its effect. These two strategies are important for they are natural
for people, although it does not mean that they constitute the only way possible when trying
to estimate Pr(cause|effect). Other ways can be imagined, for example, we can think about the
use of strategies based on negated components in the shape of a defeasible Modus Tollens (effect!,
If ¬ cause, then Pr(¬ effect) ⇒ Pr(cause))8 or of a defeasible denying the antecedent (effect!, If ¬
effect, then Pr(¬ cause) ⇒ Pr(cause)). It belongs to future research to test the psychological
reality of these kinds of strategies and to report their potential usage by people.
In this chapter, we studied the estimation strategies of Pr(cause|effect) for basic diagnostic
inferences produced from a cause → effect structure known a priori. Yet, we believe that two other
kinds of diagnostic reasoning seem to be of importance and, to our knowledge, no study exists
on their estimation strategies. Figure 14.1 at the beginning of this chapter suggests an articula-
tion between the basic diagnostic reasoning and these two complementary forms of inferences.
The first is called causal attribution (Cheng & Novick, 2005; Hilton & Slugoski, 1986) and
intervenes in the context where the causal structure is unknown. In this situation, the diagnosis
relies on the link between the cause and the effect, and the diagnostic probability of interest will
be written Pr(cause → effect | effect). The second kind of diagnostic inference that seems to
us to be of importance is a bit more complex. It consists in inferring the mechanism explaining
the nature of the causal link. A minimalistic mechanism can be reduced to a single mediator (in
Figure 14.2, the concept of mechanism is represented by dashed arrows without actually being
clarified). A mediator is a variable allowing the transmission of the influence of the cause to the
effect. For example, we know today that smoking (cause) can trigger lung cancer (effect), through
the tar (mediator) accumulating in lung alveoli (the role of tar hasn’t always been clear in the
past). In the context where the structure cause → effect is known a priori, we can, for example, ask
ourselves if a mediator transmitting the influence of the cause to the effect exists. In this situ-
ation, the probability Pr(cause → mediator → effect | cause → effect) constitutes the diagnostic
probability of interest. In statistics, a very interesting literature exists on the topic of mediation.
For a reference, we can read VanderWeele (2015). Many examples of mediation can also be
found in MacKinnon (2008).
Finally, we will note that the causal attribution Pr(cause → effect | effect) consists in the
elaboration of a causal structure by definition unknown and that the inference of a medi-
ator Pr(cause → mediator → effect | cause → effect) consists in the modification of the causal

237
238

Jean Baratgin and Jean-Louis Stilgenbauer

structure known a priori. In the literature dedicated to causal cognition, these processes are gen-
erally understood as learning processes allowing on the one hand the elaboration of the struc-
ture (defining the variables as cause and effect), and on the other, the estimation of its parameters
(defining the probability distributions of the structure’s variables). There is a third class of
processes traditionally distinguished in the literature regrouping all the causal reasoning that
can be done from a completely defined and parameterized structure (we can refer to Hastie,
2015; Rottman, 2017; and Rottman & Hastie, 2014 for a general review). Yet, we believe that
the strategies used by people in order to estimate the diagnostic probability cannot be classified
in the traditional way in effect in the field of causal cognition. The main reason for this is that
in the field of diagnosis (at least) learning a causal link is often concomitant to the diagnostic
reasoning itself. This is also what Meder et al. (2014) suggest with the model of structure induc-
tion combining learning of the causal structure with the estimation of Pr(cause|effect) (see also
Meder et al., 2009). Having said that, we believe that the result or the product of the process of
diagnostic inference (the potential cause, the causal link or the mediator) constitutes the central
element from which a typology of diagnostic reasoning strategies can be elaborated. A future
research program dedicated to the use of diagnostic estimation strategies by people will come,
we hope, to bring answers to this important question.

Acknowledgments
The authors would like to thank Frank Jamet and Baptiste Jacquet for much discussion and
other help in their research.

Notes
1 For readers interested in the different conceptions of causality, we recommend reading Beebee,
Hitchcock, and Menzies (2009), especially Parts 2 and 3.
2 The Bayes’s rule is a rule of belief updating when new convincing information is learned. It is the
only possible rule of revision of beliefs in the situation of revision called focusing (the situation where
the message concerns an object drawn at random from a population of objects which constitutes a
certain stable universe, see Baratgin & Politzer, 2010; Walliser & Zwim, 2011). In the context of
causal diagnostic, if we write c = cause and e = effet, the rule would be written in the following
Pr(e|c) × Pr (c )
way: Pr (c|e ) = . The term Pr (c|e ) corresponds to the diag-
Pr(e|c) × Pr (c ) + Pr(e|¬c) × Pr ( ¬c )
nostic probability, which is the probability of the cause when the effect is present. The quantity Pr ( e|c )
is generally called likelihood, yet in the situation of causal diagnostic, we would rather use the rate of true
positives, which is the probability of observing the effect when the cause is present. The quantity Pr (c )
is the prior probability of the cause. From an objective point of view, it corresponds to the base rate (or
prevalence) of the cause within the reference population. The term Pr ( e|¬c ) refers to the influence of
alternative causes, in other words, the probability of observing the effect without observing the cause of
interest. It can also be called the rate of false positives. Finally, the quantity Pr ( ¬ c ) corresponds to the
base rate of all of the alternative causes.
3 From the historical point of view, from the very beginning of the rise of modern statistics during the
nineteenth century, has started to emerge a radical separation between this discipline, on the one hand
and causal concepts, on the other (such as those of strength, of mechanisms or of a causal model).
Chapter 2 of Pearl and Mackenzie (2018) tells the history of this divorce which, beyond statistics, has
produced considerable negative effects for all data-driven sciences. Unmistakably, this separation also
had consequences for research on causal cognition. The excessive focus of the heuristics and bias school
on statistical concepts to the detriment of the introduction of causal concepts likely stems from this
divorce. Fortunately, we see today the return of causality. The “causal revolution,” according to Pearl is
on the way and has started planting its seeds into all sciences and in particular in cognitive psychology.

238
239

Defeasible inference in causal judgment

Waldmann and Hagmayer (2013) wrote an excellent review describing the evolution of causal cog-
nition theories and in particular the shift from purely statistical models to models integrating causal
concepts.
4 Meder and Mayrhofer (2017) have recently shown an interest in the more sophisticated forms of
diagnostic reasoning that might be produced from causal structures containing multiple causes and
effects.
5 Today, many results demonstrating that people interpret the probability of an indicative conditional such
as Pr(if A, then C) as the conditional probability Pr(C|A) are available, see for example Baratgin and
Politzer (2016). Yet in this work, we study strategies relying on conditionals in the shape of “if A then
Pr(C)” and this type of rule constitutes the natural interpretation of Pr(if A, then C), see for this specific
point Over, Douven, and Verbrugge (2013).
6 Predictive probability plays an important part in the estimation of the diagnostic probability. It influences,
in particular, the accuracy of the estimations of the diagnostic probability. For example, when we ask
participants to estimate from a set of data the value of the diagnostic probability, they will weight their
estimations with the perceived value of the predictive probability. The lower the predictive probability,
the more under-estimated the diagnostic probability is, compared to its real value contained in the
data (Meder et al., 2009, 2014; see also Meder & Mayrhofer, 2017; Stilgenbauer & Baratgin, 2019;
Stilgenbauer, Baratgin, & Douven, 2017).
7 Yet, in the experiment described in the previous section, we have defined different values of
Pr(cause|effect) by changing the number of alternative causes (see Figure 14.4).
8 The symbol “⟹” represents a defeasible consequence relationship.

References
Baratgin, J., & Politzer, G. (2006). Is the mind Bayesian? The case for agnosticism. Mind and Society, 5(1),
1–38. http://doi.org/10.1007/s11299–006–0007–1.
Baratgin, J., & Politzer, G. (2010). Updating: A psychologically basic situation of probability revision.
Thinking & Reasoning, 16(4), 253–287. http://doi.org/10.1080/13546783.2010.519564.
Baratgin, J., & Politzer, G. (2016). Logic, probability and inference: A methodology for a new paradigm.
In L. Macchi, M. Bagassi, & R. Viale (Eds.), Cognitive unconscious and human rationality (pp. 119–142).
Cambridge, MA: MIT Press.
Barbey, A. K., & Sloman, S. A. (2007). Base-rate respect: From ecological rationality to dual processes.
Behavioral and Brain Sciences, 30(3), 241–254. http://doi.org/10.1017/S0140525X07001653.
Bar-Hillel, M. (1980). The base-rate fallacy in probability judgments. Acta Psychologica, 44(3), 211–233.
http://doi.org/10.1016/0001–6918(80)90046–3.
Bechtel, W., & Abrahamsen, A. (2005). Explanation: A mechanist alternative. Studies in History and
Philosophy of Biological and Biomedical Sciences, 36, 421–441.
Beebee, H., Hitchcock, C., & Menzies, P. (Eds.) (2009). The Oxford handbook of causation. Oxford: Oxford
University Press. http://doi.org/10.1093/oxfordhb/9780199279739.001.0001.
Cheng, P. W., & Novick, L. R. (2005). Constraints and nonconstraints in causal learning: Reply to
White (2005) and to Luhmann and Ahn (2005). Psychological Review, 112(3), 694–707. http://doi.org/
10.1037/0033–295X.112.3.694.
Cohen, A. L., & Staub, A. (2015). Within-subject consistency and between-subject variability
in Bayesian reasoning strategies. Cognitive Psychology, 81, 26–47. http://doi.org/10.1016/
j.cogpsych.2015.08.001.
Cummins, D. (1995) Naïve theories and causal cognition. Memory and Cognition, 23(5), 646–659.
Cummins, D. D., Lubart, T., Alksnis, O., & Rist, R. (1991). Conditional reasoning and causation. Memory
& Cognition, 19(3), 274–282.
Douven, I. (2011). Abduction. In E. N. Zalta (Ed.), The Stanford encyclopedia of philosophy (Spring 2011
ed.). Available at: https://stanford.library.sydney.edu.au/entries/abduction/
Douven, I. (2016). Inference to the best explanation: What is it? And why should we care? In T. Poston
& K. McCain (Eds.), Best explanations: New essays on inference to the best explanation. Oxford: Oxford
University Press.
Eddy, D. M. (1982). Probabilistic reasoning in clinical medicine: Problems and opportunities. In D.
Kahneman, P. Slovic, & A. Tversky (Eds.), Judgment under uncertainty: Heuristics and biases (pp. 249–
267). Cambridge: Cambridge University Press. http://doi.org/10.1017/CBO9780511809477.019.

239
240

Jean Baratgin and Jean-Louis Stilgenbauer

Fernbach, P. M., & Rehder, B. (2012). Cognitive shortcuts in causal inference. Argument & Computation,
4(1), 64–88. http://doi.org/10.1080/19462166.2012.682655.
Griffiths, T. L., & Tenenbaum, J. B. (2005). Structure and strength in causal induction. Cognitive Psychology,
51(4), 334–384.. http://doi.org/10.1016/j.cogpsych.2005.05.004.
Griffiths, T. L., & Tenenbaum, J. B. (2009). Theory-based causal induction. Psychological Review, 116(4),
661–716. http://doi.org/10.1037/a0017201.
Hahn, U., Bluhm, R., & Zenker, F. (2017). Causal argument. In M. R. Waldmann (Ed.), Oxford
handbook of causal reasoning. Oxford: Oxford University Press. http://doi.org/10.1093/oxfordhb/
9780199399550.013.26.
Hastie, R. (2015). Causal thinking in judgments. In G. Keren & G. Wu (Eds.), The Wiley Blackwell hand-
book of judgment and decision making (pp. 590–628). New York: Blackwell. http://doi.org/10.1002/
9781118468333.ch21.
Hayes, B. K., Ngo, J., Hawkins, G. E., & Newell, B. R. (2018). Causal explanation improves judgment
under uncertainty, but rarely in a Bayesian way. Memory and Cognition, 46(1), 112–131. http://doi.org/
10.3758/s13421–017–0750-z.
Hilton, D. J., & Slugoski, B. R. (1986). Knowledge-based causal attribution. The abnormal conditions
focus model. Psychological Review, 93(1), 75–88. http://doi.org/10.1037/0033–295X.93.1.75.
Kahneman, D., & Tversky, A. (1973). On the psychology of prediction. Psychological Review, 80(4), 237–
251. http://doi.org/10.1037/h0034747.
Koehler, J. J. (1996). The base rate fallacy reconsidered: Descriptive, normative, and methodological
challenges. Behavioral and Brain Sciences, 19(1), 1–17. http://doi.org/10.1017/S0140525X00041157.
Krynski, T. R., & Tenenbaum, J. B. (2007). The role of causality in judgment under uncertainty. Journal
of Experimental Psychology General, 136(3), 430–450. http://doi.org/10.1037/0096–3445.136.3.430.
Lipton, P. (2004). Inference to the best explanation. London: Routledge.
Machamer, P., Darden, L., & Carver, C. F. (2000). Thinking about mechanisms. Philosophy of Science,
67(1), 1–25. http://doi.org/10.1086/392759.
MacKinnon, D. (2008). Introduction to statistical mediation analysis. New York: Lawrence Erlbaum Associates.
Marr, D. (1982). Vision: A computational investigation into the human representation and processing of visual infor-
mation. San Francisco: Freeman.
Meder, B., & Mayrhofer, R. (2017). Diagnostic reasoning. In M. R. Waldmann (Ed.), Oxford handbook
of causal reasoning (pp. 433–458). Oxford: Oxford University Press. http://doi.org/10.1093/oxfordhb/
9780199399550.013.23.
Meder, B., Mayrhofer, R., & Waldmann, M. R. (2009). A rational model of elemental diagnostic infer-
ence. In N. A. Taatgen & H. van Rijn (Eds.), Proceedings of the 31st Annual Conference of the Cognitive
Science Society (pp. 2176–2181). Austin, TX: Cognitive Science Society.
Meder, B., Mayrhofer, R., & Waldmann, M. R. (2014). Structure induction in diagnostic causal reasoning.
Psychological Review, 121(3), 277–301. http://doi.org/10.1037/a0035944.
Mercier, H., & Sperber, D. (2011). Why do humans reason? Arguments for an argumentative theory.
Behavioral and Brain Sciences, 34(2), 57–111. http://doi.org/10.1017/S0140525X10000968.
Over, D., Douven, I., & Verbrugge, S. (2013). Scope ambiguities and conditionals. Thinking & Reasoning,
19(3), 284–307. http://doi.org/10.1080/13546783.2013.810172.
Park, W. (2017). Abduction in context: The conjectural dynamics of scientific reasoning. Berlin: Springer. http://
doi.org/10.1007/978–3-319–48956–8.
Pearl, J. (1988). Probabilistic reasoning in intelligent systems: Networks of plausible inference. San Mateo,
CA: Kaufmann. http://doi.org/10.1016/0004–3702(91)90084-W.
Pearl, J. (2000). Causality: Models, reasoning, and inference. New York: Cambridge University Press. http://
doi.org/10.1017/CBO9780511803161.
Pearl, J., & Mackenzie, D. (2018). The book of why: The new science of cause and effect. New York: Basic Books.
Politzer, G., & Bonnefon, J.-F. (2006). Two varieties of conditionals and two kinds of defeaters help reveal
two fundamental types of reasoning. Mind and Language, 21(4), 484–503.
Protzko, J. (2018). Disentangling mechanisms from causes: And the effects on science. Foundations of
Science, 23(1), 37–50. http://doi.org/10.1007/s10699–016–9511-x.
Rottman, B. M. (2017). The acquisition and use of causal structure knowledge. In M. R. Waldmann
(Ed.), Oxford handbook of causal reasoning (pp. 1–55). Oxford: Oxford University Press. http://doi.org/
10.1093/oxfordhb/9780199399550.013.10.
Rottman, B. M., & Hastie, R. (2014). Reasoning about causal relationships: Inferences on causal networks.
Psychological Bulletin, 140, 109–139. http://doi.org/doi.org/10.1037/a0031903.

240
241

Defeasible inference in causal judgment

Spirtes, P., Glymour, C., & Scheines, R. (2000). Causation, prediction, and search. New-York: Springer-Verlag.
Stilgenbauer, J.-L., & Baratgin, J. (2018). Étude des stratégies de raisonnement causal dans l’estimation
de la probabilité diagnostique à travers un paradigme expérimental de production de règle. Canadian
Journal of Experimental Psychology/Revue Canadienne de Psychologie Expérimentale, 72(1), 58–70. http://
doi.org/doi.org/10.1037/cep0000108.
Stilgenbauer, J.-L., & Baratgin, J. (2019). Assessing the accuracy of diagnostic probability estima-
tion: Evidence for defeasible modus ponens, International Journal of Approximate Reasoning, 105, 229–240.
https://doi.org/10.1016/j.ijar.2018.11.015.
Stilgenbauer, J.-L., Baratgin, J., & Douven, I. (2017). Reasoning strategies for diagnostic probability
estimates in causal contexts: Preference for defeasible deduction over abduction. In Proceedings of the
4th International Workshop on Defeasible and Ampliative Reasoning (DARe-17) Co-located with the 14th
International Conference on Logic Programming and Nonmonotonic Reasoning (LPNMR 2017) (pp. 29–43).
Espoo, Finland. Available at: http://ceur-ws.org/Vol-1872/.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science,
185(4157), 1124–1131.
Van der Weele, T. (2015). Explanation in causal inference: Methods for mediation and interaction.
New York: Oxford University Press.
Waldmann, M. R., & Hagmayer, Y. (2013). Causal reasoning. In D. Reisberg (Ed.), Oxford handbook of
cognitive psychology (pp. 733–752). New York: Oxford University Press.
Waldmann, M. R., & Holyoak, K. J. (1992). Predictive and diagnostic learning within causal models: asym-
metries in cue competition. Journal of Experimental Psychology: General, 121(2), 222–236. Available
at: www.ncbi.nlm.nih.gov/pubmed/1534834.
Walliser, B., & Zwirn, D. (2011). Change rules for hierarchical beliefs. International Journal of Approximate
Reasoning, 52(2), 166–183. http://doi.org/10.1016/j.ijar.2009.11.005.
Williamson, J. (2007). Causality. In D. M. Gabbay & F. Guenthner (Eds.), Handbook of philosophical logic
(vol. 14, pp. 95–126). Dordrecht: Springer.

241
242

15
ATTRIBUTE- BASED CHOICE
Francine W. Goh and Jeffrey R. Stevens

Introduction
When sitting at your favorite café, you face the choice between a small cup of your favorite
beverage for $3.50 or a large cup for $3.75. Many of us would choose the large cup. But how
might we arrive at this choice? One method could generate a value for each option by com-
bining the amount of beverage received and the cost of that beverage (Figure 15.1). After
repeating this process for the second option, one could then compare the two options’ values.
This is an example of alternative-based processing because information is primarily processed
within each alternative or option and options are processed in a sequential manner (Payne,
Bettman, & Johnson, 1993).
Another way to make the choice is to compare within attributes and across options. Attribute-
based processing primarily processes information within attributes or dimensions of information
used in choice. In the beverage example, the amount of beverage and cost are the two rele-
vant attributes. This method would compare the prices and may determine that they are quite
similar. However, the amounts are noticeably different, and you prefer more to less beverage, so
you choose the larger beverage (Figure 15.1).
These two types of processing use the same information, but they use them in different
ways. Alternative-wise processing integrates within alternatives and considers options sequen-
tially, while attribute-wise processing compares within attributes and considers options sim-
ultaneously. Though many models of choice do not explicitly define the decision-making
process, they typically imply either alternative- or attribute-wise processing.
Herbert Simon (1957) proposed the notion of bounded rationality, which states that decision
makers face constraints on information availability, time to make a decision, and computa-
tional abilities. Including all information or processing it in complicated ways may be difficult
for decision makers. Instead, they may use heuristics that ignore information and use simpler
computations to make the best possible choice given the constraints of their current situation
(Payne et al., 1993; Gigerenzer, Todd, & the ABC Research Group, 1999; Selten, 2002).
Models of choice include both optimization models that use all information and combine it in
complex ways, as well as heuristic models. Many alternative-based models of decision making
use all informational cues and combine them in a way to generate something akin to a sub-
jective value for each option. But many heuristics are attribute-based models. Though there

242
243

Attribute-based choice

Figure 15.1 Approaches to choice. Alternative-wise processing integrates attributes (in this case, by
applying some function f to the attributes) within alternatives to generate a composite value. Attribute-
wise processing compares attributes across alternatives using a process such as similarity.

are exceptions in both directions, alternative-based approaches tend to use optimization and
attribute-based approaches tend to be heuristics. Here, we briefly review alternative-based
approaches to choice, then explore attribute-based approaches and what they can offer the
study of decision making.

Alternative-based choice
Alternative-based models are often compensatory because they make tradeoffs across attributes: high
values in low-weighted attributes can compensate for low values in high-weighted attributes.
Though alternative-based models can sometimes involve complex calculations, these models
can lead to sharper distinctions between choice options and thus facilitate optimal decision
making (Russo & Dosher, 1983). Thus, most normative models of choice are alternative-based
models. These models have been used to account for a range of types of decisions, including
multi-attribute choice, risky choice, and intertemporal choice.
Multi-attribute choice refers to situations in which decision makers must choose between
two or more options, and each option has values for a number of different attributes. For
example, one might choose between apartments that differ in their price, location, security
deposit, and amenities. A normative model of optimal multi-attribute choice is the weighted
additive (WADD) rule (or weighted sum model) (Keeney & Raiffa, 1976). WADD is effect-
ively a regression model that generates an overall value for each option in a choice set based
on attributes that are weighted by their importance to the decision maker (Payne et al., 1993).
The overall value for an option is determined by multiplying the value of each attribute by that
attribute’s weight and summing all of the weighted attribute values for that option. All possible
options are then compared and the option with the highest overall value is selected. WADD,
therefore, uses all available information, involves complex computational steps, and requires a
high degree of cognitive effort (Payne et al., 1993). The equal weight rule (or Dawes’ rule)
is a variation of WADD in which all attributes are equally weighted (Dawes, 1979), thereby
simplifying the decision-making process (Payne et al., 1993). Satisficing is another decision-
making strategy that involves finding an option that satisfies a threshold or set of thresholds
(Simon, 1955). Decision makers first determine minimally acceptable threshold values (or
aspiration levels) for each attribute. Each option is then considered sequentially by comparing
the option’s attribute values to their corresponding predetermined thresholds. Options that
contain any attributes that do not meet the thresholds are excluded and the first option that
contains attributes which satisfy all of the attribute thresholds is selected (Payne et al., 1993).

243
244

Francine W. Goh and Jeffrey R. Stevens

Though satisficing is considered a type of heuristic because it does not necessarily assess all
options, it uses alternative-wise processing by evaluating options in the sequence in which they
occur in the choice set.
Risky choice refers to situations in which options include different outcomes occurring with
different probabilities. For example, would you prefer a 100 percent chance of receiving $100
or a 50 percent chance of $200 and a 50 percent chance of $0? Researchers have proposed many
models of risky choice, and most of them are alternative-based approaches using a modification
of expected value. The expected value approach multiplies two attributes—the probability of
an outcome and the reward amount of that outcome—and sums these over all outcomes within
an option to generate an expected value (Pascal, 1654, as cited in Smith, 1984). Many other
models modify the expected value approach by applying a function to the probability and/or
outcome (e.g., expected utility, subjective utility, prospect theory; see Stott, 2006). The key
feature of these models is that each option is summarized into a value that is compared across
options.
Intertemporal choice refers to sets of options that differ in the reward amount and time delay
to receiving that reward. For example, would you prefer $100 today or $150 in one year? Like
risky choice models, most intertemporal choice models integrate two attributes to generate
a value for each option. For intertemporal choice, the attributes are reward amount and the
time delay to receiving the reward. Models of intertemporal choice apply different functions to
the reward amounts and time delays and different operations to combine them (Doyle, 2013;
Regenwetter et al., 2018). These operations have the effect of discounting the value of the
reward amount based on the time delay. These discounting models (e.g., exponential, hyper-
bolic, quasi-hyperbolic, additive) generate discounted values for each option and compare them
to select the least discounted option.

Attribute-based choice
Attribute-based models have also been developed to account for multi-attribute, risky, and
intertemporal choice. Many attribute-based models are non-compensatory because they do not
use all available information and therefore can avoid tradeoffs across attributes. Low-weighted
attributes may be ignored and, therefore, cannot compensate for low values in high-weighted
attributes. Attribute-based models can also allow for intransitive preference cycles in which
option A is preferred to B, B is preferred to C, and C is preferred to A. Though alternative-
based models have gained the majority of interest in the field, research on attribute-based
models has grown. Here, we survey a subset of attribute-based models of multi-attribute, risky,
and intertemporal choice.

Lexicographic heuristic
In the lexicographic heuristic, decision makers first decide on the attribute that is most
important to them. They then compare the values of that attribute across all choice options
before selecting the option with the highest value on that attribute. In instances where two
options have the same value on the most important attribute, individuals will have to compare
the options on the next most important attribute. This comparison process continues until one
option is deemed to be better than the other option on an attribute of importance (Fishburn,
1974). Several studies have found empirical support for the usage of the lexicographic heur-
istic during decision making (Slovic, 1975; Tversky, Sattah, & Slovic, 1988; Kohli & Jedidi,
2007; Yee, Dahan, Hauser, & Orlin, 2007). A variation of the lexicographic heuristic is the

244
245

Attribute-based choice

lexicographic semi-order, where the ranking of each option’s value on an attribute depends on
a just noticeable difference (Tversky, 1969). Specifically, if the values of options fall within the
just noticeable threshold for the target attribute, the attribute values of these options will be
ranked as equal and the decision maker will have to consider the next most important attribute
to break the tie.

Elimination-by-aspects heuristic
The elimination-by-aspects (EBA) heuristic combines the lexicographic heuristic with
the conjunctive rule. The conjunctive rule states that decision makers make a choice by
establishing minimally acceptable threshold values for attributes and then eliminating choice
options that do not meet these threshold values (Dawes, 1964; Einhorn, 1970). Similar to
the lexicographic heuristic, decision makers using the EBA heuristic first select the attribute
(or aspect) that is most important to them. A threshold value for each attribute is then
determined and the value of each option on that attribute is compared to the threshold value.
Choice options that do not meet the threshold value for the attribute are eliminated and the
process continues with the next most important attribute until there is only one option that
meets the threshold values for all of the attributes (Tversky, 1972). Additionally, the EBA
heuristic has been suggested as a heuristic used by decision makers to reduce cognitive effort
when they have to make a decision from several choice options (Payne, 1976). The EBA
heuristic is considered to violate the principle of rational choice because the final decision is
determined by a single attribute. On the other hand, the EBA heuristic is also considered a
rational heuristic because it comprises the ranking of attributes in order of their importance
(Tversky, 1972; Payne et al., 1993).

Proportional difference model

The proportional difference (PD) model was initially developed to predict decision-making
behavior in a risky choice setting. The PD model posits that individuals compare the values
of options along the same attributes in a proportional manner (i.e., the monetary outcome
and probability of receiving that outcome of one option relative to those of the other option).
During this comparison process, individuals add the advantages and subtract the disadvantages
for each option to obtain an adjusted difference variable. To reach a decision, individ-
uals compare their difference variable to a decision threshold that reflects the importance
of each attribute (González-Vallejo, 2002). Such decision thresholds can vary according to
individual wealth status and the context of the situation (González-Vallejo, 2002; González-
Vallejo, Reid, & Schiltz, 2003; González-Vallejo & Reid, 2006). The PD model has since
been extended to the domain of intertemporal choice by replacing the probability dimension
with the time delay to receiving the monetary outcome (Cheng & González-Vallejo, 2016).
In addition, the PD model accounts for the magnitude effect (where individuals exhibit less
discounting when values are larger), violations of stochastic dominance (which states that
when two options are similar on one attribute, individuals will choose the option that is dom-
inant on the differing attribute), transitivity of preferences, reflection effect (which states that
individuals are risk-averse when they have to make a choice among gains and risk-seeking
when they have to make a choice among losses), and additivity (which holds that preference
for a delayed option should be consistent regardless of how the delay period is segmented)
(González-Vallejo, 2002; González-Vallejo et al., 2003; González-Vallejo & Reid, 2006;
Cheng & González-Vallejo, 2016).

245
246

Francine W. Goh and Jeffrey R. Stevens

Tradeoff model
The tradeoff model proposes that individuals choose from options by weighing the advan-
tage of the monetary outcome of one option against the advantage of the time value of the
other option in intertemporal choice (Scholten & Read, 2010). The tradeoff model thus also
suggests that time can be converted to the same scale of measurement used for monetary
outcome (Scholten, Read, & Sanborn, 2014). The tradeoff model accounts for the mag-
nitude effect, the common difference effect (the tendency for individuals to exhibit more
discounting when a delay period begins sooner compared to later), violations of transitivity of
preferences, additivity, and inseparability (which states that individuals consider the value of
the time delay of an option based on the value of that option’s monetary outcome). In contrast
to the PD model which uses the absolute monetary and time delay values of options to carry
out comparisons between two options, the tradeoff model includes additional parameters that
calculate value- and time-weighing functions that capture individuals’ subjective perceptions
of monetary values and time delays respectively (Scholten & Read, 2010; Cheng & González-
Vallejo, 2016). Despite the use of a more complex formula to predict choice, the ability of the
tradeoff model to predict intertemporal choice is similar to that of the PD model (Cheng &
González-Vallejo, 2016).

Difference-ratio-interest-finance-time (DRIFT) model

The difference-ratio-interest-finance-time (DRIFT) model suggests that decisions in
intertemporal choice are affected by how choice options are framed. Specifically, the weighted
average of the absolute difference between monetary values, relative difference between mon-
etary values, experimental interest rate offered, and extent to which individuals view the
experimenter’s offer as an investment rather than a consumption are balanced against the
importance assigned to time (Read, Frederick, & Scholten, 2013). Read and colleagues
(2013) found in their analysis of the DRIFT model that the framing of monetary outcomes
as investments increased individuals’ patience for small monetary values ($700) but reduced
patience for large monetary values ($70,000), which suggests that the manner in which
intertemporal choices are framed affects decision-making behavior. The DRIFT model can
account for the magnitude effect and delay effect (which states that individuals discount less
when the time delay is described in a calendar date format instead of units of delay) when the
difference and ratio between monetary values and the experimental interest rates are varied
(Read et al., 2013).

Intertemporal choice heuristic (ITCH) model

The intertemporal choice heuristic (ITCH) model makes use of arithmetic operations to pre-
dict choice. Individuals compare available options by first subtracting and dividing option values
along their respective monetary and time dimensions to obtain absolute and relative differences.
Weights that reflect the level of importance assigned to these dimensions are then added to each
of the four variables and their sum calculated to arrive at a decision (Ericson, White, Laibson,
& Cohen, 2015).
Although the ITCH model is similar to the DRIFT model, the two models differ in that the
ITCH model calculates both absolute and relative differences in time delay whereas the DRIFT
model calculates only absolute differences (Ericson et al., 2015). The ITCH model accounts for
the property of additivity and the magnitude, common difference, and delay effects.

246
247

Attribute-based choice

Similarity model
The similarity model was initially developed to study decision making in a risky choice setting
(Rubinstein, 1988; Leland, 1994). Rubinstein (1988) suggested that individuals who have to
make a decision between two lottery options will do so by comparing the similarity of mon-
etary outcomes and similarity of probability of receiving those outcomes for both options. The
similarity model has also been extended to the strategic choice domain where it suggests that
individuals make a decision by comparing the similarity of payoff options (Leland, 2013) and to
the intertemporal choice domain where it posits that individuals arrive at a decision by com-
paring the similarity of monetary outcomes and time delays of choice options (Leland, 2002;
Rubinstein, 2003; Stevens, 2016).
In the domain of intertemporal choice, the version of the similarity model developed by
Leland (2002) suggests that individuals compare the similarity of monetary outcomes and simi-
larity of time delays for options. These similarity comparisons can then result in one of three
decision consequences: (1) a choice is made because one option dominates the other option
on one attribute but has similar values on the other attribute, (2) the choice between the two
options is inconclusive because both options offer similar values on the monetary and time
delay attributes, or (3) the choice between the two options is inconsequential because one
option dominates the other option on one attribute but is dominated on the other attribute.
When a decision is either inconclusive or inconsequential, Leland (2002) proposed that indi-
viduals will proceed to make a choice at random, whereas Rubinstein (2003) suggested that a
choice must be made using another (unspecified) criterion.
Stevens (2016) added a second stage of existing discounting models if similarity analysis
was inconclusive or inconsequential. The two-stage similarity models predicted individual
choice better than Leland’s (2002) similarity model and other discounting models alone. Finally,
the similarity model accounts for the magnitude, reflection, and common difference effects,
violations of stochastic dominance, and transitivity of preferences (Leland, 1994, 1998, 2002;
Stevens, 2016).

Fuzzy-trace theory
Fuzzy-trace theory posits that individuals encode information presented to them in both
verbatim and gist representations (Reyna & Brainerd, 1995, 2011). Verbatim representations
refer to the exact remembrance of information, such as remembering a note word-for-word
or the exact digits of a telephone number. On the other hand, gist representations refer to
memory for the general meaning of concepts, such as one’s principles or cultural norms
(Reyna & Brainerd, 2011; Reyna, 2012). Several studies have shown that individuals tend to
rely on gist compared to verbatim representations when they have to make decisions and that
individuals prefer to make use of the simplest gist level (or categorical distinctions) when-
ever possible (Reyna & Farley, 2006; Reyna & Lloyd, 2006; Reyna et al., 2011; Reyna &
Brainerd, 2011).
Fuzzy-trace theory suggests that decisions made in risky choice and intertemporal choice
settings depend on the gist, or core, principles evoked based on the context of the situation.
The evoked gist principles then make one option more salient than the other, resulting in a
decision. When specifically applied to the context of intertemporal choice, the evoked gist
principles will make either the smaller-sooner option or the larger-later option more salient
which will in turn affect the intertemporal choice made (Reyna, 2012; Rahimi-Golkhandan,
Garavito, Reyna-Brainerd, & Reyna, 2017; Reyna & Wilhelms, 2017). For example, an

247
248

Francine W. Goh and Jeffrey R. Stevens

individual presented with a choice of either receiving $5 today or $7 in three days may think of
these options in the gist representations of “receiving some money today” and “receiving some
money later” respectively. According to fuzzy-trace theory, if this individual possesses a gist
principle akin to “living in the moment,” his gist principle would make the option to receive
$5 today more attractive than the other option, resulting in him choosing the smaller-sooner
option. Fuzzy-trace theory can thus be considered an attribute-based model since it compares
choice options across attributes.

Process data
While many researchers have compared alternative-based and attribute-based models in terms
of their ability to accurately predict choice, the two classes of models also provide an important
distinction in the process of choice—that is, the cognitive steps required to make a choice.
Therefore, it can be useful to investigate process data to explore these models, particularly data
that reveal how decision makers acquire information. Researchers have used eye tracking and
mouse tracking to measure information acquisition with respect to alternative- or attribute-
based information processing (Schulte-Mecklenbeck, Kühberger, & Ranyard, 2010).

Eye tracking
Eye-tracking techniques monitor how individuals attend to different types of information
(Duchowski, 2017). As individuals look at visual stimuli, an eye tracker records the direction
and path of eye movements, thus allowing researchers insight into which information individ-
uals consider to be important (Figure 15.2). In the first use of eye tracking to study the decision-
making process, Russo and Rosen (1975) studied how individuals chose in a multi-alternative
choice setting. The researchers concluded that participants compared the options presented to
them in pairs and preferred to use options that were similar to one another to form these pairs
whenever they could. In a similar vein, Russo and Dosher (1983) found that participants pre-
ferred to compare options along attributes in a multi-attribute choice setting. Participants also
made use of attribute-wise comparisons in gambles that were expected to follow an expected
value rule where the expected monetary outcome and probability of receiving that outcome
are combined to calculate an overall value for each option. These eye-tracking studies thus
support an attribute-based decision-making process.

Figure 15.2 Example eye-tracking screenshot. Eye trackers record the direction and path of eye
movements when individuals view information presented on a screen.

248
249

Attribute-based choice

Arieli, Ben-Ami, and Rubinstein (2011) used a risky choice task to investigate whether
individuals used an alternative-based or similarity-based approach to make decisions. They
hypothesized that participants who used an alternative-based approach when evaluating lottery
options would move their eyes within options because they were formulating an overall value
for each option (see Figure 15.1). On the other hand, participants who used a similarity-based
approach would move their eyes within attributes because they were comparing the options
along each attribute (i.e., monetary outcome and probability).
Arieli and colleagues (2011) found that participants used attribute-based processing to make
decisions in a risky choice setting when the values of the monetary outcomes and probabilities
made the calculation of the overall value of each option using the alternative-based process
difficult. A follow-up study conducted by Aimone, Ball, and King-Casas (2016) supported
Arieli et al.’s (2011) finding and also found that attribute-based processing was associated with
risk-aversive choice preferences. Further, Arieli et al. (2011) found that participants maintained
attribute-based processing to make a decision when the context of the lottery options was
changed from risky choice (i.e., with the dimensions being monetary outcome and probability
of receiving the monetary outcome) to intertemporal choice (i.e., with the dimensions being
monetary outcome and time delay of monetary outcome).

Mouse tracking
Mouselab is a computer program that allows researchers to monitor how individuals acquire
information in the decision-making process by using a computer mouse as a tracking tool
(Johnson, Payne, Bettman, & Schkade, 1989; Willemsen & Johnson, 2009). The informa-
tion for each attribute for all options in the choice set is presented in a matrix in which the
cells are covered by overlays. When individuals hover their mouse over a cell of the matrix,
the overlay disappears to reveal the underlying information; when the mouse moves out of
the cell, the overlay reappears and covers the information again (Figure 15.3). Mouselab also
records the amount of time that individuals spend viewing each “opened” section, the order
in which sections are viewed and the number of times each section is viewed, thus allowing

Figure 15.3 Recreated screenshot of MouselabWEB. Individuals move their mouse over each section of
the matrix to display information for each option that is hidden by the overlay. The overlay reappears—
and the information disappears again—when the mouse is moved out of the section.

249
250

Francine W. Goh and Jeffrey R. Stevens

researchers to ascertain the importance of each attribute to individuals during the decision-
making process.
In a comparison of the Mouselab and eye-tracking process-tracing techniques, Lohse and
Johnson (1996) found that for both techniques, participants used attribute-based processing
when the number of options in a choice set was increased in both risky choice and multi-attribute
choice settings. By contrast, participants used alternative-based processing when the number of
attributes for each option in the choice set was increased (Lohse & Johnson, 1996). Reeck, Wall,
and Johnson (2017) used Mouselab and eye tracking to study variations in search strategies in an
intertemporal choice setting. They found that participants who used an attribute-based search
strategy to compare options had a higher tendency to choose the larger-later option and were
more susceptible to option-framing effects compared to participants who used an alternative-
based search strategy. Additionally, they found that participants who spent more time looking
at the amount dimension compared to time dimension of options were more patient as they
chose the larger-later option more often than other participants in the study. In a study that
combined risky and intertemporal choices, Konstantinidis, van Ravenzwaaij, and Newell
(2017) used a computer program similar to Mouselab to study individuals’ decision making
in risky intertemporal choice. Participants in their study had to choose between two lottery
options that differed in the dimensions of monetary outcome, probability of receiving the mon-
etary outcome, and delay in receiving the monetary outcome. Process data demonstrated that
participants preferred to use an attribute-based approach over an alternative-based approach
when making their decisions.
Research using eye tracking and mouse tracking has further elucidated our understanding
of how individuals acquire and process information when making decisions. Evidence from
process data has shown that individuals make use of both alternative-based and attribute-based
models of choice processing, and that the choice strategy that they use depends on the context
of the situation at hand.

Conclusion
While the study of decision making has historically focused on alternative-based models,
attribute-based models have experienced a resurgence of interest from researchers for a
number of reasons. First, they follow from Simon’s notion of bounded rationality because
they often reflect real-world limitations faced by decision makers by using less information
and simpler computations. This is especially pertinent in instances where decision makers
have to make a choice from myriad options or when there is risk involved in the decision-
making process. Second, they capture choice data quite well, predicting multi-attribute, risky,
intertemporal, and strategic choices while accounting for, or bypassing, anomalies regularly
encountered in the use of alternative-based models. Third, in addition to capturing choice
data, attribute-based models can capture the decision process by making predictions about
eye tracking and information acquisition data. Combined, these lines of evidence suggest
that attribute-based models provide a fruitful class of decision-making models that warrant
continued investigation.

Acknowledgments
This work was funded in part by the National Science Foundation (SES-1658837).

250
251

Attribute-based choice

References
Aimone, J. A., Ball, S., & King-Casas, B. (2016). It’s not what you see but how you see it: Using eye-
tracking to study the risky decision-making process. Journal of Neuroscience, Psychology, and Economics,
9(3–4), 137–144. doi:10.1037/npe0000061.
Arieli, A., Ben-Ami, Y., & Rubinstein, A. (2011). Tracking decision makers under uncertainty. American
Economic Journal: Microeconomics, 3(4), 68–76. doi:10.1257/mic.3.4.68.
Cheng, J., & González-Vallejo, C. (2016). Attribute-wise vs. alternative-wise mechanism in intertemporal
choice: Testing the proportional difference, trade-off, and hyperbolic models. Decision, 3(3), 190–215.
doi:10.1037/dec0000046.
Dawes, R. M. (1964). Social selection based on multidimensional criteria. The Journal of Abnormal and
Social Psychology, 68(1), 104–109. doi:10.1037/h0047832.
Dawes, R. M. (1979). The robust beauty of improper linear models in decision making. American
Psychologist, 34(7), 571–582. doi:10.1037/0003-066X.34.7.571.
Doyle, J. R. (2013). Survey of time preference, delay discounting models. Judgment and Decision Making,
8(2), 116–135. doi:10.2139/ssrn.1685861.
Duchowski, A. T. (2017). Eye tracking methodology: Theory and practice (3rd ed.). New York: Springer.
Einhorn, H. J. (1970). The use of nonlinear, noncompensatory models in decision making. Psychological
Bulletin, 73(3), 221–230. doi:10.1037/h0028695.
Ericson, K. M. M., White, J. M., Laibson, D., & Cohen, J. D. (2015). Money earlier or later? Simple
heuristics explain intertemporal choices better than delay discounting. Psychological Science, 26(6), 826–
833. doi:10.1177/0956797615572232.
Fishburn, P. C. (1974). Lexicographic orders, utilities and decision rules: A survey. Management Science,
20(11), 1442–1471. doi:10.1287/mnsc.20.11.1442.
Gigerenzer, G., Todd, P. M., & the ABC Research Group (Eds.) (1999). Simple heuristics that make us smart.
New York: Oxford University Press.
González-Vallejo, C. (2002). Making trade-offs: A probabilistic and context-sensitive model of choice
behavior. Psychological Review, 109(1), 137–155. doi:10.1037//0033-295X.109.1.137.
González-Vallejo, C., & Reid, A. A. (2006). Quantifying persuasion effects on choice with the decision
threshold of the stochastic choice model. Organizational Behavior and Human Decision Processes, 100(2),
250–267. doi:10.1016/j.obhdp.2006.02.001.
González-Vallejo, C., Reid, A. A., & Schiltz, J. (2003). Context effects: The proportional difference
model and the reflection of preference. Journal of Experimental Psychology, Learning, Memory & Cognition,
29(5), 942–953. doi:10.1037/0278-7393.29.5.942.
Johnson, E. J., Payne, J. W., Bettman, J. R., & Schkade, D. A. (1989). Monitoring information processing and
decisions: The Mouselab system (Report No. 89–4). Available at: www.dtic.mil/docs/citations/ADA205963
Keeney, R. L., & Raiffa, H. (1976). Decisions with multiple objectives: Preferences and value tradeoffs.
New York: John Wiley & Sons.
Kohli, R., & Jedidi, K. (2007). Representation and inference of lexicographic preference models and their
variants. Marketing Science, 26(3), 380–399. doi:10.1287/mksc.1060.0241.
Konstantinidis, E., van Ravenzwaaij, D., & Newell, B. R. (2017). Exploring the decision dynamics of risky
intertemporal choice. In G. Gunzelmann, A. Howes, T. Tenbrink, & E. J. Davelaar (Eds.), Proceedings of
the 39th Annual Conference of the Cognitive Science Society (pp. 694–699). Austin, TX: Cognitive Science
Society.
Leland, J. W. (1994). Generalized similarity judgments: An alternative explanation for choice anomalies.
Journal of Risk and Uncertainty, 9(2), 151–172. doi:10.1007/BF01064183.
Leland, J. W. (1998). Similarity judgments in choice under uncertainty: A reinterpretation of the
predictions of regret theory. Management Science, 44(5), 659–672. doi:10.1287/mnsc.44.5.659.
Leland, J. W. (2002). Similarity judgments and anomalies in intertemporal choice. Economic Inquiry, 40(4),
574–581. doi:10.1093/ei/40.4.574.
Leland, J. W. (2013). Equilibrium selection, similarity judgments, and the “nothing to gain/nothing to
lose” effect. Journal of Behavioral Decision Making, 26(5), 418–428. doi:10.1002/bdm.1772.
Lohse, G. L., & Johnson, E. J. (1996). A comparison of two process tracing methods for choice tasks.
Organizational Behavior and Human Decision Processes, 68(1), 28–43. doi:10.1006/obhd.1996.0087.
Payne, J. W. (1976). Task complexity and contingent processing in decision making: An information search
and protocol analysis. Organizational Behavior and Human Performance, 16(2), 366–387. doi:10.1016/
0030-5073(76)90022-2.

251
252

Francine W. Goh and Jeffrey R. Stevens

Payne, J. W., Bettman, J. R., & Johnson, E. J. (1993). The adaptive decision maker. New York: Cambridge
University Press.
Rahimi-Golkhandan, S., Garavito, D. M. N., Reyna-Brainerd, B. B., & Reyna, V. F. (2017). A fuzzy-trace
theory of risk and time preferences in decision making: Integrating cognition and motivation. In J. R.
Stevens (Ed.), Impulsivity: How time and risk influence decision making, Nebraska Symposium on Motivation
(vol. 64, pp. 115–144). New York: Springer. doi:10.1007/978-3-319-51721-6_4.
Read, D., Frederick, S., & Scholten, M. (2013). DRIFT: An analysis of outcome framing in intertemporal
choice. Journal of Experimental Psychology, Learning, Memory & Cognition, 39(2), 573–588. doi:10.1037/
a0029177.
Reeck, C., Wall, D., & Johnson, E. J. (2017). Search predicts and changes patience in intertemporal choice.
Proceedings of the National Academy of Sciences, 114(45), 11890–11895. doi:10.1073/pnas.1707040114.
Regenwetter, M., Cavagnaro, D. R., Popova, A., Guo, Y., Zwilling, C., Lim, S. H., & Stevens, J. R.
(2018). Heterogeneity and parsimony in intertemporal choice. Decision, 5(2), 63–94. doi:10.1037/
dec0000069.
Reyna, V. F. (2012). A new intuitionism: Meaning, memory, and development in fuzzy-trace theory.
Judgment and Decision Making, 7(3), 332–359.
Reyna, V. F., & Brainerd, C. J. (1995). Fuzzy-trace theory: An interim synthesis. Learning and Individual
Differences, 7(1), 1–75. doi:10.1016/1041-6080(95)90031-4.
Reyna, V. F., & Brainerd, C. J. (2011). Dual processes in decision making and developmental neurosci-
ence: A fuzzy-trace model. Developmental Review, 31(2–3), 180–206. doi:10.1016/j.dr.2011.07.004.
Reyna, V. F., Estrada, S. M., DeMarinis, J. A., Myers, R. M., Stanisz, J. M., & Mills, B. A. (2011).
Neurobiological and memory models of risky decision making in adolescents versus young adults.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 37(5), 1125–1142. doi:10.1037/
a0023943.
Reyna, V. F., & Farley, F. (2006). Risk and rationality in adolescent decision making: Implications for
theory, practice, and public policy. Psychological Science in the Public Interest, 7(1), 1–44. doi:10.1111/
j.1529-1006.2006.00026.x.
Reyna, V. F., & Lloyd, F. J. (2006). Physician decision making and cardiac risk: Effects of knowledge,
risk perception, risk tolerance, and fuzzy processing. Journal of Experimental Psychology: Applied, 12(3),
179–195. doi:10.1037/1076-898X.12.3.179.
Reyna, V. F., & Wilhelms, E. A. (2017). The gist of delay of gratification: Understanding and predicting
problem behaviors. Journal of Behavioral Decision Making, 30(2), 610–625. doi:10.1002/bdm.1977.
Rubinstein, A. (1988). Similarity and decision-making under risk (is there a utility theory reso-
lution to the Allais paradox?). Journal of Economic Theory, 46(1), 145–153. doi:10.1016/
0022-0531(88)90154-8.
Rubinstein, A. (2003). “Economics and psychology?” The case of hyperbolic discounting. International
Economic Review, 44(4), 1207–1216. doi:10.1111/1468-2354.t01-1-00106.
Russo, J. E., & Dosher, B. A. (1983). Strategies for multiattribute binary choice. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 9(4), 676–696. doi:10.1037/0278-7393.9.4.676.
Russo, J. E., & Rosen, L. D. (1975). An eye fixation analysis of multialternative choice. Memory &
Cognition, 3(3), 267–276. doi10.3758/BF03212910.
Scholten, M., & Read, D. (2010). The psychology of intertemporal tradeoffs. Psychological Review, 117(3),
925–944. doi:10.1037/a0019619.
Scholten, M., Read, D., & Sanborn, A. (2014). Weighing outcomes by time or against time? Evaluation
rules in intertemporal choice. Cognitive Science, 38(3), 399–438. doi:10.1111/cogs.12104.
Schulte-Mecklenbeck, M., Kühberger, A., & Ranyard, R. (2011). A handbook of process tracing methods for
decision research: A critical review and user’s guide. New York: Psychology Press.
Selten, R. (2002). What is bounded rationality? In G. Gigerenzer & R. Selten (Eds.), Bounded ration-
ality: The adaptive toolbox (pp. 13–36). Cambridge, MA: MIT Press.
Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics, 69(1),
99–118. doi:10.2307/1884852.
Simon, H. A. (1957). Models of man: Social and rational. New York: John Wiley & Sons.
Slovic, P. (1975). Choice between equally valued alternatives. Journal of Experimental Psychology: Human
Perception and Performance, 1(3), 280–287. doi:10.1037/0096-1523.1.3.280.
Smith, D. E. (1984). A source book in mathematics. Mineola, NY: Dover.
Stevens, J. R. (2016). Intertemporal similarity: Discounting as a last resort. Journal of Behavioral Decision
Making, 29(1), 12–24. doi:10.1002/bdm.1870.

252
253

Attribute-based choice

Stott, H. P. (2006). Cumulative prospect theory’s functional menagerie. Journal of Risk and Uncertainty,
32(2), 101–130. doi:10.1007/s11166-006-8289-6.
Tversky, A. (1969). Intransitivity of preferences. Psychological Review, 76(1), 31–48. doi:10.1037/h0026750.
Tversky, A. (1972). Elimination by aspects: A theory of choice. Psychological Review, 79(4), 281–299.
doi:10.1037/h0032955.
Tversky, A., Sattah, S., & Slovic, P. (1988). Contingent weighting in judgment and choice. Psychological
Review, 95(3), 371–384. doi:10.1037/0033-295X.95.3.371.
Willemsen, M. C. & Johnson, E. J. (2009). MouselabWEB: Monitoring information acquisition processes
on the web. Available at: www.mouselabweb.org/
Yee, M., Dahan, E., Hauser, J. R., & Orlin, J. (2007). Greedoid-based noncompensatory inference.
Marketing Science, 26(4), 532–549. doi:10.1287/mksc.1060.0213.

253
254
255

PART III

Occam’s razor
Mental monism and ecological rationality
256
257

16
BOUNDED REASON IN A
SOCIAL WORLD
Hugo Mercier and Dan Sperber

Introduction
Herbert Simon (1983, pp. 34–35) distinguished three “visions of rationality”: (1) the “Olympian
model,” which “serves, perhaps, as a model of the mind of God, but certainly not as a model
of the mind of man;” (2) the “behavioral” model, which “postulates that human rationality is
very limited, very much bounded by the situation and by human computational powers;” and
(3) the “intuitive model,” which “is in fact a component of the behavioral theory.” Bounded
rationality, with its intuitive component, is to be explained, Simon adds, in an evolutionary
perspective. Our joint work on reasoning and in particular our book The Enigma of Reason
(Mercier & Sperber, 2017) describes mechanisms of intuitive inference in general and the
mechanism of reason in a way that is quite consistent with Simon’s defense of a “bounded
rationality” approach to human reason. Like other evolutionary psychologists (in particular,
Leda Cosmides and John Tooby, see Tooby & Cosmides, 1992) and like Gerd Gigerenzer’s
‘adaptive toolbox’ approach (Gigerenzer, 2007; Gigerenzer, Todd, & ABC Research Group,
1999), we don’t see bounded rationality as an inferior version of Olympian rationality, nor do
we think that human or other animal inferences should be measured against abstract rationality
criteria. Our distinct contribution is to argue that there is an evolved mechanism that can rea-
sonably be called “reason,” the function of which is to address problems of coordination and
communication by producing and evaluating reasons used as justifications or as arguments in
communicative interactions.
Other approaches to human reasoning make an important but less comprehensive use of the
idea of bounded rationality. Tversky and Kahneman’s heuristics and biases program (Gilovich,
Griffin, & Kahneman, 2002) is a case in point. The heuristics studied by Tversky and Kahneman
rely on regularities in the environment to make broadly sound decisions. For instance, every-
thing else equal, more salient information is more likely to be relevant, making the availability
heuristic sensible. However, this tradition has mostly focused on the biases and errors that result
from the use of heuristics (Kruger & Savitsky, 2004), suggesting that an alternative way of pro-
cessing information could lead to superior results.
The heuristics and biases program has now become largely integrated into the dual pro-
cess paradigm, along with other strands of research, from social psychology to reasoning—
and the dual process theory has become dominant (see Melnikoff & Bargh, 2018). In this

257
258

Hugo Mercier and Dan Sperber

paradigm, the mind is split into two types of processes. System 1 processes (intuitions) are fast,
effortless, unconscious, and prone to systematic mistakes. System 2 processes (reasoning) are
slow, effortful, conscious, and able to correct System 1’s mistakes (e.g., Evans, 2007; Kahneman,
2003; Stanovich, 2004; for a more recent take on dual process models, see, e.g., Evans &
Stanovich, 2013).
By and large, System 1 behaves as would be expected by models of bounded rationality: satis-
ficing, relying on heuristics. System 2, by contrast, is closer to the ideal rational agent, able to
correct any type of mistake made by System 1, to follow normative guidelines and yield strictly
rational decisions.
However, it’s far from clear how System 2 could possibly be a maximizer rather than a mere
satisficer. The arguments put forward by Simon in favor of a view of rationality as bounded
should apply to both System 1 and System 2. Relatedly, evolutionary psychologists have
pointed out that cognitive mechanisms tend to specialize, allowing them to incorporate relevant
environmental regularities and function more efficiently. System 2, by contrast, would be the
ultimate generalist, able to fix the mistakes made by countless different System 1 processes, from
social to statistical heuristics. Unsurprisingly, then, dual process models have been plagued by
conceptual difficulties. How does System 2 know when to override System 1? How is System 2
able to find the appropriate reasons to counteract each and every System 1 heuristic? (see, e.g.,
De Neys, 2012; Osman, 2004).
Moreover, talking, for the sake of argument, in terms of Systems 1 and 2, it is not hard
to show that System 2 fails at performing a function it couldn’t fairly be expected to per-
form. By and large, solitary reasoning doesn’t correct mistaken intuitions. The large share of
participants—often the majority, sometimes the overwhelming majority—that do not provide
the normatively correct answer to simple decision making or reasoning problems reflect a
massive System 2 failure. After all, in most cases, the answer should be easily available to the
participants.
Consider the well-known bat and ball problem (Frederick, 2005):

A bat and a ball cost $1.10 together.

The bat costs $1 more than the ball.
How much does the ball cost?

Most participants—college students well able to solve this trivial mathematical problem—
give the intuitive but wrong answer of 10c, a blatant System 2 failure. Recent experiments
have revealed an even more dire picture for standard dual process models, showing that of
those participants who end up providing the correct answer (5c), the majority was able to
do so immediately after seeing the problem, and so (from the point of view of dual process
theory) through System 1 processes. Overall, only a few percent of the participants behaved
as expected in standard dual process model, making an intuitive mistake and later correcting it
thanks to System 2 (Bago & De Neys, 2019). The same pattern has been observed using other
classic reasoning and decision-making problems and other methods (Bago, 2018; Bago & De
Neys, 2017).
The reason why System 2, as a rule, fails to correct mistaken intuitions is even more of an
indictment of standard dual process models. Instead of looking for reasons why our intuitions
might be mistaken, or to look for reasons supporting alternative answers, System 2 suffers from
a massive myside bias (or confirmation bias, see Mercier, 2016a). Once someone has an intu-
ition about what the correct decision is, their System 2 mostly finds reasons supporting this
intuition. Moreover, System 2 is lazy when it comes to evaluating our own reasons, not making

258
259

Bounded reason in a social world

sure that the reasons we find to support our intuitions are particularly foolproof (Mercier,
Bonnier, & Trouche, 2016; Trouche, Johansson, Hall, & Mercier, 2016). Confirmation bias and
laziness were supposed to be System 1, not System 2 features, but the evidence says otherwise.
In light of these results, and of the fundamental problems that affect standard dual process
models, we have suggested an alternative theory that is much more in line with the ideas of
bounded rationality (Mercier & Sperber, 2011, 2017). Instead of dividing the mind between
System 1—which would mostly abide by the dictates of bounded rationality—and System
2—which wouldn’t—we suggest that it is intuitions all the way up (see also Kruglanski &
Gigerenzer, 2011; Osman, 2004). However, some of these intuitions bear on reasons, as
explained presently. In this theory, reason heavily relies on its cognitive and social environment
to solve problems—when to kick in, how to figure out if something is a good reason, and how
to find relevant reasons.

An interactionist view of reason

Our account—dubbed the interactionist view of reason—differs from most existing accounts
of reason—in particular, standard dual process models—on two grounds: what reason is, and
what reason is for.
Following the arguments put forward by evolutionary psychologists (e.g., Sperber, 1994;
Tooby & Cosmides, 1992), we suggest that the mind is a collection of cognitive mechanisms,
or modules, that function each in an autonomous manner and that are related to one another
by input/output relationships. Many of these modules perform inferences in which they
process an input and informationally enrich or transform it in an epistemically sound way.
Some of these inferential processes are completely unconscious—neither the input nor the
output is consciously accessed. For others, the output is conscious and is experienced as an
intuition (Thomson 2014). For example, the attribution to a speaker of an ironical intent is
the conscious output of intuitive mechanisms of verbal comprehension (Wilson & Sperber,
2012). As suggested by this example, some intuitions are metarepresentational: they take as
input, and/or deliver as output, representations of representations (see Sperber, 2000). The
two best-studied such mechanisms deal, one, with attribution of mental states such as beliefs
and desire and, the other, with comprehension, that is, with the attribution of speaker’s
meaning.
We suggest that reason is another of these metarepresentational mechanisms. It takes several
representations as inputs—the premises and conclusion of an argument—and delivers as output
an intuitive metarepresentational judgment of the quality of the support relation between
premises and conclusion. Crucially, this means that reason is ‘just’ another cognitive mechanism
among many others, which shares all the typical traits of intuitions, being fast, effortless, and
involving no consciousness of the process through which the outputs are arrived at. Consider
an everyday argumentative discussion—about which restaurant to go to, or which printer to
buy. Reasons are generated and evaluated quickly and effortlessly, and, most of the time, with
no insight as to why such and such reason was chosen, or why a given reason was found to
be good or bad. Sometimes we have higher-order reasons for finding a lower reason good or
bad. Such higher-order insight into the strength of lower-order reasons is, however, not to be
confused with an introspective access to the inferential process itself.
When we entertain higher-order reasons for our lower-order ones, or when we use a chain
of reasons, this typically involves an effort of working memory. This role of working memory
has sometimes be taken to be diagnostic of reasoning proper (e.g., Evans, 2003). This kind of
memory effort, however, is not really different from that of some non-reasoning tasks such as

259
260

Hugo Mercier and Dan Sperber

looking for one’s keys left somewhere in the flat and trying to avoid both overlooking a possible
place and looking twice in the same place. Unlike keeping in mind a hierarchy or a chain of
reasons, the production and evaluation of each individual reason in such a hierarchy or chain
are relatively effortless, as are all individual intuitive processes.
When sequences of reasons are deployed in a dialogic form, they are distributed among
interlocutors, doubly diminishing the memory task: on the one hand, the individual who
first invokes a given reason is motivated to bring it back into the discussion as long as it is still
relevant; on the other hand, in the mind of the interlocutors, each reason is associated with its
producer, providing, for remembering each individual argument, an analog of the method of
loci in classical mnemonics. By contrast, in solitary reasoning these memory props are not really
available. True, a solitary reasoner can imaginatively produce an internal debate of sorts. Such
an ersatz is unlikely to offer the same mnemonic and dialectic benefits as a real debate. Even
more relevantly here, the fact that solitary reasoning typically resorts to imagining a dialogue
(just as solitary sexuality typically resorts to imagining interactive sex), underscores the second
originality of our view: the claim that reason not only performs social functions (as suggested,
for instance, by Billig, 1996; Gibbard, 1990; Haidt, 2001; Kuhn, 1992; Perelman & Olbrechts-
Tyteca, 1958) but actually evolved to do so.
For a very long time, humans and their ancestors have evolved not only in very social
environments, but more specifically in very cooperative environments, collaborating to per-
form a variety of tasks, from hunting to raising children. This high degree of collaboration
was sustained by partner choice: people competing to be seen as reliable cooperation part-
ners, so they would be included in more collaborations (Baumard, André, & Sperber, 2013).
This means that individuals had strong incentives to maintain a good reputation as competent
and diligent collaborators (Sperber & Baumard, 2012). One way of doing so is by justifying
our actions. Whenever we do something that might look irrational or immoral, we can offer
justifications to change the audience’s mind. Figuring out whether other people’s actions really
are irrational or immoral involves taking into account their point of view. Hence, the audience
has an incentive to listen to these justifications and, if they are found to have some merit, adapt
their judgments accordingly.
Another consequence of humans’ high degree of cooperation is their unprecedented
ability to communicate. However, communication is a risky business, as one might be misled,
manipulated, or lied to (see Maynard Smith & Harper, 2003). Humans have evolved a suite
of cognitive mechanisms to limit these risks, for instance by calibrating their trust in different
people as a function of their competence and diligence (Mercier, 2017, 2020; Sperber et al.,
2010). However, trust calibration has limits. There are situations in which a piece of informa-
tion is too important to be accepted just on trust; there are situation where communicating a
piece of genuine information that could be misinterpreted may involve too much of a risk to
one’s reputation of trustworthiness. The best way to safeguard communication in such situ-
ations is to appeal not to authority and trust but to properties of the content of the information
itself: in particular, its consistency with what is already accepted as true. Displaying this con-
sistency is precisely what reasons do. In an exchange of arguments, speakers provide reasons to
support their point of view, and their interlocutors evaluate these reasons to decide whether or
not they should change their minds.
The interactionist view of reason claims that human reason is a metarepresentational cog-
nitive mechanism, or module, used to produce reasons and to evaluate them (for a detailed
account of this modularist approach, see Mercier & Sperber, 2017; for a defense against some
common objections, see, Sperber & Mercier, 2018). Reason evolved because it allowed indi-
viduals to find justifications and arguments, and to evaluate the justifications and arguments

260
261

Bounded reason in a social world

offered by others, thereby overcoming the limitation of trust and allowing better coordination
and communication.

When is reason triggered?

In our model, the most basic triggers of reason are social. When it comes to evaluating others’
reasons, the trigger is simply being presented with a statement as a reason for some conclusion.
That a statement should be understood as a reason for a given conclusion can be made explicit
(for example, with connectives) but is often left implicit. When it comes to producing reasons,
the basic trigger is the expression by others of criticism or doubt about our own actions or
opinions (which triggers a search for justifications) or of disagreement (which triggers a search
for arguments). Again, criticism or disagreement can be expressed more or less strongly and
explicitly (from “that was stupid” to a frown, or even a lack of clear agreement).
While these basic triggers may be entirely external, with experience, we learn to internalize
them, in particular by anticipating the need to offer reasons. For instance, adults almost always
justify actions before being questioned about them (Malle, 2004). Even young children are, to
some extent, able to anticipate the need for justifications. Three- and five-year-olds are more
likely to offer an explicit reason supporting an unconventional choice than a conventional one
(Köymen, Rosenbaum, & Tomasello, 2014). The mechanisms that allow us to anticipate the
need to offer reasons are distinct from reason itself, encompassing a variety of social cognitive
mechanisms.
Looking for reasons in anticipation of having to provide them can have several consequences.
If we start from a strong intuition, this anticipatory search of reasons typically yields a series of
poorly examined reasons supporting our initial intuition, making us more confident or driving
us towards more extreme views (Koriat, Lichtenstein, & Fischhoff, 1980; Tesser, 1978). If we
start from weak or conflicting intuitions, this anticipatory search of reasons is likely to drive us
toward the intuition which is easiest to justify, whether it is also the best by others’ standards or
not (a phenomenon known as reason-based choice, see Simonson, 1989; Shafir, Simonson, &
Tversky, 1993; and Mercier & Sperber, 2011 for review).
Another consequence of our tendency to anticipate the need for reasons is to make us aware
that some pieces of information we encounter may become relevant as reasons. For example, if
you often discuss politics with your friends, you’ll find information that supports your political
views relevant as reasons. This anticipation of the need for reasons is likely the main mechanism
going against our general preference for information that clashes with our priors (and which is,
by definition and everything else being equal, more informative). Anticipatory uses of reason
would thus be driving selective exposure (our tendency, in some contexts, to look for informa-
tion that supports our beliefs, see, e.g., Smith, Fabrigar, & Norris, 2008).
A substantial amount of evidence shows that anticipatory reason is triggered, and reinforced
by social cues. For example, being accountable—when participants are told they will have to
publicly justify their decisions—tends to reinforce reason-based choice (e.g., Simonson, 1989).

How does reason recognize good reasons?

Once reason has been triggered by the encounter with a reason—when someone provides us
with an argument to change our mind, or when we read something in the newspaper that we
could remember as a reason to defend our views—its task is to evaluate how good a reason it
is: does the premise effectively support the conclusion? What makes this task difficult is that
reasons can be about anything: whether abortion should be forbidden, which restaurant to go

261
262

Hugo Mercier and Dan Sperber

to, whether plate tectonics is an accurate theory, and so on and so forth. There cannot be gen-
eral rules for what makes a good reason. The two most popular explicit means of evaluating
arguments—logic and argumentation fallacies—are at best rough approximations, and at worst
misleading tools (Boudry, Paglieri, & Pigliucci, 2015; Mercier, Boudry, Paglieri, & Trouche,
2017; Mercier & Sperber, 2017).
Instead, the evaluation of reasons has to be made on a case-by-case basis. To do this, reason
relies neither on a general logic nor on a general probability calculus, but on a metacognitive
capacity. Reason is both a metarepresentational and a metacognitive device (Sperber & Mercier,
2018). Whenever we encounter a piece of information presented as a reason, we check whether
we would intuitively derive from what is offered as a reason the conclusion the reason purports
to support. For example, if your colleague tells you: “I’m sure Bill is here, I saw his car in the
car park not five minutes ago,” you have intuitive access to the fact that you would draw the
conclusion that Bill is here from his car having been seen in the car park five minutes ago.
The more confident you would be of this conclusion, the stronger you intuit the reason given
by your colleague to be.
It might seem, then, that reason doesn’t add anything of value. After all, if you are disposed
to infer that Bill is here from the information that your colleague saw his car in the car park,
what benefit is there, over and above drawing this conclusion, in interpreting your colleague’s
utterance as a reason for this conclusion and moreover in evaluating this reason? Actually, there
are several benefits. First, from the fact that your colleague saw Bill’s car five minutes ago, you
could have inferred many diverse conclusions, for instance, that Bill’s car is not in the garage,
that your colleague looked at the car park, that the car park is not empty, and so on indefinitely.
By offering this fact as a reason for this particular conclusion, your colleague indicated the way
in which this fact was relevant in the situation.
Second, reason allows you to check whether a reason would be good independently of
whether you accept the premise as true or not. For example, you might not be convinced that
your colleague is telling the truth in claiming to have seen Bill’s car—he might be in bad faith
or he might have confused someone else’s car for Bill’s—but still recognize that if what he says
were true, it would be a good reason for the conclusion that Bill is here. Obviously, this ability
to evaluate how good an argument would be if its premise were true is critical in science, as
it allows us to tell how evidence not yet acquired would bear on various hypotheses. Third,
evaluating the quality of reasons as reasons allows us to draw inferences about the speaker’s
competence. People who offer poor reasons don’t just fail to change their interlocutor’s mind,
they might also be perceived as less competent—and conversely for people who offer par-
ticularly good reasons. Fourth, by using reasons, speakers commit, within limits, to what
they deem to be good reasons. Someone who uses a given reason to support their ideas,
but refuses a similar reason challenging their ideas, appears inconsistent. Fifth, understanding
that a statement is presented as a reason helps guide our own search for reasons, as explained
presently.

How does reason find reasons?

In the interactionist view, reason fulfills its function by making it possible to produce and
to evaluate reasons. Reasons are produced to justify or to convince. Should we then expect
reason to be able to find, from the get-go, very strong justifications and arguments, to be able
to form long, sophisticated, well-formed pleas that anticipate most potential rebuttals? As
psychologists (and others) have long noted, this is not what typically happens (see, e.g., Kuhn,

262
263

Bounded reason in a social world

1991). Instead, when asked to justify their positions, people tend to produce relatively shallow,
superficial reasons—the first thing that comes to mind, or close to it. Why aren’t we better
at finding good reasons? The interactionist approach provides a twofold answer (see Mercier
et al., 2016).
First, it would be tremendously difficult for an individual to reliably anticipate on her own
how effectively her reasons would sway her interlocutor (see Mercier, 2012). In most cases,
good reasons are highly context-dependent. Imagine you’re trying to convince a friend to
go to a given restaurant. An ideal argument would take into account your friend’s culinary
preferences, the kind of restaurant they’ve recently been to, how much money they are willing
to spend, how far they’d go, and so on and so forth. Anticipating which argument in favor of
your choice of restaurant will prove convincing and which will be rebuffed is hard.
Second, fortunately, our interlocutors can help us find the most relevant arguments to con-
vince them, obviating the need to do so on our own. In a typical informal discussion, the
cost of having our first argument rejected is low (unless the argument is particularly dim).
For instance, you might try to convince a friend by telling them “this restaurant makes great
cocktails,” but they reply that they don’t feel like drinking tonight. Not only did you bear
no costs for failing to convince them, but thanks to this information, you can narrow down
your search for arguments, or offer a direct counter-argument—“they also make great virgin
cocktails.” The production of reasons, as we describe it, is a paradigmatic example of a satisficing
process. The criterion to be reached is that of producing reasons good enough to convince
one’s interlocutor.
For the interactionist account, reason evolved by being used chiefly in dialogic contexts, in
which we can benefit from the back and forth of discussion to refine our arguments, instead
of attempting to anticipate through extraordinary computational force what the silver bullet
might be. This means, however, that people are not well prepared to produce strong reasons in
the absence of feedback (except in the relatively rare circumstances where such feedback can
be reliably anticipated). In ordinary circumstances, people are likely to come up with the same
kind of relatively shallow reasons that work well to open a discussion, rather than attempting to
imagine a variety of potential counter-arguments and trying to pre-empt them. This explains
why people tend to be lazy—as mentioned above—when evaluating their own reasons, and
why, partly as a result, they often fail to correct their mistaken intuitions when reasoning on
their own.
Besides providing us with direct feedback guiding our search for reasons, the social envir-
onment offers other opportunities for improving one’s production of reasons. One simple
opportunity consists in recycling reasons provided by others, and that we found to be good
reasons. Pupils have been observed to pick up on the argument forms used by their classmates
(Anderson et al., 2001). Participants who have been convinced to accept the logical answer to a
reasoning problem (such as the bat and ball) are then able to reconstruct the argument in order
to convince other participants (Claidière, Trouche, & Mercier, 2017).
Still, even if the social environment provides many learning opportunities, producing
good reasons to justify oneself and convince others is harder than evaluating reasons provided
by others. Individual variations are greater in the production than in the evaluation of
reasons. The production of reasons can sometimes benefit from their evaluation “off-line”
done to pretest their strength, but the converse doesn’t seem to be true; to evaluate reasons,
you don’t have to imagine yourself producing them. In general, the study of the production
of reasons presents a stronger challenge to research on reason than the evaluation of reason
(Mercier, 2012).

263
264

Hugo Mercier and Dan Sperber

Reason with limited resources works well in the right social setting
In the interactionist account, reason is not a mechanism or a system superior to intuition.
It is a mechanism of intuition about reasons. Like all cognitive mechanisms, its benefits are
weighted against its cost. It evolved under a pressure to optimize not its cognitive benefits but
its cost-benefit ratio relative to that of other cognitive mechanisms with which it competes
for processing resources. Outside of communicative interaction, we claim, the deployment of
reason is unlikely to provide an adequate cost-benefit ratio. The social environment provides
cues regarding when reason should be triggered, and how to find reasons appropriate in the
situation.
Only in the right social setting can one expect reason to function well: people should
help each other find progressively better reasons (see Resnick, Salmon, Zeitz, Wathen, &
Holowchak, 1993), bad reasons should be shot down, and good ones carry the day. What is
the right social setting? People who have time to talk with each other, ideally in small groups
(Fay, Garrod, & Carletta, 2000), who share some common goals—otherwise, like poker players,
they have no incentive to communicate in the first place—people, who share many inferential
procedures—otherwise they can’t understand each other’s reasons—and who disagree on some
point—otherwise reasons supporting the consensual view risk piling up unexamined, leading
to group polarization or to groupthink (Janis, 1982).
A considerable amount of evidence shows that, in such social setting, reason does indeed
function well. In particular, the best ideas present in a group can spread, through discussion,
until everyone accepts them. Good insights can even be combined to form a better conclu-
sion than what even the best group members would have been able to reach on their own.
These positive outcomes are well-established when small groups discuss accessible logical or
mathematical problems, such as the bat and ball (Moshman & Geil, 1998; Laughlin, 2011;
Trouche, Sander, & Mercier, 2014; Claidière et al., 2017). They extend well to a variety of
other problems: inductive problems, any sort of academic problem faced by students in schools,
but also economic predictions, lie detection, medical diagnoses, and more (for reviews, see
Mercier & Sperber, 2011; Mercier, 2016b). Even when ascertaining what the best answer is can
be difficult, the exchange of reasons seems to point in the right direction. Juries (mock juries
at least) make better-informed verdicts, more in line with the opinion of specialists (Hastie,
Penrod, & Pennington, 1983). Citizens debating policies in the context of deliberative dem-
ocracy experiments usually end up more enlightened, with a better grasp of the issues, and
more moderate opinions (e.g., Fishkin, 2009; for reviews, see Mercier & Landemore, 2012;
Gastil, 2018).

Conclusion: a bounded reason mechanism?

We share with Cosmides and Tooby and Gigerenzer the view that human rationality is bounded
through and through: there is no System 2 or other superior mechanism in the human mind
that aims at approximating Olympian rationality. We don’t believe, however, that an evolu-
tionary approach that recognizes the fully bounded character of human rationality is committed
to seeing the faculty of reason hailed by philosopher as a wholly non-existent mechanism, a
kind of psychological phlogiston. Where philosophers erred was in overestimating the power of
reason and in misrepresenting its function as a prodigious enhancement of individual cognition.
There is, we have argued, a specialized mechanism that produces metarepresentational and
metacognitive intuitions about reasons and that, in spite of major differences, is the best true
match for reason as classically understood. Hence we call this mechanism “reason.” Reasons,

264
265

Bounded reason in a social world

that is, articulated representations of facts together with the conclusion they support, are a
very rare and peculiar object in the universe. This makes reason a highly domain-specific
cognitive mechanism. Even so, reason indirectly provides a form of virtual domain generality.
While the intuitions provided by reason are only about reasons, these reasons themselves can
be about anything that humans can think about. In this respect, reason is comparable to lin-
guistic competence: a very specialized competence that, however, makes it possible to produce
and understand utterances about anything. This similarity between language and reason is not
an accident: reason is mostly deployed by linguistic means and exploits the virtual domain-
generality of language itself.
Reasons, we argue, are not tools for individual, solitary thinking; they are tools for social
interaction aimed at justifying oneself or at convincing others and at evaluating the justifications
and arguments others present to us. In performing these functions, reason is bounded not only
by limited internal resources and external opportunities, but also by obstacles to the social flow
of information. In situations of cooperative dialogue, reason can help overcome these obstacle
and foster convergence. In the case of entrenched antagonisms, on the other hand, reason can
foster polarization.

Acknowledgments
Hugo Mercier’s work is supported by grants from the Agence Nationale de la Recherche
(ANR-10-LABX-0087 to the IEC and ANR-10-IDEX-0001–02 to PSL). Dan Sperber’s
work is supported by the European Research Council under the European Union’s Seventh
Framework Programme (FP7/2007–2013) / ERC grant agreement n° [609819], SOMICS.

References
Anderson, R. C., Nguyen-Jahiel, K., McNurlen, B., Archodidou, A., Kim, S., Reznitskaya, A., …
Gilbert, L. (2001). The snowball phenomenon: Spread of ways of talking and ways of thinking across
groups of children. Cognition and Instruction, 19(1), 1–46.
Bago, B. (2018). Testing the corrective assumption of dual process theory in reasoning (PhD thesis). Paris
Descartes, Paris.
Bago, B., & De Neys, W. (2017). Fast logic?: Examining the time course assumption of dual process
theory. Cognition, 158, 90–109.
Bago, B., & De Neys, W. (2019). The smart System 1: Evidence for the intuitive nature of correct
responding on the bat-and-ball problem. Thinking & Reasoning 25(3), 257–299.
Baumard, N., André, J. B., & Sperber, D. (2013). A mutualistic approach to morality: The evolution of
fairness by partner choice. Behavioral and Brain Sciences, 36(1), 59–78.
Billig, M. (1996). Arguing and thinking: A rhetorical approach to social psychology. Cambridge: Cambridge
University Press.
Boudry, M., Paglieri, F., & Pigliucci, M. (2015). The fake, the flimsy, and the fallacious: Demarcating
arguments in real life. Argumentation, 29(4), 431–456.
Claidière, N., Trouche, E., & Mercier, H. (2017). Argumentation and the diffusion of counter-intuitive
beliefs. Journal of Experimental Psychology: General, 146(7), 1052–1066.
De Neys, W. (2012). Bias and conflict: A case for logical intuitions. Perspectives on Psychological Science,
7(1), 28–38.
Evans, J. S. B. T. (2003). In two minds: Dual-process accounts of reasoning. Trends in Cognitive Sciences,
7(10), 454–459.
Evans, J. S. B. T. (2007). Hypothetical thinking: Dual processes in reasoning and judgment. Hove: Psychology Press.
Evans, J. S. B. T., & Stanovich, K. E. (2013). Dual-process theories of higher cognition advancing the
debate. Perspectives on Psychological Science, 8(3), 223–241.
Fay, N., Garrod, S., & Carletta, J. (2000). Group discussion as interactive dialogue or as serial mono-
logue: The influence of group size. Psychological Science, 11(6), 481–486.

265
266

Hugo Mercier and Dan Sperber

Fishkin, J. S. (2009). When the people speak: Deliberative democracy and public consultation. Oxford: Oxford
University Press.
Frederick, S. (2005). Cognitive reflection and decision making. Journal of Economic Perspectives, 19(4), 25–42.
Gastil, J. (2018). The lessons and limitations of experiments in democratic deliberation. Annual Review of
Law and Social Science, 14, 271–292.
Gibbard, A. (1990). Wise choices: Apt feelings. Cambridge: Cambridge University Press.
Gigerenzer, G. (2007). Fast and frugal heuristics: The tools of bounded rationality. In D. Koehler &
N. Harvey (Eds.), Handbook of judgment and decision making. Oxford: Blackwell.
Gigerenzer, G., Todd, P. M., & ABC Research Group. (1999). Simple heuristics that make us smart.
Oxford: Oxford University Press.
Gilovich, T., Griffin, D., & Kahneman, D. (2002). Heuristics and biases: The psychology of intuitive judgment.
Cambridge: Cambridge University Press.
Haidt, J. (2001). The emotional dog and its rational tail: A social intuitionist approach to moral judgment.
Psychological Review, 108(4), 814–834.
Hastie, R., Penrod, S., & Pennington, N. (1983). Inside the jury. Cambridge, MA: Harvard University Press.
Janis, I. L. (1982). Groupthink (2nd rev. ed). Boston: Houghton Mifflin.
Kahneman, D. (2003). A perspective on judgment and choice: Mapping bounded rationality. American
Psychologist, 58(9), 697–720.
Koriat, A., Lichtenstein, S., & Fischhoff, B. (1980). Reasons for confidence. Journal of Experimental
Psychology: Human Learning and Memory and Cognition, 6, 107–118.
Köymen, B., Rosenbaum, L., & Tomasello, M. (2014). Reasoning during joint decision-making by pre-
school peers. Cognitive Development, 32, 74–85.
Kruger, J., & Savitsky, K. (2004). The “reign of error” in social psychology: On the real versus imagined
consequences of problem-focused research. Behavioral and Brain Sciences, 27(3), 349–350.
Kruglanski, A. W., & Gigerenzer, G. (2011). Intuitive and deliberate judgments are based on common
principles. Psychological Review, 118(1), 97.
Kuhn, D. (1991). The skills of arguments. Cambridge: Cambridge University Press.
Kuhn, D. (1992). Thinking as argument. Harvard Educational Review, 62(22), 155–178.
Laughlin, P. R. (2011). Group problem solving. Princeton, NJ: Princeton University Press.
Malle, B. F. (2004). How the mind explains behavior: Folk explanations, meaning, and social interaction.
Cambridge, MA: The MIT Press.
Maynard Smith, J., & Harper, D. (2003). Animal signals. Oxford: Oxford University Press.
Melnikoff, D. E., & Bargh, J. A. (2018). The mythical number two. Trends in Cognitive Sciences, 22(4),
280–293.
Mercier, H. (2012). Looking for arguments. Argumentation, 26(3), 305–324.
Mercier, H. (2016a). Confirmation (or myside) bias. In R. Pohl (Ed.), Cognitive illusions (2nd ed.,
pp. 99–114). London: Psychology Press.
Mercier, H. (2016b). The argumentative theory: Predictions and empirical evidence. Trends in Cognitive
Sciences, 20(9), 689–700.
Mercier, H. (2017). How gullible are we? A review of the evidence from psychology and social science.
Review of General Psychology, 21(2), 103.
Mercier, H. (2020). Not Born Yesterday: The Science of Who we Trust and What we Believe. Princeton, NJ:
Princeton University Press.
Mercier, H., Bonnier, P., & Trouche, E. (2016). Why don’t people produce better arguments? In
L. Macchi, M. Bagassi, & R. Viale (Eds.), Cognitive unconscious and human rationality (pp. 205–218).
Cambridge, MA: MIT Press.
Mercier, H., Boudry, M., Paglieri, F., & Trouche, E. (2017). Natural-born arguers: Teaching how to make
the best of our reasoning abilities. Educational Psychologist, 52(1), 1–16.
Mercier, H., & Landemore, H. (2012). Reasoning is for arguing: Understanding the successes and failures
of deliberation. Political Psychology, 33(2), 243–258.
Mercier, H., & Sperber, D. (2011). Why do humans reason? Arguments for an argumentative theory.
Behavioral and Brain Sciences, 34(2), 57–74.
Mercier, H., & Sperber, D. (2017). The enigma of reason. Cambridge, MA: Harvard University Press.
Moshman, D., & Geil, M. (1998). Collaborative reasoning: Evidence for collective rationality. Thinking
and Reasoning, 4(3), 231–248.
Osman, M. (2004). An evaluation of dual-process theories of reasoning. Psychonomic Bulletin and Review,
11(6), 988–1010.

266
267

Bounded reason in a social world

Perelman, C., & Olbrechts-Tyteca, L. (1958). The new rhetoric: A treatise on argumentation. Notre Dame,
IN: University of Notre Dame Press.
Resnick, L. B., Salmon, M., Zeitz, C. M., Wathen, S. H., & Holowchak, M. (1993). Reasoning in con-
versation. Cognition and Instruction, 11(3/4), 347–364.
Shafir, E., Simonson, I., & Tversky, A. (1993). Reason-based choice. Cognition, 49(1–2), 11–36.
Simon, H. A. (1983). Reason in human affairs. Stanford, CA: Stanford University Press.
Simonson, I. (1989). Choice based on reasons: The case of attraction and compromise effects. The Journal
of Consumer Research, 16(2), 158–174.
Smith, S. M., Fabrigar, L. R., & Norris, M. E. (2008). Reflecting on six decades of selective exposure
research: Progress, challenges, and opportunities. Social and Personality Psychology Compass, 2(1),
464–493.
Sperber, D. (1994). The modularity of thought and the epidemiology of representations. In L. A.
Hirschfeld & S. A. Gelman (Eds.), Mapping the mind: Domain specificity in cognition and culture (pp. 39–
67). Cambridge: Cambridge University Press.
Sperber, D. (2000). Metarepresentations in an evolutionary perspective. In D. Sperber (Ed.),
Metarepresentations: A multidisciplinary perspective (pp. 117–137). Oxford: Oxford University Press.
Sperber, D., & Baumard, N. (2012). Moral reputation: An evolutionary and cognitive perspective. Mind
& Language, 27(5), 495–518.
Sperber, D., Clément, F., Heintz, C., Mascaro, O., Mercier, H., Origgi, G., & Wilson, D. (2010). Epistemic
vigilance. Mind and Language, 25(4), 359–393.
Sperber, D., & Mercier, H. (2018). Why a modular approach to reason? Mind & Language, 131(4), 496–501.
Sperber, D., & Wilson, D. (1995). Relevance: Communication and cognition. New York: Wiley-Blackwell.
Stanovich, K. E. (2004). The robot’s rebellion. Chicago: Chicago University Press.
Tesser, A. (1978). Self-generated attitude change. In L. Berkowitz (Ed.), Advances in experimental social
psychology (pp. 289–338). New York: Academic Press.
Thompson, V. A. (2014). What intuitions are... And are not. In B. H. Ross (Ed.), The psychology of learning
and motivation (vol. 60, pp. 35–75). New York: Academic Press.
Tooby, J., & Cosmides, L. (1992). The psychological foundations of culture. In J. H. Barkow, L. Cosmides,
& J. Tooby (Eds.), The adapted mind: Evolutionary psychology and the generation of culture (p. 19).
New York: Oxford University Press.
Trouche, E., Johansson, P., Hall, L., & Mercier, H. (2016). The selective laziness of reasoning. Cognitive
Science, 40(8), 2122–2136.
Trouche, E., Sander, E., & Mercier, H. (2014). Arguments, more than confidence, explain the good per-
formance of reasoning groups. Journal of Experimental Psychology: General, 143(5), 1958–1971.
Wilson, D., & Sperber, D. (2012). Meaning and relevance. Cambridge: Cambridge University Press.

267
268

17
RATIONALITY
WITHOUT OPTIMALITY
Bounded and ecological rationality from a
Marrian perspective

Henry Brighton

Does rationality imply optimality?

The idea that rationality implies optimality is so widely assumed as to seem barely worth
discussing. Optimal Bayesian decision makers in the cognitive sciences, optimal foragers in
biology, and Bayesian maximizers of expected utility in economics are different faces of the
same interdisciplinary orthodoxy. Does bounded rationality offer something different? I will
argue that ecological rationality, a development of Herbert Simon’s (1957) notion of bounded
rationality, provides a fully-fledged alternative to the view that rational decisions under environ-
mental uncertainty are optimal probabilistic decisions (Gigerenzer and Selten, 2001; Gigerenzer
and Brighton, 2009; Brighton, 2019). For those familiar with the concept of ecological ration-
ality, the study of simple heuristics will likely spring to mind. The study of simple heuristics
examines how minimalist decision making algorithms that ignore information can, depending
on the structure of the environment, outperform familiar and more sophisticated algorithms
that weigh and integrate information, or explicitly optimize a rationally justified criterion
(e.g., Gigerenzer et al., 1999; Brighton, 2006; Brighton and Gigerenzer, 2007; Gigerenzer and
Brighton, 2009). Rather than focus on these findings, I will scrutinize an argument often made
by critics of ecological rationality: Insights arising from the study of ecological rationality are,
after the fact, formally reducible to instances of optimal Bayesian inference and, furthermore,
require principles of Bayesian inference to fully explain them. The claim is that ecological
rationality neither undermines the use of optimality nor challenges orthodox rationality.
I will use Marr’s (1982) levels of analysis to challenge this claim and argue that ecological
rationality does not imply optimality. Marr’s levels organize and relate functional, algorithmic,
and implementational theories of information processing systems. In Marr’s terms, simple
heuristics studied under the rubric of ecological rationality address algorithmic level concerns,
where the question is how organisms make decisions, not the functional question of what
constitutes a rational decision and why. For some theorists, this elementary distinction between
function and mechanism clarifies why ecological rationality is subservient to orthodox ration-
ality (e.g., Chater et al., 2003; Gintis, 2012). My counterargument will be developed in three
stages: (1) ecological rationality involves more than a conjecture about simple heuristics, it also
rests on a statistical formulation of the problem of inductive inference under unquantifiable

268
269

Rationality without optimality

uncertainty that neither assumes nor implies the existence of an optimal solution (Breiman,
2001; Brighton and Gigerenzer, 2007, 2012; Brighton, 2019); (2) this statistical perspective
exposes the inability of optimality to characterize functional responses to ill-posed problems,
and undermines the idea that function and mechanism can be studied independently; and
(3) ecological rationality is incompatible with and exposes the limitations of Marr’s (1982)
levels of analysis, but finds support in Marr’s (1977) largely overlooked views on the broader
functioning of biological information processing systems.

Marr and Poggio’s three levels of analysis

David Marr’s book Vision (1982) set out an innovative approach to understanding vision as an
information processing problem. As well as detailing computational models of visual processing,
Marr argued that a deeper understanding of information processing systems in general, not just
the visual system, requires three levels of analysis. Originally developed with Tomaso Poggio
(Marr and Poggio, 1976; Poggio, 2012), these three levels map onto fundamental distinctions in
the study of rationality that will recur throughout this discussion. At the highest level, termed
the computational level, the goal is to identify an appropriate formal system, a calculus, used
to define the problem and derive its solution. This abstract rendering of the problem details
what problem the system attempts to solve and why. For example, a pocket calculator solves the
problem of performing arithmetic calculations and, so the story goes, human decision makers
solve the problem of maximizing expected utility when making decisions under uncertainty.
Calculi used to develop computational level theories include Bayesian statistics (Griffiths and
Tenenbaum, 2006), deductive logic (Baggio et al., 2014), and the mathematics of signal pro-
cessing (Marr and Hildreth, 1980).
Moving down to Marr’s second level, the algorithmic level, the goal is to specify the
algorithms and data structures needed to compute the solution. For example, if part of the
computational level theory of a pocket calculator is to calculate factorials, then at the algo-
rithmic level this calculation, for
n
reasons of efficiency, might be implemented using Stirling’s
 n
approximation, n ! ≈ 2πn   , which incurs a loss in precision but requires fewer operations
e
to compute. Similarly, it has been proposed that simple heuristics might be used to approximate
a full Bayesian computation specified at the computational level (Chater et al., 2006; Oaksford
and Chater, 2009).
Marr’s third and final level of analysis is the implementation level, which considers the
constraints of physically instantiating the proposed algorithms and data structures in, say, bio-
logical machinery or a conventional digital computer. Combined, Marr’s levels consider what
problem the system attempts to solve, what the solution to this problem is, how this solution is
computed, and how this computation is realized in a physical device. Notice that each Marrian
level corresponds to a different level of analysis rather than a different level of organization.
Levels of organization in nature typically reflect part-whole relationships such as those that sep-
arate proteins from cells and individuals from groups, whereas levels of analysis separate different
forms of understanding (Bechtel and Shagrir, 2015; McClamrock, 1991). For Marr, studying an
information processing system without a formal understanding of the problem being addressed
should be avoided, and he advocated a top-down approach to developing theories of bio-
logical information processing systems. This methodology, though, does not exclude a theorist
from adopting alternative priorities when working within the Marrian framework. Constraints
from implementation and algorithmic levels can, depending on the question, also drive theory
development.

269
270

Henry Brighton

Why place Marr’s levels at the center of this discussion? To begin with, Marr’s levels of ana-
lysis are seen as “a core tenet of cognitive science” (Peebles and Cooper, 2015, p. 188), defining
the de facto categories used to orientate cognitive models in relation to rationality claims.
Bayesian optimality modeling in the cognitive sciences, for instance, seeks computational level
theories that formulate the problems faced by the cognitive system and what constitutes a
rational solution (Tenenbaum and Griffiths, 2001; Chater et al., 2006; Griffiths et al., 2012).
Rarely are these models interpreted as hypotheses about how the cognitive system computes.
Cognitive process models such as simple heuristics, on the other hand, are categorized as algo-
rithmic level theories. Both kinds of theory can be interpreted as models of human behavior,
but only the second kind involves an explicit conjecture about cognitive processing. From this
Marrian perspective, bounded rationality is typically seen as an attempt to inform computa-
tional level theory development by importing constraints arising from the algorithmic and
implementation levels. For example, Anderson’s (1990) rational analysis of cognition and the
more recent study of resource rationality (Griffiths et al., 2015) use these constraints to revise
the problem specification, and consequently, what problem to which the computational level
theory specifies an optimal response. Now, on the one hand, if we assume that simple heuristics
are to Bayesian rationality what Stirling’s approximation is to arithmetic, then as critics as have
argued, studying ecological rationality without appeal to Bayesian optimality is like trying to
understand how birds fly without a theory of aerodynamics (Chater and Oaksford, 1999) or
understanding how people play billiards without a theory of the physics of motion (Gintis,
2012). On the other hand, both Simon (1991) and advocates of ecological rationality have
rejected the proposal that the study of bounded and ecological rationality reduces to studying
optimality (e.g., Gigerenzer and Selten, 2001; Brighton, 2019). What is the basis of this dis-
agreement, and can Marr’s levels be used to clarify and resolve the issue?

The statistical foundations of ecological rationality

Ecological rationality is, and always has been, more than the study of simple heuristics. The
goal is to understand the adaptive fit between organisms and the structure of the environment
by examining three interacting components: (1) algorithmic models of how organisms make
inductive inferences, with a particular focus on simple heuristics; (2) the properties of natural
environments whose probabilistic structure is either uncertain or unknown; and (3) a for-
mulation of the problem of statistical inference that defines and quantifies the meaning of an
adaptive fit. To focus this discussion, I will sidestep experimental research examining the use of
simple heuristics in humans and other animals, and instead focus on the overarching hypoth-
esis: Simple heuristics that ignore information are a vital part of how organisms successfully
cope with the uncertainty of the natural world.
Results supporting this hypothesis are termed less-is-more effects, and they detail how min-
imalist processing strategies improve the accuracy of decisions relative to more complex and
supposedly sophisticated strategies commonly assumed in the cognitive sciences and beyond.
For example, when deciding which of two objects in the environment scores higher on some
criterion of interest, such as which of two food items has a higher calorific content, a user of
the recognition heuristic will choose the recognized object if the other object is unrecognized.
A partly ignorant decision maker using this heuristic can make more accurate inferences than a
more informed decision maker who weighs and integrates all available information (Goldstein
and Gigerenzer, 2002). Similarly, a user of the simple heuristic take-the-best ignores correlations
between cues and uses a single cue to make an inference, yet can make more accurate inferences
than a decision maker who attempts to model these correlations and weighs and integrates all

270
271

Rationality without optimality

available cues (e.g., Gigerenzer and Goldstein, 1996; Czerlinski et al., 1999; Brighton, 2006;
Brighton and Gigerenzer, 2007; Gigerenzer and Brighton, 2009; Şimşek and Buckmann, 2015).
In both cases, a simple heuristic (component 1 of the interaction above) confers a function by
exploiting properties of natural environments (component 2 of the interaction above). Of abso-
lutely crucial importance to the coming discussion is component 3 of the interaction described
above: The formulation of the statistical problem facing the organism. This third component
also requires theoretical commitments to be made, and it is these commitments that establish,
most directly, ecological rationality as a substantive alternative to orthodox rationality.

The role of statistics in constructing the rationality problem

Simple heuristics like take-the-best are supervised learning algorithms. They tackle the problem
of making inductive inferences from observations of the environment where the goal is to dis-
tinguish systematic patterns that govern the observations from any noisy and accidental patterns.
What is a rational solution to this problem? If we possess a complete probabilistic understanding
of the environmental processes that generate the observations, then the answer is simple: The
optimal Bayesian response relative to a prior and a hypothesis space expressing this probabilistic
knowledge. But what if we lack what Binmore (2009) termed this “metaphysical hotline to the
truth?” Or, what if the decision maker is not in what Savage (1954), the inventor of Bayesian
decision theory, termed a “small world” where identifying a mutually exclusive and exhaustive
set of future states of the world, consequences, and the actions that map between them is a
realistic possibility? Put simply, how should an optimal response be formulated when partial or
complete ignorance precludes probabilistic quantification of the uncertainties we face?
Orthodox responses to these questions cite uninformed priors, second-order probabilities,
imprecise probabilities, or some other technique that assumes and proposes how we should
quantify the varying degrees of uncertainty we face. Another response is to devise alternatives
to Bayesian decision theory but retain the goal of optimality (e.g., Gilboa, 1987; Gilboa and
Schmeidler, 1989). A third option is to accept that “it is sometimes more rational to admit
that one does not have sufficient information for probabilistic beliefs than to pretend that
one does” (Gilboa et al., 2012, p. 28). This third option represents the fork in the road where
orthodox rationality and ecological rationality part ways. The orthodox view proceeds under
the assumption that we can, by hook or by crook, always quantify environmental uncertainty.
The study of ecological rationality sets out to examine the implications of unquantifiable
uncertainty. The statistician Leo Breiman (2001) characterized two cultures of thought under-
lying these issues, and they map directly onto the divergent practices of orthodox rationality
and ecological rationality. Figure 17.1 depicts these relationships, which are worth considering
in detail because they prove critical in constructing the formal nature of the problem.

Orthodox rationality and the statistical culture of data modeling

If we start with observations, each relating a set of independent variables to a dependent vari-
able, then the environment can be seen as a black box containing data-generating machinery
that determines the joint distribution over the inputs to the black box (independent variables)
and outputs (the dependent variable) shown in Figure 17.1(a). Much of traditional statistical
inquiry, and this includes the practice of formulating optimal probabilistic responses to envir-
onmental uncertainty, requires that we make a conjecture about the contents of this black box,
depicted in Figure 17.1(b). We might work from the assumption that dependent and inde-
pendent variables are linearly related, for example. Taking a Bayesian perspective, we might

271
272

Henry Brighton

(a) The problem of statistical inference

x nature y

(b) Data modeling (c) Algorithmic modeling

x mθ y x unknown y

algorithm 1,
algorithm 2,
algorithm 3, ...

(d) Bayesian optimality modeling (e) Study of ecological rationality

x mθ y x unknown y
cognitive model 1,
optimal response cognitive model 2,
cognitive model 3, ...

Figure 17.1 How should the problem of statistical inference be formulated? We start with observations
of an unknown functional relationship between input vectors x (the independent variables) and output y
(the dependent variable) determined by some aspect of nature, shown in (a). Following Breiman (2001),
the statistical culture of data modeling views the conjecture of a stochastic model y = mθ(x) of nature’s
black box as an essential step in statistical inference, shown in (b). Rather than attempt to model the
contents of nature’s black box, algorithmic modeling is an incremental search for learning algorithms
that can, to varying degrees, accurately predict the input-output relationship, shown in (c). Bayesian
optimality modeling conducts data modeling in order to define an optimal response (d), while the study
of ecological rationality conducts algorithmic modeling and interprets predictive models as potential
cognitive models, shown in (e).
Source: Diagrams (a–c) adapted from Breiman (2001).

formulate an hypothesis space, prior distribution, and various parameters that we fit using the
available observations. The defining characteristic of this approach is that, at some point, a
conjecture is made about the contents of the black box, and it is relative to this conjecture that
subsequent claims about rationality and optimality are made. For instance, Anderson’s rational
analysis of cognition is a methodology for formulating the problems that the environment
poses the cognitive system, and using rational solutions to these problems to derive theories of
cognitive processing (Anderson, 1990, 1991; Chater and Oaksford, 1999). In both Anderson’s
methodology and Bayesian optimality modeling, formulating a probabilistic model of the task
environment is a key step, illustrated in Figure 17.1(d) (Tenenbaum and Griffiths, 2001; Chater
et al., 2006; Griffiths et al., 2012). Breiman (2001) used the term “data modeling” to describe
this general approach to statistical inquiry.

Ecological rationality and the statistical culture of algorithmic modeling

An alternative approach to statistical modeling is what Breiman termed algorithmic modeling.
When algorithmic modeling, we refrain from making a conjecture about the contents of the
black box and instead try to predict its behavior. As shown in Figure 17.1(c), the observations

272
273

Rationality without optimality

are used to estimate the predictive accuracy of competing models of inductive inference,
which in practice usually means comparing machine learning algorithms using a model selec-
tion criterion, such as cross-validation (Stone, 1974), where algorithms are judged on their
ability to generalize to unseen cases (Hastie et al., 2001; Bishop, 2006). Crucially, learning
algorithms and the probabilistic assumptions they imply tend not to be seen as models of the
environment, but rather inductive biases that may explicitly introduce model infidelities in
order to reduce a component of prediction error known as variance, described below (such
techniques are part of the theory of regularization; see, e.g., Chen and Haykin, 2002). Among
the algorithms being considered, one or more algorithm will achieve the greatest reduction in
prediction error. Findings like these in no way license an optimality claim. Instead, they pro-
vide an indication of the kinds of algorithmic design decisions or statistical assumptions that
improve our ability to reduce prediction error, thereby suggesting further algorithms worth
evaluating.
This process is fundamentally exploratory, yields a functional understanding of the
algorithms being considered, yet in no way invokes the concept of optimality to explain model
performance. Specifically, one can seek to improve predictive accuracy relative to existing
models without assuming that an optimal model can ever be identified, or, given the com-
plexity and uncertainty of the natural world, meaningfully exists. How, though, is ecological
rationality related to the culture of algorithmic modeling? From the outset, key results in
the study of ecological rationality have rested almost exclusively on algorithmic modeling
studies, where the ability of simple heuristics to reduce prediction error relative to other
models has been analyzed in environments for which the optimal response is unknown and
arguably unknowable (Czerlinski et al., 1999; Martignon and Laskey, 1999; Brighton, 2006;
Gigerenzer and Brighton, 2009). Such studies typically proceed by investigating model per-
formance in new environments, or new heuristics generated by combining a set of basic
building blocks that include rules for information search, rules for stopping search, and rules
for making decisions (Gigerenzer et al., 1999). Combining these building blocks, which are
often inspired by experimental studies of decision making in humans and other animals, is one
way of steering the algorithmic modeling process. In short, the study of ecological rationality,
schematized in Figure 17.1(e), proceeds by studying algorithmic functioning in environments
composed on observations rather than environments rendered as probabilistic models that
imply optimal responses. Next I will argue that in theory too, ecological rationality is not in
the business of studying optimality.

Bayesian reductionism and the limitations of optimal function

Despite the preceding points, critics have argued that when a heuristic works, one still needs
a rational explanation for why it works, and so the concept of ecological rationality is ultim-
ately subservient to Bayesian rationality. Oaksford and Chater (2009), for example, argue that
Bayesian optimality modeling “cannot be replaced by, but seeks to explain, ecological ration-
ality” (p. 110). This view extends to critics of Bayesian optimality modeling, such as Jones
and Love (2011), who argue that the two approaches are “highly compatible” because “any
inference algorithm implicitly embodies a prior expectation about the environment” (p. 186).
Thus, even though ecological rationality does without strong probabilistic conjectures and
optimality claims in practice, in theory, these concepts will be required to explain functional
success. These arguments, which express a kind of Bayesian reductionism, are worth dissecting
in turn.

273
274

Henry Brighton

Argument 1: Bayesian thinking is required to explain functional success

When a learning algorithm performs well in a particular environment, incurring lower pre-
diction error than the other algorithms being considered, how can this behavior be explained?
Seen through the lens of probabilistic optimality, there is a strong tendency to view this func-
tional success as indicating a close match between the prior distributional assumptions of
the algorithm and the probabilistic structure of the environment. An alternative pattern of
explanation that has proven critical to understanding the success of simple heuristics is to
analyze the prediction error of competing algorithms using the bias/variance decomposition
(Gigerenzer and Brighton, 2009). Specifically, we can decompose prediction error into two
controllable components, bias and variance, and one uncontrollable component, noise, such
that they combine additively as follows (O’Sullivan, 1986; Geman et al., 1992; Hastie et al.,
2001; Bishop, 2006):

Total error = (bias)2 + variance + noise (17.1)

The bias component reflects the inability of the model to capture what is systematic in the
observations. The variance component, in contrast, reflects the sensitivity of the model’s
predictions to different observations of the same problem, such as a different sample from
the same population. All algorithms face what is termed the bias/variance dilemma because
methods for reducing variance tend to increase bias, whereas methods for reducing bias tend
to increase variance (Geman et al., 1992). For example, a heuristic may suffer from high bias
due to having a single free parameter, but this simplicity can lead to a greater reduction in
variance when learning from small samples (Brighton and Gigerenzer, 2015). A competing
algorithm may strike a different trade-off between bias and variance, and whether this point
proves functional will depend on the sample size and the properties of the environment. Many
of the algorithmic tricks used by simple heuristics and other learning algorithms to incur low
prediction error are methods for reducing variance. In short, the idea that Bayesian thinking is
required to explain the performance of learning algorithms is inaccurate, both in theory and
in practice.

Argument 2: Bayesian explanations of functional success should be preferred

Even if the principles of Bayesian inference are not required to explain functional success,
perhaps, given their elegance and foundations in probability theory, they should be pre-
ferred. A counterexample to this intuition is the naïve Bayes classifier, a simple learning
algorithm that makes a strong assumption that the features (also termed cues, attributes, or
independent variables) are conditionally independent of each other, given the class. Despite
this assumption being highly unlikely to be met in practice, the naïve Bayes classifier has
a long history of performing surprising well, particularly when learning from sparse data.
Why is this? Research spanning several decades has repeatedly invoked the concepts of
bias and variance to explain why the naïve Bayes classifier performs well relative to other
algorithms: Despite introducing a strong bias, the conditional independence assumption is
often highly effective at reducing variance (e.g., Domingos and Pazzani, 1997; Friedman,
1997; Hand and Yu, 2001; Ng and Jordan, 2002; Webb et al., 2005). Thus, even when we
have a clear probabilistic formulation of a learning algorithm, it can prove more insightful
to explain functional success using the bias/variance perspective than a perspective rooted in
probabilistic optimality.

274
275

Rationality without optimality

Argument 3: All learning algorithms imply an optimal Bayesian response

Given any supervised learning algorithm, let us assume that a probabilistic reformulation of this
algorithm is always possible, such that the probabilistic conditions under which the algorithm
makes Bayes optimal decisions are made explicit. Wouldn’t this then prove that ecological ration-
ality reduces to the study of optimal Bayesian inference? No. All this proves is that one aspect of
algorithmic functioning is optimal functioning. When algorithmic modeling, learning algorithms
can, and typically do, confer a function without being optimal. They also confer a function by
virtue of their performance relative to other learning algorithms. Indeed, if the optimal response is
indeterminable, then relative function is all we have to work with. More generally, the conditions
under which an algorithm is optimal and the conditions under which it confers a function outside
these optimality conditions cannot be assumed to be directly related. Reductions in prediction
arise from a complex interaction between the learning algorithms being considered, the number
of observations available, and properties of the environment (e.g., Perlich et al., 2003). And these
complexities mean that the optimality conditions of most learning algorithms are simply not
known, and attempts to identify them tend to be only partially successful (e.g., Domingos and
Pazzani, 1997; Kuncheva, 2006). To summarize, optimality results are (1) of limited practical use
outside the study of idealized environments for the simple reason that we tend not to know
the probabilistic structure of natural environments; and (2), limited in their ability to provide a
reliable indication of how well an algorithm will perform outside its optimality conditions, and
relative to other algorithms.

The theory-dependence of Marrian decomposition

To recap, my argument is that Breiman’s (2001) statistical cultures of data modeling and algo-
rithmic modeling differ in how they view and respond to uncertainty, these differences have
direct implications for the study of rationality, and crucially, these differences are reflected
in the practices of Bayesian optimality modeling and the study of ecological rationality. The
fly in the ointment is optimality. Given all this, how can the study of ecological rationality
be reconciled with the Marrian perspective? Resolving this issue is of central importance
because any incompatibility between ecological rationality and Marr’s levels of analysis would
leave us in the precarious position of advocating a view on rationality at odds with “the
canonical scheme for organizing formal analyses of information processing systems” (Griffiths
et al., 2015, p. 217) and “one of the most well-known theoretical constructs of twentieth
century cognitive science” (Willems, 2011, p. 1). On the other hand, if ecological rationality
proves to be entirely compatible with Marr’s levels, then the existence of a computational-
level theory would suggest, as is typically assumed, that the whole approach reduces to an
optimality claim.

Ecological rationality from a Marrian perspective

Recall that the objective of a computational level theory is to identify an appropriate formal
system, a calculus, used to define the problem and derive a solution. Ultimately the computa-
tional level should define the input-output mappings to be computed, perhaps approximately,
by an algorithmic level theory. The problem we face now is that the statistical “goal” of algo-
rithmic modeling, and hence ecological rationality, is both exploratory and algorithmic in
nature, echoing Selten’s (2001) point that “bounded rationality cannot be precisely defined.
It is a problem that needs to be explored” (p. 15). In contrast to data modeling, algorithmic

275
276

Henry Brighton

modeling doesn’t supply a solution derived using a calculus but instead a relative understanding
of the efficacy of competing algorithms. The problem facing the organism remains, only this
problem is not well posed in Hadamard’s (1902) sense of requiring that a solution exists, is
unique, and is stable with respect to small perturbations to the problem definition. Nature
doesn’t always, or even often, present well-posed problems in this sense, and it is imperative
that our conception of rationality should reflect this. Taken in conjunction with Marr’s (1977)
proposal that a computational level theory should be “expressed independently of the particular
way it is computed” (pp. 37–38), the conclusion that ecological rationality is incompatible with
the demands of a computational level theory is hard to avoid.

The limits of Marrian decomposition: type-1 and type-2 theories

This conclusion, that Marr’s levels of analysis fall short as a canonical system and imply a limited
view of algorithmic functioning, is in fact consistent with Marr’s thinking when considered
in light of his 1977 article published in Artificial Intelligence (Marr, 1977). This article addresses
some of the issues discussed above, yet is largely overlooked in contemporary discussions of
Marr’s work. Here, Marr drew the distinction between biological information processing
systems with type-1 theories (those which yield to a computational-level analysis) and those
with type-2 theories (those which don’t). An archetypal type-1 theory for Marr is the Fourier
Transform, a mathematical operation central to edge detection that can be approximated at the
algorithmic level in a number of ways (Marr and Hildreth, 1980). Marr’s examples of archetypal
type-2 theories include protein unfolding and the grammar of natural language. Neither of
these problems, Marr speculates, are likely to have useful mathematical abstractions. He warns,
for instance, that “an abstract theory of syntax may be an illusion” (p. 42), such that productive
study at the computational level is likely to be questionable.
Marr’s overarching point is that we should prioritize the study of problems with type-1 the-
ories but not expect that all interesting problems will have them. This suggests three possibilities
that need to be considered. First, the theorist seeks a computational level theory, finds one,
and then invokes Marr’s three levels of analysis to guide further theory development. Second,
the theorist seek a computational level theory, fails to find one, and then pursues an alternative
form of functionalist inquiry. Third, the theorist fails to adopt a functionalist perspective in the
first place, overlooks a potentially insightful computational theory, and misses the opportunity
to arrive at the kind of understanding that Marr advocates. Marr (1977) reserves his criticism
for this third scenario, which he saw as a “waste of time” because it “amounts to studying a
mechanism not a problem” (p. 46). Advocates of ecological rationality find themselves in the
second scenario, and set out to examine what implications these problems have for the study
of rationality. In summary, the idea that theories of cognitive processing that make function-
alist claims, including claims about rationality, must be reconcilable with Marr’s three levels of
analysis rests on a narrow conception of the kinds of problem that the cognitive system is likely
to face. This category of problem was one that Marr undoubtedly saw as a priority, but he also
recognized that this category offers only a partial view of the kinds of problem tackled by bio-
logical information processing systems.

Rationality without optimality

It is commonly assumed, outside the cognitive sciences as well as within them, that “the only
viable formulation of perception, thinking, and action under uncertainty is statistical inference,
and the normative way of statistical inference is Bayesian” (Edelman and Shahbazi, 2011, p. 198).

276
277

Rationality without optimality

Furthermore, despite the existence of alternative methods for formulating the problem of stat-
istical inference, the corollary that “all such methods would ultimately reduce to Bayesian infer-
ence” (p. 198), if true, leaves little room for maneuver in the study of rationality. This is a form
of Bayesianism, the view that the uncertainty surrounding the problems faced by the cognitive
system is quantifiable, and that rational solutions to these problems are optimal probabilistic
solutions. One can dispute this view without questioning Bayesian methods, probability theory,
or mathematical statistics. The dispute concerns the application of these methods outside what
Savage (1954), the founder of Bayesian decision theory, termed “small worlds.” Savage considered
it “utterly ridiculous” (p. 16) to apply his theory in large worlds where uncertainty surrounds
which states, actions, and consequences are relevant, and where we suffer from varying degrees
of ignorance that undermine our ability to quantify the uncertainties we face.
Deductive logics in their various guises are used to define rationality in domains that demand
deductive rather than inductive reasoning, so the expectation that our conception of rationality
must adapt to fit the nature of the problem should be both familiar and uncontroversial. The
claim here is that ecological rationality is an adaptation to the problem of decision making in
large worlds. These are environments that we can observe but our partial ignorance precludes
them from being probabilistically quantifiable (Binmore, 2009; Brighton and Gigerenzer, 2012),
and a key implication of this claim is that optimality ceases to be a meaningful characteristic of
rational decisions under these conditions. Critics, in contrast, have argued that because eco-
logical rationality involves a conjecture about simple heuristics, it necessarily focuses on Marr’s
algorithmic rather than the computational level. And given that rationality claims are tradition-
ally made at the computational level, the assumption is that ecological rationality must therefore
inherit principles of Bayesian rationality to explain the success of simple heuristics. In response,
I have argued that Marr’s levels of analysis are incapable of resolving this debate because the stat-
istical commitments driving the study of ecological rationality make the problem of statistical
inference exploratory, algorithmic, and ill-posed. Marr’s levels of analysis, on the other hand,
are tailored to well-posed problems with a precise mathematical formulation enabling the der-
ivation of an optimal solution. As Marr (1977) himself noted, not all of the problems faced by
biological systems are well-posed in this way.
In many areas of science, optimality modeling is seen as a heuristic of discovery that
introduces abstractions that may or may not prove insightful when formulating problems,
deriving solutions, and understanding the functioning of organisms and other complex systems
(Dupré, 1987; Parker and Maynard Smith, 1990; Schoemaker, 1991). Seen in this way, it
becomes imperative to consider the boundary conditions of optimality modeling, and rec-
ognize when the abstractions and assumptions needed to formulate optimality claims cease to
be insightful and instead provide elegant solutions to problems of a different nature. Is the use
of optimality in the study of rationality any different? I have argued that it isn’t, and the basis
for my argument is a statistical one rooted in the uncertainty arising from what Simon (1957)
referred to as informational bounds, rather than the more commonly assumed computational,
cognitive, or biological bounds that motivate the study of bounded and ecological rationality.
Seen from this statistical perspective, the study of ecological rationality can cast new light on the
limited ability of probabilistic optimality to capture what is functional outside idealized worlds,
those where all aspects of uncertainty are probabilistically quantifiable.

References
Anderson, J. R. (1990). The adaptive character of thought. Lawrence Erlbaum, Hillsdale, NJ.
Anderson, J. R. (1991). Is human cognition adaptive? Behavioral and Brain Sciences, 14, 471–517.

277
278

Henry Brighton

Baggio, G., van Lambalgen, M., and Hagoort, P. (2014). Logic as Marr’s computational level: Four case
studies. Topics in Cognitive Science, 7(2), 287–298.
Bechtel, W. and Shagrir, O. (2015). The non-redundant contributions of Marr’s three levels of analysis for
explaining information-processing mechanisms. Topics in Cognitive Science, 7(2), 312–322.
Binmore, K. (2009). Rational decisions. Princeton University Press, Princeton, NJ.
Bishop, C. M. (2006). Pattern recognition and machine learning. Springer, New York.
Breiman, L. (2001). Statistical modeling: The two cultures. Statistical Science, 16, 199–231.
Brighton, H. (2006). Robust inference with simple cognitive models. In C. Lebiere, and R., Wray (Eds.),
Between a rock and a hard place: Cognitive science principles meet AI-hard problems (AAAI Technical Report
SS-02–06) (pp. 189–211). AAAI Press, Menlo Park, CA.
Brighton, H. (2019). Beyond quantified ignorance: Rebuilding rationality without the bias bias.
Economics Discussion Papers No 2019–25, Kiel Institute for the World Economy.
Brighton, H. and Gigerenzer, G. (2007). Bayesian brains and cognitive mechanisms: Harmony or disson-
ance? In N. Chater, and M., Oaksford (Eds.), The probabilistic mind: Prospects for Bayesian cognitive science
(pp. 189–208). Cambridge University Press, Cambridge.
Brighton, H., and Gigerenzer, G. (2012). Are rational actor models “rational” outside small worlds? In
S. Okasha and K., Binmore (Eds.), Evolution and rationality: Decisions, co-operation and strategic behaviour
(pp. 84–109). Cambridge University Press, Cambridge.
Brighton, H. and Gigerenzer, G. (2015). The bias bias. Journal of Business Research, 68, 1772–1784.
Chater, N. and Oaksford, M. (1999). Ten years of the rational analysis of cognition. Trends in Cognitive
Sciences, 3(2), 57–65.
Chater, N., Oaksford, M., Nakisa, R., and Redington, M. (2003). Fast, frugal, and rational: How rational
norms explain behavior. Organizational Behavior and Human Decision Processes, 90, 63– 86.
Chater, N., Tenenbaum, J. B., and Yuille, A. (2006). Probabilistic models of cognition: Conceptual
foundations. Trends in Cognitive Sciences, 10(7), 287–291.
Chen, Z. and Haykin, S. (2002). On different facets of regularization theory. Neural Computation, 14(12),
2791–2846.
Czerlinski, J., Gigerenzer, G., and Goldstein, D. G. (1999). How good are simple heuristics? In
G. Gigerenzer, P. M. Todd, and the ABC Research Group (Eds.), Simple heuristics that make us smart
(pp. 119–140). Oxford University Press, New York.
Domingos, P. and Pazzani, M. (1997). On the optimality of the simple Bayesian classifier under zero-one
loss. Machine Learning, 29, 103–130.
Dupré, J., (Ed.) (1987). The latest on the best: Essays on evolution and optimality. MIT Press, Cambridge, MA.
Edelman, S. and Shahbazi, R. (2011). Survival in a world of probable objects: A fundamental reason for
Bayesian enlightenment. Behavioral and Brain Sciences, 34(4), 197–198.
Friedman, J. H. (1997). On bias, variance, 0/1-loss, and the curse-of-dimensionality. Data Mining and
Knowledge Discovery, 1, 55–77.
Geman, S., Bienenstock, E., and Doursat, R. (1992). Neural networks and the bias/variance dilemma.
Neural Computation, 4, 1–58.
Gigerenzer, G. and Brighton, H. (2009). Homo heuristicus: Why biased minds make better inferences.
Topics in Cognitive Science, 1, 107–143.
Gigerenzer, G. and Goldstein, D. G. (1996). Reasoning the fast and frugal way: Models of bounded ration-
ality. Psychological Review, 103(4), 650–669.
Gigerenzer, G. and Selten, R. (2001). Bounded rationality: The adaptive toolbox. MIT Press, Cambridge, MA.
Gigerenzer, G., Todd, P. M., and the ABC Research Group (Eds.) (1999). Simple heuristics that make us
smart. Oxford University Press, New York.
Gilboa, I. (1987). Expected utility with purely subjective non-additive probabilities. Journal of Mathematical
Economics, 16, 65–88.
Gilboa, I., Postlewaite, A., and Schmeidler, D. (2012). Rationality of belief or: Why Savage’s axioms are
neither necessary or sufficient for rationality. Synthese, 187, 11–31.
Gilboa, I. and Schmeidler, D. (1989). Maxmin expected utility with a non-unique prior. Journal of
Mathematical Economics, 18, 141–153.
Gintis, H. (2012). An evolutionary perspective on the unification of the behavioral sciences. In S. Okasha
and K., Binmore (Eds.), Evolution and rationality: Decisions, co-operation and strategic behaviour (pp. 213–
245). Cambridge University Press, Cambridge.
Goldstein, D. G., and Gigerenzer, G. (2002). Models of ecological rationality: The recognition heuristic.
Psychological Review, 109, 75–90.

278
279

Rationality without optimality

Griffiths, T. L., Lieder, F., and Goodman, N. D. (2015). Rational use of cognitive resources: Levels of
analysis between the computational and the algorithmic. Topics in Cognitive Science, 7(2), 217– 229.
Griffiths, T. L. and Tenenbaum, J. B. (2006). Optimal predictions in everyday cognition. Psychological
Science, 17(9), 767–773.
Griffiths, T. L., Vul, E., and Sanborn, A. N. (2012). Bridging levels of analysis for probabilistic models of
cognition. Current Directions in Psychological Science, 21(4), 263–268.
Hadamard, J. (1902). Sur les problèmes aux dérivées partielles et leur signification physique. Princeton
University Bulletin, 13, 49–52.
Hand, D. J. and Yu, K. (2001). Idiot’s Bayes: Not so stupid after all? International Statistical Review, 69,
385–398.
Hastie, T., Tibshirani, R., and Friedman, J. (2001). The elements of statistical learning: Data mining, inference,
and prediction. Springer, New York.
Jones, M. and Love, B. (2011). Bayesian fundamentalism or enlightenment? On the explanatory status and
theoretical contributions of Bayesian models of cognition. Behavioral and Brain Sciences, 34, 169–231.
Kuncheva, L. I. (2006). On the optimality of naïve Bayes with dependent binary features. Pattern Recognition
Letters, 27, 830–837.
Marr, D. (1977). Artificial intelligence: A personal view. Artificial Intelligence, 9, 37–48.
Marr, D. (1982). Vision. Freeman, San Francisco, CA.
Marr, D. and Hildreth, E. (1980). Theory of edge detection. Proceedings of the Royal Society of London B,
207, 187–217.
Marr, D. and Poggio, T. (1976). From understanding computation to understanding neural circuitry. AI
Memo 357.
Martignon, L. and Laskey, K. B. (1999). Bayesian benchmarks for fast and frugal heuristics. In
G. Gigerenzer, P. M. Todd, and the ABC Research Group (Eds), Simple heuristics that make us smart
(pp. 169–188). Oxford University Press, New York.
McClamrock, R. (1991). Marr’s three levels: A re-evaluation. Minds and Machines, 1(2), 185–196.
Ng, A. Y. and Jordan, M. I. (2002). On discriminative vs. generative classifiers: A comparison of logistic
regression and naive Bayes. In T., Dietterich, S., Becker, and Z. Ghahramani (Eds.), Advances in neural
information processing systems (NIPS) 14. MIT Press, Cambridge, MA.
Oaksford, M. and Chater, N. (2009). The uncertain reasoner: Bayes, logic, and rationality. Behavioral and
Brain Sciences, 32, 105–120.
O’Sullivan, F. (1986). A statistical perspective on ill-posed inverse problems. Statistical Science, 1, 502–518.
Parker, G. A. and Maynard Smith, J. (1990). Optimality theory in evolutionary biology. Nature, 348, 27–33.
Peebles, D. and Cooper, R. P. (2015). Thirty years after Marr’s vision: Levels of analysis in cognitive
science. Topics in Cognitive Science, 7(2), 187–190.
Perlich, C., Provost, F., and Simonoff, J. S. (2003). Tree induction vs. logistic regression: A learning curve
analysis. Journal of Machine Learning Research, 4, 211–255.
Poggio, T. (2012). The levels of understanding framework, revised. Perception, 41(9), 1017–1023.
Savage, L. J. (1954). The foundations of statistics. Wiley, New York.
Schoemaker, P. J. H. (1991). The quest for optimality: A positive heuristic of science? Behavioral and Brain
Sciences, 14, 205–245.
Selten, R. (2001). What is bounded rationality? In G. Gigerenzer and R. Selten (Eds.), Bounded ration-
ality: The adaptive toolbox (pp. 13–36). MIT Press, Cambridge, MA.
Simon, H. A. (1957). Models of man. John Wiley & Sons, New York.
Simon, H. A. (1991). Cognitive architectures and rational analysis: Comment. In K. Van Lehn (Ed.),
Architectures for intelligence (pp. 25–39). Erlbaum, Hillsdale, NJ.
Şimşek, O. and Buckmann, M. (2015). Learning from small samples: An analysis of simple decision
heuristics. In D. D. Lee, M. Sugiyama, and R. Garnett (Eds.), Advances in Neural Information Processing
Systems (NIPS) 28. MIT Press, Cambridge, MA.
Stone, M. (1974). Cross-validatory choice and assessment of statistical predictions. Journal of the Royal
Statistical Society B, 36, 111–147.
Tenenbaum, J. B. and Griffiths, T. L. (2001). Some specifics about generalization. Behavioral and Brain
Sciences, 24, 762–778.
Webb, G. I., Boughton, J. R., and Wang, Z. (2005). Not so naive Bayes: Aggregating one dependence
estimators. Machine Learning, 58, 5–24.
Willems, R. (2011). Re-appreciating the why of cognition: 35 years after Marr and Poggio. Frontiers in
Psychology, 2, 244.

279
280

18
THE WINDS OF CHANGE
The Sioux, Silicon Valley, society,
and simple heuristics

Julian N. Marewski and Ulrich Hoffrage

Introduction
Imagine the following: Born in 1860, Leaning-Bear, a Sioux (Lakota), grows up in the West of
North America. His life unfolds amidst traditional and season-dependent tribal activities, such
as hunting and gathering or moving from one camp site to another. There is regular contact
with other tribes, and occasional ones with ‘white’ settlers. Then, suddenly, Leaning-Bear’s
world begins to change: More ‘white’ intruders enter the prairies and the Black Hills, and
occasional skirmishes turn into an increasingly fierce struggle for survival (see e.g., Brown,
1971, 1974; Brown & Schmitt, 1976). In 1876, aged 16, Leaning-Bear fights in the legendary
battle of the Little Big Horn; he sees George Armstrong Custer and his soldiers of the 7th
U.S. Cavalry Regiment die. But shortly afterwards, the U.S. army gains control. At the age of
30, Leaning-Bear and his family live, on ‘reservation land’, under surveillance by government
agents, no longer moving on horseback freely across the seemingly endless space of the Great
Plains. Famine and government rations replace buffalo meat; and the memories of the massacre
of Wounded Knee replace those of the victory of Little Big Horn. Violence, chaos, death, and
misery surround him, and as time passes, alcohol enters his life. That is the end of Leaning-
Bear’s old world. But it is not the end of changes in his new world: Only a few years later,
he finds himself surrounded by the impenetrable walls of a ‘white’ man’s prison cell; and after
release, he lives on to hear about cars and even airplanes; he sees factories and learns about a
great war raging on the other side of the ocean between 1914–1918. Leaning-Bear, a man who
was born in a teepee and who grew up in a world of buffalo hunting with bows and arrows,
dies in 1933, at the age of 73 and in the same year Adolf Hitler seizes power in Germany.1,2
The story of Leaning-Bear’s life is one about drastic changes in a person’s environment. It is
actually not only a story of change in one individual’s environment, but also of change in the
environment of a social collective: the Sioux nation. Changes in collective environments are
common throughout human history: Genocides, wars, economic conflicts between groups,
climatic catastrophes, epidemics, and technological revolutions can all lead to change.
What, then, characterizes the change in the story of Leaning-Bear more precisely? (1) The
change is profoundly disruptive; it has terribly negative consequences for an individual’s—here,
Leaning-Bear’s—well-being. (2) The change is abrupt; it is sudden, taking place within only a few
years of an individual’s life-time. (3) The change is thorough; it cuts across many, if not all, dimensions

280
281

The winds of change

of life. It affects, for instance, what Leaning-Bear and his fellow tribe members eat, which tools and
artifacts they use, how they dress, who is part of their social network, and whose laws are enforced.
(4) Finally, from the perspective of the individual, at the moment of such sudden, profoundly nega-
tive, and all-encompassing change, the environment becomes deeply uncertain: Known regularities
seem to break down and the environment continues to evolve at rapid pace. In short, Leaning-
Bear’s story is that of a revolution breaking out upon individuals. The violent winds of change that
were blowing over the prairies in the 1870s seem fundamentally different to all those storms the
Sioux and other Native American tribes had encountered throughout their entire history.
Today, one might argue, we are also living through a period of rapid, extensive change.
A different type of change––not produced by war, invasion, genocide, or migration, but by
technology. The contemporary winds of change we are referring to are not blowing over the
prairies, but across the entire planet. One of their epicenters––to the extent that is possible
to locate its roots geographically––is Silicon Valley with its many technology giants (for the
philosophical and historical roots of digital technologies, see Emberson, 2009; Hoffrage, 2019).
Digital technology has had and will continue to have a powerful impact across many dimensions
of our lives: It affects what many of us do on a daily basis to make a living, what is considered
to be wrong and what is considered to be right, how we search for and how we find informa-
tion, and how we organize our social lives––or, more precisely, how digital technologies affect
this organization. These technologies determine, and will increasingly continue to determine,
what tools we use and have access to, and it is uncertain what other aspects of our lives they will
change. The digital winds of change might shape some individual’s or collectives’ lives later, with
remnants or pockets of old, non-digital environments being at least partially conserved within
and by more-or-less isolated groups. Older generations who might have scant contact with
digital technology offer one example. In that pockets subsist for some time, the digital winds of
change are no different from other changes in human history, including those that impacted on
the Native American nations over the last centuries. A stunning historical example is the exist-
ence of the last free Apaches, bearing their own tragic history by hiding in the Sierra Madre in
the North of Mexico until at least as late as the 1930s (see Goodwin & Goodwin, 2018).
Importantly, while the digital winds of change will blow into different individuals’ faces sooner
or later, the ongoing digital revolution has great potential to do much harm to individuals and
societies as a whole. In our view, those potential negative side-effects of the digital winds of
change need to be understood scientifically, such that we, to the extent that we are still in control,
might be able to counteract them. That is, reflection about how sudden, profoundly negative,
and all-encompassing changes affect the ways in which individuals think, decide, and act seems
warranted, alongside reflection about how individuals and societies can manage such changes.
By focusing on the possible negative aspects of digitalization, this chapter considers how
humans understand and manage abrupt, disruptive, and thorough changes through the the-
oretical lens of a psychological framework: the fast-and-frugal heuristics approach to decision
making, developed by the ABC (Adaptive Behavior and Cognition) Research Group around
Gerd Gigerenzer and colleagues (e.g., Gigerenzer, Todd, & the ABC Research Group, 1999;
Gigerenzer, Hertwig, & Pachur, 2011). This theoretical lens, also known as the simple heuristics
approach, lends itself particularly well to trying to understand how boundedly rational humans–
–individuals who are neither “omniscient” nor endowed with computational “omnipotence”
(e.g., Gigerenzer, 2008a, p. 5 and p. 4, respectively)––behave and decide in an uncertain world.
In addition to this, the framework aids reflecting about normative, prescriptive aspects of
adapting to and shaping a fundamentally uncertain world.
We proceed as follows: Upon offering an overview of the fast-and-frugal heuristics research
program, we sketch out what future aversive digital environments might look like, and speculate

281
282

Julian N. Marewski and Ulrich Hoffrage

what strategies individuals and societies might rely upon in order to manage those aversive
changes in their environments. We close by (1) pointing to a series of research questions about
how digital environments might differ from other environments we humans have encountered
both in our more recent history and over the course of our evolution as well as (2) turning to
questions about children and education.
Before we start, a commentary is warranted. We acknowledge and believe that digital tech-
nologies can do, do do, and will do much good (and we are both, by no means, technology-
aversive and/or change-resistant, reactionary scientists). Yet, we have decided to play devil’s
advocate and to focus, in this chapter, exclusively on digitalization’s potential negative aspects,
because we believe it is important for more people to adopt a healthily skeptical view in a
time when ‘going digital’ has become a mantra for businessmen, politicians, administrators,
and others. In so doing, we are intellectually indebted to and join others who have underlined
the negative aspects of digitalization (e.g., Helbing, 2015, 2019; O’Neil, 2016; Helbing et al.,
2017; SVRV, 2018).

The fast-and-frugal heuristics framework

The fast-and-frugal heuristics framework asks how the structure of environments shapes
behavior and cognition. The assumption is that people can select inferential and behavioral
mechanisms from a repertoire, commonly referred to as the adaptive toolbox (Gigerenzer et al.,
1999; for an overview, see e.g., Gigerenzer & Gaissmaier, 2011). By using different mechanisms
(i.e., tools) from that toolbox as a function of the environment, people can make accurate
inferences about unknown elements of their world, prevent being excluded from important
social groups, find mates, or otherwise behave adaptively. Relying on tools that do not fit
the environment through which people are navigating can produce maladaptive behavior and
potentially catastrophic outcomes.
The idea that performance and behavior depend on an organism’s environment is common
to a number of research programs on human cognition. For instance, Gibson’s (1986) theory of
visual perception stresses that neither the perceptual system nor that system’s environment can be
understood in isolation; instead what matters is their interplay. According to Gibson, the envir-
onment comes with functional properties––dubbed affordances––that organisms can perceive and
act upon. The ACT-R theory of cognition (e.g., Anderson et al., 2004) and the rational analysis (e.g.,
Anderson, 1991) on which parts of ACT-R are based, place emphasis on how human memory,
by being adapted to the statistical structure of the environment, achieves its processing goals (see
Anderson & Schooler, 1991). To give a final example, Brunswik’s approach to studying human
perception assumes that achievement (i.e., performance) depends on how individuals integrate
cues available to them in the environment. According to Brunswik (e.g., 1955), it is the statistical
structure of information in an environment (i) that needs to be studied in order to understand
behavior and cognition, and (ii) that needs to be maintained in experiments and in observational
studies aimed at assessing performance.
The fast-and-frugal heuristics research program has been influenced, to a large extent, by
Brunswik’s conceptual framework of probabilistic functionalism, and by Simon’s seminal work on
bounded rationality (e.g., Gigerenzer, Hoffrage, & Kleinbölting, 1991, and Gigerenzer & Goldstein,
1996, respectively). According to Simon (1990), “[h]uman rational behavior ... is shaped by a
scissors whose two blades are the structure of task environments and the computational capabil-
ities of the actor” (p. 7). In emerging from the interplay between mind and environment, human
rationality does not correspond to that prescribed by classic, content-blind norms (e.g., Bayes
rule, expected utility maximization, logic). Instead, rationality is ecological: Rational behavior

282
283

The winds of change

is that kind of behavior that allows an individual to achieve its goals in a given task environment
(for a volume on the fast-and-frugal heuristics framework that focuses on ecological rationality,
see, e.g., Todd, Gigerenzer, & the ABC Research Group, 2012).
The fast-and-frugal heuristics framework’s emphasis on the environment (i.e., the first blade
of Simon’s scissors) contrasts with the focus on traits as explanations of behavior, as it is typical
for personality psychology. For instance, imagine that Leaning-Bear, at 13 had ‘celebrated as
victory’ the slaughter of Pawnee families in Massacre Canyon, and as a 14- or 15-year old, he
had joined war parties attacking ‘white’ intruders, and participated in raids against other tribes.
Later, he had stolen horses on the reservation and then beaten up, with several other Sioux, a
former army scout. Rather than attributing what some today might call ‘anti-social’ behavior to
stable traits, deeply entrenched in Leaning-Bear’s ‘pathological’ personality, the fast-and-frugal
heuristics framework would ask how the interplay between his decision-making strategies and
his environment may have led Leaning-Bear to commit those acts. Candidate mechanisms in
the repertoire may be imitation strategies: Raids against others and the stealing of horses may
represent behaviors that were common in Leaning-Bear’s old world, so one hypothesis may read.
In committing these acts, one could speculate, Leaning-Bear might have simply been imitating
the behavior of respected peers. Even in our contemporary world, not participating in certain
acts can signal distancing from social groups, with the risk of social isolation. Belongingness to
a group, in turn, can be key to an individual’s well-being and, ultimately, survival––especially in
societies that depend on cooperative hunter-gatherer activities in harsh environments.3
Another assumption of the fast-and-frugal heuristics framework, adapted from Simon’s
work (i.e., the second blade of his scissors), is that human information-processing capacities are
bounded: We do not have unlimited computational power, infinite knowledge, or endless time.
In trying to achieve their goals, humans do not try to optimize or maximize, which are descrip-
tive and/or normative building blocks in both classic and contemporary theories of human and
animal behavior (e.g., Becker, 1976; Stephens & Krebs, 1986). Instead humans need to strive
for solutions that are sufficient to achieve those goals in a given environment––they satisfice, as
Simon (e.g., 1990) put it.
To illustrate the point, a utility-maximization approach assumes that, when deciding
whether or not to steal a horse, Leaning-Bear would take into account all the numerous pos-
sible consequences of both stealing and not stealing (e.g., relating to the former: being arrested,
getting killed; having a vision after being injured by a soldier’s gunshot; gaining prestige among
his tribe members, and even counting a coup; see Hassrick, 1992, on the role of coups in Sioux
warfare). Moreover, he would have had to assess not only the probabilities that each of these
consequences would occur, but also the utility of each one. These utilities would then have
to be brought into the same currency or scale (e.g., being killed, gaining a horse, and so on
would all need to be assigned a numeric value). Finally, with all that in mind, Leaning-Bear
would have had to calculate which option would yield the largest expected utility and decide
for that course of action. While it is certainly possible to assume that Leaning-Bear made these
calculations (perhaps unconsciously), there might be a simpler way of explaining his decision
which involves only one goal: to increase social prestige, thereby ignoring anything else: As far
back as Leaning-Bear could remember, one way for warriors to gain recognition had been to
participate in raids on other tribes, and, notably, to steal their horses (see Hassrick, 1992, on
how the successful theft of horses mattered for a Sioux’s social prestige and economic well-
being). Indeed, prestige played and plays a role as a motor for action in other communities as
well, including the Comanche of the Southern Plains (Hämäläinen, 2012), the Tiwi people of
northern Australia (Hicks, Burgman, Marewski, Fidler, & Gigerenzer, 2012), or modern-day
scientists, as an own particular tribe. Such a hypothesis about a driving motive could then be

283
284

Julian N. Marewski and Ulrich Hoffrage

combined with others, focusing on the behavioral strategies relied upon to achieve a goal (e.g.,
imitation of others’ behavior as a route to prestige; see above).
In taking up Simon’s notion of bounded rationality, the fast-and-frugal heuristics framework
assumes that behavioral strategies do not need to depend on massive amounts of information and
complex information integration (see e.g., Marewski, Gaissmaier, & Gigerenzer, 2010). Instead,
the fast-and-frugal heuristics research program stresses that people can make smart decisions
by relying on particularly simple mechanisms, dubbed heuristics. Heuristics are strategies that
ignore information (hence frugal), and that can, in so doing, enable efficient behavior, such as
making decisions quickly (hence fast; see e.g., Gigerenzer & Gaissmaier, 2011). For instance,
the simple take-the-best heuristic (Gigerenzer & Goldstein, 1996) bases inferences on just one pre-
dictor variable. In computer simulations, this heuristic has yielded, on average, more accurate
inferences than multiple regression when tested, out-of-sample, in 20 different real-world envir-
onments (e.g., Czerlinski, Gigerenzer, & Goldstein, 1999; see also Brighton, 2006; Gigerenzer &
Brighton, 2009; Katsikopoulos, Schooler, & Hertwig, 2010). Surprising performance has been
reported for other heuristics, too––be it in marketing, criminal profiling, or forecasting of polit-
ical election results, and such like (Gaissmaier & Marewski, 2011; Goldstein & Gigerenzer, 2009;
Hafenbrädl, Waeger, Marewski, & Gigerenzer, 2016). Note that this is not to say that simple
heuristics will always and necessarily yield smart decisions with favorable outcomes. Such a guar-
antee does not exist, because the relative efficacy of a heuristic depends on the environment. For
example, Leaning-Bear’s environment changed in many dimensions and continuing to use the
same heuristics, such as imitation, despite environmental change, can lead to disaster, including
punishment for horse theft, or alcoholism when imitating the alcohol consumption of others.

The research questions of the science of simple heuristics

The fast-and-frugal heuristics framework asks a series of research questions that can be grouped
into four categories: descriptive, ecological, applied, and methodological (e.g., Gigerenzer,
2008b; Gigerenzer, Hoffrage, & Goldstein, 2008).

(1) Descriptive: What are the heuristics people have in their adaptive toolbox that they are able
to rely upon––be it when it comes to making simple decisions or when navigating through
a complex social world? What elements of our cognitive architecture (e.g., memory and
perception) do these heuristics exploit? When and how do people choose to use which
heuristic?
(2) Ecological: To which aspects of environmental texture is each heuristic from the toolbox
adapted? How can those environments be described (e.g., in terms of statistical regularities
or affordances)? Answering this question aids understanding of how selecting a heuristic in
a fitting environment can aid people to make clever decisions; or, conversely, how failure
can result if a heuristic is mismatched to an environment. This is a question of ecological
rationality.
(3) Applied: How can human performance be improved by better matching heuristics to envir-
onments? Two ways of boosting performance are possible. One would be to lead people
(e.g., through education) to change the heuristics they rely upon in a given environment;
another would be to change (e.g., through policy-making) the environment in which
people rely on given heuristics.
(4) Methodological: How can people’s use of heuristics (the descriptive question) and the fit of
the heuristics to different environments (the ecological question) be studied, and how can
the effectiveness of interventions (the applied question) be tested?

284
285

The winds of change

Instructive answers to these four types of questions have been formulated in numerous fields and
disciplines to which the fast-and-frugal heuristics program has been applied (e.g., Gigerenzer
et al., 1999, 2011). How can these four questions aid our understanding of drastic environ-
mental change, notably, the ways in which individuals think, decide, and act?
Descriptive questions would ask which heuristics people rely upon when they sense such
massive changes: Are there heuristics for managing environmental change? For instance, when
an individual does not know what a right or clever course of action is, relying on others and
imitating their behavior are typically viable strategies: Groups can furnish resources in situations
of scarcity and unpredictable access (e.g., through the pooling of food), and observing what
others do can facilitate learning of what works well and what does not, thereby avoiding repeti-
tion of costly errors. But how are people’s willingness, propensity, or readiness to imitate others
affected by the stability of the environment?
Ecological questions would ask: Which characteristics of massively changing environments
would likely increase the performance of different strategies––be they heuristics or other tools—
and which characteristics of changing environments would limit those strategies’ performance?
For example, when it comes to making accurate inferences about unknown or future events,
imitating others is likely helpful when these others can be expected to make at least minimally
accurate judgments. If the future is fully unpredictable or if all individuals in a group are likely
to be led astray, then relying on others will not aid making accurate inferences (but could still
help inclusion in the social group; on the ecological rationality of imitating the majority of
others, see Gigerenzer, 2008a). If environments differ with respect to which strategy performs
best––and they may vary over time, or space, or both––then an ecological analysis becomes
quite complex and its conclusions cannot be summarized in one sentence (for a review and a
simulation study that compared the performance of various individual and social learning strat-
egies depending on the stability of the environment, see McElreath, Wallin, & Fasolo, 2013).
Applied questions might prescribe asking, for instance, how imitation strategies could be
turned into clever ones in a new environment; for example, by endowing at least certain individ-
uals from a collective with training and insights (e.g., through schooling), this way creating leaders
or role-models others might benefit from by imitating. Applied questions might also ask how
heuristics could be used as tools (Hafenbrädl et al., 2016), and how they could be taught to foster
insights and boost people’s decision-making competencies (see also Grüne-Yanoff & Hertwig,
2016, on boosting; Kozyreva, Lewandowsky, & Hertwig, in press, on nudging versus boosting
with a focus on internet users; and Gigerenzer, 2014, on heuristic principles for teaching).
Finally, the methodological perspective would prescribe tackling those descriptive, eco-
logical, and applied questions by means of observational studies, laboratory experiments or,
as is often done in studies of fast-and-frugal heuristics, through mathematical analysis (e.g.,
Martignon & Hoffrage, 2002) and computer simulation (e.g., Gigerenzer & Goldstein, 1996;
Schooler & Hertwig, 2005). General guiding principles for asking methodological questions
can be found in Marewski, Schooler, and Gigerenzer (2010).
The ecological focus of its research questions allows differentiating the fast-and-frugal
heuristics research program further from other frameworks. Take applied questions on policy
making and approaches that locate explanations for performance largely inside individuals, such
as in psychopathic personality structures, lack of intelligence, or other traits. Corresponding
to those inside-explanations, those frameworks’ prescriptions might look very different than
those of the fast-and-frugal heuristics program. For instance, a trait-approach to eradicate horse-
stealing and other anti-social behavior might start from the basis that traits are difficult to change;
hence corresponding individuals need to be ‘managed,’ not their environment or the changes that
occurred therein. A simple way of managing individuals with undesirable traits is to eliminate

285
286

Julian N. Marewski and Ulrich Hoffrage

them, say, by locking them away. To give another, more modern example, consider individ-
uals involved in a scandal, such as Dieselgate or a case of industry bribery. Rather than seeking
simplistic “one-word explanations” (Gigerenzer, 1998, p. 196) for wrong-doing (e.g., greed,
recklessness, or egoism), which in the worst case can border re-description or tautology (e.g., a
person behaves egoistically because she is an egoist), one could ask what behavioral strategies lead
such individuals to do what they do, and what the ecological rationality of adopting those strat-
egies is. To turn to the latter, ecological question: Could it be, for instance, that an unexpected
change in their environment (e.g., how pollution should be assessed) in the past casts them and
their behavior in a bad light (e.g., being reckless, behaving not particularly smart), but only today,
that is, with the benefit of hindsight? This example also illustrates that, when answering the eco-
logical question, no universal (here: time-invariant) norm for assessing rationality exists. While
the fast-and-frugal heuristics framework is by no means the only program that shifts the focus
from people to their interaction with their context (for an example from ethics, see e.g., Palazzo,
Krings, & Hoffrage, 2012), this shift is not about sides in debates on meaningless dichotomies
(e.g., inside versus outside explanations, genes versus culture and such like; on dichotomies in
psychology more generally, see e.g., Newell, 1973), but rather about describing the interplay
between environments and decision making mechanisms with precision.

On environments and heuristics

The functioning of heuristics cannot be assessed independently of the environment (Todd et al.,
2012). Likewise, answering the four research questions set out above hinges on building adequate
models of heuristics and environments. Environments can be characterized in multiple ways (e.g.,
Bullock & Todd, 1999; Czerlinski et al., 1999). For instance, they may change over time in sys-
tematic and hence predictable manners. The environment of the Plains tribes looked very different
depending on the season, but the seasons were periodic. The Sioux (Lakota) of the nineteenth
century were nomads and they could exploit such regularities: They knew when to be where in
order to harvest ripe wild berries or to hunt (see Hassrick, 1992). More generally, environments
can be described in terms of their regularities, with past research on fast-and-frugal heuristics
having focused on, for instance, the statistical regularities found in modern-day informational
environments (e.g., the news media, Goldstein & Gigerenzer, 2002; Schooler & Hertwig, 2005).
Moreover, environments differ between and within people. ‘White’ contemporaries of
Leaning-Bear, living in the same area, did not live in his environment. And Leaning-Bear’s
biography offers an example of a difference within a person: there is a before and after the
breakdown of Sioux (Lakota) military power.4
Furthermore, not only the objective, but also the perceived environment determines
behavior and performance and such perceptions may change over time. Hence, a description
of the environment in terms of its objective properties (e.g., cue inter-correlations) can be
complemented by the environment’s inner (e.g., mnemonic, emotional) representation (e.g.,
Marewski & Schooler, 2011).
Finally, an environment offers opportunities but also imposes challenges and confronts
one with different tasks, such as finding a mate, maintaining friendships, or securing food
and shelter. Consequently, environments can be described alongside different task-relevant
dimensions. Gibson’s (1986) notion of affordances, which highlights environments’ functional
properties (e.g., what actions does the property X afford?), offers a complement to this level of
description.
Still another dimension by which an environment can be described is how much uncer-
tainty it entails. Since the notion of uncertainty is particularly important in the context of

286
287

The winds of change

this chapter––with its focus on uncertainty-generating changes and our digital future––let us
discuss that notion in a bit more detail. In the literature on fast-and-frugal heuristics, uncer-
tainty typically implies a world in which the full range of behavioral options is unknown, or
in which the consequences of acting in a certain way are unknown, or the probabilities of
these consequences occurring are inaccessible (see e.g., Mousavi & Gigerenzer, 2014, 2017,
for discussions). In such a world, unforeseeable and/or surprising, fully unexpected events
can ruin plans, for better or worse.5 Savage (1972 [1954]) referred to such situations as large(r)
worlds and contrasted them with small(er) worlds, a dichotomy that corresponds, roughly, to
Knight’s (1921) distinction between uncertainty and risk (see Binmore, 2007). Small worlds,
characterized by risk, are roulette and card games: Here all options, their consequences, and
probabilities of occurring are known or can be estimated with sufficient accuracy––assuming
nobody cheats. From the perspective of the fast-and-frugal heuristics framework, it is only
in small, risky worlds that Bayesian, subjective expected utility-maximizing, and other classic
rational approaches to modeling human behavior, and performance might work well and
represent adequate benchmarks for gauging human rationality (e.g., Gigerenzer, 2014;
Hafenbrädl et al., 2016). In contrast, in large worlds, characterized by uncertainty, optimiza-
tion is not feasible. If the decision maker does not know (1) what the consequences of behav-
ioral options might be, or (2) what the probabilities of these consequences are (e.g., because
the probabilities of relevant future events are not known, and eventually not even the events
themselves are known––events that will, jointly with the chosen behavioral option, determine
these consequences), or (3) eventually does not even know how any of these consequences
should be evaluated, then optimization is pointless.
As a side note, optimization, as a mindset, is hence also not useful when reflecting about pol-
icies for shaping digitalization. For example, chains of unforeseen, distant future consequences,
including changes in values letting current evaluations of foreseeable consequences diverge
from future evaluations of the same consequences, might all be brought about by digitalization,
resulting in massive uncertainty.
Importantly, the construction of models of heuristics hinges on the type and amount of
uncertainty in the environment, an insight that is, actually, not always stressed in the litera-
ture on fast-and-frugal heuristics. Historically, the fast-and-frugal heuristics program started
with inferential heuristics (for a brief history of this program, see Hoffrage, Hafenbrädl,
& Marewski, 2018). The first heuristic formulated by members of the ABC Research
Group dates back to before the group came into existence: the PMM-algorithm (which was
a central component of a Brunswikian model on overconfidence; Gigerenzer et al., 1991).
Gigerenzer and Goldstein (1996) renamed it take-the-best, referred to it as a heuristic, and
determined its performance in a complete-paired comparison between all German cities
with more than 100,000 inhabitants. Like other heuristics for inference, take-the-best can
be described in terms of three algorithmic rules: A search rule and a stopping rule, which
describe how specific cues are searched for and when information searching stops, respect-
ively, and a decision rule formalizing how the cues identified are relied upon to arrive at a
decision. Such heuristics are sufficiently precise in order to be implemented in computer
simulations testing the accuracy of the heuristics in predicting the criterion against that of
other models (e.g., other heuristics or more information-greedy statistical models, including
regressions). A well-defined reference class with a finite set of objects (cities), one criterion
(number of inhabitants), and nine dichotomous or dichotomized cues (e.g., a soccer team
in the national league, an industrial belt)––this was, in fact, a relatively ‘small’ ‘large world’
and not a situation that would be characterized by the massive uncertainty that is typical of
large (e.g., social) worlds in real life.

287
288

Julian N. Marewski and Ulrich Hoffrage

A first, albeit, small, step in the direction of a ‘larger’ world was taken by Czerlinski et al. (1999).
These authors tested strategies, including heuristics, in cross-validation: Each strategy estimated its
parameters in a given sample, but was then tested on another sample, taken from the same popu-
lation. In this set-up, the world is still clearly defined, namely, in terms of a known criterion, and
known cues, but there is noise when it comes to the assignment of objects to learning and test
samples. Arguably, such variation through sampling does not entail a strong form of uncertainty.6
Stronger forms of uncertainty emerge, for example, when predictions have to be made
out-of-population––situations where the learning sample might have little, if anything, in
common with the world that one wishes to generalize. Even more uncertainty can arise
when distant or quite general goals (e.g., survival, reproduction, well-being) can only be
achieved if multiple intermediate goals are met and when it is far from clear how these inter-
mediate goals should be prioritized and/or considered when deciding among options. Such
complexity is likely to be found in social environments (Hertwig, Hoffrage, & the ABC
Research Group, 2013).
Whereas some simple heuristics for navigating such very large, fuzzy worlds can still be cast
into algorithmic search, stopping, and decision rules, others resemble guidelines, principles,
or routines, tailored to specific situations or domains. An example is “hire well and let them do
their jobs,” a simple rule that the former president of Florida International University, Modesto
Maidique (2012), gleaned from CEOs whom he interviewed (cited in Gigerenzer, 2014, p. 116).
Other examples, from the domain of leadership, are: “If a person is not honest and trustworthy, the rest
does not matter,” “First listen, then speak,” “You can’t play it safe and win. Analysis will not reduce uncer-
tainty,” or “When judging a plan, put as much stock in the people as in the plan” (cited from Gigerenzer,
2014, p. 117, italics added). To return to the Sioux, Hassrick (1992) points out that a lemma of the
Sioux was, “It is better to die on the battlefield than to live to be old” (p. 47, italics added), could this be
thought of as a heuristic that would have eliminated, akin to a non-compensatory decision rule,
any concerns about injury or death, such as when deciding whether to steal a horse?7
These heuristics for a large world characterized by massive uncertainty bear resemblance
to the simple rules discussed by others (e.g., Manimala, 1992). Emphasizing their frugal (but
not their ecological) nature, Sull and Eisenhardt (2015), for instance, describe such rules as
“shortcut strategies that save time and effort by focusing our attention and simplifying the way
we process information” (p. 5). They “allow people to act without having to stop and rethink
every decision” (p. 5).
Such rules can be categorized in different ways. Sull and Eisenhardt, for example, distin-
guish among boundary rules (which “define the boundaries of inclusion or exclusion, … like
the ‘Thou shalt not’ of the Ten Commandments,” p. 50), prioritizing rules (which “can help you
rank a group of alternatives competing for scarce money, time, or attention,” p. 57), stopping rules
(e.g., “stop eating when I start feeling full,” p. 67), how-to rules (that “address the basics of getting
things done without prescribing every detail of what to do,” p. 81), and coordination rules (which
“guide interactions among members who intermingle in a complex system,” p. 84).
Complementary ways to classify heuristics for massively uncertain worlds would be in terms
of the functions they can serve (e.g., imitate-the-majority can allow for increasing accuracy
of inferences, inclusion in a group, avoiding conflict, protecting oneself). Considering such
functions may aid understanding of how people select between different heuristics. Conversely,
the same function may be achieved by various heuristics, with such redundancies potentially
enhancing the robustness of human decision-making capabilities (on Brunswik’s related notion
of vicarious functioning, see e.g., Wolf, 1999; on strategy selection and robustness, see e.g.,
Marewski & Schooler, 2011). Notwithstanding their name and classification (see also e.g.,
Guercini, 2019), a common feature of all such rules is that they constrain the space of possible

288
289

The winds of change

behavioral options or decisions, and hence facilitate finding solutions to problems––and indeed,
the term heuristic comes from ancient Greek and means “to discover” or “to find.” This is why
we conceive––and refer to––such rules or recommendations also as heuristic principles. In our
view, when it comes to coping with the challenges offered by the winds of digitalization, these
heuristic principles are more likely to be used (and be helpful) than the algorithms the fast-and-
frugal heuristic program studied in its early days.

Modern-day dramatic change: from buffalo hunting on the

Great Plains to the fruits of Silicon Valley
Jenny was born in 1983. When she grew up, cell phones were virtually non-existent in her
world. The internet played no role in her youth and if she wanted to contact a friend she would
have to call her either from home or from a public speaker phone. Calling someone in another
country was expensive, the way to stay in touch with someone abroad would be to send a letter
via postal mail. Twenty years later, in 2005, at the age of 22, Jenny uses the internet to find sci-
entific articles during her university studies. She regularly looks up other information on the
web, too, be it about political issues, medical conditions, or subject-matter related to her great
passion: horses. She becomes excited about the prospect of a society where information is freely
available to everyone, she engages in social networks, puts her CV and photos online, and writes
her emails at work as if she were having a normal casual conversation with somebody, a mode
of address and interaction that migrates to WhatsApp a few years later. That informal commu-
nication style is consistent with the netiquette of the time (a word used to refer to the rules of
behavior in electronic communication). As a matter of fact, in becoming an internet actor, via
email and other digital technologies, she behaves almost in the same way as she would if she were
interacting with other people personally (be it face-to-face or on the phone). Jenny’s life in the
2010s continues with her starting a job at a major car manufacturer, giving birth to two children,
taking out the usual insurances, and buying the usual commodities. She has a customer discount
card for her preferred supermarket chain, she has installed the latest version of Windows on her
computers, and she uses a smartphone like most of her peers. She pays for things electronically
and is enthusiastic about getting paperless when it comes to managing all administrative aspects
of her life, ranging from health insurance to writing a will. Cameras ensure the security of her
mansion and that of most streets in the city she lives in. Jenny lives a happy life.
But then, in 2030, everything changes. Jenny’s husband develops a disease, and his health
insurance provider refuses to pay because the disease was allegedly present prior to him taking
out the insurance policy. The insurance company ‘discovered’ this with the help of an algo-
rithm which––when Jenny’s husband asked for reimbursement of treatment costs––had searched
through all accessible electronic records on him, pooled from multiple sources, including other
companies, numerous individuals (e.g., private pictures of Jenny’s husband floating around on
the internet), and government databases. Jenny’s husband is devastated––he does not remember
having suffered from or even having been tested for that disease when he started his policy
30 years ago. Furthermore, he also has not saved any documents from that time that would
allow him to prove the insurance company is wrong. But things get worse: Jenny gets arrested,
charged for threatening national security. The accusation is based on emails Jenny had sent over
the years, containing criticism of the government and on her having watched too many videos
on the internet considered to be supportive of certain conspiracy theories. None of these activ-
ities alone would have been enough for such an accusation, but the evidence accumulated over
the years such that, in 2030, the algorithms that scrutinized her electronic traces classified her as a
threat. Suspicious gestures captured by cameras that filmed her when reading the email accusing

289
290

Julian N. Marewski and Ulrich Hoffrage

her of criminal activity, as well as recordings of phone conversations with her friends (automat-
ically saved by her phone company) provide further butter for the prosecutor’s bread: She suffers
from a psychopathological personality, claims the prosecution based on the available digital evi-
dence. Once the charges become public, Jenny’s friends drop her on social networks, she is fired
from her job, and even her family avoids talking to her. Nobody wants to be associated with
somebody like Jenny. In 2032, she commits suicide after coming to the conclusion that there is
no way to escape this vicious circle. The same year, in Jenny’s hometown alone, hundreds of citi-
zens receive similar treatment, suffering public shaming and stigmatization. Some were caught
watching porn, for others there were clear indications that they supported a prohibited political
movement fighting against digital surveillance, some were in favor of abortion, and at least ten
were caught making derogative comments about a leading political party in private (captured by
their smartphones). None of those behaviors are tolerated in Jenny’s world of the 2030s. And it
is virtually impossible to keep behavior, preferences, or interests as a secret––the citizen of that
time is entirely transparent.8 Helbing (2015) offers a comprehensive overview of what kind of
personal data can be collected in the digital age, and in Box 18.1 we let Google’s privacy policy
as it stands at the time of writing, alongside a recent statement from the European Commission
on creating rules for so-called “E-evidence,” speak for themselves. Can both documents be seen
as signposts of small but decisive steps toward Jenny’s world? After all, the technology is already
operational. The more interesting question, however, is who will use it, when and for what pur-
pose, and how will such use be legitimated, enforced, and controlled?

Box 18.1 The beginning of the end of privacy: Google and E-evidence

We collect information … from figuring out … which language you speak, to … which ads
you’ll find most useful, the people who matter most to you online, or which YouTube videos you
might like …
… We also collect the content you create, upload, or receive from others when using our
services. This includes … email you write and receive, photos and videos you save, docs and
spreadsheets you create, and comments you make on YouTube videos …
… We collect information about the apps, browsers, and devices you use to access Google ser-
vices … The information we collect includes unique identifiers, browser type and settings, device
type and settings, operating system, mobile network information including carrier name and phone
number, and application version number. We also collect information about the interaction of your
apps, browsers, and devices with our services, including IP address, crash reports, system activity,
and the date, time, and referrer URL of your request. …
…We collect information about your activity in our services, which … may include:

• Terms you search for

• Videos you watch
• Views and interactions with content and ads
• Voice and audio information when you use audio features
• Purchase activity
• People with whom you communicate or share content
• Activity on third-party sites and apps that use our services
• Chrome browsing history you’ve synced with your Google Account

290
291

The winds of change

… we may collect telephony log information like your phone number, calling-party number,
receiving-party number, forwarding numbers, time and date of calls and messages, duration of calls,
routing information, and types of calls…
… We collect information about your location …
... In some circumstances, Google also collects information about you from publicly accessible sources ...
We may also collect information about you from trusted partners, including marketing … and
security partners ….
Extracts from Google’s Privacy policy9

***

To make it easier and faster for law enforcement and judicial authorities to obtain the electronic
evidence they need to investigate and eventually prosecute criminals and terrorists, the Commission
proposed on 17 April 2018 new rules in the form of a Regulation and a Directive, which will:

• create a European Production Order: this will allow a judicial authority in one Member State
to obtain electronic evidence (such as emails, text or messages in apps, as well as information to
identify a perpetrator as a first step) directly from a service provider or its legal representative in
another Member State, which will be obliged to respond within 10 days, and within 6 hours in
cases of emergency …
Extract from the European Commission’s online statement on “E-evidence”10

Why did we include two fictional characters in this chapter? What do they have in common,
Leaning-Bear, the Sioux born more than 150 years ago, and Jenny, born at the end of the last
millennium? First, we believe that digital technology has the potential to lead to a change in
our environment that might, in a certain way, end up being almost as dramatic as the change
we proposed Leaning-Bear experienced 150 years ago. Note that Jenny, born in 1983, belongs
to the last generation who has a few childhood memories of a world in which computers have
not yet entered the daily life of the masses. Only a few decades later, today’s born-digitals find
it hard to imagine a time without smartphones and internet––and no one knows how techno-
logical advances will continue. So, second, as much as Leaning-Bear’s and Jenny’s stories are fic-
tional ones, the following discussion about how digitalization may change our environment is,
by necessity, a speculative narrative concerning an uncertain future. We deliberately chose these
fictional characters to highlight this speculative nature. However, we also believe that there is
sufficient basis for our speculation and know that some of the worries we formulate are shared
by others (e.g., Helbing, 2015, 2019; Helbing et al., 2017; O’Neil, 2016).
In the remainder of this chapter, we discuss what insights the fast-and-frugal heuristic
framework might offer when it comes to understanding and eventually managing changes
related to the digital revolution. We speculate what future aversive environments might look
like. We reflect on how such environments might consequently influence behavior by leading
people to adopt different heuristics. We then ask how such heuristics might, in turn, shape
environments, namely, the societies in which they are enacted. Subsequently, we explore how
aversive environmental changes can be managed, first, by an individual navigating through a
changing environment and, second, by societies as a whole undergoing environmental change.
Finally, we turn from changes in evolutionary history to the children of evolution: the digital
natives.

291
292

Julian N. Marewski and Ulrich Hoffrage

In writing these sections, we have followed common wisdom––a meta-heuristic of sorts––

for dealing with potentially aversive future environments: In order to prepare for challenges,
one has to try to identify them, knowing that the future is uncertain and that surprises may
occur. But even when leaving these caveats aside, it is clear that future digitalized worlds might
take many forms. The goal is not to speculate about every form; rather, we focus on three
aspects of digitalized worlds that we take to be particularly disruptive of individual behavior
and performance (see Helbing, 2015; Helbing et al., 2017; Kozyreva et al., in press; O’Neil,
2016; SVRV, 2018).

What might future aversive digital environments look like?:

interconnectedness, influenceability, and traceability
First, whereas Leaning-Bear’s world was one of relatively small groups, our contemporary
modern world is one in which small groups intermingle with larger groups and contexts. Thanks
to email, social media, and so on, we can easily reach out to people who live far away. When
reflecting on how digitalization might affect our social environment, one thesis that might be
formulated is that those growing social worlds make individuals more reachable by other people,
including those who live on the other side of the planet. In some ways, new ‘tribes,’ united via
digital technology over potentially thousands of miles might emerge. Such ‘tribes’ might share
common (e.g., political) beliefs and worldviews, indeed, technology itself might contribute to
creating and cementing those ‘tribes’ (e.g., through the selective posting of information via
algorithms, resulting in echo-chambers; see Helbing et al., 2017). Needless to say, any increased
ability to reach out to others can also offer more opportunities for collaboration, the sharing
of knowledge and insights into warranted political and societal reforms. However, like a two-
faced Janus head, the accessibility of individuals can also offer increased opportunities for attack
and social destruction (e.g., Helbing, 2015). Virtual bullying and cybercrime are cases in point.
Twenty years ago, a child that was bullied at one school might have moved to another school
(possibly in another town) in order to be placed out of reach of his aggressors. Nowadays, the
bullying can continue online through social media. Likewise, in order to steal money, robbers
20 years ago had to physically enter houses or banks. Nowadays, online theft and other types of
cybercrime are possible over thousands of kilometers. The new digital world is an interconnected
one. But note that not all connections are visible, at least not for John Q. Public: Individuals may
have a hard time figuring out where social attacks come from––one cannot see enemies’ actions
on social media and unless one is part of a chat, one does not know who said what––uncertainty
reigns. Indeed, it is worth asking, who, actually, is the adversary in a vastly interconnected world
full of digital ‘tribes’? At best, one can only sense that something is going wrong, giving rise,
potentially, to paranoia and even vicious cycles. A manager’s authority at a company, for instance,
may be undermined by anonymous electronic messages spread widely among employees. Such
messages may transform a supportive or neutral work environment into an aversive one that the
manager no longer understands and, hence, cannot adequately respond to, potentially resulting
in behavior that further contributes to the decline of the manager’s authority.
Second, increased potential for attack comes with increased potential for influence: Be it
news about shark attacks in Australia or local government corruption in mid-Nevada, in a
digitalized world, information from all parts of the globe can easily spread, and hence influ-
ence individuals’ beliefs, emotions, and behaviors. How would you feel if, after watching 20
videos of shark attacks somewhere on the other side of the planet, you snorkeled over deep blue
water in the Mediterranean Sea? Would you experience some kind of anxious arousal, even if
you knew that any fear is completely unwarranted? Or if you hear stories, repeated over and

292
293

The winds of change

over again, thanks to social media, about waves of violent immigrants, stealing, beating, and
raping: Would you call for stricter immigration laws or even applaud the building of walls at the
border of your country? Over-informed, over-aroused, and over-reactive are three words that
characterize our modern, interconnected world.
A third aspect of our world that is changing is the traceability of individuals’ past
behavior: Individuals are likely now to leave behind far more traces of their past actions and
intentions than ever before in human history. If an individual coughs on the street, laughs out loud
or farts, gives a polemical speech, wears crocodile skin trousers, drives through a red traffic light,
visits a brothel, does not pay her taxes, commits fraud, or hacks the bank account of his neighbor,
chances are there is a trace of those actions somewhere. When we interact with a digital device, it
is presumably even more likely that records of this interaction remain, and this holds true not only
for our own devices but also for those beyond our own control; for instance, the devices of those
to whom we sent an email, the owners of websites we visit, or the owner of a camera installed
on the street or in a car. This, in turn, might create additional affordances for social destruction
(see Helbing, 2015). Obviously, not all traces are visible to everyone; and, indeed, those who can
access them are also those who have power––and so everything hinges on what their interests and
intentions are. Moreover, the individual is not necessarily in control of the recorded traces; she
does not ‘own’ the data and cannot destroy them. The individual might not even know what data
were recorded or when. And the amount of data that can be traced back in time exceeds what the
individual would be able to remember about her own actions. All of this generates uncertainty.
In short, in digitalized societies, individuals lose control of what is knowable about their
own past and present behavior. At the same time, individuals can be aggressed over any distance
and by any other individual on the planet, and similarly they can also be influenced over any
distance (see Helbing, 2015).

What heuristics might people rely on to navigate through

aversive digital environments?
Individuals who realize, ahead of time, how traceable and attackable behavior will become,
might react in various ways. One might be defensive.

Heuristic principle: If you have to have to make a decision, act in a way that allows you to
defend your decision.

Defensive decision-making has been described in managerial and medical environments (e.g.,
Artinger, Artinger, & Gigerenzer, 2019; Marewski & Gigerenzer, 2012)––that is, in situations
where decision makers fear punishment for their actions, such as a physician who might be
afraid of being sued for a wrong diagnosis or unsuccessful treatment. Indeed, what is commonly
subsumed under the label paranoia can possibly be conceived of as the end of the spectrum of
cognitive and behavioral outcomes of such defensive approaches to decision making.
Increased social conformity might be another behavioral outcome displayed by individuals
who think defensively about their future (see Helbing, 2015). Conformity might be produced
by genuine defensive decision-making principles (“If you act like everybody else acts, then your
action can likely not be wrong”) or by mere (not necessarily intentionally defensive) imitation
heuristics. Those social decision strategies can allow dealing with situations where the ‘correct’
or desirable course of action will never or only later reveal itself (Gigerenzer, 2008a; Hertwig
et al., 2013). More generally, they may also aid an actor who has little or no knowledge about
the situation at hand (Gigerenzer & Gaissmaier, 2011).

293
294

Julian N. Marewski and Ulrich Hoffrage

Heuristic: “Imitate the majority” (Gigerenzer, 2008a, p. 31)––behave as most others do.

It takes two to tango, but three to make a list. A third example of how individuals might adapt
their behavior once they become aware of the increased traceability of actions and the poten-
tial for destruction resulting from those traces is that those individuals resort to offense as the
best defense. An example of this is dash cameras, nowadays being increasingly installed in cars.
The rationale is that these cameras can record the potentially hazardous behavior of other drivers,
offering important legal proof in case of an accident. The downside: If an accident occurs and only
one party has a dash camera, then only that party will have access to the images, allowing them to
potentially control which images enter the judicial stage and which do not. Hence, those who do
not have a dash camera installed might find themselves in a less advantageous situation compared to
those who have, thereby eventually leading to all drivers recording each other’s behavior constantly.

Heuristic principle: If your behavior can potentially be recorded, record the behavior of every-
body else who you interact with.

What might be thought of as a digital arms race can also take place when trust among different
members of an institution erodes. For instance, the prevailing communication style might grad-
ually shift from verbal commitments and promises to the sending out of emails that summarize
the contents of conversations, to eventually all details being put in writing by each party, each
one trying to be the one with the best documentation––just in case. Putting others’ cc on
emails in work contexts––ideally higher in the hierarchy––is a notorious escalation signal, too.
Such arms races do not depend on digital media for their existence, but digital media can
fuel them. Digital media facilitate the storing, copying, and sharing of content, regardless of
whether it comes to individuals, companies, public administrations, governments, or secret ser-
vices. Some of these agents may fear that having less information than others may be disadvan-
tageous in a complex, competitive, or potentially threatening social world. A world that comes,
moreover, with seemingly little margin for error, because behavior that represents, either in
foresight or in hindsight, a ‘wrong-doing,’ might have a greater chance of being detected or
leading to negative consequences (see also Helbing, 2015). One does not need to understand
much of game theory to anticipate that corresponding fears, whether justified or not, can spiral
into a desire to collect data and to store information.

What might future aversive digital societies, shaped by defensive,

social, and offensive heuristics, look like?
The heuristics people might rely upon to navigate through a digital environment might, in
themselves, contribute to shaping those environments. What kinds of societies could a con-
sistent enactment of those and other defensive and offensive heuristics create? Possibly, societies
governed by fear. Perhaps societies governed by mistrust in social relations. And maybe soci-
eties characterized by less individualism and less individual freedom, with the masses defining,
through the likes of Twitter, Facebook, and other social media, which behaviors are acceptable
for certain collectives (digital ‘tribes’), and which are not. Indeed, changes in value systems
might be an additional consequence (Helbing, 2015). For example, it is likely to be no coinci-
dence that in many European countries the word “transparency” has become frequently used in
public discourses: Transparency is needed, so the public outcry goes, when yet another scandal
is uncovered. But calls for transparency might also transform into beliefs and the spread of
heuristic principles such as “He who has nothing to hide has no problem with sharing all information

294
295

The winds of change

about him,” a lack of transparency being equated with a lack of trustworthiness, or even worse,
criminal intent. Still another consequence might be a lack of willingness for decision makers
to take risks and potentially make mistakes (Helbing, 2015). Those decision makers might be
politicians who opt for defending what they take to be ‘safer’ positions or entrepreneurs who
opt not to promote a new technology because if things go awry, it might be easy to make a case
against them thanks to the massive amounts of data collected in the brave new digital world.
The lack of willingness to make mistakes, in turn, might translate into reduced opportunities
for learning: after all, only those who make mistakes can learn. To avoid misunderstandings: In
both analogous and digital worlds, mistakes can occur; but the difference is that they might be
less likely to be detected in an analogous world and, if detected, they might be less likely to stay
visible––people forget, but computers and big data do not. (For more detail, see Helbing, 2015,
who discusses these and related observations, pointing, for instance, to the emergence of vicious
cycles, herding effects, and the loss of independent judgment of others.)
If social relations, value systems, and the permissibility of mistakes change on a societal level,
what else can happen to make things even worse? Totalitarianism, not ‘just’ dictated by new
norms and controlled, bottom-up, by the masses through digital media, but totalitarianism put
into place top-down by institutions. Institutional totalitarianism can start small, with individ-
uals being assigned digital scores if their behavior conforms to a certain target, say, customers
buying certain products in a supermarket being awarded a score that offers them a discount. This
is already a reality today and might represent a slippery, habituating slope into “Super-scores”
(SVRV, 2018, p. 7). Indeed, institutional totalitarianism can also start bigger, with China’s citizen
scoring system being located, perhaps, still at the lower end of the spectrum (e.g., Helbing,
2015; Helbing et al., 2017; Hoffrage & Marewski, 2020). At the upper end of the spectrum,
there might be room for total surveillance and control over all areas of life, including of life itself.

How can heuristics aid individuals to manage aversive change?

Above, we speculated about the heuristics that individuals might resort to in aversive digital envir-
onments. Are there other heuristics that could aid individuals to manage aversive change? A first
move to managing aversive changes might be to recognize that they may occur––and in dramatic
ways. A follow-on step may then be to reflect upon how those changes might look, much like
we did above. For instance, the authors of this chapter tend to believe that some of the most dra-
matic changes brought about by the digital revolution might concern people’s values: Behavior
that is deemed acceptable or even desirable today might, within a few years, be thought of as
being unacceptable (Helbing, 2015). Why? Digital technology allows actors to reach out to many
individuals at once, offering tremendous potential to influence what people believe they know
(Kozyreva et al., in press). Whether such inter-personal influence is steered, top-down, by indi-
viduals trying to impose their views (we firmly believe that Joseph Goebbels, Hitler’s propa-
ganda minister, would have been excited if he had had the internet at his disposal and, perhaps
more importantly, under his control) or whether it emerges bottom-up through chats and social
networks does not matter. If that hypothesis is correct, then it becomes potentially risky for indi-
viduals to be unable to quickly adapt their values and behavior to the new trends.

Heuristic principle: Avoid getting caught in a spider-web of old-world values and behavior as
your world changes.

The German expression “Die Fahne nach dem Wind hängen” (loosely translated: hanging the
flag as the wind blows) seems to reflect this heuristic principle, or simple rule. Following this

295
296

Julian N. Marewski and Ulrich Hoffrage

principle could have helped Leaning-Bear to realize that stealing horses is not acceptable in a
reservation environment,11 such a strategy might not always be easy to enact in a digital world
in which behavior is fully traceable, offering potential for destruction even 20 years later. This
is akin to stating that it might be impossible to avoid failure. If failure is unavoidable, then
another set of strategies might come into play: Strategies that allow dealing with failure once
it has occurred. A classic representative of this is the simple heuristic of “not putting all your eggs
into one basket” but rather to diversify.

Heuristic principle: Split resources up across different, non-connected targets.

For instance, this diversification heuristic might suggest that income should be earned not just
through one job (which might be lost) but through at least two, and that those jobs should ideally
come with unrelated failure modes (e.g., losing one job will not entail losing another). A simple
way to decrease connectivity might be to distribute jobs, and hence income, across different indi-
viduals, that is, to pool them. Partnerships where both individuals work in largely unconnected
domains (e.g., different sectors) are a case in point. If one partner loses her job, the income from
the other might still offer enough resources to make for a living (e.g., until a new job is found).
Factoring in redundancy while avoiding common failure modes are simple engineering
principles for creating robust systems, heuristics for resilient design. Modern passenger aircrafts
are examples of this, as most important components typically exist in at least two versions that
are unconnected and made in different ways. This way, a failure in one component is less likely
to imply a failure in its disconnected counterpart (see e.g., Kitano, 2004).

How can heuristics aid societies to manage aversive change?

Just as there are heuristic principles for building robust machines, could there also be principles
for creating societies that are potentially more robust to digital change? We believe that this
might be possible (see also e.g., SVRV, 2018). For instance, Helbing et al. (2017) stress how
important the encouragement of diversity might be at the societal level, the co-existence of
different goals and opinions aids the creation of robust, resilient societies (see also Helbing,
2015, 2019, on robustness and digitalization). Here are three additional heuristic principles
that, when implemented together, might further help to create what we would take to be more
resilient digitalized societies.

Principle of disconnectedness: Implement policies that disconnect people from each other
(see also Helbing, 2015).

There may be different forms––weak and strong––of disconnecting people. At the upper end,
there would be legal prohibitions and digital censorship. The lower end comprises much less
dramatic forms. For example, users of Skype cannot receive messages or calls from others unless
they explicitly approve it, technically, by having accepted a contact request. Other communi-
cation services could follow this model, leaving it to the market, that is, ultimately, to users,
whether they want to have such protection against an avalanche of incoming information. Such
decisions require that users are aware of the problems, potentially inviting policy makers to design
corresponding educational interventions to further boost John Q. Public’s digital expertise (e.g.,
Kozyreva et al., in press). Equally important, not only receivers but also senders should be aware
of problems generated by the high degree of connectedness in modern societies. For instance,
the forwarding of emails, often even without the original sender’s knowledge, can cause quite a

296
297

The winds of change

lot of harm. We, personally, think legal prohibition of such practices is neither feasible nor desir-
able, but we posit that more mindfulness might be healthy (e.g., with regard to a higher threshold
for forwarding emails and, eventually also, for sending them in the first place).

Principle of deceleration: Implement policies that slow down the spread of information among
people.

Rather than speeding up the flux of information via ever faster transmission speed, it might be
worth slowing down the spread of information (see Helbing, 2015); for example, by setting the
speed of transmission of electronically-represented information to the equivalent speed of regular
long-distance travel: Let trivial news about shark attacks in Australia travel at the same speed as it would
take an individual to, physically, deliver such news! Building time lags into a system can make the system
less reactive (and e.g., foster forgetting or emotional tranquilization). Of course, there is uncertainty
over what counts as trivial. For instance, information that may be trivial at t1, might turn out to
be nontrivial at t2, or what is trivial for one person might not be for someone else. Another way
to reduce the spread and, indirectly, the speed of information transmission may be to tax (e.g.,
Helbing, 2015) or otherwise charge for electronic communication, placing a burden on either the
receivers or the senders of information. The challenge lies in mitigating the side-effects of such an
economic approach to steering behavior, such as wealthy individuals or companies being allowed
to increasingly dominate the information flux, simply because they have more resources.
While the first and second principles tackle the interconnectedness and influenceability of
digital environments discussed above, the third principle tackles the traceability of individuals.

Principle of information loss: Implement policies that prevent the unlimited storage of
behavioral data.

There are different ways to implement information loss. One might be to prohibit storing
behavioral data such as browsing on the internet or messages sent out publicly via Twitter or
other social media. Another way might be to implement systematic data loss via decay over
time: Just as human memory forgets information, digital information should systematically fade away!
What information should, over time, be lost is an extremely difficult question, with the most
radical approach prescribing the forgetting, over time, of all information, and others allowing
for the selective forgetting of certain (e.g., private) contents (for a legal basis in the European
Union, see Article 17 GDPR).

Principle of non-sharing: Records of individuals’ behavior when interacting with digital

devices cannot be shared beyond the original purpose of the recording.

Of course, companies, non-profit organizations, governments, and other institutions do have

to share data in order to function well in an interconnected, borderless world, at least that is
a frequently made argument. The tricky question is to determine what kind of data ought
not to be shareable. Why is this an issue? In digital environments, increased complexity is
caused by the massive trading of information about individuals: By letting algorithms pool
customer ratings with their credit score, health-related data, and other records, available in
big data lakes, even minute details about an individual can become knowable and exploitable
for commercial and political ends. That, in turn, can contribute to increasing traceability
and influenceability. To diminish the traceability and influenceability of individuals, policies
that prevent the trading and pooling of data could be put into place, including laws that

297
298

Julian N. Marewski and Ulrich Hoffrage

prohibit such actions (see also Helbing et al., 2017, on making punishable the unauthorized
use of data).
Note that this principle of non-sharing would be undermined if individuals were given the
possibility to voluntarily share their data. Such voluntary authorization can be instigated by
offering opt-ins and opt-outs on websites (often not very user-friendly or transparent.)12 An
example of this is to alert internet users that continuing to surf on a website constitutes cookie
consent; a parallel from outside the digital realm is to inform customers that walking into a shop
implies agreement with being filmed. While there are different ways to induce individuals to
authorize the recording and potential sharing of their behavior, from an ethical point of view it
is highly problematic that, as of today, many individuals might not understand what can be done
with their data (for Germany, see SVRV, 2018). And they might never have reflected upon the
possibility that their world might change. The possibility that non-sharing could be something
highly desirable for them might simply not be on many people’s radars and so they do not care.
Moreover, understanding how data from multiple sources can be used to classify individuals,
for instance, in their role as patients, customers, or citizens, into different categories requires
knowledge of statistics. Corresponding statistical knowledge might simply be insufficient in the
general population or in certain collectives. Those, in turn, who are aware of the risks might
be averse to buying into yet other risks associated with non-sharing: Even if social exclusion or
limited access to information (triggered by, for instance, shunning digital networks, messaging
tools and internet search engines) do not spell doom for individuals who want to avoid one’s
behavioral data being shared, life will certainly not get easier for them if they try to prevent this
data-sharing, if only because personalized digital services (e.g., web search, shopping) permit
people to save time and effort. Indeed, the recording and subsequent selling of digital behavior
have become a business in the digital world. Clearly, policy makers might find it hard to simply
prohibit the pooling, selling, and other forms of sharing of behavioral data given that there is a
market with numerous individuals consenting to such activities.
***
Changing tack. Above, we pointed to difficult questions arising when trying to implement each
of the listed principles, such as the tricky question how to decide what information ought to
be classified as trivial (or harmful), warranting its spread to be slowed down, what information
to forget, or what information to share. Those questions boil down, in one way or the other,
to classification problems. In any classification problem, two types of mistakes can occur: false
positives (e.g., sharing information that should better not be shared) and false negatives (e.g.,
not sharing information that should be shared). Could such classification tasks be tackled with
heuristics? An obvious candidate model for classification tasks, one might think, are fast-and-
frugal trees (e.g., Woike, Hoffrage, & Martignon, 2017).
Yet, contrary to other (smaller-world) classification tasks where the criterion is certain and
reveals itself (e.g., classifying a mushroom as poisonous or not, and then eating the mush-
room), in these kind of large-world classification tasks, the criterion is fundamentally uncer-
tain, and perhaps even unknowable. Counterfactuals, too, may be impossible to observe (e.g.,
what would have happened had I not sent this email to my boss or written that tweet about my
neighbor?), making learning more difficult than in tasks where the counterfactual is accessible
to the decider (e.g., inferring, in a TV show, which of two cities has more inhabitants, while
being told after each decision what the correct answer would have been). Moreover, in those
larger worlds, the criterion might be multi-dimensional and criterion values might vary over
time. Hence, fast-and-frugal classification heuristics might actually not represent adequate tools
to tackle those types of tasks. Instead, we believe that simple rules that foster a maximum of
diversity of classification outcomes might be a better bet: If we do not know what classifications

298
299

The winds of change

are desirable, the best option might be not to “put all your eggs into the same basket.” Diversity
can, too, foster robustness, to the extent, for instance, that issues arising from different classifi-
cation outcomes are unconnected (see also Helbing et al., 2017).
What could these diversity-maintaining principles be? And can we entrust humans to enact
them? A common belief is that we humans may suffer from systematic biases, causing not only
reasoning fallacies, but also other undesirable behavioral outcomes, including discriminative
behavior (e.g., against certain collectives) and/or self-serving actions. Classification algorithms
are, nowadays, often seen as better alternatives to those ‘biased,’ ‘egoistic’ humans (Hoffrage
& Marewski, 2015). Machines are seemingly stripped of all undesirable subjectivity; they are
objective in that they can make decisions independently from humans who, in one way or the
other, have to be mere objects of the decisions (e.g., what messages, written by a human, are
to be deleted.)13
Yet, there is no bias-free classifier, whether human or algorithm-based. Any policy, explicitly
or implicitly, enacts values, reflected, for instance, in how it trades off one mistake (false positives)
against the other (false negatives). Moreover, human classifiers differ, not only between individ-
uals (e.g., in values) but also over time (i.e., individuals change their views, they make mistakes in
enacting their own worldviews, they forget). That is, humans do not only vary in terms of their
biases, human behavior can also be conceived of as exhibiting a noise component. As a conse-
quence, human classifiers, with all their idiosyncrasies and resulting variation in behavior, could,
in principle, be their own solution to the problem they created, with the heuristic principle
being, “Put humans in full power,” that is, let humans make their own decisions about, for instance,
what information to share massively, to quickly pass on, or to save on the public cloud! We write
“could.” Democracies can maintain diversity, but totalitarianism can synchronize human behavior.
Gleichschaltung (i.e., making equal or uniform) is a word capturing how the Nazis reduced vari-
ation, trying to bring all elements of society in line with their ideology: a single horrific bias.
The disturbing question is: Can a few digital technologies (e.g., email, social media, Google’s
search algorithm) dramatically reduce the variation in behavior across human classifiers, a digital
Gleichschaltung, not necessarily orchestrated top-down by authorities, but emerging bottom-up,
as a vast assimilation or acculturation process, caused by the massive pooling of information?14
If so, a first step for policy makers may be to regulate the magnitude of those technologies’
outreach. Anti-trust laws and other rules that might break up the market share of (dominating)
companies might go in that direction, but such approaches fall short of solving the problem,
because different companies, as well as different digital technologies that are nonetheless com-
patible with each other, can all support the massive spread of the same information and thereby
continue reducing diversity. In short, the paradox is: Regulators should implement policies that
reduce the speed, spread, and saving of information, and the resulting classification problem
(e.g., what information to slow down, not share, or forget) might be best tackled by heuristic
principles that foster diversity––diversity which may be destroyed by the forces unleashed (e.g.,
speed, spread of information) that make tackling the classification problem necessary in the first
place. But that is not the end of it: a trade-off must be managed between fostering diversity and
limiting it enough, such that, to put it bluntly, people can still talk to each other.
How to foster diversity? Help people construct a toolbox of heuristics for critical judgment,
would be our recommendation (see Gigerenzer, 2014, who puts forward a toolbox of simple
heuristics for health care, business, and other domains). This toolbox would include guiding
principles such as push oneself to understand a subject matter at hand (e.g., a classification
problem), even if one is not an expert in the topic (e.g., if one has no expertise in statistics) or
to play devil’s advocate on dominant views. Box 18.2 lists examples of heuristics for managing
another important challenge, brought about by digitalization: scoring.

299
300

Julian N. Marewski and Ulrich Hoffrage

Box 18.2 Teaching scoring with heuristics

One of us teaches executives and other individuals, potentially affected by and/or interested in
leading digitalization in companies, non-profit organizations, and other institutions. One basis of
that teaching is the book, Weapons of Math Destruction, by O’Neil (2016), which invites us to extract,
from context-rich stories, more general insights that can, in our view, be cast as heuristic principles.
A few of these heuristic principles are, alongside related points (see also e.g., Helbing, 2015), listed
below; they are meant to help people to reflect on scoring (e.g., of customers or citizens), such as
when algorithms are used by companies (e.g., banks) or public institutions (e.g., administrations)
to make decisions (e.g., which customers to offer credit to; which citizen is allowed to buy a high-
speed train ticket):

1. Convert probabilities into natural frequencies when trying to comprehend classification problems
(e.g., Gigerenzer, Gaissmaier, Kurz-Milcke, Schwartz, & Woloshin, 2007; Hoffrage, Lindsey,
Hertwig, & Gigerenzer, 2000). To understand what natural frequencies are, consider the
following fictional example. Imagine an algorithm is used to grade convicts; if the score is
positive, the person is not let out of prison. The overall recidivism rate in that population is
5 percent, the true positive and false positive rates are 80 percent and 30 percent, respectively.
What is the probability of recidivism if the score happens to turn out positive for a given
criminal? Many people might find answering that question difficult and, as a result, they may
blindly trust the positive score. Now, take 1,000 criminals. Of those 1,000, 50 (5 percent) will
commit another crime and of these, 40 (80 percent) will score positive. Of the 950 who do
not commit another crime, 285 (9 percent) get a false positive score. Since the total number
of positive scores is 40 + 285 = 329, the probability of recidivism if the score is positive is only
40/329 ≈ 12 percent!
2. Recognize that no decision––be it made by a human or a scoring algorithm––is free of values (e.g.,
how to trade off false positives and false negatives, such as when building scoring algorithms
like the one described in point 1’s fictional example). As O’Neil (2016, p. 21) herself puts it,
“models, despite their reputation for impartiality, reflect goals and ideology”.
3. Recognize that even scoring algorithms you try to program to act in line with your values (e.g., to not
discriminate on the basis of race or gender) might end up deciding against your values. For instance,
few people would want credit-worthiness scores to include race as a variable; however, let us
assume that ZIP code is predictive of income and you decide to include that variable in the
score. Voilà! To the extent that people of the same race are poor and live close to each other
(e.g., in ghettos), the score may end up being racist. Another way to frame this heuristic prin-
ciple is that one needs to realize that there is uncertainty with respect to hidden confounds.
4. Realize that scoring systems can create vicious circles (e.g., criminals get poorer and even more
criminal). This can happen, especially if the systems upscale (see point 8 below). Such vicious
cycles can be another constitutive element of uncertainty; they can create new unknown
confounds, and, without you realizing it, can lead to decisions that do not align with your
values.
5. Transparency is not the solution; at worst it helps only a bit. Even very simple, transparent scores
(e.g., produced by equal-weighting heuristics, laid open to the general public) can generate
dynamic, hard-to-foresee, and non-transparent outcomes; for instance, by creating the afore-
mentioned vicious circles on a societal level.

300
301

The winds of change

6. Make sure that you constantly monitor the performance of your scoring algorithms. Because envir-
onments can quickly and unexpectedly change, the performance (e.g., accuracy) of scoring
algorithms can also change.
7. Do not think of scoring in the singular. Never implement just one scoring algorithm. Instead,
approach scoring experimentally and implement several diverse algorithms. For instance,
randomly assign different customers to different scoring algorithms and monitor and com-
pare performance, such as the accuracy of the scores, across treatment groups. In so doing,
if possible (e.g., ethically acceptable) also try to implement counterintuitive (out-of-the-
box) algorithms (e.g., if need be, as merely hypothetical secondary scoring rules kept
in the background, without consequences for people). Diversity can provide a buffer to
sudden changes in environments; and diversity also aids learning about those changes
(e.g., if counter-intuitive algorithms suddenly start to work well). The insight to test
counter-intuitive algorithms in model comparisons (and, ideally, out-of-sample or popu-
lation) reflects the development of the fast-and-frugal heuristics framework: many would
not believe that less information and computation can be more (see e.g., Hertwig &
Todd, 2003).
8. Think about scale. An algorithm with an outreach to 100 people is, even if that algorithm
discriminated against, say, all 50 women among those 100 (e.g., none of them gets a job; prob-
ability of discriminating decisions: 100 percent) far less devastating than an algorithm with an
outreach to 1,000,000 citizens which discriminates only with a probability of 1 percent against
women (i.e., 10,000 women do not get a job).
9. Understand that you may have to make difficult decisions. For instance, if one of your algorithms
has scaled up and obtained a huge market share, then that’s good for business (short term) but
can be bad for society (long term). If you voluntarily give up on market share (short term),
the competition may step in (and, in the worst case scenario, you may have to sacrifice your
business or job).

In line with the environmental focus of the fast-and-frugal heuristics research program, per-
haps law makers should also reflect in how far legal frameworks warrant adaptation, moving
away from allocating agency and responsibility to individuals and more toward making the
environment surrounding individuals the central unit of analysis (for other legal issues, see
Helbing et al., 2017). In other words: When, why and to what extent is the individual to be
held accountable (corresponding, roughly, to an inside-explanation for behavior and perform-
ance) as opposed to her/his digital environment (corresponding to an ecological explanation)?
What else could one do to aid societies to manage aversive digital change? Perhaps the best rec-
ommendation is to try to slow that change down: People need time to understand new environ-
ments in order to be able to shape them. The lemma of a “digital Enlightenment” (Helbing, 2019)
might be: Let people learn about the changes (heuristic principle: Boost John Q. Public’s digital know-
ledge!), and create opportunities for feedback that are forgiving of mistakes: digital error-cultures
in times of uncertainty. That recommendation is in line with the notion of the adaptive toolbox
of the fast-and-frugal heuristics research program. Decision makers need to be aware of the tools
(i.e., decision strategies) in their repertoire; they need to be ready and able to create new ones and,
equally importantly, they need to acquire expertise when it comes to selecting from among the
different tools available to them. Strategy selection, calls for, at the end of the day, not a (seem-
ingly unbiased) algorithm, but good human judgment (Hoffrage & Marewski, 2015). Of course,
there may be other principles, too––principles referring to aspects of digitalized environments

301
302

Julian N. Marewski and Ulrich Hoffrage

we did not explicitly discuss above (for more aspects, concerns and recommendations, see e.g.,
Gigerenzer, 2014; Helbing, 2015, 2019; Helbing et al., 2017; Kozyreva et al., in press).

Digitalization: from evolution to the children of evolution

In this chapter, we have looked at environments mainly from a life-history perspective. Another
perspective might be an evolutionary one. The resilience of our make-up allowed homo sapiens
to survive and spread over millennia. Yet, the robustness of a system is always relative to the
environment in which it acts; after all, the environment forms part of the system. Without
wanting to evoke catastrophic scenarios borrowed from apocalyptic science-fiction, we feel that
it is paramount to consider the degree to which digital technology is able to transform human
environments. Is there a possibility that, after a rapid transformation process, environments will
emerge that are very different from those that our cognitive architecture evolved to be adapted
to over the course of evolutionary history? Does it make sense to speculate about the extent to
which design features that can make information processing systems robust might be present in
the cognitive make-up of the mind and in the structure of pre-digital environments, but not in
the texture of digitalized worlds?
Arguably, features of human environments that played a role in human cognitive evolution,
notably, our information-processing mechanisms, were constituted by humans themselves: It
is the social world with its affordances for cooperation and mating that may have shaped not
only imitation and other heuristics for a social world, but more generally the mechanisms that
determine how we perceive, think, feel, and act, including the ways how we can, potentially,
perceive, think, feel, and act. According to this view, cognition is shaped and constrained by
our phylogenetic history. And it is this heritage from the environment of evolutionary adaptedness
(e.g., Tooby & Cosmides, 1992) that now stands to meet our new environments. As discussed
above, digital environments may have a number of characteristics that distinguish them from
pre-digital environments. All of them concern the social world:

• In digital environments, social interactions over long distances (across the globe)
are possible, and they are instantaneous, that is, without a time delay. Long-distance
interactions, of course, have always been possible in non-digital environments, but
never throughout human evolutionary history has it been as easy, and never has it
been as instantaneous.
• Humans can interact with everyone who is connected to the digital network, poten-
tially resulting in a very large number of interaction partners. Compared to non-
digital worlds, the potential number of partners has thus tremendously increased.
• In a digital world, humans do not interact with each other face-to-face, but by means
of technical mediators such as Twitter, Facebook, WhatsApp and email. Such mediated
interactions were also possible in the pre-digital world (e.g., by means of letters or
messengers transmitting oral messages), but the ease of access to such indirect forms
of interaction, and hence their frequency have increased in digital worlds. Equally
importantly, since social information is now maintained as digital records, it is no
longer forgotten.
• In contrast to the pre-digital world, in digital environments, information can be pre-
processed, modified, and even fully created by non-human agents, that is, by com-
puter programs, without either human or artificial recipients (e.g., other computer
programs) of the information necessarily being aware of these manipulations. This
seems to be new in human evolutionary history.

302
303

The winds of change

Easier, faster, longer-distance, and more frequent machine-mediated interactions might not at all
represent significant changes in evolutionary terms. They might not pose, per se, any threats to
the robustness of human-information processing mechanisms in social worlds. But do they create
indirect potential for harm, for example, by fueling interpersonal conflict and aggression? In a
digital world, there can be more connections between different elements (humans) in a social
system (a group). As a consequence of this, false information (e.g., rumors about hostile agents,
past or intended future aggressions) and other types of disturbance (e.g., fake news) might have a
greater potential to spread (Kozyreva et al., in press). Do these disturbances also have more poten-
tial to endure, just as information is not simply sorted out by digitalized social systems over time
through decay or other forms of information loss (see Helbing, 2015)? Is cooperation aided by for-
giving, and is forgiving aided by forgetting, such as the forgetting of past aggressions (e.g., Stevens,
Marewski, Schooler, & Gilby, 2016)? How would challenges likely to have been important to
human cognitive evolution (e.g., cooperating, mating) in the past be transformed if, in the future,
human information-processing systems only received information via digital technology? What
are the implications for society and human interaction if all information transmitted becomes fully
detached from the context in which the information was produced, including the smiles, smells,
and sounds emitted by the sender (another human being) and the texture of the environment in
which that sender is located (a group of anxious, angry, or happy humans)?
Rather than trying to reach back into our evolutionary history, one might adopt a more
modest historical perspective; for example, by asking how different technological artifacts
shaped behavior and cognition in different epochs. For instance, one might believe that digital
communication comes with less trust than face-to-face communication; yet is there evidence
that interpersonal trust decreased in collectives when individuals started to communicate by
means of paper (e.g., writing letters)? What affordances does paper offer when it comes to
deciding what and how to communicate? What affordances are special to digital media? Both
allow for long-distance communication; yet only the latter allows for delivery within seconds.
Paper might retain the exciting smell of the beloved one (even if it is only because perfume can
be sprayed on paper); an email, in turn, affords attaching a picture and a video-conference even
affords watching that lover in real time, making things like cybersex possible. Touching (or even
actual sexual reproduction) is afforded by neither artifact.
To turn to another technological artifact, when the TV was invented and became available
in households, these new machines afforded making many different decisions, including the
decision to access information or to entertain one’s children. Modern-day smartphones seem
to come with these affordances, too; albeit watching a movie in front of a TV is possible for a
group of up to a dozen people, watching a movie on a smartphone gets more difficult if there
are more than three or four persons crowded in front of the small display. Moreover, carrying
a TV around (a heavy item at the time and dependent on a mains electric power supply) was
impossible, hence TVs did not pose affordances for entertaining, say, small children while
waiting in line in the supermarket or in the waiting room of a doctor’s practice. In contrast,
smartphones afford children entertainment in any location and at any time. Moreover, TVs
afforded receiving pre-selected information (e.g., a movie listed in a TV program in the ana-
logue age) at certain points in time; smartphones, in turn, allow receiving self-selected informa-
tion (e.g., any movie) at any point in time, and they also afford the sending of information. If
these ‘new’ affordances are conceived of as decisional options that are systematically acted upon,
what, then, might be the possible longer-term consequences? Could it be, for instance, that
children are more likely to get used to being entertained, less used to active play, more easily
bored (e.g., when there is nothing to play with), or more likely to participate in familial conflict
(e.g., when boredom and impatience result in behavior that mummy and daddy are not willing

303
304

Julian N. Marewski and Ulrich Hoffrage

to tolerate)? Will those children’s upbringing and education prepare the ground for yet more
environmental changes, namely when those children, as adults, shape environments themselves?
Another gift of digitalization that is entering households these days is artificial intelligence
and the internet of things. Like TV and smartphones, both are entirely new from an evolutionary
point of view. In contrast to adults, children do not know that these are very recent gadgets,
after all, for them, everything is, at one time, new anyway. But still: What happens psychologic-
ally if a child is exposed to an electronic assistant at home, a machine one can talk to and that
responds? Imagine you give a child a toy, say, a car, a doll or a teddy, that has a microphone, a
motion sensor, and other built-in devices. Nowadays such toys can transmit information, ranging
from the toy’s position to the child’s utterances (down on earth), (up) to a central server on the
cloud. That information, in turn, allows inferences to be made, such as when the child sleeps
or what the child likes. How do children know that a doll is more than just a doll? How should
parents teach them what they can tell their doll and what not (e.g., Mum and Dad fight? We go
on vacation and the house will be empty? My uncle has debts? I will be alone at the bus station
tomorrow after school?). In the analogue age, children could be taught principles such as “Do
not trust strangers! Do not talk to others about X, Y, and Z (e.g., your family) at school!” But the doll
is not a stranger––it is not even an “other.” Likewise, what principles can we teach children for
interacting with an electronic home assistant, a device that transmits information to the cloud
and that is just a box; that is, something that does not even have the shape of a being (see Box
18.3)? Manches (2019) and Manches, Duncan, Plowman, and Sabeti (2015) raise and discuss
such and related issues. In particular, Manches et al. (2015) ask three questions:

How do IoT [the Internet of Things] devices influence children’s interpretation of,
and interaction with, everyday things? (p. 81) How is the data that is captured by
IoT devices being used? (p. 82) How cognizant are children and their carers of IoT
designs? (p. 82)

As they point out,

The nature of IoT is that everyday things become interfaces for digital technology.
As technology becomes more seamless, children may become less aware of how their
actions are being captured and used. Without explanation, children may be unaware
that hugging a teddy is providing a range of adults with intimate knowledge of their
biometric data. Clearly, parents need to be informed if they are to give consent on
behalf of their children, but the public’s awareness of the implications of the IoT for
data protection is currently quite low.
Manches et al., 2015, p. 77

Box 18.3 Teaching children with narratives

During a presentation at the workshop of the Herbert Simon Society in November 2019, Manches
pointed to the misfit of the rule “Don’t talk to strangers!” to a world of IoT devices, including teddies
with microphones. This rule is insufficient to protect children against the dangers that come with
the digitalization. What rules can we teach them that are adapted to these new technologies with
their new affordances for others (e.g., who have access to the cloud)? Rules for behavior are taught,
across cultures, with narratives including fairy tales and holy books. Fairy tales and the bible contain

304
305

The winds of change

mysteries and one should be careful not to destroy these with an intellectual mindset. However,
one can ask what they reveal about human nature and how such an enriched understanding allows
one to better understand elements of the Zeitgeist and technology coming with digitalization. Can
the witch in Hänsel und Gretel, in her perfidy, reveal more about an Internet-of-Things-teddy
than we think? Can we use the mirror in Schneewittchen to talk to our children about omniscience,
self-reflection, and vanity by relating it to modern affordances such as spying, stalking, ‘liking’ or
‘dis-liking’ and cyberbullying? Who is the most beautiful, both here and at the other side of the
world? What is the correspondence between a mirror, reflecting an image of oneself, and the echo
in digital echo-chambers? “Smartphone, Smartphone in der Hand, wer hat die meisten ‘Likes’ im
Land?” (Martignon & Hoffrage, 2019, p. 185).

These issues become even more disturbing when one takes into account that such devices do
not need to be ‘just’ dolls or teddys collecting data on a toddler’s movements or a young child’s
‘secrets’; the devices could also be digital learning games gathering inferred IQ. There could
be computer games that gauge a teenager’s sleeping habits or behavior in brutal virtual fights.
Imagine that the collected data were indicative of (e.g., later) academic aptitude, job perform-
ance, or health and consider that such data might stay on the cloud for decades, and be shared
with others (e.g., insurance companies, job application portals) for an unlimited time.15
We want to stress that while such issues are certainly important to consider, they may actually
be miniscule compared to those related to building proper representations of the physical and
social world. This is already hard enough (Hirschfield & Gelman, 1994) and it is an open question
how children can and will acquire concepts, such as person, agency, intelligence, intention, or
responsibility if they grow up in an environment in which the winds of change have eradicated
the border between man and machine. Applying a precautionary principle suggests that we should
withhold digital technology from younger children for as long as possible, and this is, in fact, what
many managers working for digital poster child companies in Silicon Valley do (Dwyer, 2011),
but this requires some understanding, awareness, discipline, and courage (for a more in-depth
discussion of the risks of digitalization for children and adolescents, see Martignon & Hoffrage,
2019). On the other hand, learning about the dangers stemming from digital devices may require
exposure, or even some direct experience of them, and, if so, through vicarious learning or story-
telling. Perhaps we could create tales of almighty, all-knowing beings, who can, through a spider
web composed of other people, know everything about us and our lives. Will angry, almighty
gods and mythological figures (e.g., the Hydra) come back into our lives, albeit in new guises?
Can we even learn from millennia-old stories, religious beliefs, and other forms of wisdom,
how to teach our children how to interact with these new-old ‘beings’? Will there be new-old
heuristics, such as “You cannot hide anything from the gods!” or “Don’t make the gods angry!” Perhaps
the cloud is a strikingly accurate name for gods, who, in a material sense, are new and a human
creation. We find it extremely thought-provoking that Dicke and Helbing (2017) entitled their
science-fiction novel, which is centered around the interaction between the protagonist Alex and
a world-encompassing artificial intelligence that uses Alex’s smart-home assistant as an interface,
iGod. And interestingly enough, perhaps also some of those old stories foreshadow the cataclysmic
drama of the current attempt to build a digital version of a new God. After all, the Tower of Babel,
which was meant to be tall enough to reach heaven, was stopped thanks to diversity (through
language), which halted the flux of information among its builders. It remains to be seen what
will happen to the modern towers (shall we say, server farms accessed by artificial intelligence?).

305
306

Julian N. Marewski and Ulrich Hoffrage

We want to remind readers that the “Wind of Change”––a song by the Scorpions––became
a kind of hymn of the fall of the Berlin Wall. This fall closed a chapter in history (Fukuyama,
2006, even declared this event marked the “end of history”), and opened a new one. What will
fall and what will rise through the winds of digitalization? These and many other issues might be
worth reflecting upon, not only for scientific reasons, but also when it comes to actively shaping
the brave new digital world—and thereby potentially paving the way for what might follow.

Conclusion: compassion in the winds of change

Leaning-Bear and Jenny experienced drastic aversive changes in their environments; changes
that they did not foresee, but had to cope with by relying on the behavioral and cognitive
mechanisms available to them. They lived through them, and they suffered through them.
Their life histories, despite being fictitious, can help us to think about such changes. We close
this chapter with the words of a non-fictional character, a person who experienced dramatic
change in his environment for real. Changes that, with Gleichschaltung and an all-dominant
ideology, brought horror and devastation to large collectives. A person who decided to act
against those changes, and who, in stepping out of the line dictated by that ideology, lost his
life. Dietrich Bonhoeffer was executed for acting against the Nazi regime just a few days before
the regime collapsed in 1945. The following lines by Bonhoeffer could be taken as a general
statement on drastic aversive change:

One has to expect that most humans will only become savvy when they themselves
experience suffering. This way one can explain, firstly, most human’s stunning inability
to execute any kind of preventive action – one simply tends to believe to still be able
to avoid the danger, until it is finally too late; [and this way one can explain] secondly
the insensibility with respect to others’ suffering; compassion emerges proportional
to the growing fear of the threatening proximity of doom … Deedless waiting and
obtuse standing on the sidelines are not Christian postures. It is not, at first, his proper
sufferings, but the sufferings of his brothers, for whose sake Christ suffered, which
calls the Christian to action and to co-suffering.
Bonhoeffer, 1998 [1942], pp. 33–34, translated from German, additions in brackets

As Bonhoeffer points out, the route to wisdom may be paved by first-hand experiences. Indeed,
as we have discussed in this chapter, we humans are not “omniscient,” we cannot boast of com-
putational “omnipotence” (see e.g., Gigerenzer, 2008a, p. 5 and p. 4, respectively), rather, our
rationality is bounded, as is our ability to foresee the digital future. Without adequate foresight,
preventive action might seem pointless. Yet, as Bonhoeffer further remarks, we are capable of
something else: Compassion, as well as action and co-suffering. Compassion can be produced
by environmental cues signaling growing threats to oneself. But action and co-suffering can
come without threats from values. It is beholden on us to reflect whether we want to follow
the heuristic “Act once you feel compassionate!” or its counterpart, namely, “Act prior to feeling com-
passionate!” as we encounter the winds of change.

Epilogue
In his memoirs, likely written in 1939, Bruno Stange, the great-grandfather of one of the
authors, describes the following encounter with Hitler, which presumably took place in 1933:
“The ‘Führer’ wanted to speak to the workforce in the Dynamowerk of the Siemenswerke.

306
307

The winds of change

What enthusiasm ... Special admission tickets were issued, multiple checks were carried out,
then we were allowed to wait for three hours until the splendid Führer appeared in the presence
of his paladins. For a short time, I had the opportunity to look into Hitler’s eyes. A horror
overcame me and back at home I said to ... my dear wife: ‘I have looked into the eyes of a
madman’.” Today, with a benefit of hindsight and greater knowledge, we know what happened
after this “madman” came to power in 1933. We wonder what our great-grandchildren will
say when they read this essay after the winds of digital change have had another few decades to
marry humans with machines.

Acknowledgments
We would like to thank Riccardo Viale, Benjamin Müller, and the GABE reading group at the
Faculty of Business and Economics at the University of Lausanne for many helpful comments
on earlier versions of this chapter. We also thank those attendees of the 2019 Workshop of the
Herbert Simon Society (“The digital world, cognition and behavior”) in Turin who shared
their thoughts with us on digitalization. We especially owe Dirk Helbing a large debt: His
warnings and notably his books Thinking ahead: Essays on big data, digital revolution, and partici-
patory market society (Helbing, 2015) and Towards digital enlightenment: Essays on the dark and light
sides of the digital revolution (Helbing, 2019) have been a major source of inspiration for us. We
are also indebted to Gerd Gigerenzer, founder of the fast-and-frugal heuristics approach, and to
several members of the former ABC Research Group with whom we have discussed the poten-
tial negative aspects of digitalization over the course of two working retreats. Finally, we would
like to thank Barnaby Dicker and Susan Dunsmore for their careful copy-editing and Caroline
Plumez and Jérémy Orsat for additional checks.

Notes
1 The character of Leaning-Bear is fully fictional; yet the story depicted here may have happened in one or
the other form to individuals from Native American nations. They all experienced the dramatic break-
down of their world (see e.g., Brown, 1974 for an encompassing recent account of the history of the
Lakota, see Hämäläinen, 2019; for first-hand accounts of Lakota ways of life, see Waggoner, 2013). In
general, there have been numerous instances of dramatic changes during history that resulted from vio-
lence, war, or some sort of clash of civilizations. When discussing a previous version of this chapter with
colleagues, some pointed to the possibility that our example of the Sioux might elicit negative emotions,
in particular, among the descendants of Native Americans. However, when contemplating using another
example, we realized that this concern is quite general and could be made, presumably, for all other his-
torical examples of sudden, dramatically negative, and thorough changes in an individual’s environment.
These changes for the worse set the stage for the topic of this chapter. Our hope is that by looking at a
grim picture of the past, we might aid reflection on new types of potentially grim future changes.
2 After having finished this chapter, we learned about Luther Standing Bear (Ota Kte, Plenty Kill), a
Sioux born in the 1860s. His fascinating autobiography (Standing Bear, 2006 [1928]) tells the story
of the life of a man who grew up when the world of the Sioux was changing dramatically––a man
who had experienced traditional ways of life on the Great Plains, who then attended Carlisle Indian
Industrial School in Pennsylvania, who later in his life traveled with Buffalo Bill to Europe where he
“had the honor being introduced to King Edward the Seventh, the monarch of Great Britain” (Standing
Bear, 2006 [1928], p. 256), who was injured in a train accident in 1903, and who came to write his own
books. He died in 1939. The following lines, taken from his autobiography, speak for themselves about
the change in his world:
Our scouts, who had gone out to locate the buffalo, came back and reported that the plains were
covered with dead bison. These had been shot by the white people. The Indians never were such
wasteful, wanton killers of this noble game animal. We kept moving, fully expecting soon to
run across plenty of live buffalo; but we were disappointed. I saw the bodies of hundreds of dead

307
308

Julian N. Marewski and Ulrich Hoffrage

buffalo lying about, just wasting, and the odor was terrible ... Now we began to see white people
living in dugouts, just like wild bears, but without the long snout. These people were dirty. They
had hair all over their faces, heads, arms, and hands. This was the first time many of us had ever
seen white people, and they were very repulsive to us. None of us had ever seen a gorilla, else
we might have thought that Darwin was right concerning these people.
Standing Bear, 2006 [1928], p. 67
3 Take the (real) story of Plenty Horses, a Lakota born in 1869, who was brought from the West to a
boarding school (Carlisle Indian Industrial School) in the East (Pennsylvania), where he stayed from
1883–1888. When he went back to his tribe, a few days after the Wounded Knee Massacre (which
happened on December 29, 1890), he killed an army Lieutenant by firing a bullet through the back of
his head, a man with whom he had just previously “chatted” (Gitlin, 2011, p. 96). Plenty Horses was
put on trial. He explained:
Five years I attended Carlisle and was educated in the ways of the white man. When I returned
to my people, I was an outcast among them. I was no longer an Indian. I was not a white man.
I was lonely. I shot the lieutenant so I might make a place for myself among my people. I am now
one of them. I shall be hung, and the Indians will bury me as a warrior.
cited in Fear-Segal & Rose, 2016, p. 3
4 The Sioux documented events and experiences in pictorial form. These chronicles, painted on hides,
are referred to as winter counts; they represent a valuable source for contemporary historians. As
Hämäläinen (2019) points out, “[t]here is a tangible change in winter counts during and after the
painful years between 1877 and 1890. Several counts stopped in the late 1870s, ending with Crazy
Horse’s death, with the U.S. Army confiscating horses, or with children sent to boarding schools. “Big
Foot killed,” reads one of the few winter counts that refers to the Wounded Knee Massacre, the silence
capturing the enormity of the shock and trauma.”
5 Also Plenty Horses’ life took a surprising turn. At the end of his trial, he was not found guilty of
murder and not hanged, but acquitted and released. The argument was that a state of war had existed;
without a state of war also the soldiers who participated in the Wounded Knee massacre a few days
beforehand could have been accused of murder, too. Assuming a state of war helped to exonerate them
(Fear-Segal & Rose, 2016). Plenty Horses died in 1933, like our fictional character Leaning-Bear.
6 To the extent to which there are repeated learning opportunities (e.g., repeated model calibration and
testing) and the compositions of the learning samples and the test samples only differ at random across
repetitions and that average (rather than one-shot performance) matters to an organism, this kind of
uncertainty might even invite attempts to ‘optimize’, that is, to identify the strategy in the repertoire that
performs, on average, best across many test samples. While past research on fast-and-frugal heuristics has
stressed assessing absolute accuracy when making inferences out-of-sample, one might, alternatively, want
to identify the strategy that suffers the least when generalizing from the calibration to the test samples.
Note that these two different goal functions may lead to the selection of different strategies, which, in
turn, offers an interesting link to the psychology of happiness: A person’s happiness, after a loss, might
not depend only on the final state after the loss (e.g., health, prestige, fortune), but on comparisons with
reference points in the past, that is, the magnitude of the loss. Do people make comparative evaluations
when selecting their decision making strategies in environments hit by winds of change?
7 Interestingly, that lemma seems to have propelled Luther Standing Bear, as a young boy, directly into
the ‘white’ man’s world, eventually boosting his ability to navigate that new environment in later
phases of his life. In his autobiography (Standing Bear, 2006 [1928]), he discusses his motivations to
go East (to Pennsylvania to Carlisle Indian Industrial School) in 1879: “I was thinking of my father,
and how he had many times said to me, ‘Son, be brave! Die on the battle-field if necessary away from
home. It is better to die young than to get old and sick and then die.’ When I thought of my father,
and how he had smoked the pipe of peace, and was not fighting any more, it occurred to me that
this chance to go East would prove that I was brave if I were to accept it.” (p. 124) He continues by
pointing out: “I had come to this school merely to show my people that I was brave enough to leave
the reservation and go East, not knowing what it meant and not caring.” (p. 135) “It did not occur to
me at that time that I was going away to learn the ways of the white man.” (p. 128)
8 While our fictitious Jenny only appears in one section of this chapter, Dicke and Helbing’s (2017) science
fiction novel iGod is entirely centered around their protagonist Alex. The life of Alex is a projection of
the authors’ speculations, or, shall we say, informed guesses, given that both authors are scientists, and the

308
309

The winds of change

reader, wondering whether he has a utopia or a dystopia in his hands, is invited to imagine how a world
shaped by digitalization, big data, artificial intelligence, and social scoring might soon look.
9 Available at: https://policies.google.com/privacy?hl=en#infocollect (accessed January 21, 2020).
10 Available at: https://ec.europa.eu/info/policies/justice-and-fundamental-rights/criminal-justice/e-
evidence-cross-border-access-electronic-evidence_en (accessed January 21, 2020).
11 Tragically, ‘horse-stealing’ seemed actually acceptable to American policy-makers: In an attempt to
dismount the Sioux (Lakota), their horses were taken away in the aftermath of the battle of the Little
Big Horn (Waggoner, 2013).
12 Whoever has read, in detail, the long privacy statements that come with digital services knows what
we are talking about: Page-long, seemingly transparent disclosure statements alerting users that their
data can be pooled, sold, or otherwise shared with other parties, including other companies or gov-
ernment agencies (see also Box 18.1).
13 Traces of this view can be found in the decision sciences, bibliometrics, and statistics, with null-
hypothesis significance testing or metric-based science evaluations representing examples of seemingly
automatic, objectifying decision procedures (see Marewski & Bornmann, 2020; Gigerenzer, 2018).
14 Gleichschaltung brought about by digital technologies represents just one way of reducing diversity.
Another way to reduce diversity is to set nudges (Thaler & Sunstein, 2009) that lead a large number
of people to behave in the same way, namely, in line with the behavioral options championed by the
policy maker. In that sense, nudges can be thought of as potentially inducing “behavioral biases” (see
also Kozyreva et al., in press, and Helbing et al., 2017, on “big nudging”).
15 If such devices collected data that, furthermore, would aid in making or justifying judgments about a
child’s descendants, then issues related to surveillance, privacy, security, and (e.g., heath-related) dis-
crimination would become inter-generational. Note that such inter-generational judgments would
not need to be justifiable, scientifically. What would matter would be that, for instance, people in
power would be able to believe or pretend that attributes (e.g., inferred IQ or psychopathic person-
ality traits) were informative across generations. Numeric scores may be particularly susceptible to such
exploitation as they may be able to lend seeming objectivity to dubious claims and/or policies, as the
history of IQ and other indicators, used to score individuals, illustrates (see e.g., Young, 1922; Gould,
1981; Gigerenzer, Swijtink, Porter, Daston, Beatty, & Krüger, 1989; Severson, 2011).

References
Anderson, J. R. (1991). Is human cognition adaptive? Behavioral and Brain Sciences, 14, 471–517.
Anderson, J. R., Bothell, D., Byrne, M. D., Douglass, S., Lebiere, C., & Qin, Y. (2004). An integrated
theory of the mind. Psychological Review, 111, 1036–1060.
Anderson, J. R., & Schooler, L. J. (1991). Reflections of the environment in memory. Psychological Science,
2, 396–408.
Artinger, F. M., Artinger, S., & Gigerenzer, G. (2019). C. Y. A.: frequency and causes of defensive decisions
in public administration. Business Research, 12, 9–25.
Becker, G. S. (1976). The economic approach to human behavior. Chicago: The University of Chicago Press.
Binmore, K. (2007). Rational decisions in large worlds. Annales d’ Économie et de Statistique, 86, 25–41.
Bonhoeffer, D. (1998 [1942]). Rechenschaft an der Wende zum Jahr 1943 [Accountability at the turn
towards the year 1943]. In C. Gremmels, E. Bethge, & R. Bethge (Eds.), Bonhoeffer: Widerstand und
Ergebung. Briefe und Aufzeichnungen aus der Haft [Bonhoeffer: Resistance and surrender. Letters and
records from the time of arrest] (pp. 19–39). Gütersloh, Germany: Kaiser/Gütersloher Verlagshaus.
Brighton, H. (2006). Robust inference with simple cognitive models. In C. Lebiere & R. Wray (Eds.),
A.A.A.I. Spring Symposium: Cognitive science principles meet AI-hard problems (pp. 17–22). Menlo Park,
CA: American Association for Artificial Intelligence.
Brown, D. (1971). The Fetterman massacre. Lincoln, NE: University of Nebraska Press.
Brown, D. (1974). Begrabt mein Herz an der Biegung des Flusses [Bury my heart at the turn of the river].
München, Germany: Knaur. (Original work published in 1970 by Holt Rinehart, and Winston,
New York, as “Bury My Heart at Wounded Knee”.)
Brown, D., & Schmitt, M. F. (1976). Fighting Indians of the West. New York: Ballantine Books.
Brunswik, E. (1955). Representative design and probabilistic theory in a functional psychology. Psychological
Review, 62, 193–217.
Bullock, S., & Todd, P. M. (1999). Made to measure: Ecological rationality in structured environments.
Minds and Machines, 9, 497–541.

309
310

Julian N. Marewski and Ulrich Hoffrage

Czerlinski, J., Gigerenzer, G., & Goldstein, D. G. (1999). How good are simple heuristics? In G.
Gigerenzer, P. M. Todd, & The ABC Research Group, Simple heuristics that make us smart (pp. 97–118).
New York: Oxford University Press.
Dicke, W., & Helbing, D. (2017). iGod. Wroclaw, Poland: Amazon Fullfillment.
Dwyer, L. (2011). Why are Silicon Valley executives sending their kids to a tech-free school? Available
at: www.good.is/articles/why-are-silicon-valley-executives-sending-their-kids-to-a-tech-free-school
(accessed January 22, 2020).
Emberson, P. (2009). From Gondhishapur to Silicon Valley. Tobermory, Scotland: Etheric Dimensions Press.
Fear-Segal, J. & Rose, S.D. (2016). Introduction. In J. Fear-Segal & S.D. Rose (Eds.), Carlisle Indian
Industrial School: Indigenous histories, memories, and reclamations (pp. 1–34). Lincoln, NE: University of
Nebraska Press.
Fukuyama, F. (2006). The end of history and the last man. New York: Simon & Schuster. (Original work
published in 1992.)
Gaissmaier, W., & Marewski, J. N. (2011). Forecasting elections with mere recognition from small, lousy
samples: A comparison of collective recognition, wisdom of crowds, and representative polls. Judgment
and Decision Making, 6, 73–88.
Gibson, J. J. (1986). The ecological approach to visual perception. Hillsdale, NJ: Lawrence Erlbaum.
(Original work published in 1979 by Houghton Mifflin, Boston.)
Gigerenzer, G. (1998). Surrogates for theories. Theory & Psychology, 8, 195–204.
Gigerenzer, G. (2008a). Rationality for mortals: How people cope with uncertainty. New York: Oxford
University Press.
Gigerenzer, G. (2008b). Why heuristics work. Perspectives on Psychological Science, 3, 20–29.
Gigerenzer, G. (2014). Risk savvy: How to make good decisions. New York: Viking.
Gigerenzer, G. (2018). Statistical rituals: The replication delusion and how we got there. Advances in
Methods and Practices in Psychological Science, 1, 198–218.
Gigerenzer, G., & Brighton, H. (2009). Homo heuristicus: Why biased minds make better inferences.
Topics in Cognitive Science, 1, 107–143.
Gigerenzer, G., & Gaissmaier, W. (2011). Heuristic decision making. Annual Review of Psychology, 62,
451–482.
Gigerenzer, G., Gaissmaier, W., Kurz-Milcke, E., Schwartz, L. M., & Woloshin, S. (2007). Helping
doctors and patients to make sense of health statistics. Psychological Science in the Public Interest, 8, 53–96.
Gigerenzer, G., & Goldstein, D. G. (1996). Reasoning the fast and frugal way: Models of bounded ration-
ality. Psychological Review, 103, 650–669.
Gigerenzer, G., Hertwig, R., & Pachur, T. (Eds.). (2011). Heuristics: The foundations of adaptive behavior.
New York: Oxford University Press.
Gigerenzer, G., Hoffrage, U., & Goldstein, D. G. (2008). Fast and frugal heuristics are plausible models
of cognition: Reply to Dougherty, Franco-Watkins, and Thomas (2008). Psychological Review, 115,
230–239.
Gigerenzer, G., Hoffrage, U., & Kleinbölting, H. (1991). Probabilistic mental models: A Brunswikian
theory of confidence. Psychological Review, 98, 506–528.
Gigerenzer, G., Swijtink, Z., Porter, T., Daston, L. J., Beatty, J., & Krueger, L. (1989). The empire of
chance: How probability changed science and everyday life. Cambridge: Cambridge University Press.
Gigerenzer, G., Todd, P. M., & the ABC Research Group (1999). Simple heuristics that make us smart.
New York: Oxford University Press.
Gitlin, M. (2011). The Wounded Knee massacre. Santa Barbara, CA: ABC-CLIO.
Goldstein, D. G., & Gigerenzer, G. (2002). Models of ecological rationality: The recognition heuristic.
Psychological Review, 109, 75–90.
Goldstein, D. G., & Gigerenzer, G. (2009). Fast and frugal forecasting. International Journal of Forecasting,
25, 760–772.
Goodwin, G., & Goodwin, N. (2018). Les guerriers silencieux: Journaux Apaches [The silent warriors: Apache
diaries]. Monaco: Groupe Elidia. (Original work published in 2000 by the University of Nebraska
Press, Lincoln, as “The apaches diaries. A father-son journey”.)
Gould, S. J. (1981). The mismeasure of man. New York: Penguin.
Grüne-Yanoff, T., & Hertwig, R. (2016). Nudge versus boost: How coherent are policy and theory?
Minds and Machines, 26, 149–183.
Guercini, S. (2019). Heuristics as tales from the field: The problem of scope. Mind and Society, 18,
191–205.

310
311

The winds of change

Hafenbrädl, S., Waeger, D., Marewski, J. N., & Gigerenzer, G. (2016). Applied decision making with fast-
and-frugal heuristics. Journal of Applied Research in Memory and Cognition, 5, 215–231.
Hämäläinen, P. (2012). L’Empire Comanche [The Comanche empire]. Toulouse, France: Anacharis
Éditions. (Original work published in 2008 by Yale University Press, New Haven and London, as
“The Comance Empire”.)
Hämäläinen, P. (2019). Lakota America: A new history of indigenous power. New Haven, CT: Yale
University Press.
Hassrick, R. B. (1992). Das Buch der Sioux [The book of the Sioux]. Augsburg, Germany: Weltbild
Verlag. (Original work published in 1964 by the University of Oklahoma Press, Norman, as “The
Sioux”.)
Helbing, D. (2015). Thinking ahead: Essays on big data, digital revolution, and participatory market society.
Cham, Switzerland: Springer Nature Switzerland.
Helbing, D. (2019). Towards digital Enlightenment: Essays on the dark and light sides of the digital revolution.
Cham, Switzerland: Springer Nature Switzerland.
Helbing, D., Frey, B. S., Gigerenzer, G., Hafen, E., Hagner, M., Hofstetter, Y., ... Zwitter, A. (2017). Will
democracy survive big data and artificial intelligence? Available at: www.scientificamerican.com/art-
icle/will-democracy-survive-big-data-and-artificial-intelligence/ (accessed January 22, 2020).
Hertwig, R., Hoffrage, U., & the ABC Research Group (2013). Simple heuristics in a social world.
New York: Oxford University Press.
Hertwig, R., & Todd, P. M. (2003). More is not always better: The benefits of cognitive limits. In
D. Hardman & L. Macchi (Eds.), The psychology of reasoning and decision making: A handbook (pp. 213–
231). Chichester: Wiley.
Hicks, J. S., Burgman, M. A., Marewski, J. N., Fidler, F., & Gigerenzer, G. (2012). Decision making in a
human population living sustainably. Conservation Biology, 26, 760–768.
Hirschfeld, L. A., & Gelman, S. A. (Eds.) (1994). Mapping the mind. Cambridge: Cambridge University Press.
Hoffrage, U. (2019). Digitalisierung: Verheissung und Verhängnis. AGORA – in geänderter Zeitlage, 5/6/
7, 25–29.
Hoffrage, U., Hafenbrädl, S., & Marewski, J. N. (2018). The fast-and-frugal heuristics program. In L. J.
Ball & V. A. Thompson (Eds.), The Routledge international handbook of thinking & reasoning (pp. 325–
345). Abingdon: Routledge.
Hoffrage U., Lindsey S., Hertwig R., & Gigerenzer G. (2000). Communicating statistical information.
Science, 290, 2261–2262.
Hoffrage, U., & Marewski, J. (2015). Unveiling the Lady in Black: Modeling and aiding intuition. Journal
of Applied Research in Memory and Cognition, 4, 145–163.
Hoffrage, U., & Marewski, J. N. (2020). Social Scoring als Mensch-System-Interaktion. In O. Everling
(Ed.), Social Credit Rating: Reputation und Vertrauen beurteilen (pp. 305–329). Springer Verlag.
Katsikopoulos, K. V., Schooler, L., & Hertwig, R. (2010). The robust beauty of ordinary information.
Psychological Review, 117, 1259–1266.
Kitano, H. (2004). Biological robustness. Nature Reviews Genetics, 5, 826–837.
Knight, F. H. (1921). Risk, uncertainty and profit. New York: Houghton Mifflin.
Kozyreva, A., Lewandowsky, S., & Hertwig, R. (in press). Citizens versus the internet: Confronting digital
challenges with cognitive tools. Psychological Science in the Public Interest.
Maidique, M. (2012). The leader’s toolbox. Available at: https://lead.fiu.edu/resources/news/archives/
the-leaders-toolbox-by-dr-modesto-maidique.html (accessed January 27, 2020).
Manches, A. (2019). The Internet of Toys for children: Risk or opportunity? Paper presented at the
International Workshop of the Herbert Simon Society, The Digital World, Cognition and Behavior,
Turin, Italy.
Manches, A., Duncan, P., Plowman, L., & Sabeti, S. (2015). Three questions about the Internet of things
and children. TechTrends, 59, 76–83.
Manimala, M. (1992). Entrepreneurial heuristics: A comparison between high PI (pioneering-innovative)
and low PI ventures. Journal of Business Venturing, 7, 477–504.
Marewski, J. N., & Bornmann, L. (2020). Opium in science and society: Numbers. Unpublished
manuscript.
Marewski, J. N., Gaissmaier, W., & Gigerenzer, G. (2010). Good judgments do not require complex cog-
nition. Cognitive Processing, 11, 103–121.
Marewski, J. N., & Gigerenzer, G. (2012). Heuristic decision making in medicine. Dialogues in Clinical
Neuroscience, 14, 77–89.

311
312

Julian N. Marewski and Ulrich Hoffrage

Marewski, J. N., & Schooler, L. J. (2011). Cognitive niches: An ecological model of strategy selection.
Psychological Review, 118, 393–437.
Marewski, J. N., Schooler, L. J., & Gigerenzer, G. (2010). Five principles for studying people’s use of
heuristics. Acta Psychologica Sinica, 42, 72–87.
Martignon, L., & Hoffrage, U. (2002). Fast, frugal and fit: Simple heuristics for paired comparison. Theory
and Decision, 52, 29–71.
Martignon, L., & Hoffrage, U. (2019). Wer wagt, gewinnt? Wie Sie die Risikokompetenz von Kindern und
Jugendlichen fördern können [He who dares wins? How you can foster the risk competence of children
and adolescents]. Bern, Switzerland: Hogrefe.
McElreath, R., Wallin, A., & Fasolo, B. (2013). The evolutionary rationality of social learning. In R.
Hertwig, U. Hoffrage, & the ABC Research Group, Simple heuristics in a social world (pp. 381–408).
New York: Oxford University Press.
Mousavi, S., & Gigerenzer, G. (2014). Risk, uncertainty and heuristics. Journal of Business Research, 67,
1671–1678.
Mousavi, S., & Gigerenzer, G. (2017). Heuristics are tools for uncertainty. Homo Oeconomicus, 34, 361–379.
Newell, A. (1973). You can’t play 20 questions with nature and win: Projective comments on the papers of
this symposium. In W. G. Chase (Ed.), Visual information processing (pp. 283–310). New York: Academic
Press.
O’Neil, C. (2016). Weapons of math destruction: How Big Data increases inequality and threatens democracy.
New York: Penguin Random House.
Palazzo, G., Krings, F., & Hoffrage, U. (2012). Ethical blindness. Journal of Business Ethics, 109, 323–338.
Savage, L. J. (1972 [1954]). The foundation of statistics. New York: Dover Publications. (Original work
published in 1954 by John Wiley & Sons.)
Schooler, L. J., & Hertwig, R. (2005). How forgetting aids heuristic inference. Psychological Review, 112,
610–628.
Severson, K. (2011). Thousands sterilized, a state weighs restitution. New York Times, December 9.
Simon, H. A. (1990). Invariants of human behavior. Annual Review of Psychology, 41, 1–19.
Standing Bear, L. (2006 [1928]). My people the Sioux. Lincoln, NE: University of Nebraska Press.
Stephens, D. W., & Krebs, J. R. (1986). Foraging theory. Princeton, NJ: Princeton University Press.
(Original work published in 1928 by Houghton Mifflin Company.)
Stevens, J. R., Marewski, J. N., Schooler, L. J., & Gilby, I. C. (2016). Reflections of the social environ-
ment in chimpanzee memory: Applying rational analysis beyond humans. Royal Society Open Science,
3, 160293.
Sull, D. N., & Eisenhardt, K. M. (2015). Simple rules: How to thrive in a complex world. New York: Houghton
Mifflin Harcourt.
SVRV (2018). Verbrauchergerechtes Scoring. Gutachten des Sachverständigenrats für Verbraucherfragen
[Consumer-friendly scoring. Expert report of the Advisory Council for Consumer Affairs]. Berlin:
Sachverständigenrat für Verbraucherfragen. Available at: www.svr-verbraucherfragen.de/wp-content/
uploads/SVRV_Verbrauchergerechtes_Scoring.pdf (accessed December 13, 2018.)
Thaler, R. H., & Sunstein, C. R. (2009). Nudge: Improving decisions about health, wealth, and happiness.
London: Penguin Books. (Original work published in 2008 by Yale University Press, New Haven,
CT.)
Todd, P. M., Gigerenzer, G., & the ABC Research Group (2012). Ecological rationality: Intelligence in the
world. New York: Oxford University Press.
Tooby, J., & Cosmides, L., (1992). The psychological foundations of culture. In J. H. Barkow, L. Cosmides,
& J. Tooby (Eds.), The adapted mind: Evolutionary psychology and the generation of culture (pp. 19–136).
New York: Oxford University Press.
Waggoner, J. (2013). Witness: A Hunkpapha historian’s strong-heart song of the Lakotas. Lincoln, NE:
University of Nebraska Press.
Woike, J. K., Hoffrage, U., & Martignon, L. (2017). Integrating and testing natural frequencies, Naïve
Bayes, and fast-and-frugal trees. Decision, 4, 234–260.
Wolf, B. (1999). Vicarious functioning as a central process-characteristic of human behavior. Available at: http://
brunswik.org/notes/essay4.html (accessed February 5, 2020).
Young, K. (1922). Intelligence tests of certain immigrant groups. The Scientific Monthly, 15(5), 417–434.

312
313

19
ECOLOGICAL RATIONALITY
Bounded rationality in an evolutionary light

Samuel A. Nordli and Peter M. Todd

Herbert Simon famously characterized bounded rationality as being “shaped by a scissors

whose two blades are the structure of task environments and the computational capacities of
the actor” (1990, p. 7). Illuminated by the “light of evolution,” research on bounded rationality
in humans can be seen as a proximate cross-sectional analysis of our species’ current point in
its evolutionary trajectory—that is, the study of human decision making and problem solving
as interactions between (A) our present capacities, needs, and desires/goals, and (B) the struc-
ture of the present environmental contexts in which such decision making takes place. But
the evolutionary perspective enables us to do more, looking at the ultimate selection pressures
that operated over time in our and other species’ evolutionary trajectories—that is, the study
of any organism’s decision making as interactions between (A) its evolved capacities, needs,
and desires/goals, in terms of (B) how those aspects fit the structure of the past environmental
decision-making contexts, and (C) how they match or mismatch the structure of the present
environmental contexts. The study of ecological rationality extends the concept of bounded
rationality by situating it within an evolutionary perspective: Like other Darwinian approaches
to psychology (e.g., Barkow, Cosmides, & Tooby, 1992), ecological rationality emphasizes the
role of the past environment to which we are adapted, the present environment in which we
make decisions, and the ecology and structure of information as it is processed by decision-
making mechanisms (e.g., Todd, Gigerenzer, & the ABC Research Group, 2012). Ecological
rationality seeks to understand decision making in much the same way as the traditional
bounded rationality approach, but posits that this understanding will necessarily be species-
specific (revealing both commonalities and differences between humans and other organisms),
and must be informed by each species’ evolutionary history (including past environments) as
well as what we know of evolutionary processes in general. In this way, the ecological ration-
ality approach follows in the investigative tradition of behavioral ecology as championed by
Tinbergen (1963), who argued that proximate (mechanistic and developmental) and ultimate
(adaptive and evolutionary) analyses are each required in order to fully understand observed
behavior.

313
314

Samuel A. Nordli and Peter M. Todd

Bounded rationality as proximate, ecological rationality

as ultimate analyses
As formulated by Simon, bounded rationality is essentially concerned with mechanistic
answers to proximate questions about problem solving and decision making. What decision-
making mechanisms or strategies are used in a particular task environment and what are their
outcomes? How do these mechanisms function? What are their practical limits? What are the
relevant structures of the task environment, and how do those variables (and variation therein)
affect operations and outcomes? Answers to these and other such what and how questions are
integral to a well-developed understanding of human decision making and problem solving,
and they often help to shape and answer the ultimate questions of why and whence. However,
much of decision research tends to overlook ultimate analyses and simply settles for mechan-
istic descriptions as explanations. This lack of evolutionary contextualization can leave unreal-
istic notions of rationality unchallenged, engendering a tendency to blame irrational biases or
systematic inadequacies when human behavior does not match the optimal performance of
normative behavioral prescriptions (e.g., Tversky & Kahneman, 1974). From this proximate
perspective, the failure of boundedly-rational behavior to reach classically rational levels of per-
formance is just that: a failure.
Ecological rationality research charts a different course. It does not predefine rationality
based on idealized expectations of how people should act in order to optimize performance
(given infinite resources and unbounded capacities). Instead, it seeks both to describe how
people actually act in specific contexts, and to understand how decisions and actions fit together
with particular environmental contexts in terms of their evolutionary provenance and adaptive
significance. Judgment and decision making are unavoidable necessities for most species, whose
very existence suggests that they are adapted to be effective decision makers: Individuals who
were ill-equipped to respond to basic fitness challenges—such as finding and selecting good
food, a mate, and a habitat—were less likely to survive and reproduce than their more-capable
contemporaries. When an organism’s decision-making behavior works well in a given context,
ecological rationality research explores how evolved cognitive capacities and evolved or learned
decision strategies fit with environment structure to produce the adaptive behavior; conversely,
when the behavior in a given context runs contrary to what might be expected (e.g., because
it results in error or appears to be maladaptive), ecological rationality asks questions from an
evolutionary perspective to uncover how an ultimate understanding could explain why the
observed behavior occurs in that context.

Decision making: driven by goals, shaped by ecological structure

and variability
For an individual decision maker (and ultimately for its species), a decision is a means to an end.
In this sense, decisions and their associated goals (whether proximate or ultimate) are insepar-
able, as a decision represents the selection and application of behavior in service of a goal within
some environment. The pursuit of a specific goal by an environmentally-situated decision
maker constitutes a particular ecological context. One way that such contexts can be divided
systematically is according to whether their associated goals can be achieved via the repetition
or modulation of past behavior (i.e., ‘past’ in multiple senses, being drawn from the life his-
tory of a decision-making organism, the cultural history of its community, or the evolutionary
history of its species). This can be framed in terms of exploration and exploitation: Does the
structure of the present environment allow the application of some form of past behavior to

314
315

Ecological rationality

achieve the goal associated with a particular context? If yes, that possibility can be exploited; if
no (or if the answer is unknown), exploration may uncover new or existing means of achieving
that goal, which may then be exploited for similar contexts in the future. For example, a habit
is a past behavior being exploited in a current situation, while trial-and-error learning is explor-
ation to find a new behavior for a new situation. (Furthermore, when considered as a search
strategy that is applied repeatedly in various exploratory contexts, trial-and-error learning is
itself a form of exploiting past behavior, in an exploratory context.) Ecological rationality can
be conceptualized as the study of how individuals and species are adapted to identify and take
advantage of opportunities in which past behavior can be exploited to achieve their goals in
present contexts, including cases in which the exploited behavior is a strategic approach to
exploration.
The use of ‘past’ behavior in present contexts means reusing behavior that matches particular
structural features of environments that co-occur across some range of contexts over time (Todd
& Brighton, 2016). For example, shifting an automobile into drive is a recurrent goal for many
people; different instances of exploiting past shifting behavior will be nearly identical from day
to day across slight variations in the context of driving a family car (e.g., if one’s spouse has
adjusted the seat), but shared features in a rented pickup truck may also permit the exploit-
ation of past shifting behavior in a new driving context. Similarly, the anatomy, physiology, and
environment of Caenorhabditis elegans worms share enough cross-generational features to permit
an inherited mechanism to guide stereotypical patterns of locomotion during foraging (Hills,
Brockie, & Maricq, 2004), which effectively allows individual worms in present environments
to exploit an ancestral search strategy that was similarly successful in past environments. Indeed,
as these divergent examples suggest, such exploitation is accomplished by a wide variety of
mechanisms and strategies that may be learned or evolved, specific or general. Contexts (goal
pursuit in different environment structures) can be organized in terms of the decision strategies
that can exploit them, in a way that bridges behavioral ecology and ecological rationality (see
Hutchinson & Gigerenzer, 2005), as much work in these areas (illustrated in the next section)
can be characterized under this framework as studying various manifestations of the same gen-
eral exploitative phenomenon.

Cue-based behavior: fast and frugal exploitation of statistical regularity

If time is not a strong pressure and/or the outcome of a decision depends on a number of
important factors, it can be worthwhile collecting and integrating information regarding the
specifics of those factors before making the decision. For example, as a colony prepares to
relocate, ants of the species Leptothorax albipennis appear to collect and differentially weight
information regarding a number of prospective nest site cavity characteristics—including
the area, entrance size, ceiling height, and amount of ambient light for each space—before
choosing the best location (Franks, Mallon, Bray, Hamilton, & Mischler, 2003). However,
time is typically a significant factor for many decision contexts that animals face in the world,
whether directly (e.g., a prey animal deciding in the moment how to react to a predator in
pursuit) or indirectly (e.g., the more time an animal spends courting, the less time it will have
for foraging), whereby animals will generally be at an advantage if they can make effective
decisions faster than competitors (Todd, 2001). Thus, for contexts in which goals are achiev-
able via some reiteration of past (i.e., recalled, inherited, or imitated) behavior, it is typ-
ically advantageous to identify those contexts and decide to implement those behaviors as
quickly and efficiently as possible, assuming outcomes are equivalent. One way to achieve
such efficiency is to ignore as much irrelevant information as possible, which can streamline

315
316

Samuel A. Nordli and Peter M. Todd

the search for relevant information as well as minimize the time and energy spent processing it
(Gigerenzer & Todd, 1999; see also the effort-reduction framework of Shah & Oppenheimer,
2008). By relying on only the subset of environmental cues that are sufficient to distinguish
a given context and characterize its structure, an organism can quickly make decisions that
exploit the structural regularities that are indicated by those cues. The following examples
illustrate the ecologically rational use of contextual cues to exploit predictable features of
environmental structure.
In behavioral ecology, fixed action patterns—stereotypical sequences of unlearned behavior
that are triggered by certain contextual cues—are exhibited by species whose individuals have
effectively inherited specific exploitative behaviors to be used in specific environmental settings
that ancestral generations encountered sufficiently often for selection to occur. For instance,
when greylag geese (Anser anser, a ground-nesting species) are brooding, the cue of visually
perceiving an egg outside the nest elicits a scripted sequence of behavior whereby a goose
extends its neck and uses its beak to roll the egg toward itself and into the nest (Alcock, 2005).
Other exploitative strategies often studied in behavioral ecology are of rules of thumb, which are
evolved algorithmic decision-making mechanisms used by many species to achieve particular
contextualized goals. Scouts of the above-mentioned ant species, Leptothorax albipennis, use a
rule-of-thumb algorithm to determine whether the area of potential nesting site cavities is large
enough for their colonies (one of the factors in their relocation decision)—ants explore such
sites twice, first laying a pheromone trail as they walk a random path of fixed length, and then
walking the same distance along a new random path and assessing the rate at which it intersects
its prior trail as a reliable indication of the approximate area of the cavity (Mugford, Mallon, &
Franks, 2001).
Paralleling fixed action patterns, humans (and other vertebrates) form habits when they
repeat goal-oriented behaviors in recurrent contexts: With continued and consistent repetition,
a learned behavioral sequence will coalesce into a singular ‘chunk’ of behavior that comes to be
executed automatically in its entirety upon being triggered by contextual cues (Graybiel, 2008).
Habit formation is essentially a mechanism through which procedurally learned patterns of
behavior come to mimic the cue-driven efficiency and efficacy of evolved behavioral sequences
such as fixed action patterns (Nordli & Todd, 2017). This connection is not merely concep-
tual: Learned habits and inherited fixed action patterns appear to be controlled by the same
recursive neural circuitry in vertebrates (Graybiel, 1995; Berridge, Aldridge, Houchard, &
Zhuang, 2005), and—when observed externally—behaviors of one type can be indistinguish-
able from those of the other, without prior knowledge of the behaviors’ ontogeny or phylogeny
(Burkart, Schubiger, & van Schaik, 2017).
Likewise, rules of thumb are paralleled by the simple heuristics that are often studied by
researchers in ecological rationality (Gigerenzer, Todd, & the ABC Research Group, 1999).
A heuristic, as defined by Gigerenzer and Gaissmaier (2011), “is a strategy that ignores part of
the information, with the goal of making decisions more quickly, frugally, and/or accurately
than more complex methods” (p. 454). Rephrased in the framework of exploitable contexts,
heuristics are cue-guided decision strategies that are used to make choices and pursue goals
in particular contexts by exploiting regularities in environment structure with particular
variations of past decision-making behavior. Heuristics operate in a manner that is “fast and
frugal”—fast because the decision mechanism generating behavior in a given environment
uses little computation, and frugal because it operates on little information. Perhaps the sim-
plest example is the recognition heuristic, which states if you recognize only one of two possible
alternatives, choose the one you recognize. When German students were asked which city was more
populous, San Diego or San Antonio, 100 percent of them responded correctly (San Diego)

316
317

Ecological rationality

compared to about 66 percent of American students answering the same question (Goldstein
& Gigerenzer, 2002). Having heard of San Diego, the German students—mostly not recog-
nizing San Antonio—were able to use the recognition heuristic to answer the question cor-
rectly in this context because larger cities are generally more famous, hence more commonly
recognized (Todd, 2007). Here, recognition is information about past situations that the indi-
vidual has encountered, and that crucially has a reliable structure to it: recognized options are
likely to be higher on the decision criterion (here, city size). This structure paired with the
simple recognition heuristic together exploit the past experience of the individual to yield
good decisions in current situations.
Fixed action patterns, rules of thumb, habits, and heuristics are all examples of decision-
making tools that are ecologically rational, i.e., these tools are efficient means of exploiting
effective pairings of particular strategies and particular environment structures that have been
discovered through past behavior to achieve goals in recurrent contexts. In the field of ecological
rationality, the adaptive toolbox is the conceptual collection of decision-making mechanisms that
an individual draws from to help them successfully navigate the world (Gigerenzer, 2001).
Broadly, ecological rationality is the study of such decision-making tools, the structure of the
environments in which those tools are used to make decisions, and, crucially, the specifics of
how and why those tools and environments fit together to serve the ultimate interests of deci-
sion makers (Todd & Brighton, 2016). This section has stressed that ecological rationality is
species-specific and pertains to the decision-making ecology of many different species, allowing
work from behavioral ecology to be conceptually appreciated as ecological rationality research;
however, since most of this research so far has focused on human heuristic decision making and
search strategies, the remainder of this chapter will primarily focus on the ecological rationality
of our own species, concluding with directions for future work.

Fitting the right tool to the right context: ecological rationality in action
Research in ecological rationality often proceeds by identifying different decision contexts,
in terms of environmental information structures, that fit well with particular categories of
heuristics, in terms of their information processing structures. Contextual specificity is crit-
ical: Identifying and characterizing the tools that feature in individuals’ adaptive toolboxes must
be done by studying their use in specific situations. This is because effective simple heuristics
and other such tools are highly context-sensitive, working well in a limited set of environmental
structures but possibly producing poor decisions when used in others (Gigerenzer, Todd, & the
ABC Research Group, 1999).
Many heuristic strategies consist of three basic elements: a search rule, a stopping rule, and
a decision rule—that is, a heuristic algorithm can often be described by how it searches for
cues, how it stops that search, and how it decides based on the outcome of the search process
(Gigerenzer, 2001). In addition to divisions by context, heuristics are often grouped according
to overlap in their specific search, stopping, and decision rules. For example, one-reason deci-
sion heuristics are those that make decisions based on a single cue, evaluating all options (two
or more) on the basis of that cue and then stopping comparisons and making a decision if one
option scores higher than the others on that criterion. Take-the-best and the minimalist heur-
istic are examples of heuristics that base decisions on a single reason or factor, differing only in
how they search through cues (Todd & Gigerenzer, 2003). Take-the-best searches through cues
in a particular order, ranked according to their past successes, while the minimalist heuristic
searchers through cues in a random order. Potential choices are evaluated according to their
score on the first cue, only going on to consider the next cue if no option surpasses any other

317
318

Samuel A. Nordli and Peter M. Todd

on the first, and so on, for instance, deciding between two entrees at a restaurant according
to whichever is cheaper, or if both are the same price, according to whichever dish features
the greatest number of different vegetables, etc. When used in task contexts with appropri-
ately exploitable structure, the decisions made by take-the-best are at least as accurate as more
complex processing-intensive decision strategies such as linear regression models, and typic-
ally achieve such accuracy using much less information (e.g., Gigerenzer & Goldstein, 1996;
Czerlinski, Gigerenzer, & Goldstein, 1999). Figuring out which environment structures fit well
with one-reason heuristics such as take-the-best is a central component of ecological rationality
research.
For situations in which one-reason decision heuristics are insufficient to differentiate among
available options (as in some contexts with three or more alternatives), multiple-cue decision
heuristics can make appropriate choices based on two or more cues (Todd, 2007). The elimination
by aspects heuristic (Tversky, 1972) is one such strategy that culls options according to selected
criteria, narrowing down to a reduced set of alternatives, for instance, when deciding among
multiple restaurants to visit, first eliminating all those that are more than 10 miles away, and then
all those that are non-vegetarian. This process of elimination can be continued down to a single
option, or a reduced set of options can be decided among by some other means of discrimin-
ation (such as switching to a one-reason heuristic, if the context allows).
In contexts in which choice alternatives have yet to be identified or encountered, or in
which no known behavior can be exploited to achieve a specific goal, sequential search heuristics
are appropriate (Todd & Gigerenzer, 2003). For example, when seeking a long-term romantic
partner, how does an individual determine whether they have found a good possibility if a
number of other prospective mates remain available? After being passed over, potential partners
may pursue other relationships or may become uninterested in the meantime; environment
structure constraints such as these are typical of some sequential search contexts, for which
targets are encountered one at a time, and often cannot be retrieved after they have been passed
by (e.g., job search and housing search). One solution to such sequential search problems is
a satisficing strategy that sets an aspiration level for some criterion, and then, if an option is
discovered that meets or exceeds the aspiration level, stops searching and chooses that option
(Simon, 1955; Todd & Miller, 1999). Similarly, in the context of tasks with unknown behav-
ioral solutions, a strategy of trial-and-error learning can be used to search for behaviors that
successfully achieve goals.
How is a given heuristic selected for use in a particular context? This question highlights
an aspect of ecological rationality research that is not well-specified: The manner in which
heuristics and other decision tools in the adaptive toolbox are determined to appropriately
match the structure of a given situation, otherwise known as the strategy selection problem
(Marewski & Link, 2014). One possible solution to this problem is the development of a
mapping between context structure and appropriate heuristics to use (Marewski & Schooler,
2011). This can be done in part by a “strategy selection learning” procedure as proposed by
Rieskamp and Otto (2006), which suggests that strategies are applied according to predictions
regarding their expected success in specific scenarios. For any given instance of strategy selec-
tion, after applying the strategy with the greatest predicted success, a process of reinforcement
learning updates that expectation for future situations based on the current outcome. This
process can lead to occasional errors in which the selection of a strategy results in a mismatch
between behavior and environmental structure. Readers may recall examples where one of
their own habits was selected inappropriately, for instance, when needing to consciously locate
the gear stick in a rented car after unconsciously grasping at the air where the gear stick is found
in one’s own vehicle.

318
319

Ecological rationality

When behavior and beliefs diverge from expectations

Bounded and ecological rationality consider behavior as the product of actor-environment
interactions, both those that are occurring now and those that have occurred in the past and
shaped the current decision mechanisms and environmental structures. This means that species
and their environments are subject to change. When environmental change takes place, mul-
tiple generations of selection may need to occur before a species can evolve to be adapted to
new fitness challenges—thus, behavioral traits and mechanisms that are observed in present
environments are typically adapted to past environmental structure, which an ecological ration-
ality perspective prompts us to consider. If past and present environments differ in important
ways, this can lead to a mismatch of old cognitive mechanisms or strategies applied to modern
settings, which can result in poor decisions. An example of this is the strong evolved human
preference for sugary, salty, and fatty foods, which can lead to unhealthy food choices that
contribute to obesity and other chronic diseases (Cordain et al., 2005). The possibility of such
mismatches—whether between past adaptations and present environments in which they are
maladaptive, or between a strategy that works well in one modern setting but fails when applied
to another—serves to reiterate the importance of considering particular environments and
contexts when evaluating the (ecological) rationality of a particular decision strategy. Strategies
are not ecologically rational in and of themselves, but only in the context of specific situations
with particular environmental structure. A decision strategy such as if there is sweet food available,
eat it, is ecologically rational in an environment where foods of high caloric density are scarce or
difficult to obtain; but in wealthy modern environments where readily available foods are often
specifically manufactured to satisfy this human preference for sweets, an eat sweet food strategy
leads to bad decisions and undesirable consequences.
In addition to cases in which environments and strategies are mismatched, there are
examples where ostensibly irrational behavior (according to normative prescriptive models)
may be considered perfectly reasonable when evaluated from an ecological rationality perspec-
tive. Consider the “error tendencies” illustrated by Thagard (2011), in which the interaction of
emotion and cognition leads to beliefs that are purportedly irrational because they are unduly
influenced by motivated reasoning. Thagard suggests that the integration of cognitive and emo-
tional information has evolved to ensure that cognitive capacities are efficiently directed toward
achieving goals that are important for survival and reproduction (see also Gallese, Mastrogiorgio,
Petracca, & Viale, Chapter 23 in this volume); but despite this adaptive provenance, he asserts
that such integration often induces irrational beliefs when faced with the complexity of modern
environments (Thagard, 2011). This suggests that errors of motivated inference may simply be
further examples of environment-mechanism mismatch, but consider, for example, cases of sour
grapes (see e.g., Elster, 1983; Kay, Jimenez, & Jost, 2002), wherein the inaccessibility of a desired
result leads one to conclude that the result was not particularly desirable after all (e.g., I couldn’t
reach the grapes, but they were probably sour anyway). Sour grapes and other such instances
of motivated reasoning do appear to lead to beliefs that are logically unsound; however, when
considered from an evolutionary perspective, such processes may ultimately produce adaptive
behavior that reflects an ecologically rational strategy for balancing the risks, rewards, and prob-
abilities that characterize the environmental structure of contexts in which motivated reasoning
occurs—even in complex modern environments.
To illustrate, the logically unfounded belief that grapes are not any good because they cannot
be reached may simply be the product of an effective proximate mechanism (i.e., emotional
biases in specific contexts) that ultimately serves to discourage the pursuit of goals in situations
where contextual cues indicate that goal achievement is unlikely. Similarly, cases of fear-driven

319
320

Samuel A. Nordli and Peter M. Todd

inference (e.g., I don’t recognize this mole—I must have cancer!) may not be irrational “gut
overreactions … in which a feeling that something is wrong is erroneously taken as evidence
that something really is wrong” (Thagard & Nussbaum, 2014, p. 43); instead, they may reflect
an adaptive way to encourage cautious over-preparation when facing possible scenarios with
severe fitness consequences, such as for situations in which multiple Type I errors are still less
costly than a single fatal Type II error. Put another way, even if the result is a false belief, it may
ultimately be ecologically rational to tend to underestimate the value of positive-but-unlikely
outcomes (as in sour grapes) and to overestimate the probability of unlikely-but-costly scenarios
(as in fear-driven inference) if those motivational biases respectively serve to minimize fruitless
investments of time and energy and increase the chances of avoiding or mitigating the nega-
tive impact of an unlikely scenario that happens to come to pass (Haselton & Nettle, 2006).
Such examples of motivated reasoning appear to be qualitatively in line with what would be
expected from an adaptive system that uses emotional biases to influence behavior in response
to specific contextual cues. These examples reinforce the point that prescriptive proximate-only
definitions of rationality provide little explanatory insight when they classify behavior in the
world as irrational because it is contrary to normative prescriptions. The biases of motivated
reasoning may lead to false beliefs, but such biases should not be considered ‘irrational’ if they
are the result of an adaptive (e.g., ecologically rational) mechanism that yields decisions that
tend to appropriately balance the inherent tradeoff between risk and reward (see e.g., Pleskac
& Hertwig, 2014); rather, we should understand how such mechanisms function and under
what circumstances they operate, in order to combat false beliefs in cases where they may have
negative practical consequences.

Future directions in research on ecological rationality

Our general understanding of behavior will continue to develop as we construct a more com-
plete account of (1) the decision mechanisms that underlie behavior; (2) the ecological contexts
in which they are applied; and (3) the structural correspondences between decision mechanisms
and environments that result in successful (and unsuccessful) outcomes.
First, with respect to mechanisms, a full theoretical framework of ecological rationality
will be promoted by an understanding of the neurophysiological mechanisms that support
and produce decision behavior and cognition. For example, integrating work across cognitive
neuroscience and neurophysiology with research on ecological rationality will help solve the
problem of strategy selection. In this regard, one promising avenue to explore involves the
potential role played by neural circuitry involving the basal ganglia, a set of subcortical nuclei
in vertebrates. The reinforcement learning-based mechanism of strategy selection, as proposed
by Otto and Rieskamp (2006), is consistent with accounts that tie reinforcement learning
processes to recurrent cortico-basal circuitry (e.g., Yin & Knowlton, 2006), which is the very
same circuitry that is critical to the exploitation of past behavior in the form of habits and fixed
action patterns. Indeed, this circuitry may be critical to most ecologically rational behavior.
Graybiel (2008) explains how “repetitive behaviors, whether motor or cognitive, are built up
in part through the action of basal ganglia-based neural circuits that can iteratively evaluate
contexts and select actions and can then form chunked representations of action sequences”
(p. 361), which aptly characterizes the general process of exploiting past behavior in present
contexts, and offers a solution to strategy selection that is grounded in neurophysiology. Under
this neurological framework, an evaluative process determines the structure of the present con-
text, whereby a connection is made to past behavior, which is selected by the system, allowing

320
321

Ecological rationality

a specific behavioral sequence to attempt to exploit the contextual structure and achieve an
associated goal.
Second, more work is needed on systematic approaches to uncovering characteristics of
environment structure that are important in decision making. As mentioned above, behavioral
ecology (and other ecological perspectives) can supply useful theoretical frameworks (see e.g.,
Wilke et al., 2018). The study of situations in social psychology may also provide a different
unified approach (Todd & Gigerenzer, 2020). This perspective builds on Lewin’s (1936) pro-
posal of field theory, which describes behavior (B) as the product of some function (f) operating
on a person (P) and their environment (E), formulated by the simple equation

B = f(P, E)
A person’s perceptions of their current environment then goes into defining a given situation.
One of the challenges ahead is to determine whether different types of situations, as studied by
social psychology, can be described as different patterns of information that particular decision
heuristics can exploit, as studied in ecological rationality.
Finally, although ecological rationality seeks to understand behavior from appropriate prox-
imate and ultimate perspectives—to provide accurate descriptions of how and why behavior
occurs in specific contexts—the field also aims to facilitate accurate predictions of what behaviors
will occur under specific conditions and, ultimately, to use such a comprehensive understanding
to help people and institutions make better decisions (Gigerenzer & Todd, 2012).

References
Alcock, J. (2005). Animal behavior: An evolutionary approach (8th ed.). Sinauer Associates, Inc.
Barkow, J. H., Cosmides, L., & Tooby, J. (Eds.) (1992). The adapted mind: Evolutionary psychology and the
generation of culture. Oxford: Oxford University Press.
Berridge, K. C., Aldridge, J. W., Houchard, K. R., & Zhuang, X. (2005). Sequential super-stereotypy of
an instinctive fixed action pattern in hyper-dopaminergic mutant mice: A model of obsessive compul-
sive disorder and Tourette’s. BMC Biology, 3(4).
Burkart, J. M., Schubiger, M. N., & van Schaik, C. P. (2017). The evolution of general intelligence.
Behavioral and Brain Sciences, 40.
Cordain, L., Eaton, S. B., Sebastian, A., Mann, N., Lindberg, S., Watkins, B. A., … & Brand-Miller,
J. (2005). Origins and evolution of the Western diet: Health implications for the 21st century. The
American Journal of Clinical Nutrition, 81(2), 341–354.
Czerlinski, J., Gigerenzer, G., & Goldstein, D. G. (1999). How good are simple heuristics? In
G. Gigerenzer, P. M. Todd, & the ABC Research Group (Eds.), Simple heuristics that make us smart
(pp. 97–118). New York: Oxford University Press.
Dobzhansky, T. (1964). Biology, molecular and organismic. Integrative and Comparative Biology, 4(4),
443–452.
Elster, J. (1983). Sour grapes. Cambridge: Cambridge University Press.
Franks, N. R., Mallon, E. B., Bray, H. E., Hamilton, M. J., & Mischler, T. C. (2003). Strategies for
choosing between alternatives with different attributes: Exemplified by house-hunting ants. Animal
Behaviour, 65, 215–223.
Gigerenzer, G. (2001). The adaptive toolbox. In G. Gigerenzer & R. Selten (Eds.), Bounded rationality: The
adaptive toolbox (pp. 37–50). Cambridge, MA: MIT Press.
Gigerenzer, G., & Gaissmaier, W. (2011). Heuristic decision making. The Annual Review of Psychology, 62,
451–482.
Gigerener, G., & Goldstein, D. G. (1996). Reasoning the fast and frugal way: Models of bounded ration-
ality. Psychological Review, 103(4), 650–669.
Gigerenzer, G., & Todd, P. M. (1999). Fast and frugal heuristics: The adaptive toolbox. In G. Gigerenzer,
P. M. Todd, & the ABC Research Group (Eds.), Simple heuristics that make us smart (pp. 3–36).
New York: Oxford University Press.

321
322

Samuel A. Nordli and Peter M. Todd

Gigerenzer, G., Todd, P. M., & the ABC Research Group (Eds.). (1999). Simple heuristics that make us smart.
New York: Oxford University Press.
Goldstein, D. G., & Gigerenzer, G. (2002). Models of ecological rationality: The recognition heuristic.
Psychological Review, 109(1), 75–90.
Graybiel, A. M. (1995). Building action repertoires: Memory and learning functions of the basal ganglia.
Current Opinion in Neurobiology, 5, 733–741.
Graybiel, A. M. (2008). Habits, rituals, and the evaluative brain. The Annual Review of Neuroscience, 31,
359–387.
Haselton, M. G., & Nettle, D. (2006). The paranoid optimist: An integrative evolutionary model of cog-
nitive biases. Personality and Social Psychology Review, 10(1), 47–66.
Hills, T., Brockie, P. J., & Maricq, A. V. (2004). Dopamine and glutamate control area-restricted search
behavior in Caenorhabditis elegans. The Journal of Neuroscience, 24(5), 1217–1225.
Hutchinson, J. M. C., & Gigerenzer, G. (2005). Simple heuristics and rules of thumb: Where psychologists
and behavioural biologists might meet. Behavioural Processes, 69, 97–124.
Kay, A. C., Jimenez, M. C., & Jost, J. T. (2002). Sour grapes, sweet lemons, and the anticipatory rational-
ization of the status quo. Personality and Social Psychology Bulletin, 28(9), 1300–1312.
Lewin, K. (1936). Principles of topological psychology. New York: McGraw-Hill.
Marewski, J. N., & Link, D. (2014). Strategy selection: An introduction to the modeling challenge. Wiley
Interdisciplinary Reviews: Cognitive Science, 5, 39–59.
Marewski, J. N., & Schooler, L. J. (2011). Cognitive niches: An ecological model of strategy selection.
Psychological Review, 118, 393–437.
Mugford, S. T., Mallon, E. B., & Franks, N. R. (2001). The accuracy of Buffon’s needle: A rule of thumb
used by ants to estimate area. Behavioral Ecology, 12(6), 655–658.
Nordli, S. A., & Todd, P. M. (2017). Habit formation generates secondary modules that emulate the effi-
ciency of evolved behavior. Behavioral and Brain Sciences, 40, 41.
Pleskac, T. J., & Hertwig, R. (2014). Ecologically rational choice and the structure of the environment.
Journal of Experimental Psychology General, 143(5), 2000–2019.
Rieskamp, J., & Otto, P. E. (2006). SSL: A theory of how people learn to select strategies. Journal of
Experimental Psychology General, 135(2), 207–236.
Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics, 69(1),
99–118.
Simon, H. A. (1990). Invariants of human behavior. Annual Review of Psychology, 41, 1–20.
Shah, A. K., & Oppenheimer, D. M. (2008). Heuristics made easy: An effort-reduction framework.
Psychological Bulletin, 134(2), 207–222.
Thagard, P. (2011). Critical thinking and informal logic: Neuropsychological perspectives. Informal Logic,
31(3), 152–170.
Thagard, P., & Nussbaum, A. D. (2014). Fear-driven inference: Mechanisms of gut overreaction. In
L. Magnani (Ed.), Model-based reasoning in science and technology (pp. 43–53). Berlin: Springer-Verlag.
Tinbergen, N. (1963). On aims and methods of ethology. Zeitschrift für Tierpsychologie, 20, 410–433.
Todd, P. M. (2001). Fast and frugal heuristics for environmentally bounded minds. In G. Gigerenzer &
R. Selten (Eds.), Bounded rationality: The adaptive toolbox (pp. 51–70). Cambridge, MA: MIT Press.
Todd, P. M. (2007). How much information do we need? European Journal of Operational Research, 177,
1317–1332.
Todd, P. M., & Brighton, H. (2016). Building the theory of ecological rationality. Minds & Machines,
26, 9–30.
Todd, P. M., & Gigerenzer, G. (2003). Bounding rationality to the world. Journal of Economic Psychology,
24, 143–165.
Todd, P. M., and Gigerenzer, G. (2020). The ecological rationality of situations: Behavior = f(Adaptive
Toolbox, Environment). In J. F. Rauthmann, R. A. Sherman, & D. C. Funder (Eds.), The Oxford hand-
book of psychological situations (pp. 143–158). New York: Oxford University Press.
Todd, P. M., Gigerenzer, G., & the ABC Research Group (Eds.) (2012). Ecological rationality: Intelligence in
the world. Oxford: Oxford University Press.
Todd, P. M., & Miller, G. F. (1999). From pride and prejudice to persuasion: Satisficing in mate search.
In G. Gigerenzer, P. M. Todd, & the ABC Research Group (Eds.), Simple heuristics that make us smart
(pp. 287–308). Oxford: Oxford University Press.
Tversky, A. (1972). Elimination by aspects: A theory of choice. Psychological Review, 79(4), 281–299.

322
323

Ecological rationality

Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science,
185(4157), 1124–1131.
Wilke, A., Lydick, J., Bedell, V., Dawley, T., Treat, J., ... Langen, T. A. (2018). Spatial dependency in local
resource distributions. Evolutionary Behavioral Sciences, 12(3), 163–172.
Yin, H. H., & Knowlton, B. J. (2006). The role of the basal ganglia in habit formation. Nature
Reviews: Neuroscience, 7, 464–476.

323
324

20
MAPPING HEURISTICS AND
PROSPECT THEORY
A study of theory integration

Thorsten Pachur

One of the most valuable methodological tools in the behavioral science repertoire is to build
models of behavior. The predictions of the models can then be contrasted with data, either
to refute a model or to refine it. Formal models often have adjustable parameters representing
behaviorally relevant constructs, and variability on these constructs can be measured by fitting
the models to empirical data.
Herbert Simon’s (1956, 1990) idea of bounded rationality—according to which successful
and adaptive behavior is shaped by the mind’s natural limits with respect to information pro-
cessing and computational power—has inspired two directions in the modeling of decision
making. First, prospect theory (Kahneman & Tversky, 1979) and its subsequent elaboration
and formal specification in cumulative prospect theory (CPT; Tversky & Kahneman, 1992)
assume that people’s sensitivity to differences in the outcomes and probabilities of risky
options diminishes the further away those magnitudes are from natural reference points
(such as zero, impossibility, or certainty) and that losses carry more psychological weight
than gains. These notions are described algebraically with psycho-economic functions that
translate objective magnitudes of outcomes and probabilities into subjective ones. CPT is
typically considered an “as-if ” model, in the sense these functions are not meant to represent
the actual cognitive processes underlying choices. The second modeling tradition rooted
in bounded rationality, by contrast, has developed heuristics, that are intended as cogni-
tive process models. Heuristics implement bounded rationality by describing choices as
being based on simplifying information processing operations, such as lexicographic search,
difference thresholds, stopping rules, and limited search (e.g., Brandstätter, Gigerenzer, &
Hertwig, 2006; Payne, Bettman, & Johnson, 1993; Thorngate, 1980; Venkatraman, Payne,
& Huettel, 2014).
Despite their common theoretical roots, CPT and heuristics thus rely on very different con-
ceptual languages—algebraic functions vs. simple information processing operations—to model
boundedly rational decision making. Potentially due to these differences, there have been only
few attempts to systematically connect the two approaches.1 As a consequence, theoretical
concepts for characterizing risky choice—such as probability weighting, risk aversion, and
loss aversion—exist side-by-side with theoretical concepts characterizing heuristic cognitive
processes—such as lexicographic search, difference thresholds, stopping rules, and one-reason

324
325

Mapping heuristics and prospect theory

decision making (see below for details), with only little understanding of how these concepts
relate to each other (but see Willemsen, Böckenholt, & Johnson, 2011). Moreover, it is little
understood how heuristics might relate to empirical choice phenomena that have critically
shaped the development of CPT. Take, as an example, the fourfold pattern of risk attitudes
(other phenomena include the common ratio effect, the certainty effect, and reflection effect).
The fourfold pattern refers to the finding that whether people appear risk averse or risk seeking
shifts depending on whether the options offer positive or negative outcomes and whether the
probability of the risky outcome is high or low. CPT accounts for the fourfold pattern by
assuming distorted probability weighting, where rare events are overweighted and common
events are underweighted (relative to their objective probabilities). But there might be other,
more process-oriented accounts (e.g., heuristics) that can explain the fourfold pattern—a pos-
sibility that has hardly been pursued.
In this chapter I will illustrate how CPT and heuristics can be brought together to their
mutual benefit. The analyses presented contribute to theory integration (e.g., Gigerenzer, 2017)
by clarifying the relationship between the different theoretical concepts of CPT and heuristic
decision making. For instance, I show that the typical assumption in CPT of distorted prob-
ability weighting can arise from a hallmark of heuristic processing: limited search. Conversely,
CPT’s conceptual language can be used to isolate and measure how properties of the choices
produced by heuristics change across choice ecologies.
It should be emphasized that the primary goal of this theory integration is not to even-
tually merge the approaches in a unified theory. Instead, it aims to overcome the disintegra-
tion between alternative theoretical accounts of decisions under risk by contributing to a
better understanding of how the frameworks overlap and how they complement each other
in describing behavior. The field of risky choice is teeming with theories that together pro-
vide a rich conceptual repertoire to account for decisions under risk: in addition to heuristics
and prospect theory, there are, for instance, the transfer-of-attention exchange model (e.g.,
Birnbaum, 2008), the security-potential aspiration model (Lopes & Oden, 1993), the pro-
portional difference model (González-Vallejo, 2002), and decision field theory (Busemeyer
& Townsend, 1993). Work on these theories, however, has mostly either focused on the
performance of a single theory, or pitted some of these theories against each other. As a con-
sequence, we know little of the extent to which the conceptual languages of these theories
cover separate parts of the terrain or whether they overlap in their explanatory contribu-
tion. Importantly, the theory integration pursued here can also lead to new predictions. For
instance, if heuristic choices can be reflected in nonlinear probability weighting, which in
turn accounts for a particular choice phenomenon, insights into the boundary conditions of
the use of the heuristics can used to predict moderating conditions for the emergence of the
choice phenomenon. Eventually, a better understanding of the network between the theories
might contribute to a unification of existing approaches, but this is not necessary for theory
integration to be useful.
In the following, I first describe CPT and heuristics in more detail before presenting an
analysis that shows how CPT accommodates the choices produced by various heuristics in the
shapes of its weighting and value functions. I then illustrate how prominent choice phenomena
that have shaped critical assumptions in CPT could arise from heuristic information processing.
Finally, I highlight an important added value from this theory integration, namely, that insights
about the contingent nature of heuristic information processing (e.g., Payne, 1976) can be used
to derive novel predictions regarding moderating conditions for the occurrence of the choice
phenomena.

325
326

Thorsten Pachur

Two modeling approaches for boundedly rational risky choice

Cumulative prospect theory
CPT was developed as an attempt to account for empirical violations of expected value (EV)
and expected utility (EU) theory, such as the common ratio effect (also known as the Allais
paradox; Allais, 1953), the fourfold pattern of risk attitudes, and the certainty effect. In EV
theory, the valuation of a risky option A with outcomes

xm > … > x1 ≥ 0 > y1 > … > yn

and corresponding probabilities

pm ... p1 and q1 ... qn

follows from summing the outcomes, each weighted by its probability, i.e.,

N
EV = ∑ xi pi .
i =1

In CPT, by contrast, it is assumed that outcomes and probabilities are subject to a nonlinear
transformation. The transformation of outcomes is described by a value function that is defined
as shown in Equation (20.1):

v( x ) = x α
v( y ) = −λ( −y )α

The parameter α governs how strongly differences in objective outcomes are reflected in
differences in subjective value and thus indicates outcome sensitivity (with lower values indi-
cating less sensitivity). Parameter λ reflects the relative weighting of losses and gains; with values
of λ larger than 1, a higher weight is given to losses, indicating loss aversion. The left panel of
Figure 20.1 depicts value functions for different values of α and λ.
The transformation of the outcomes’ probabilities (more precisely, the rank-dependent,
cumulative probability distribution) is described with a probability weighting function. A com-
monly used parameterization of the weighting function (see Goldstein & Einhorn, 1987)
separates the curvature of the probability weighting function from its elevation and is defined
as shown in Equation (20.2):

δ + pγ +
w+ = for x
δ + p + (1 − p )γ +
γ+

δ − pγ −
w− = for y,
δ − p + (1 − p )γ −
γ−

with γ+ and γ– governing the curvature of the weighting function in the gain and loss
domains, respectively (in the following analyses, a common γ for both domains is estimated).
Lower values on γ+ and γ– indicate greater curvature and thus lower sensitivity to probabil-
ities. The parameters δ+ and δ– govern the elevation of the weighting function for gains and
losses, respectively, and are often interpreted as indicating the degree of optimism thus risk
seeking (e.g., Gonzalez & Wu, 1999). Higher values on δ+ indicate higher risk seeking in the
gain domain, higher values on δ– indicate higher risk aversion in the loss domain. The right

326
327

Mapping heuristics and prospect theory

Figure 20.1 CPT’s value function for different values of the outcome sensitivity (α) and loss aversion (λ)
parameters (left); and the probability weighting function for different values of the probability-sensitivity
(γ) and elevation (δ) parameters (right)

panel of Figure 20.1 depicts probability weighting functions for different values of the γ and δ
parameters. Assuming these transformations of the objective outcomes and probabilities of risky
options, as formalized in the value function and the weighting function, CPT can account for
the violations of people’s choices of EV and EU theory mentioned above (for a more extensive
discussion of CPT, see Wakker, 2010).

Heuristics
Whereas CPT’s value and weighting function are not meant to describe the cognitive processes
leading to a choice, researchers following the heuristics approach use conceptual building
blocks that acknowledge which mental operations actual decision makers are able to perform.
These include lexicographic search (i.e., the sequential consideration of attributes, which in
risky choice are the possible outcomes and their probabilities), within-attribute comparison
of options, difference thresholds (i.e., boundaries indicating when two attribute values are
to be treated as different), stopping rules (i.e., conditions under which information search is
truncated), and one-reason decision making (i.e., basing a choice on a single attribute).
Table 20.1 gives an overview of five heuristics of risky choice. Some involve extremely
simple processing and consider only a single attribute. Take, for instance, the minimax heuristic
(e.g., Savage, 1954). In order to choose between options—each described by possible outcomes
and their probabilities—minimax focuses exclusively on each option’s worst possible outcome
and chooses the option whose worst outcome is more attractive. All other attributes—that is,
the other outcomes and all probabilities—are ignored. The least-likely heuristic (Thorngate,
1980) is somewhat more complex. Having likewise identified the worst possible outcome of
each option in a first step, it bases its decision on the probability of that outcome, choosing the
option whose worst possible outcome is least probable. The most-likely heuristic (Thorngate,
1980) also has a two-step structure: It first identifies each option’s most likely outcome and then

327
328

Thorsten Pachur

Table 20.1 Five heuristics of risky choice

Heuristic Description
Minimax Choose the option with the highest minimum outcome.1
Maximax Choose the option with the highest outcome.1
Least-likely Identify each option’s worst outcome. Then choose the option with the lowest
probability of the worst outcome.1
Most-likely Identify each option’s most likely outcome. Then choose the option with the
highest most likely outcome.1
Priority Go through attributes in the following order: minimum gain, probability of
heuristic minimum gain, and maximum gain. Stop examination if the minimum gains
differ by 1/10 (or more) of the maximum gain; otherwise, stop examination if
the probabilities differ by 1/10 (or more) of the probability scale. Choose the
option with the more attractive gain (probability). For options with more than
two outcomes, the search rule is identical, apart from the addition of a fourth
reason: probability of maximum gain. For loss lotteries, “gains” are replaced by
“losses.” For mixed lotteries, “gains” are replaced by “outcomes.”

Note: 1If the decisive reason does not discriminate, the heuristic choses randomly. Heuristics are taken
from Thorngate (1980) and Brandstätter et al. (2006).

chooses the option whose most likely outcome is more attractive. Finally, some heuristics imple-
ment difference thresholds (also known as aspiration levels) and involve conditional processing
of attributes. The priority heuristic (Brandstätter et al., 2006), for example, first considers the
options’ minimum gain or loss, depending on the domain. If the two options differ sufficiently
on that attribute (formalized by a difference threshold, defined as 10 percent of the maximum
gain or loss), it chooses the option with the more attractive minimum outcome. If the diffe-
rence between the options does not exceed the difference threshold, the priority heuristic moves
on to the next attribute, the probabilities of the outcomes. If the probabilities differ by at least
10 percent (the difference threshold for the probability attribute), the heuristic chooses the
option with the lower probability of the minimum outcome; otherwise, it moves on to the next
attribute, the maximum outcome, and chooses the option that is most attractive on that attribute.

Mapping heuristics onto CPT

How could CPT and heuristics—despite using very different modeling frameworks to describe
people’s decisions under risk—be related to each other? Gigerenzer (2017) has sketched out an
approach to theory integration. Rather than submitting theories to critical tests (e.g., Popper,
[1934] 1959) or pitting them against each other, the goal is to build networks between theories,
integrating and connecting empirical phenomena and theoretical concepts, in order to achieve
greater coherence in a field, identifying alternative routes to explain the same behavior or to
see how they complement each other.
So how can heuristics and CPT be meaningfully integrated? Different heuristics make
different assumptions about which attributes are considered, the order in which attributes are
considered, and which attribute determines a choice. As a consequence, they also differ in the
priority they give to probability information and to the minimum and maximum outcomes.
The crucial idea is now to use CPT to measure theoretical properties of the heuristics’ choices,

328
329

Mapping heuristics and prospect theory

such as the degree of probability sensitivity, risk aversion, and loss aversion. Differences between
the heuristics on these theoretical properties can then be linked to differences in the heuristics’
information processing architectures. Elaborating these connections therefore helps to illu-
minate how CPT’s theoretical constructs, and key empirical phenomena (e.g., fourfold pattern)
that have shaped CPT, might be related to information processing principles. I next describe
such a theory-integration analysis of CPT and heuristics.

What shapes of CPT’s weighting and value function do heuristics produce?

In order to understand how heuristics map onto the conceptual framework of CPT, Pachur,
Suter, and Hertwig (2017) investigated how the choices produced by the heuristics described
in Table 20.1 would be reflected in CPT’s weighting and value functions. To that end, they
estimated the shapes of functions that reproduce, as best as possible, the heuristics’ choices. The
results, shown in Figure 20.2, indicate that the five heuristics gave rise to very distinct value
and weighting functions.2 For instance, the choices of the minimax and the maximax heuristics
were accommodated in CPT by assuming very strongly curved weighting functions (upper
row of Figure 20.2), indicating very low probability sensitivity. Importantly, this result is con-
sistent with the heuristics’ information processing architecture, which completely ignores prob-
ability information. For minimax, moreover, the elevation of the resulting weighting function
was very low for gains and very high for losses, indicating generally high risk aversion. For
maximax, the pattern was reversed, indicating generally high risk seeking. For the priority
heuristic, the curvature of the weighting function was not quite as pronounced as for minimax
and maximax, indicating higher probability sensitivity—again consistent with the fact that the
heuristic considers probability information when the first-ranked attribute does not discrim-
inate between the options (this was the case in around 20 percent of the choice problems used
in the analysis). The weighting functions estimated for the choices of the least-likely and most-
likely heuristics, finally, showed the least pronounced curvature, indicating high probability
sensitivity and therefore accurately reflecting that these heuristics always consider probability
information (though in different ways).
The lower row of Figure 20.2 shows CPT’s value functions that best describe the choices
produced by the heuristics. Here as well there were considerable differences between the
heuristics, and these differences were consistent with the differences between the heuristics in
terms of how their information processing architectures treat outcome information: the least-
likely heuristic, which never bases its choices on outcomes, yields a very strongly curved value
function, indicating very low outcome sensitivity. The minimax, maximax, and most-likely
heuristics, for all of which outcomes are the decisive attribute, have a less curved value function,
indicating high outcome sensitivity.
These results highlight two things. First, they demonstrate that CPT’s weighting and value
functions, though not meant to represent cognitive mechanisms, can meaningfully measure
characteristics of the cognitive mechanism underlying choices: Choices produced by heuristics
whose information processing architecture implement different degrees of attention to prob-
abilities and outcomes are accommodated in CPT by systematic differences in the curvature
of the weighting and value functions, respectively, with lower attention giving rise to stronger
curvatures. Pachur et al. (2017) showed the CPT profiles produced by different heuristics are
sufficiently distinct from each other to allow differentiating which heuristic produced a set of
choices. Second, it is not necessary to assume differential distributions of attention at the high
and low ends of the probability scale to produce overweighting of rare and underweighting
of common events (e.g., Bordalo, Gennaioli, & Shleifer, 2012; Hogarth & Einhorn, 1990).

329
 newgenrtpdf
330
Thorsten Pachur
330

Figure 20.2 CPT’s weighting function (upper row) and value functions (lower row) estimated for the choices of the five heuristics described in
Table 20.1
331

Mapping heuristics and prospect theory

Instead, this pattern can arise from a single cognitive mechanism, namely, reduced attention
to any probability information (for other proposals, see Bhatia, 2014; Johnson & Busemeyer,
2016). This means that the apparent overweighting of low-probability events can, ironically,
result from underattending to low probabilities.

Characterizing changes in the behavior of heuristics across environments

In some heuristics, the character of the choices they produce depends on the structure of the
environment. For instance, in the priority heuristic, whether a choice is made based on a
probability attribute depends on whether the minimum outcomes in a decision problem differ
sufficiently. In the most-likely heuristic, whether the option it chooses has a very attractive or
a rather unattractive outcome depends on which of the two outcomes is more likely—and thus
whether and how outcomes and probabilities are correlated. In the following, I show how the
CPT framework can make visible and measure these interdependencies between environmental
structures and the properties of heuristic choices.
One fundamental property of choice behavior is the strength of the tendency to pick or to
avoid the more risky option—that is, how risk-seeking or risk-averse choices are. A common
measure of an option’s risk is the coefficient of variation (CV; e.g., Weber, Shafir, & Blais, 2004),
with a higher CV indicating higher risk.3 In CPT, the tendency to choose the riskier option
is reflected in the elevation of the weighting function. To see how the apparent risk attitude—
defined as the willingness to choose the option with a higher CV—produced by heuristics
might differ across environments, let us first consider the minimax heuristic. As mentioned
above, minimax produces strongly risk-averse (risk-seeking) choices in the gain (loss) domain;
the opposite holds for maximax. The high level of risk aversion and risk seeking, respectively,
generated by these heuristics is due to their focus on extreme outcomes (either the minimum
or the maximum outcome). The option with the more attractive minimum outcome (which
minimax chooses) is usually the less risky one, because (unless it is a dominating option) its
maximum outcome is usually rather unattractive (and the range of outcomes of that option is
usually smaller, yielding a lower CV). The option with the more attractive maximum outcome
(which maximax chooses) is usually the more risky one, because its minimum outcome is usu-
ally rather unattractive (and the range of outcomes thus rather large, yielding a higher CV).
Minimax and maximax give rise to risk aversion and risk seeking, respectively, irrespective of
the degree to which outcomes and probabilities are correlated, a key property of decision envir-
onments (e.g., Pleskac & Hertwig, 2014). But consider the most-likely heuristic. Recall that it
first identifies each option’s most likely outcome and then picks the option where that outcome
is more attractive. In environments in which the magnitudes of outcomes and their probabil-
ities are negatively correlated, the most likely outcome of the option chosen by the most-likely
heuristic is usually also the minimum one in the gain domain and the maximum one in the loss
domain. As a consequence, the heuristic should produce risk-averse choices in the gain domain
and risk-seeking choices in the loss domain in such environments. Importantly, the regularity
that the most likely outcome is also the minimum gain (or the maximum loss) does not hold
in an uncorrelated environment, where the magnitude of the outcomes and their probabilities
are not correlated; here, the heuristic should thus produce neither systematically risk-averse or
risk-seeking, but risk-neutral choices. In other words, the risk attitude produced by the most-
likely heuristic is contingent on the structure of the environment.
CPT can make visible such contingent risk attitudes of heuristics through the elevation
of the probability weighting function, governed by the parameter δ (see Equation 20.2). In
Pachur et al. (2017), we measured the dependency of the most-likely heuristic’s degree of risk

331
332

Thorsten Pachur

Figure 20.3 Probability weighting functions estimated for choices produced by the most-likely
heuristic, separately for an environment in which the magnitudes of outcomes and probabilities were
uncorrelated (left panel) and an environment in which the magnitudes of outcomes and probabilities
were negatively correlated (right panel).

aversion on the structure of the environment by estimating CPT’s weighting and value function
for choices produced by the heuristic, separately for an environment in which outcomes and
probabilities were negatively correlated (within a decision problem) and an environment in
which outcomes and probabilities were uncorrelated. Figure 20.3 shows the resulting weighting
functions (the elevation parameter δ was estimated separately for the probabilities of gain and
loss outcomes). As can be seen, the most-likely heuristic produced distinct shapes in the two
environments, mainly driven by differences in the elevation parameter. In the uncorrelated
environment, the weighting function was approximately linear (and δ close to 1), indicating
that the heuristic produced risk-neutral choices. In the correlated environment, by contrast, the
elevation was very low (and δ rather small), indicating strong risk aversion in the gain domain
and strong risk seeking in the loss domain. The other heuristics in Table 20.1 did not show such
contingency in the risk attitudes (see Pachur et al., 2017, for details).
To summarize, the parametric measurement framework of CPT makes it possible to track
how the behavior of heuristics is affected (or not) by the structure of the environment. Although
the qualitative shift in the risk attitude of the most-likely heuristic across environments that we
observed in the present analysis could, in principle, also be derived without invoking CPT,
CPT has the advantage that it also captures more gradual changes in behavior in response to
more gradual changes in the structure of the environment (e.g., correlations between outcomes
and probabilities of different strengths).

Connecting phenomena
The previous section has shown how theory integration between CPT and heuristics can be
achieved by mapping the two theoretical frameworks onto each other. Another route to theory
integration is to connect empirical phenomena that are rooted in different theoretical traditions
(Gigerenzer, 2017). In this section, I apply this approach for CPT and heuristics, considering
the aforementioned fourfold pattern of risk attitudes (Tversky & Kahneman, 1992; Tversky &
Fox, 1995). The fourfold pattern has played an important role in the development of CPT and
is illustrated in Table 20.2. When the risky option offers a gain with a relatively high probability,
people display risk aversion, with the majority choosing a safe option over a risky option.

332
333

Mapping heuristics and prospect theory

Table 20.2 The fourfold pattern of risk attitudes

Probability level Domain

Gains Losses

Low $32, .1; $0, .9 vs. $3, 1 −$32, .1; $0, .9 vs. −$3, 1
Risk seeking (48%)a Risk aversion (36%)
High $4, .8; $0, .2 vs. $3, 1 −$4, .8; $0, .2 vs. −$3, 1
Risk aversion (36%) Risk seeking (72%)

Notes:
Typical risk attitude observed for the lottery problems, containing a safe option and a risky option
with the specific combination of probability level and domain. the percentages in brackets indicate the
proportions of choices of the risky option in Hertwig, Barron, Weber, and Erev (2004).
a Although this particular choice proportion observed in Hertwig et al. (2004) does not represent risk
seeking, risk seeking is often observed for other choice problems of the same type—in particular,
when risk attitudes are estimated with certainty equivalents (e.g., Tversky & Fox, 1995; Tversky &
Kahneman, 1992; for a discussion, see Harbaugh, Krause, & Vesterlund, 2010).

When the risky gain has a low probability, choices are more risk seeking, in that the willingness
to choose the risky option is increased. In the loss domain, the opposite holds (i.e., people are
risk seeking for low-probability losses and risk averse for high-probability losses).
CPT accounts for the fourfold pattern with its assumption of an inverse S-shaped weighting
function, where low-probability events are overweighted and high-probability events are
underweighted relative to their objective probabilities. As Brandstätter et al. (2006) showed, how-
ever, the fourfold pattern is also consistent with heuristic information processing. Specifically,
it can follow from the heuristic principles of lexicographic processing, difference thresholds,
stopping rules, and one-reason decision making implemented by the priority heuristic. Here
is how, starting with the choice problem with a low-probability gain. Comparing a safe gain
of $3 with a 1 percent chance of winning $32, the priority heuristic starts by comparing the
options’ minimum gains: $3 for the safe option and $0 for the risky option. Because the diffe-
rence between these two outcomes does not exceed the difference threshold of 10 percent of the
maximum gain ($3 < $32 * .1 = $3.2), the priority heuristic next compares the probabilities of
the minimum gains. Because for the risky option this probability is lower than for the safe option
(.9 < 1), the risky option is chosen, giving rise to risk-seeking behavior. For the choice problem
with a high-probability gain, the choice is made on the first attribute, the minimum gain, because
the difference between the options on that attribute exceeds the difference threshold ($3 > $4
* .1 = $0.4), and the safe option is chosen, giving rise to risk-averse behavior. In the problem
with a low-probability loss, the priority heuristic first compares the minimum losses. Because this
attribute does not discriminate between the options ($3 < $32 *.1 = $3.2), the heuristic next
considers the probabilities of the minimum losses. The safe option has a higher probability (1 > .9)
and is thus picked, leading to risk aversion. In the problem with a high-probability loss, the heur-
istic makes a decision on the first attribute, the minimum loss, because the difference between the
options on that attribute exceeds the difference threshold ($3 > $4 * .1 = $0.4); the risky option
is chosen because its minimum loss is smaller ($0 < −$3), leading to risk seeking.
This example illustrates how empirical phenomena that have shaped key assumptions in
CPT—such as an inverse S-shaped weighting function—could also arise from heuristic infor-
mation processing. It thereby also points to the cognitive mechanisms behind these hallmark

333
334

Thorsten Pachur

features in CPT. Specifically, the apparent overweighting of low-probability events and

underweighting of high-probability events can result from key principles of heuristic infor-
mation processing: lexicographic search, difference thresholds, stopping rules, and one-reason
decision making. Brandstätter et al. (2006) illustrate how also the possibility effect, the common
ratio effect, the reflection effect, and the certainty effect are consistent with the information-
processing architecture of the priority heuristic.
One interesting implication of building bridges between heuristics and these empirical
choice phenomena is that insights about the boundary conditions of heuristic information pro-
cessing can also cast light on potential moderators of the choice phenomena, allowing respective
hypotheses to be formulated. For instance, Pachur, Hertwig, and Wolkewitz (2014; see also
Suter, Pachur, & Hertwig, 2016) showed that the neglect of probability information in risky
choice is more pronounced when outcomes are affect-rich (e.g., averse medical side-effects)
than when they are affect-poor (e.g., moderate amounts of monetary losses). Higher probability
neglect is associated with a more pronounced curvature of CPT’s weighting function, which is
in turn related to the strength of the fourfold pattern. It follows that the fourfold pattern could
be more pronounced when outcomes are affect-rich than when they are affect-poor—a pre-
diction that only emerges once CPT and heuristics have been connected.

Conclusion
This chapter has illustrated the potential of bringing together CPT and heuristics—two
models of boundedly rational decision making formulated in very different concep-
tual languages and with very different modeling goals. In addition to building a network
between CPT and heuristics, a further benefit of this theory integration is that it forges
connections between the influential parametric framework of CPT and process-tracing
methods that originated in work on heuristics (see Pachur, Schulte-Mecklenbeck, Murphy,
& Hertwig, 2018). But theory integration can also accentuate aspects in which the two
approaches diverge. For instance, whereas the algebraic framework of CPT, based on con-
tinuous functions, enforces transitive choice patterns, the conditional nature of the informa-
tion search of lexicographic heuristics, such as the priority heuristic, can lead to intransitive
choices (see also Tversky, 1969).
It is important to emphasize again that theory integration does not seek to replace one
approach with the other—even if one approach may in some respects be descriptively superior
to the other. Rather, theory-integration analyses are complementary, taking into account that
each approach has its relative strengths and limitations. CPT has been shown to be often more
versatile than heuristics in predicting choices across a large range of choice problems (e.g.,
Glöckner & Pachur, 2012; but see Mohnert, Pachur, & Lieder, 2019). Conversely, process
analyses have obtained clear evidence for heuristic processes (e.g., Mann & Ball, 1994; Pachur
et al., 2013; Payne & Braunstein, 1978; Su et al., 2013; Venkatraman et al., 2014). Thus, based
on the respective merits of each approach, insights from integrating CPT and heuristics will
eventually contribute to making both approaches more complete and guide further theory
development.
In conclusion, although the development of new models is without doubt important, it
is also crucial to take stock of existing models and to elaborate the conceptual network that
connects them; as argued in this chapter, this can even lead to novel predictions, highlighting
how different models can enrich each other. Such theory integration can target qualitatively
different modeling frameworks, as presented in this chapter (see also Bhatia, 2017, for ana-
lyses relating heuristics with connectionist models), but is equally applicable and informative

334
335

Mapping heuristics and prospect theory

for relating models with similar roots (e.g., different types of diffusion models; Khodadadi &
Townsend, 2015). It is time to dissolve the boundaries that have separated accounts of bounded
rationality and to progress toward a more comprehensive perspective of how the mind, and
models of it, work.

Acknowledgments
The author thanks Veronika Zilker and Julia Groß for constructive comments on a previous
draft of this chapter, and Susannah Goss for editing the manuscript.

Notes
1 Kahneman and Tversky (1979) proposed heuristic editing operations (e.g., cancellation, segregation,
combination) that people might apply before evaluating lotteries, but these operations were barely
investigated further. Moreover, whenever CPT and heuristics of risky choice have been considered
within the same analysis, they have been treated as rival accounts (e.g., Brandstätter et al., 2006;
Fiedler, 2010; Glöckner & Herbold, 2011; Glöckner & Pachur, 2012; Pachur, Hertwig, Gigerenzer, &
Brandstätter, 2013; Su et al., 2013).
2 Unsurprisingly, CPT cannot mimic the heuristics’ choices perfectly. Moreover, there are differences
between the heuristics in how well CPT can capture their choices. For instance, CPT accommodated
the choices by minimax and maximax heuristics better than those of the least-likely heuristic. See
Pachur et al. (2017) for more details.
3 The CV is a standardized measure of risk and expresses the amount of risk per unit of return. It is a function
of the outcomes and probabilities of an option, and is larger, the larger the range and the more skewed
the distribution of possible outcomes. In addition, ceteris paribus, the CV is larger, the more the distri-
bution of probabilities is similarly skewed as the distribution of outcomes, but in the opposite direction.
2
N
N 
Specifically, the CV is defined as the standard deviation of an option, σ = ∑ x × pi −  ∑ xi × pi 
2
i
i =1  i =1 
,
with N outcomes x and probabilities p, divided by the absolute value of the option’s expected value.

References
Allais, M. (1953). Le comportement de l’homme rationel devant le risque: Critique des postulats et
axioms de l’école americaine [Rational man’s behavior in face of risk: Critique of the American
School’s postulates and axioms]. Econometrica, 21, 503–546.
Bhatia, S. (2014). Sequential sampling and paradoxes of risky choice. Psychonomic Bulletin & Review, 21,
1095–1111.
Bhatia, S. (2017). Choice rules and accumulator networks. Decision, 4, 146–170.
Birnbaum, M. H., & Chavez, A. (1997). Tests of theories of decision making: Violations of branch inde-
pendence and distribution independence. Organizational Behavior and Human Decision Processes, 71,
161–194.
Busemeyer, J. R., & Townsend, J. T. (1993). Decision field-theory: A dynamic cognitive approach to deci-
sion making in an uncertain environment. Psychological Review, 100, 432–459.
Bordalo, P., Gennaioli, N., & Shleifer, A. (2012). Salience theory of choice under risk. Quarterly Journal of
Economics, 127, 1243–1285.
Brandstätter, E., Gigerenzer, G., & Hertwig, R. (2006). The priority heuristic: Making choices without
trade-offs. Psychological Review, 113, 409–432.
Fiedler, K. (2010). How to study cognitive decision algorithms: The case of the priority heuristic. Judgment
and Decision Making, 5, 21–32.
Gigerenzer, G. (2017). A theory integration program. Decision, 4, 133–145.
Glöckner, A., & Herbold, A.-K. (2011). An eye-tracking study on information processing in risky
decisions: Evidence for compensatory strategies based on automatic processes. Journal of Behavioral
Decision Making, 24, 71–98.

335
336

Thorsten Pachur

Glöckner, A., & Pachur, T. (2012). Cognitive models of risky choice: Parameter stability and predictive
accuracy of prospect theory. Cognition, 123, 21–32.
Goldstein, W. M., & Einhorn, H. J. (1987). Expression theory and the preference reversal phenomenon.
Psychological Review, 94, 236–254.
Gonzalez, R., & Wu, G. (1999). On the shape of the probability weighting function. Cognitive Psychology,
38, 129–166.
González-Vallejo, C. (2002). Making trade-offs: A probabilistic and context-sensitive model of choice
behavior. Psychological Review, 109, 137–154.
Harbaugh, W. T., Krause, K., & Vesterlund, L. (2010). The fourfold pattern of risk attitudes in choice and
pricing tasks. The Economic Journal, 120, 595–611.
Hertwig, R., Barron, G., Weber, E. U., & Erev, I. (2004). Decisions from experience and the effect of rare
events in risky choice. Psychological Science, 15, 534–539.
Hogarth, R. M., & Einhorn, H. J. (1990). Venture theory: A model of decision weights. Management
Science, 36, 780–803.
Johnson, J. J., & Busemeyer, J. R. (2016). A computational model of the attention process in risky choice.
Decision, 3, 254–280.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica,
47, 263–291.
Kahneman, D., & Tversky, A. (Eds.). (2000). Choices, values, and frames. Cambridge: Cambridge
University Press.
Khodadadi, A., & Townsend, J. T. (2015). On mimicry among sequential sampling models. Journal of
Mathematical Psychology, 68–69, 37–48.
Mann, L., & Ball, C. (1994). The relationship between search strategy and risky choice. Australian Journal
of Psychology, 46, 131–136.
Mohnert, F., Pachur, T., & Lieder, F. (2019). What’s in the adaptive toolbox and how do people choose from
it? Rational models of strategy selection in risky choice. In A., Goel, C. Seifert, & C. Freksa, (Eds.),
Proceedings of the 41st Annual Meeting of the Cognitive Science Society (pp. 2378–2384). London: Cognitive
Science Society.
Pachur, T., Hertwig, R., Gigerenzer, G., & Brandstätter, E. (2013). Testing process predictions of models
of risky choice: A quantitative model comparison approach. Frontiers in Psychology, 4, 646.
Pachur, T., Hertwig, R., & Wolkewitz, R. (2014). The affect gap in risky choice: Affect-rich outcomes
attenuate attention to probability information. Decision, 1, 64–78.
Pachur, T., Schulte-Mecklenbeck, M., Murphy, R. O., & Hertwig, R. (2018). Prospect theory reflects
selective allocation of attention. Journal of Experimental Psychology: General, 147, 147–169.
Pachur, T., Suter, R. S., & Hertwig, R. (2017). How the twain can meet: Prospect theory and models of
heuristics in risky choice. Cognitive Psychology, 93, 44–73.
Payne, J. W. (1976). Task complexity and contingent processing in decision making: An information
search and protocol analysis. Organizational Behavior and Human Performance, 16, 366–387.
Payne, J. W., & Braunstein, M. L. (1978). Risky choice: An examination of information acquisition
behavior. Memory and Cognition, 6, 554–561.
Payne, J. W., Bettman, J. R., & Johnson, E. J. (1993). The adaptive decision maker. Cambridge: Cambridge
University Press.
Pleskac, T. J., & Hertwig, R. (2014). Ecologically rational choice and the structure of the environment.
Journal of Experimental Psychology General, 143, 2000–2019.
Popper, K. R. ([1934)] 1959). The logic of scientific discovery. London Hutchinson.
Savage, L. J. (1954). The foundations of statistics. New York: Wiley.
Simon, H. A. (1956). Rational choice and the structure of the environment. Psychological Review, 63,
129–138.
Simon, H. A. (1990). Invariants of human behavior. Annual Review of Psychology, 41, 1–19.
Su, Y., Rao, L., Sun, H., Du, X., Li, X., & Li, S. (2013). Is making a risky choice based on a weighting
and adding process? An eye-tracking investigation. Journal of Experimental Psychology: Learning, Memory
and Cognition, 39, 1765–1780.
Suter, R. S., Pachur, T., & Hertwig, R. (2016). How affect shapes risky choice: Distorted probability
weighting versus probability neglect. Journal of Behavioral Decision Making, 29, 437–449.
Thorngate, W. (1980). Efficient decision heuristics. Behavioral Science, 25, 219–225.
Tversky, A. (1969). Intransitivity of preference. Psychological Review, 76, 31–48.
Tversky, A., & Fox, C. R. (1995). Weighing risk and uncertainty. Psychological Review, 102, 269–283.

336
337

Mapping heuristics and prospect theory

Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: Cumulative representation of uncer-
tainty. Journal of Risk and Uncertainty, 5, 297–323.
Venkatraman, V., Payne, J. W., & Huettel, S. A. (2014). An overall probability of winning heuristic for
complex risky decisions: Choice and eye fixation evidence. Organizational Behavior and Human Decision
Processes, 125, 73–87.
Wakker, P. P. (2010). Prospect theory for risk and ambiguity. Cambridge: Cambridge University Press.
Weber, E. U., Shafir, S., & Blais, A. R. (2004). Predicting risk sensitivity in humans and lower animals: Risk
as variance or coefficient of variation. Psychological Review, 111, 430–445.
Willemsen, M. C., Böckenholt, U., & Johnson, E. J. (2011). Choice by value encoding and value con-
struction: Processes of loss aversion. Journal of Experimental Psychology: General, 140, 303–324.

337
338

21
BOUNDED RATIONALITY
FOR ARTIFICIAL
INTELLIGENCE
Özgür Şimşek

The capabilities of Artificial Intelligence (AI) systems have grown tremendously in recent
years. For instance, in the ancient board game of Go, which was long seen as a grand
challenge for AI, the best player in the world is now a computer program (Silver et al.,
2018). What is impressive is not only the level of performance attained in the game but also
the fact that the program learned how to play on its own, starting with only rudimentary
knowledge about the rules of the game. In fact, the same program learned to master other
board games, including chess and shogi, using the same learning algorithm. Reinforcement
learning, the AI approach employed to train the system, holds great promise for the future.
It allows artificial agents to learn from their own experience, through trial and error,
without relying on advice from human experts. Not only does this require minimal effort
from system designers but it also makes it possible for the resulting system to surpass human
experts in performance.
Reinforcement learning is not a new approach to AI. It has had many notable successes in
the past. For example, more than 20 years ago, reinforcement learning was used to develop a
program that learned how to play backgammon, another ancient board game beloved in many
regions of the world (Tesauro, 1995). This program was arguably the best backgammon player
in the world at the time.
While the successes of AI are truly exciting, it must be noted that the intelligence developed
by current systems is effective in very narrow environments. For example, the game of Go is a
highly confined environment with a very small number of well-defined rules that never change.
In contrast, people and animals function effectively in a world that is complex, dynamic, and
deeply uncertain. Furthermore, existing AI systems require very large amounts of experience
in order to achieve high levels of performance. In contrast, people and animals can learn from
much smaller amounts of experience, sometimes even from a single piece of experience. In
board games such as Go, a program can easily play millions of games with itself to improve its
performance. But, in many real-world problems, obtaining such huge amounts of experience
is simply not possible.
Why have the capabilities of AI systems grown so tremendously in recent years? The most
fundamental reason is the vast increase in the speed of computation. Going forward, fur-
ther technological advances will enable even faster computation, continuing to extend the

338
339

Bounded rationality for AI

capabilities of our AI systems. But it is unlikely that the performance gap between people and
AI can be closed by advances in computational speed only. Just like people and animals, our
machines will need to learn how to be boundedly rational, in other words, how to achieve their
objectives when time, computational resources, and information are limited.
Models of bounded rationality (Simon, 1956; Rubinstein, 1998; Gigerenzer & Selten, 2002)
provide fertile ground for developing algorithmic ideas for creating such AI systems. Early
work in AI was in fact rooted in this view of bounded rationality. In their work on problem
solving, Newell, Shaw and Simon (1958) modelled reasoning as search and suggested that the
search space could be made tractable and navigated by using heuristics. The same approach was
also used in modelling the processes of scientific discovery (Langley, Shaw, & Simon, 1987).
Since these early years, the theory of bounded rationality has advanced considerably, with
contributions from diverse fields, including cognitive psychology, economics, and operations
research. During this time, machine intelligence grew in leaps and bounds in many areas,
including probabilistic reasoning and statistical learning, but it had limited contact with new
advances in bounded rationality.
This chapter will provide an overview of some the more recent research in bounded ration-
ality and discuss its implications for AI. The discussion will start with one-shot comparison
problems and two families of simple heuristics (Gigerenzer, Todd, & the ABC Research
Group, 1999; Gigerenzer, Hertwig, & Pachur, 2011) that can be used in that context: tallying
and lexicographic decision rules. I will describe how these heuristics perfectly match certain
characteristics of natural environments that allow them to make accurate decisions using rela-
tively small amounts of information and computation. The discussion will then progress to how
this knowledge can be applied in the context of sequential decision problems, of which the
game of Go is an example, to reduce the branching factor in a simple and transparent manner.
I will describe how, using ideas from the literature on bounded rationality, it was possible to
reduce the average branching factor in the game of Tetris from 17 to 1 (Şimşek et al., 2016).1
The branching factor shows the number of alternative actions at a decision point. It is one of
the fundamental reasons why AI remains a challenge. This drastic reduction is therefore signifi-
cant, especially because the approach is not specific to Tetris but is applicable to sequential deci-
sion problems broadly. Furthermore, the reasons for eliminating the vast majority of branches
are easy to explain, making the approach transparent, a property that is becoming increasingly
more desirable, even critical, in AI systems.

The comparison problem

People and animals spend much of their time comparing alternatives available to them. For
example, a tiger may compare various prey in its vicinity to determine which one to attack, a
venture capitalist may compare various companies to determine which one to invest in, and an
ant colony may compare various nest areas to determine where to settle. In all of these cases, the
true criterion for selection is unavailable—for example, the venture capitalist does not know the
future return on investment—but a decision must still be made using the available information.
We will first examine a one-shot version of this decision problem. Specifically, we will
define the comparison problem as follows: Determine which of a given number of objects has
the highest value on a specified (but unobserved) criterion, given the values of a number of
attributes on the objects. For example, which of two stocks will bring a larger return on invest-
ment five years from now, given known attributes of the stocks such as type of industry, current
value, and return on investment in the last three months?

339
340

Özgür Şimşek

One common approach to the comparison problem is to estimate the criterion as a function
of the attributes of the object, for example, as a linear function:

yˆ = w 0 + w1x1 + … + wk xk , (21.1)

where ŷ is the estimate of the criterion, x1, …, xk are the attribute values, w0 is a constant, and
w1, …, wk are attribute weights. This estimate leads to the following decision rule in choosing
between objects A and B:

Choose object A if w 0 + w1 (x1A − x1B ) + … + w k (xkA − xkB ) > 0,

choose object B if w 0 + w1 (x1A − x1B ) + … + w k (xkA − xkB ) < 0,

choose randomly if w 0 + w1 (x1A − x1B ) + … + w k (xkA − xkB ) = 0.

Comparison using simple decision heuristics

Alternatively, one can compare objects using decision heuristics. These are simple methods
that use few pieces of information and combine the pieces in simple ways. We will focus
on two well-known and widely used families of heuristics: lexicographic decision rules and
tallying.
Lexicographic decision rules examine attributes one at a time, in a specific order, until an
attribute is found that discriminates between the two objects. An attribute is said to discriminate
between two objects if its value differs on the two objects. The first discriminating attribute,
and that attribute alone, is used to make the decision. In other words, a lexicographic decision
rule orders the attributes in importance and decides based on only the most important attribute
on which the alternatives differ.
Lexicographic decision rules are noncompensatory. That is, differences in the value of one
attribute, no matter how large, cannot compensate for any difference in the value of an attribute
ranked higher in importance. For example, if quality is ranked more important than price, a
lower price cannot compensate for a difference in quality: the selected item will be the one
with the highest quality, regardless of the cost.
While lexicographic decision rules use the attributes one at a time, starting with the most
important attribute, tallying uses all attributes simultaneously, giving them equal weight. If
attributes are binary, taking on values of 0 or 1, tallying is equivalent to a majority vote, where
each attribute either votes for one alternative over the other (if one alternative is higher in
attribute value) or abstains from voting (if alternatives are identical in attribute value).
Tallying has a long history of use and has been known to perform well in many contexts,
especially when generalizing to previously unseen decision problems (Dawes & Corrigan,
1974; Dawes, 1979; Einhorn & Hogarth, 1975). Competitive testing of lexicographic deci-
sion rules is more recent and has shown similar results (e.g., Czerlinski, Gigerenzer, &
Goldstein, 1999).
Over the years, both families of heuristics have been compared to various statistical learning
methods, including linear regression, naïve Bayes, Bayesian networks, support vector machines,
nearest-neighbor methods, decision trees, and neural networks. The result that emerges from
these studies is that decision heuristics can compete remarkably well with more complex

340
341

Bounded rationality for AI

statistical methods, outperforming them on many occasions, especially when there is a small
number of training examples (Brighton, 2006; Katsikopoulos, 2011; Şimşek & Buckmann,
2015; Buckmann & Şimşek, 2017). These results have stimulated a rich scientific literature on
how, and under what conditions, simple heuristics can yield accurate decisions.

When are decision heuristics accurate?

This question has been explored from various viewpoints in the scientific literature. One pro-
ductive viewpoint has been to examine decision heuristics as approximations to the linear
decision rule, asking the following question: If the environment is linear, in other words, if the
relationship between the criterion and the attributes is linear as reflected by Equation (21.1), do
various heuristics yield accurate decisions? Research has shown that they do, under at least three
conditions: simple dominance (Hogarth and Karelaia, 2006), cumulative dominance (Kirkwood
and Sarin, 1985; Baucells et al., 2008), and noncompensatoriness (Martignon and Hoffrage,
2002; Katsikopoulos and Martignon, 2006; Şimşek, 2013). Furthermore, evidence suggests
that these conditions are prevalent in natural environments (Şimşek, 2013). What follows is a
description of these conditions in some detail, followed by a simple illustrative example.
Assume that the relationship between the criterion and the attributes is linear as shown
in Equation (21.1). Let xiA denote the value of attribute i for object A. Let the direction of
attribute i, denoted by di, equal −1, 0, or +1, respectively, if the corresponding attribute weight
is negative, zero, or positive. Let attribute order refer to the order in which w i decrease.
Finally, recall that an attribute is said to discriminate between two objects if its value differs on
the two objects.
We will assume that the values of attribute weights are not known, so the linear decision
rule cannot be applied directly. The question we will explore is whether it is still possible to
make decisions that are identical to those of the linear decision rule. In particular, we will
explore whether it is possible to do so using simple decision heuristics such as tallying or lexico-
graphic decision rules. These heuristics do not need or use attribute weights but they do require
some knowledge about the environment. Specifically, tallying requires that attribute directions
are known while the lexicographic decision rule requires that both attribute directions and
attribute order are known.
To simplify exposition, and without loss of generality, discussion of the three conditions below
assumes that the attributes are ordered such that the sequence w1 ,…, w k is non-increasing.

1 Simple dominance. Object A simply dominates object B if ∀i (di xiA ≥ di xiB ) and
∃i (di xiA > di xiB ) . Objects A and B are dominance equivalent if ∀i (di xiA = di xiB ) . If object
A simply dominates object B, it follows that object A has the higher criterion value. If two
objects are dominance equivalent, they have identical criterion values. To check for simple
dominance, it suffices to know attribute directions; the exact values of the attribute weights or
their order are not needed. Simple dominance is a very useful condition. When it is present,
most decision heuristics choose accurately if their attribute directions are accurate. These
include tallying and the lexicographic decision rule with any ordering of the attributes.
i
2 Cumulative dominance. Let zi = ∑d j x j , i = 1,…, k, denote the cumulative profile of an object.
j =1

Object A cumulatively dominates object B if ∀i (ziA ≥ ziB ) and ∃i (ziA > ziB ) . If object
A cumulatively dominates object B, it follows that object A has the higher criterion value.
To check for cumulative dominance, it suffices to know the attribute directions and order;

341
342

Özgür Şimşek

the exact values of attribute weights are not needed. When cumulative dominance is pre-
sent, the lexicographic decision rule decides accurately if its attribute directions and order
are set accurately, and tallying decides accurately if its attribute directions are set accurately,
with one exception: tallying may find a tie even though one of the decision alternatives
actually has a higher criterion value than the other one.
3 Noncompensatoriness. Consider the following lexicographic decision rule: examine the
attributes in non-increasing absolute value of their weights; identify the first attribute
that discriminates between the objects; choose the object whose value on this attribute
is more favorable (that is, higher if attribute direction is positive, and lower if negative). If
this lexicographic decision rule decides correctly, the decision problem is said to exhibit
noncompensation (Şimşek, 2013). Noncompensation holds with probability 1 if the
attributes are binary, taking on values of 0 or 1, and the weights satisfy the set of constraints
k
wi > ∑ w j , i = 1,…, k − 1, (Martignon and Hoffrage, 2002). Such weights are called
j = i +1

noncompensatory. One example is the sequence 1, 0.5, 0.25, 0.125. If attributes are not
binary or weights are not noncompensatory, noncompensation may or may not hold,
depending on the attribute weights and the attribute values of the objects being compared.

Both simple and cumulative dominance are transitive: if object A dominates object B, and
object B dominates object C, it follows that object A dominates object C. Simple dominance
implies cumulative dominance. Both types of dominance imply noncompensation.
When one or more of these three conditions hold, the decision problem can be considered
to be easy in the following sense: the correct decision can be made without knowledge of the
exact linear relationship between the criterion and the attributes. It suffices to know only the
attribute directions, and, in two of the conditions, also the attribute order. Furthermore, deci-
sion heuristics such as lexicographic rules and tallying make accurate decisions.

An illustrative example
As an illustrative example, consider a set of coins of the United States. Let the sequence
x $1, x50¢ , x 25¢ , x10¢ , x5¢ , x1¢ denote the number of each coin type in the set. Let V denote the
total dollar value of the coins in the set, which is given by the following linear equation,
Equation (21.2):

V = x $1 + 0.5x50¢ + 0.25x 25¢ + 0.10x10¢ + 0.05x5¢ + 0.01x1¢ . (21.2)

Assume you are given two sets of coins and asked to determine which set has the higher value.
Using the linear equation above, the correct answer can be computed easily. However, this
approach would require knowing the exact value of each coin type. If this information is not
available, it is still possible to make the correct decision under the three conditions described
earlier. Below are examples from each condition.
Coin set ⟨4, 4, 4, 0, 0, 0⟩ simply dominates coin set ⟨4, 3, 2, 0, 0, 0⟩. With no knowledge
of the dollar value of the various coin types, we can correctly, and with certainty, identify the
first set as the set with the higher value. If deciding using the tallying heuristic, we would com-
pute the total number of coins in each set, obtaining 12 for the first set and 9 for the second
set, and pick the first set because of its higher count. We would therefore decide correctly.
Similarly, by using the lexicographic decision rule with any ordering of the coins, we would

342
343

Bounded rationality for AI

decide correctly. For instance, consider the accurate ordering of the coins from the most to the
least valuable. We would first compare the sets with respect to the counts of the $1 coin, the
highest-valued coin type. Because both sets have four $1 coins, we would then compare the sets
with respect to the counts of the 50¢ coin. The first set has four 50¢ coins while the second set
has only three such coins. We would then terminate the decision process, picking the first set,
without examining the remaining coin types.
Coin set ⟨4, 4, 4, 0, 0, 0⟩ cumulatively (but not simply) dominates the coin set ⟨2, 5, 1, 0, 0,
0⟩. Tallying would pick the first set because it has more coins than the second set. The lexico-
graphic decision rule with the correct coin order would also pick the first set because the first
set has more $1 coins than the second set. Both heuristics would be accurate.
Coin set ⟨1, 0, 0, 0, 0, 0⟩ neither simply nor cumulatively dominates the coin set ⟨0, 1, 1, 1, 1,
1⟩. Note, however, that the weights of the linear equation (Equation 21.2) are noncompensatory
and the attribute values are binary (there is either 0 or 1 of each coin type in each coin set).
Consequently, noncompensation holds with probability 1. Using the lexicographic decision
rule with the correct coin order, we are guaranteed to decide accurately. Specifically, the lex-
icographic decision rule would pick the first coin set because it contains more of the highest-
valued coin type than the second set, correctly reasoning that the coins of lower value will not
be able to compensate for this difference.

How prevalent are “easy” problems?

These three conditions are theoretically interesting but are they empirically relevant? In other
words, are they encountered frequently when making decisions? Evidence from a large, diverse
collection of data sets suggests that they are (Şimşek, 2013).
This evidence comes from 51 natural data sets obtained from a wide variety of sources,
including online data repositories, textbooks, research publications, and individual scientists
collecting field data. The subjects were diverse, including biology, business, computer science,
ecology, economics, education, engineering, environmental science, medicine, political science,
psychology, sociology, sports, and transportation. The datasets varied in size, ranging from 12
to 601 objects, corresponding to 66–180,300 distinct paired comparisons. The data sets also
varied in the amount of information available for making decisions, with number of attributes
ranging from 3 to 21.
Table 21.1 shows the prevalence of various conditions in two columns. Prevalence is
measured as the proportion of paired comparisons encountered in a data set where the reported
condition was observed, for example, where one of the objects dominated the other object or
the objects were dominance-equivalent, which also allows simple heuristics to decide accur-
ately. The first column shows the results from the original datasets. In this case, numerical
attributes were used as given. The second column shows the results when numerical attributes
were dichotomized around the median, assigning the median value to the category with fewer
objects. For instance, with height as a numerical attribute, the first column shows the results
when height was used as given (for example, 182 cm) while the second column shows the
results with height taking on a value of 0 or 1, indicating only whether the height of the given
object was above or below the median height in the data set.
In addition to the prevalence of exact simple and cumulative dominance, Table 21.1
also reports the prevalence of close approximations to these two conditions. Exact domin-
ance allows one to compare accurately with certainty, without knowing the exact values of the
weights. Approximate dominance allows one to compare accurately with very high probability.
For instance, consider a problem where 9 out of 10 attributes favor object A against object

343
344

Özgür Şimşek

Table 21.1 Prevalence of simple dominance, cumulative dominance, and noncompensation in 51 natural
data sets

Condition Median prevalence Median prevalence

(data used as given) (data dichotomized)

Simple dominance 0.16 0.54

Approximate simple dominance 0.31 0.59
Cumulative dominance 0.62 0.89
Approximate cumulative dominance 0.77 0.92
Noncompensation 0.85 0.96

Source: Şimşek (2013).

B. Although we cannot be certain that the correct decision is object A, that would be a very
good bet. The approximate dominance results reported in Table 21.1 were computed with an
expected accuracy rate of 0.99.
Table 21.1 shows that, on average, simple dominance held in 16 percent, cumulative dom-
inance in 62 percent, and noncompensation in 85 percent of the pairwise comparisons in a data
set. When approximations were allowed, prevalence of both simple and cumulative dominance
increased by 15 percent. Furthermore, when attributes were available only in dichotomized
form, the prevalence of all conditions increased substantially.
These results are useful in understanding why simple decision heuristics perform remark-
ably well when tested on natural data sets. They are also useful in understanding the source of
error when heuristics decide incorrectly. Specifically, when error rates are examined through
the lens of bias-variance decomposition (Geman et al., 1992; Brighton & Gigerenzer, 2008;
Gigerenzer & Brighton, 2009), the three conditions discussed here are particularly relevant
for the bias component of the error. Although simple decision heuristics have very little
flexibility in model selection, they may not be introducing much additional bias compared
to a linear model.

Sequential decision problems

Many real-world problems are sequential in nature, where a successful outcome depends not
on a single decision but a series of related decisions. For example, driving safely to a destin-
ation to arrive on time for an appointment requires a long sequence of interrelated decisions.
These types of problems can be viewed as a series of comparison problems, albeit ones that
have complex dependencies. Consequently, the three conditions that make single-shot com-
parison problems easy can also make sequential decision problems easy, if decision algorithms
are tailored to exploit them.
Here, I will discuss one particular sequential decision environment in depth: the game of
Tetris,2 one of the most popular video games of all time. Artificial players can learn to play
Tetris very well, removing on average hundreds of thousands of rows. One of the best artificial
players is BCTS (Thiery & Scherrer, 2009), which uses a linear evaluation function with eight
attributes. These attributes and their weights are shown in Table 21.2. For each new piece
(called a tetrimino) to be placed on the board, BCTS evaluates all possible placements using this
linear evaluation function and selects the placement with the highest value.

344
345

Bounded rationality for AI

Table 21.2 The attributes and weights of a linear evaluation function for playing Tetris, developed by
Thiery and Scherrer (2009)

Attribute Weight

Rows with holes –24.04

Column transitions –19.77
Holes –13.08
Landing height –12.63
Cumulative wells –10.49
Row transitions –9.22
Eroded piece cells +6.60
Hole depth –1.61

Note:
Three key concepts used in attribute definitions are holes, wells, and transitions. A hole is an empty cell
that has one or more full cells placed higher than itself in the same column. A well is a succession of
empty cells in a column such that the immediate cells on the left and on the right are full. A transition is
a switch from a full cell to an empty cell, or vice versa, when moving along a row or a column. When
computing wells and transitions, it is assumed that the outside of the grid is full.

The number of possible legal placements of a tetrimino ranges from 0–34, with a median
value of 17. For example, the square tetrimino can be placed in 9 different ways on an empty
board. Because the evaluation function is linear, the three conditions discussed earlier can be
used to reduce the number of placements under consideration without losing the placement
that BCTS considers to be the best. For example, any placement that is simply or cumulatively
dominated by one or more other placements can safely be eliminated.
The question of interest, once again, is how often the three conditions are observed when
playing Tetris. In other words, when considering various placements of the new tetrimino, how
often does one placement simply or cumulatively dominate another placement, thus allowing
the player to eliminate the dominated placement from further consideration? And, how often
do the lower-ranked attributes fail to compensate for the difference in the highest-ranked
attribute that discriminates between the alternatives? In other words, if we were to place the
tetriminos using the lexicographic decision rule rather than the linear evaluation function of
BCTS, how often would we choose the placement chosen by BCTS?
The answers to these questions depend on the states encountered while playing the game,
which depends on how one plays the game, in other words, how the placements are chosen as
the game progresses. Below is a summary of the results reported by Şimşek et al. (2016) when
the game was played by BCTS. Results were similar when the games were played by people or
by a controller that selected placements randomly.
The lexicographic decision rule selected the same placement as BCTS 68 percent of the
time. The lexicographic rule was applied as follows. The attributes were used sequentially, one
at a time, in the order listed in Table 21.2, to reduce the number of placements further and
further, until there was only one placement left. With each feature, only those placements with
the best (highest or lowest, depending on the sign of the feature weight) value for this feature
were kept, while the remaining placements were eliminated.
Filtering by simple dominance reduced the median number of decision alternatives
(placements of the tetrimino) from 17 to 3, while filtering by cumulative dominance further

345
346

Özgür Şimşek

reduced it to 1. This result can be stated as follows. Ordinarily, a player has 17 placements
to consider (on average). By simply knowing the sign of the attribute weights (or guessing
correctly, which is easy to do in this domain), the player can reduce this number to 3 (on
average). If the attribute order is also known, this number further reduces to 1 (on average).
This is remarkable. With only one placement left, there is no decision left to make! And
there is certainty that the remaining placement is the one that would have been chosen
by BCTS.
This substantial reduction in the branching factor, however, is not enough to play the game
well. Consider an artificial player that first filters the available actions by cumulative domin-
ance, then chooses randomly among the remaining actions. If the game score is increased by
one point with each eliminated row, this player reaches a median score of 510 in the game,
which corresponds to emptying the Tetris board more than 25 times. This is a considerable
achievement given that it is very difficult to eliminate rows by chance—a player who chooses
actions randomly generally finishes the game with a score of zero—but it is far below the
median score of 709,636 achieved by BCTS.
In summary, simple dominance, cumulative dominance, and noncompensation are all preva-
lent in Tetris. Şimşek et al. (2016) report substantial improvements in the learning rates and
in the game score reached when dominance filters are inserted into existing reinforcement
algorithms. An important question for AI is how these structures in the environment can be
exploited fully, for faster and more effective learning.

Concluding remarks
In the history of AI, algorithmic ideas inspired by natural intelligence have been among
the most fruitful. They include neural networks (inspired by the brain) and reinforcement
learning (inspired by animal learning), both of which were central to the development of
the artificial Go player that defeated one of the best human players in the history of this
board game. The theory of bounded rationality is a similar inspirational resource, untapped
by current AI systems. It seeks to explain, algorithmically and mathematically, how people
are able to make good decisions even though they are almost always constrained by limited
time, information, and computational resources. For fundamental shifts in the capabilities of
our AI systems, we need fundamentally new ideas in our learning algorithms. The theory of
bounded rationality is a useful foundation for developing such ideas in service of creating AI
systems that can function effectively and transparently in large, complex, uncertain decision
environments.

Notes
1 The branching factor is the number of alternative actions that need to be evaluated at a decision point
and is one of the fundamental reasons why Artificial Intelligence is intractable.
2 Tetris is played on a two-dimensional grid, which is initially empty. The grid gradually fills up as pieces
of different shapes, called tetriminos, fall from the top of the grid, one at a time, piling up on each other.
The player can control how each tetrimino lands by rotating it and moving it horizontally, to the left or
to the right, any number of times, as the tetrimino falls one row at a time. The tetrimino continues to
fall until one of its cells sits directly on top of a full cell or on the grid floor. When an entire row of the
grid becomes full, the row is deleted, creating additional space on the grid. The game ends when there
is no space at the top of the grid for the next tetrimino. The standard grid is 20 cells high and 10 cells
wide. Tetriminos each occupy four cells and have seven different shapes.

346
347

Bounded rationality for AI

References
Baucells, M., Carrasco, J. A., & Hogarth, R. M. (2008). Cumulative dominance and heuristic perform-
ance in binary multiattribute choice. Operations Research, 56(5), 1289–1304.
Brighton, H. (2006). Robust inference with simple cognitive models. In C. Lebiere & B. Wray (Eds.),
AAAI spring symposium: Cognitive science principles meet AI-hard problems (pp. 17–22). Menlo
Park, CA: American Association for Artificial Intelligence.
Brighton, H., & Gigerenzer, G. (2008). Bayesian brains and cognitive mechanisms: Harmony or disson-
ance? In N. Chater & M., Oaksford (Eds.), The probabilistic mind: Prospects for Bayesian cognitive science
(pp. 189–208). New York: Oxford University Press.
Buckmann, M., & Şimşek, Ö. (2017). Decision heuristics for comparison: How good are they? In
T. V. Guy, M. Kárný, D. Rios-Insua, & D. H. Wolpert (Eds.), Proceedings of the NIPS 2016 Workshop
on Imperfect Decision Makers, vol. 58 of Proceedings of Machine Learning Research, pp. 1–11.
Czerlinski, J., Gigerenzer, G., & Goldstein, D. G. (1999). How good are simple heuristics? In
G. Gigerenzer, P. M. Todd, & the ABC Research Group (Eds.), Simple heuristics that make us smart
(pp. 97–118). New York: Oxford University Press.
Dawes, R. M. (1979). The robust beauty of improper linear models in decision making. American
Psychologist, 34(7), 571–582.
Dawes, R. M., & Corrigan, B. (1974). Linear models in decision making. Psychological Bulletin, 81(2),
95–106.
Einhorn, H. J., & Hogarth, R. M. (1975). Unit weighting schemes for decision making. Organizational
Behavior and Human Performance, 13(2), 171–192.
Geman, S., Bienenstock, E., & Doursat, R. (1992). Neural networks and the bias/variance dilemma.
Neural Computation, 4(1), 1–58.
Gigerenzer, G., & Brighton, H. (2009). Homo heuristicus: Why biased minds make better inferences.
Topics in Cognitive Science, 1(1), 107–143.
Gigerenzer, G., Hertwig, R., & Pachur, T. (Eds.) (2011). Heuristics: The foundations of adaptive behavior.
New York: Oxford University Press.
Gigerenzer, G., & Selten, R. (2002). Bounded rationality: The adaptive toolbox. Cambridge, MA: MIT Press.
Gigerenzer, G., Todd, P. M., & the ABC Research Group (Eds.) (1999). Simple heuristics that make us smart.
New York: Oxford University Press.
Hogarth, R. M., & Karelaia, N. (2006). “Take-the-best” and other simple strategies: Why and when they
work “well” with binary cues. Theory and Decision, 61(3), 205–249.
Katsikopoulos, K. V. (2011). Psychological heuristics for making inferences: Definition, performance, and
the emerging theory and practice. Decision Analysis, 8(1), 10–29.
Katsikopoulos, K. V., & Martignon, L. (2006). Naive heuristics for paired comparisons: Some results on
their relative accuracy. Journal of Mathematical Psychology, 50(5), 488–494.
Kirkwood, C. W., & Sarin, R. K. (1985). Ranking with partial information: A method and an application.
Operations Research, 33(1), 38–48.
Langley, P., Simon, H. A., Bradshaw, G. L., & Zytkow, J. M. (1987). Scientific discovery: Computational
explorations of the creative processes. Cambridge, MA: MIT Press.
Martignon, L., & Hoffrage, U. (2002). Fast, frugal, and fit: Simple heuristics for paired comparison. Theory
and Decision, 52(1), 29–71.
Newell, A., Shaw, J. C., & Simon, H. A. (1958). Elements of a theory of human problem solving.
Psychological Review, 65(3), 151–166.
Rubinstein, A. (1998). Modeling bounded rationality. Cambridge, MA: MIT Press,
Silver, D., Hubert, T., Schrittwieser, J., Antonoglou, I., Lai, M., ... Hassabis, D. (2018). A general
reinforcement learning algorithm that masters chess, shogi, and go through self-play. Science, 362(6419),
1140–1144.
Simon, H. A. (1956). Rational choice and the structure of the environment. Psychological Review, 63(2),
129–138.
Şimşek, Ö. (2013). Linear decision rule as aspiration for simple decision heuristics. In C., Burges, L.
Bottou, M. Welling, Z. Ghahramani, & K. Weinberger (Eds.), Advances in Neural Information Processing
Systems 26, 2904–2912.
Şimşek, Ö., Algorta, S., & Kothiyal, A. (2016). Why most decisions are easy in Tetris—And perhaps in
other sequential decision problems, as well. In Proceedings of the 33rd International Conference on Machine
Learning, pp. 1757–1765.

347
348

Özgür Şimşek

Şimşek, Ö., & Buckmann, M. (2015). Learning from small samples: An analysis of simple decision
heuristics. In C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, & R. Garnett (Eds.), Advances in
Neural Information Processing Systems 28, 3159–3167.
Tesauro, G. (1995). Temporal difference learning and TD-Gammon. Communications of the ACM,
38(3), 58–69.
Thiery, C., & Scherrer, B. (2009). Building controllers for Tetris. International Computer Games Association
Journal, 32, 3–11.

348
349

22
PSYCHOPATHOLOGICAL
IRRATIONALITY AND
BOUNDED RATIONALITY
Why is autism economically rational?

Riccardo Viale

Deductive irrationality in madness?

As noted by Michel Foucault in his History of Madness (1961), during the age of the Enlightenment
in the sixteenth and seventeenth centuries, the theme of rationality and irrationality became
one of the main intellectual and political issues of the time. Reason was defined according to
normative canons and it was the center of gravity of political and social action. Any behavior
that did not adhere to those canons was considered irrational, “unreasonable,” a sign of madness.
In the Middle Ages, madness was treated as a disease of the spirit and a moral disorder to be
contained and punished. The madman was considered to be possessed in his body and soul
by the devil and for this reason he was subjected to corporal and moral punishments. During
the Enlightenment, madness came to be regarded instead as a reversible deviation of reason,
a condition that should not be punished but rather treated by segregating the madman in
the appropriate medical facility. The Enlightenment was therefore tolerant toward madness,
even though it relegated mentally disturbed individuals to the margins of society. The thread
between the Enlightenment and the psychopathology of the twentieth century is evident. The
madman deviated from the normative principles of reasons established by the canons of logical
thinking, and possibly from the main social norms and habits. In this sense, normal thinking was
expressed through correct inferences from a deductive point of view, consistent beliefs that are
not mutually contradictory, and a pursuit of the truth as correspondence to reality.
As Viale has emphasized (2013), the presence of rooted prejudices of a logicist type in the
study of human inferential performances is well illustrated by a number of traditional theories
on schizophrenic thinking. In many of the theories on abnormal and, in particular, schizo-
phrenic thinking present in treatises of psychopathology and psychiatry, we find the thesis on
conformity of normal human deductive reasoning set out along the lines of classical logic.
For years, psychopathology, in order to characterize schizophrenic psychotic reasoning, used
a normal model of reasoning which considered it aprioristically compliant with the precepts of
classical logic. According to some theories prevalent in psychiatry, the schizophrenic displays
a clear deviation from classical canons of logical reasoning. This different logical behavior was
thought to be characteristic not only of the psychotic but also of cognitive behaviors in men
who lived in archaic cultures, and it was therefore termed, by Arieti, for example, “paleologic”

349
350

Riccardo Viale

(Arieti 1963). The paleologic individual does not reason using Aristotelian logic, but uses a
sui generis logic. This type of thinking is fundamentally based on a principle described by Von
Domarus (1925, 1944). After his studies on schizophrenia, this author formulated a theory
which in a slightly modified form can be stated as follows: “Whereas the normal person accepts
identity only on the basis of identical subjects, the schizophrenic accepts identity based on
identical predicates.” For example, if the following information is given to a normal individual,
“All men are mortal; Socrates is a man,” the normal person will be able to conclude, “Socrates
is mortal.” This conclusion is valid because the subject of the major premise (all men) contains
the subject of the minor premise (Socrates).
If, on the other hand, a schizophrenic thinks, “a has the property x,” and “b has the property
x,” in some cases, he may conclude that “a is b.” This conclusion, which would seem insane to
a normal person, is reached, according to some authors, because the identity of the predicate
of the two premises, “x,” makes the schizophrenic accept the identity of the two subjects, “a”
and “b.”
Moreover, the schizophrenic attributes a much broader meaning to the predicate. There
is a tendency to identify a part with the whole (for example, a room with the house to
which the room belongs). Therefore, it can be said that a = a + b + c because the two
terms of the equation have a in common (a term that assumes the function of an “identi-
fying link”).
In the schizophrenic, these forms of paleologic reasoning are usually automatic, in the
same way that the application of some laws of Aristotelian logic is automatic for normal indi-
viduals. According to Von Domarus, the first three laws of traditional Aristotelian logic are
annulled (the principle of identity, non-contradiction, the excluded middle). Furthermore,
paleologic thinking seeks the origin and cause of an event in a different way to logical
thinking: by confusing the real world and the psychological world, it looks for the causes of
an event in personal and subjective reasons rather than in reasons of an external nature. In
other words, the causality present in objective explanations concerning the physical world
is replaced by the causality based on psychological and subjective causal factors. Here, too,
the authors identify a clear link with the thinking of young children and in particular with
primitive thinking.
Another theory linked to the traditional model of rationality and normality of reason in the
sense of conformity to the principles of classical logic is that of Matte Blanco (1981). He iden-
tifies a number of fundamental laws in schizophrenic thinking and uses them in an attempt to
explain the symptoms found.

1. Principle of generalisation: an individual thing is treated as if it were an element of a class.

This class is treated as a subclass of a more general class, and so on, to form a chain of
generalizations. This differs from normal behavior in that often higher classes are chosen
for shared characteristics of an accessory not for their fundamental nature.
2. Principle of symmetry: the schizophrenic treats the converse of a relation as being identical
to the relation itself. From this follows several important corollaries: mainly the disappear-
ance of time and any part being identical to the whole. According to Matte Blanco, these
principles can explain a number of characteristics of schizophrenic thinking, such as literal
interpretation of metaphor, displacement, condensation, concrete thinking. The principle
of symmetry has the property of eliminating any possibility of logical organization in the
sector of thinking where it is applied. The typical reasoning of schizophrenic thinking may
be defined as “bi-logical” in that in part it respects traditional bivalent logic and in part
symmetrical logic.

350
351

Psychopathological irrationality and BR

These theories are conceptual edifices whose foundations rest on an a priori definition of the
deductive performance of normal humans, nowadays empirically proved by cognitive psych-
ology (Viale, 2012). We need only take two examples highlighted by Wason and Johnson-Laird
(1972, pp. 236–238).
The thesis whereby paleologic logic is characterized in relation to normal logic, on the
grounds of being governed by the principle that two classes are identical if they have some
attribute in common, is not adequate because this particular inferential fallacy is just an example
of what Chapman and Chapman (1959) call “probabilistic inference,” an error which normal
subjects often make when an argument lacks thematic content. To give an example, if the syl-
logism is of the kind:

Some A are B
Some C are B
Then both A and C show the property of being Bs and hence they will be assumed to
be linked in some way. This will produce the conclusion “Some A are C” which is a
common error in this problem.1

Similarly, the thesis put forward by Matte Blanco whereby the root of the schizophrenic thought
process is that all relational terms are treated as if they were symmetrical is weakened by the
finding of a similar phenomenon called “illicit conversion” (Chapman & Chapman 1959),
which frequently occurs among normal individuals when the tasks are abstract. Illicit conver-
sion is the fallacy caused by the inversion of the subject and predicate in a proposition:

No P are Q
Therefore no Q are P.2

From this and other experimental observations, it can therefore be demonstrated that normal
subjects too, at times, show forms of reasoning, traditionally seen as aberrant and considered
peculiar to the cognitive symptomatology of schizophrenia. It is no longer convincing, there-
fore, to characterize schizophrenic reasoning based on the infraction of the laws of the classical
logicist ideal of deductive rationality.
Is normality irrational and abnormality rational? It is also normal subjects, and not only
schizophrenics, who show deviations in their deductive reasoning. Over the past 50 years, many
empirical studies have been presented stigmatizing the irrationality of human thought. The
most famous study in the field of the psychology of deductive reasoning was the one presented
by Wason (1966) on selection tasks. In this study, and in numerous subsequent ones, it was
demonstrated that when performing abstract tasks, individuals do not apply some fundamental
rules of logic like modus tollens and are also often exposed to confirmation bias (also known as the
“fallacy of affirming the consequent”). In the field of the psychology of probabilistic judgment
and decision making, the heuristics and biases program led by Daniel Kahneman and Amos
Tversky (for an overview, see Kahneman, 2011) has produced a vast quantity of data––obtained
mainly through abstract tests and characterized by situations involving risk (betting, lotteries,
and games)––which has outlined an alarming inclination toward bias, systematic errors, and
cognitive distortions in normal subjects. Conversely, recent studies paint a picture that it is the
exact opposite of the one outlined above concerning the deductive irrationality of the psych-
otic subject. These studies examine the rationality of judgment and decision making in patients
suffering from neurological and psychiatric impairments. Paradoxically, unlike the so-called
normal individuals, these patients seem not to be subject to a series of biases of rationality. In

351
352

Riccardo Viale

their paper, Hertwig and Volz (2013) examined a series of studies analyzing the behavior of
neurological and psychiatric patients in various decision-making contexts. Hertwig and Volz
summarize the results of these studies as follows (2013, p. 547):

(i) Patients with damage to the ventromedial prefrontal cortex (VMPFC) were more
coherent in their preferences in a consumer choice context (i.e., the Pepsi paradox)3
(Koenigs & Tranel, 2008). Likewise, they did not fall prey to the correspondence bias
and, thus, made more advantageous decisions in an investment context (Koscik &
Tranel, 2013). (ii) In moral dilemmas in which the utilitarian choice (a weaker bench-
mark of rationality) implies emotionally aversive behavior, patients with VMPFC
damage, frontotemporal dementia, or frontal traumatic brain injury showed a greater
propensity to make utilitarian judgments (Young & Koenigs, 2007; Mendez, 2009).
(iii) Patients with OFC lesions made choices between gambles that were guided more
by the expected value of the gambles than by reported or anticipated regret (Camille
et al., 2004). (iv) Patients with any of various focal lesions (including damage to
the OFC) were less subject to myopic loss aversion and, thus, made more advanta-
geous decisions (resulting in higher income) in an investment task (Shiv et al. 2005).
(v) Participants with a virtual lesion to the right dorsolateral PFC (induced through
transcranial magnetic stimulation) exhibited higher acceptance of unfair offers in the
ultimatum game (Knock et al., 2006). (vi) Patients with autism were less responsive to
the framing of monetary outcomes as either losses or gains and, thus, exhibited more
internally consistent behavior (De Martino et al., 2008).
Hertwig & Volz, 2013 p. 547

Can abnormality be conducive to rationality? In some cases, the answer is yes, if we consider
the type of rationality that is under examination and that has been made the basis for judgment
based on the rationality and irrationality of human behavior. What is it that characterizes this
type of rationality in the examples of neurological and mental disorders mentioned above, where
behavior appears to be more rational in mentally disturbed individuals compared to normal
individuals? It is substantially the lack of influence of emotion, which translates as awareness of
the consequences of one’s choices on one’s utility and, on a prosocial level, on that of others.
Hertwig and Volz (2013, p. 549) underline this point by analyzing previous cases. Humans with
damage to the dorsolateral PFC (Knock et al., 2006), which weakens “the emotional impulses
associated with fairness goals,” are free to follow their selfish impulses without restraint, thereby
maximizing material income. A lack of the “emotional associations” that are the “driving
force behind commercial advertisements” enables patients with damage to the VMPFC to
show coherent taste-based brand preferences in the Pepsi paradox (Koenigs & Travel, 2006).
Attenuated “prosocial sentiments” (Mendez, 2009, p. 614) and weaker emotional reactions to
the possibility of causing direct harm to others enable patients with damage to the VMPFC or
OFC to overcome emotional revulsion at the “means” of an action (e.g., smothering a baby to
quieten it) and focus on its “ends” (saving the lives of several others) (Young & Koenigs, 2007).
Similarly, dampened “emotional responses” to the possibility of losses liberate patients with
“deficient emotional circuitry” from myopic loss aversion (Shiv et al., 2005, p. 435, p. 436) and
the experience of regret (Camille et al., 2004). The case concerning the reduction of ambiguity
aversion and the Ellsberg paradox in patients with damage to the orbitofrontal cortex (OFC)
versus normal subjects (Hsu et al., 2005) contains yet another reference to the reduction of
the affection component.4 In point of fact, when ambiguity and uncertainty are compared to
certainty and certain probabilities, they generate anxiety and insecurity that lead one to avoid

352
353

Psychopathological irrationality and BR

them. In case of damage to the OFC, weaker emotional reactions to ambiguity prevented the
subject from falling into the Ellsberg paradox.
Autism is a psychiatric disorder in which the lack of an emotional dimension seems to result
in the debiasing of the framing effect. A failure to “integrate emotional contextual cues into
the decision making process” enables patients with autism to choose in an internally consistent
way, that is, independently of option framing (i.e. loss versus gain (De Martino et al., 2008, p.
10746). Loss aversion caused by heavier emotional content of loss vs. gain is not active in autism
and, as demonstrated by De Martino et al. (2008), behavior is less subject to the framing effect.
In short, an autistic individual does not show any greater propensity to risk in loss frames, nor
a greater aversion to risk in gain frames. The rational behavior of the autistic subject that has
been highlighted by research is emblematic of irrationality corresponding solely to a cold cal-
culating capacity without any external influences of an affective type and without the capacity
to grasp, understand, and predict empathically any altruistic and social behavior in other people.

From coherence to correspondence: the goal of cognitive success

Over the years, the evaluation of what is considered reason and the equation “normality equals
rationality” seem to have undergone a paradoxical change. Initially, the rationalist legacy of the
eighteenth-century Enlightenment resulted in a very clear separation of reason from madness.
The madman was characterized by a failure of deductive and inductive logical thinking. In the
1950s, however, this distinction disappeared when cognitive science brought to light a whole
series of supposed failures of rational behavior in the normal individual as well. What dealt a
serious blow to the theories of human decision-making and judgment formation was the fact
that, quite surprisingly, several reasoning flaws that had been attributed solely to the psychiatric
patient were, in fact, also found in the normal individual. Over the past few years, the theory
that started with the Enlightenment seems to have reversed in a circular way. If we consider
rationality as based on the logical coherence adopted in economic theory, then rationality is
more apparent in individuals suffering from certain neurological and psychiatric disorders than
among people without these diseases. Paradoxically, normality seems characterized by irration-
ality, and abnormality by rationality. Why this paradox? In my view, the reason is that adherence
to a formal concept of rationality as consistent with the norms of deductive logic, the theory
of probability, and the theory of utility, is incorrect. Moreover, the analysis of conformity of
human behavior to those norms is mostly performed in abstract tests, with little relation to
real life, and generally under conditions of risk. There is, therefore, a triple anti-ecological
bias at work here: behavioral irrationality is not assessed in its capacity to adapt to the real
environment and to solve environmental problems. Instead, it is tested in abstract and artifi-
cial situations rather than the real world. The world to which the tests pertain is characterized
by risk (in gaming, betting, and lotteries, the possibilities and probabilities are fixed), but not
uncertainty (in reality, we often don’t know all the possibilities and therefore cannot correctly
estimate all probabilities). The paradox lies in the fact that we do not consider the lesson from
the theory of bounded rationality, aptly expressed by the Herbert Simon’s metaphor (1990)
of a pair of scissors to emphasize how cognition and environment work in tandem: “Human
rational behaviour … is shaped by a scissors whose blades are the structure of task environments
and the computational capabilities of the actor” (p. 7). Often the real environments have an
unlimited space of possibilities because they are characterized by uncertainty or they have a
limited space but they are computationally intractable. Therefore, the classical model of ration-
ality, characterized by coherence with formal norms of logic, probability, and utility cannot be
applied (Simon, 1978).

353
354

Riccardo Viale

This section will analyze what it means to apply the concept of bounded rationality in the
evaluation of human rationality or irrationality. As highlighted by the metaphor of the scissors,
judgment of the rationality of human behavior does not depend on its conformity to some
formal norm, but only on its cognitive success in given specific environmental tasks.

What is cognitive success?

The judgment of rationality of human behavior depends on its ability to adapt to specific
environmental tasks. This ecological adaptability happens if the decision maker is successful
in coping with the cognitive requirements of a specific environmental task. The requirements
may be different. When one has to solve an abstract task like Tower of Hanoi or Missionaries
and Cannibals, the requirement is to find a solution. The same problem solving requirements
may be found in many tasks of everyday life, from replacing a tire to dealing with the failures of
our computer software. When one has to make a financial investment, the requirement seems
different, that is to say making good predictions about the future value of one’s investments.
The same requirement seems to apply also when one is engaged in a negotiation, or needs to
find a reliable partner or has to decide which approach may be better suited to pursue one’s
research. According to some authors (Schurz & Hertwig, 2019), cognitive success should be
analyzed only in term of predictive power of the cognitive method used to deal with a task. In
other words, according to Schurz and Hertwig (2019, p. 16):

the predictive success of a cognitive system or (more generally) a cognitive method

depends on two components that are commonly in competition and whose optimiza-
tion thus involves a trade-off. In the psychological literature, this trade-off is reflected
in the distinction between the ecological validity of a prediction method (Brunswik,
1952; Gigerenzer et al., 1999) and its applicability.

The ecological validity refers to the validity of the system in condition in which it is applic-
able. The applicability is the scope of conditions under which it can be applied. Or in other
words it “is the percentage of targets for which the method renders a prediction, among all
intended targets of prediction” (Schurz & Hertwig, 2019, p. 17). Does the reduction of cogni-
tive success to successful predictions include most successful adaptive answers to environmental
tasks? According to Schurz and Hertwig (2019, p. 16), the answer is yes:

The core meaning of the cognitive success of a system (including algorithms, heuristics,
rules) is defined in terms of successful predictions, assuming a comprehensive meaning
of prediction that includes, besides the predictions of events or effects, predictions of
possible causes (explanatory abductions) and in particular predictions of the utilities of
actions (decision problems).

In fact, this reduction of cognitive success to successful prediction excludes a number of cogni-
tive human abilities to adapt to complex environments. In 1986, Herbert Simon wrote:

The work of managers, of scientists, of engineers, of lawyers––the work that steers the
course of society and its economic and governmental organizations––is largely work
of making decisions and solving problems. It is work of choosing issues that require
attention, setting goals, finding or designing suitable courses of action, and evalu-
ating and choosing among alternative actions. The first three of these activities–fixing

354
355

Psychopathological irrationality and BR

agendas, setting goals, and designing actions––are usually called problem solving; the
last, evaluating and choosing, is usually called decision making.
Simon, 1986, p. 1

In dealing with a task, humans have to frame problems, set goals, and develop alternatives.
Evaluations and judgments about the future effects of the choice are the final stages of the
cognitive activity. This is particularly true when the task is an ill-structured problem. When
a problem is complex, it has ambiguous goals and shifting problem formulations, cognitive
success is characterized mainly by setting goals and designing actions. Herbert Simon offers the
example of design-related problems:

The work of architects offers a good example of what is involved in solving ill-
structured problems. An architect begins with some very general specifications of
what is wanted by a client. The initial goals are modified and substantially elaborated
as the architect proceeds with the task. Initial design ideas, recorded in drawings and
diagrams, themselves suggest new criteria, new possibilities, and new requirements.
Throughout the whole process of design, the emerging conception provides continual
feedback that reminds the architect of additional considerations that need to be taken
into account.
Simon, 1986, p. 15

Most of the problems of corporate strategy or government policy are as ill-structured as

problems of architectural and engineering design or of scientific activity. Reducing cognitive
success to predictive ability is the product of the decision-making tradition and in particular
of the theory of subjective expected utility (SEU). The theory of subjective expected utility
combines two subjective concepts: first, a personal utility function, and, second, a personal
probability distribution (usually based on Bayesian probability theory) that corresponds to
the predictive component. Which decision the person prefers depends on which subjective
expected utility is higher. Different people may make different decisions because they may have
different utility functions or different beliefs about the probabilities of different outcomes. The
decision-making tradition follows the syntactic structure of SEU theory. It deals solely with
analytic judgments and choices and it is not interested in how to frame problems, set goals,
and develop a suitable course of action (Viale, Macchi, & Bagassi, in preparation). On the con-
trary, cognitive success in most human activities is based precisely on the successful completion
of those problem-solving phases. In reality, decision-making, including the predictive success
of judgements, may be considered a subordinate stage of problem solving (Viale, Macchi, &
Bagassi, in preparation).

Are syntactical rules adaptive?

Rationality has always been defined in terms of coherence, assuming that a single syntactical
rule suffices to evaluate behavior. We have seen earlier how the very evaluation of irrationality
and rationality in mental disorders itself was carried out against syntactical rules. As noted by
Arkes et al. (2016, Table 1), there are many examples of coherence rules interpreted as norma-
tive for rational behavior. All the norms are solely defined in term of a syntactical rule whereas
pragmatic or functional goals do not enter the normative analysis (Arkes et al., 2016). Thus,
the question is: “How useful and feasible are these rules in individual choices?” One of the
reasons why attention turned to the coherence of economists and scholars in decision making

355
356

Riccardo Viale

was the theory that violation resulted in material costs for the violator. Arkes, Gigerenzer, and
Hertwig (2016) analyzed the validity of this theory and found no empirical evidence in this
regard. To answer this question, they conducted several systematic Web of Knowledge searches
for the major coherence rules reported in the literature, and in addition they conducted a
survey among 1,000 researchers of the Society for Judgement and Decision Making. Their
systematic literature searches suggest that there is little empirical evidence that violations
of coherence norms are costly, or if they are, that they survive arbitrage and learning. The
survey further confirmed this result. Among the phenomena of violation of coherence that
were studied were the violation of transitivity and the related “money-pump,” the viola-
tion of procedural invariance and the related preference reversal, the framing effect, and the
independence of irrelevant alternatives condition (Arkes et al., 2016). In some cases, rather
than generating material damage, the violation of coherence seems to generate an adaptive
effect. This adaptive incoherence (Arkes et al., 2016) stems from the content-blind syntactic
characteristics of formal norms. These fail to account for the pragmatic adaptive objectives of
the decision-maker that may conflict with the requirements of formal coherence. Sometimes
violating the norms based on transitivity, non-contradiction, consistency, and propositional
logic may be adaptive. A case in point is illustrated by Amartya Sen (1993) and shows that
sometimes violating the consistency of the norm of “independence of irrelevant alternatives”
(or Property Alpha)5 may be necessary. The following two choices are inconsistent because they
violate Property Alpha.

A is chosen over B given the options [A, B], and

B is chosen over A given the options [A, B, C].

Can Property Alpha guide the rationality of our choices without considering the background
knowledge of values, social norms, external environmental and contextual conditions in which
we are called to choose? According to Sen (2002), the idea of internal consistency of choice “is
essentially confused, and there is no way of determining whether a choice function is consistent
or not without referring to something external to choice behaviour (such as objectives, values,
or norms).” Let’s consider this example: John is asked to go to a club with Jane, the fiancée of
his friend Carl. Out of respect for his friend, John decides not to go out with Jane (A) instead
of going out with her (B). One hour later, he receives a phone call in which he is informed
that he can meet Carl that night at the club (C). At this point, John decides to go out with Jane
(B) in a way that will not violate the principle of respect toward his friend.
What appears to be an internal inconsistency is not, in fact, if we consider the values, social
norms, and environmental constraints in the equation. In this example the heuristic norm was
“It is not socially correct to go to a club alone with the girlfriend of a friend.” Moreover, what
often appears to be an incoherent behavior corresponds instead to a coherent strategy that
generated it. In this example the heuristic rule is coherent and generates a behavior that is at
first sight incoherent. Judgment on rationality cannot therefore be limited only to the observ-
able behavior (as is implicit in the economic theory of the revealed preference approach by Paul
Samuelson, 1938), but has to open the black box of rules and heuristics that generated it. And
it must also question the adaptive success of the behavior as compared to the context. Lastly,
the poor adaptivity of coherence norms is also rooted in their cognitive unfeasibility. Herbert
Simon’s theory of bounded rationality is part of the naturalizing programs of decisional norms
that, starting with Willard Orman Quine and continuing with Christopher Cherniak, Steven
Stich, Alvin Goldman, Steven Kornblith, and others, have addressed the issue of the visibility
of the norms for the acquisition of knowledge and for decision making in economic, social,

356
357

Psychopathological irrationality and BR

and moral areas. If a procedure is unfeasible at the cognitive or perceptive level, it cannot be
adopted as a normative reference. Coherence as a behavioral norm is unfeasible because it is
computationally intractable (Arkes et al., 2016). For example, applying the principle of non-
contradiction to 20 beliefs in social interaction between two people would require testing 1,770
copies of beliefs. Considering at least a few seconds (say, 5) per pair, it would take up to two
and a half hours to do it. If we consider that people generally have many more beliefs than 20,
and more to the point, we apply them to many more people than two, we realize how unfeas-
ible the application of this principle is. A second difficulty must be added to the feasibility of
the application of the rules of coherence in reality: the rules of coherence––such as transitivity,
consistency, and non-contradiction––can be applied when there are no conflicting goals or
preferences (Arkes et al., 2016). In reality, though there are many conflicting goals. In this case,
if you try to have a behavior that is consistent with a goal (for example, living long), you are
forced not to be consistent with another goal (enjoying unhealthy food). This is another case in
which coherence becomes an objective that is impossible at the adaptive level.

Heuristics and ecological rationality

Cognitive biases characterizing normal or abnormal thinking are identified according to a prin-
ciple of coherence. This principle represents the basis of the program “Heuristics and Biases” by
Tversky and Kahneman (1974) and lies at the heart of libertarian paternalism in public policy by
Thaler and Sunstein (2008; for a critical overview of libertarian paternalism, see Viale, forth-
coming). But there is an alternative approach that is based on the principle of correspondence
as a benchmark for judging rationality. The principle of correspondence deals with the rela-
tionship between behavior and the environment, according to what was illustrated above when
discussing bounded rationality. The measure of the success of this relationship can be assessed
not formally but pragmatically. In general, the evaluation concerns the capacity to reach the
goals of some kind of task, be it the prediction of an event, solving a practical problem, or
other objectives aimed at our happiness and environmental survival. Thus, instead of being
syntactical, behavioral rules become functional or ecological. Instead of being logical, ration-
ality becomes ecological. Without dwelling on the characteristics of ecological rationality that
are discussed in detail in this volume (for a summary, see also Gigerenzer & Gaissmaier, 2011),
the main passages of an evaluation according to ecological rationality can be broken down as
follows (Arkes, Gigerenzer, & Hertwig, 2016):

1. Identify the goal of an individual or a group.

2. Identify the set of strategies an individual or group has available to reach that goal (“the
adaptive toolbox”).
3. Identify the structural properties of the environment in which the strategies should operate.
4. Determine the “ecological rationality” of the strategies in the environment, that is, a set of
environmental conditions that specify when a given strategy is likely to reach the goal better
than a competing strategy (Arkes et al., 2016).

The “adaptive toolbox” is constituted by a series of rules of thumb, heuristics,6 and algorithms
used, according to the structural properties of the environment. The most important environ-
mental condition is uncertainty, that is to say, the situation characterized by the unpredictability
of future states of the world and, consequently, the impossibility of estimating the probability
of the consequences of one’s options (and what happens in conditions of risk). Faced with
the descriptive and normative failure of the optimizing algorithms in decision making under

357
358

Riccardo Viale

conditions of uncertainties, in the 1990s, an articulated research program was developed around
heuristics, which attempted to formalize and test them empirically in situations of uncertain
choice (Gigerenzer, Todd, & the ABC Research Group, 1999). The basic principle was simpli-
city. The greater accuracy of the results obtained in comparison with decision-making methods
of a statistical type can also be summarized by the standard of “less is more.” It was observed that
there is an inverse U relationship between the level of accuracy of the results and the quantity of
information, computation or time. Beyond a certain threshold, “more is harmful,” that is to say,
a greater quantity hinders the result (Gigerenzer & Gassmaier, 2011). The methods tested were
those of linear and nonlinear multiple regression, Bayesian analysis, etc. Simon, in his commen-
tary to Gigerenzer, Todd, and the ABC Research Group (1999), refers to a “revolution in cog-
nitive science that represents an extraordinary impulse in favour of good sense in the approach
to human rationality.” How can we explain the positive role of simplicity in decisional contexts
characterized by uncertainty? Let us consider an example of heuristic decision. For reasons of
economic and marketing planning, large commercial chains are often called to deal with the
problem of distinguishing between active customers, that is to say those who might return to
buy again, from those who are inactive. In general, these companies can rely on a database of
customer purchases. Based on this data, an inferential statistical algorithm is traditionally applied
to predict which clients will be active in the future. A recent study (Wübben & Wangenheim,
2008) highlighted how expert managers use a simple rule of “time since last purchase,” which
can be described as a “hiatus heuristic” (Gigerenzer & Gassmaier, 2011, p. 455): “If a client has
not made a purchase for a number of months (the hiatus), the customer is classified as inactive.
Otherwise, he/she is simply classified as active.”
Depending on the commercial sector, the hiatus may be different. Technological products
or those with a higher turnover (e.g., electronics) have shorter hiatuses as compared to more
traditional products with a lower turnover (e.g., furniture). This heuristic disregards basic statis-
tical information, while the Pareto/NBD model generally analyzes at least 40 weeks’ worth of
data to calculate its parameters. Wübben and Wangenheim (2008) tested the predictive success
of the two decisional models in the sectors of clothing, airlines, and online sales of CDs. On
average, the hiatus heuristic was successful in predicting the type of customers, in the next 40
weeks, in 80 percent of cases as compared to 76 percent of the Pareto/NBD model. This type
of heuristic is based solely on one indicator that is deemed valid. It is part of a category known
as “single-reason heuristics” whereby the individual, when faced with two alternatives, chooses
the one with the strongest reason (“take-the-best heuristic”).7
Another example introduces another category of heuristics based on simplicity and frugality,
“recognition-based” heuristics (Goldstein & Gigerenzer, 2009). An experiment carried out in
2006 by Serwe and Frings compared amateur tennis players (with only episodical knowledge
of tennis matches) against members of the Association of Tennis Professionals and other expert
associations (who were intimately familiar with all the international players) in their capacity to
predict the results of the 2004 Wimbledon tournament. It turned out that the amateurs cor-
rectly predicted 72 percent of the tournament’s outcomes compared to an average of 68 per-
cent among the experts. The same study was repeated for the 2006 tournament and yielded
the same results. Similar evidence emerged with regard to other predictions, for example, the
performance of shares on the stock market. Ortmann et al. (2008) verified that on average, the
returns of stock portfolios picked solely by recognition surpass the returns of managed assets
and those managed by stock exchange experts. Unlike heuristics that are based on a reason, in
this case it is the sense of recognition, of familiarity toward a certain name, event, or object that
dictates our choice. This can happen not only when we see that name, but also when that name
comes to mind first (“take-the-first heuristic”). And if we recognize more than one name, we

358
359

Psychopathological irrationality and BR

attach the greatest value to the one that we recognize fastest (“fluency heuristic”) (Schooler
& Hertwig, 2005). An example of this heuristic successfully applied to predict stock exchange
performances can be found in Alter and Oppenheimer (2006). The structure of heuristics has
also been studied for the purpose of formalization by simulation objectives. Following on from
Herbert Simon’s work, three fundamental elements have been identified as general components
making up many, but not all, heuristics (Gigerenzer et al., 1999):

1. Rules of research that specify the directions in which to search for information (e.g., the
hiatus heuristic looks for information concerning the recency of the last purchase made).
“Stopping rules” that specify when the search ends (the search ends when the information
on recency is found).
2. The rules of decision-making that specify how the final decision is made (use the threshold
of n months for the decision of whether the customer is regular or not).

The study of heuristics based on the principle of “less is more” was, in the past, mainly descrip-
tive. As the examples above indicate, though, there is an interesting prescriptive aspect to
heuristics that tries to answer the question of when their use is preferable compared to the
use of more complex strategies in uncertain environments. In other words, it is a matter of
understanding which characteristics of the decision-making environment make the use of
heuristics preferable over optimization algorithms. The question is one of understanding when
heuristics allow us to make better predictive inferences, particularly in situations of uncertainty
where we have little time and little computing power. The answer to this question allows us to
understand when a heuristic is adaptive, that is to say, improves our ecological rationality. The
structural variables of a task that seem to favour the success of a heuristic in a decision-making
environment characterized by uncertainty are the following8 (Todd et al., 2011):

1 Redundancy: This is the level of interconnection of signals in the environment. The more
interconnected they are, the more we can rely on a single signal or reason. For example,
when dealing with people with psychiatric issues or who exhibit violent behavior,
it is generally sufficient to consider one signal to reliably infer the associated behav-
ioral profile. The profiles of deviant personalities often possess a coherent structure of
interconnected psychological characteristics. This is what taxi drivers working late at
night do, for example, when they have to assess the reliability of a potential customer
quickly and intuitively.
2 Variability: This has to do with the weight and the importance of signals. When there are
some signals that are more important and representative than others, it is enough to inter-
cept one of the signals to generate accurate inferences. For example, when choosing among
different credit contract alternatives or mortgage loans, in Italy, the EAPR9 is the most
important variable that can enable one to decide quickly and accurately.
3 Size of the sample: When faced with situations of uncertainty, most samples of available
data are never enough to allow complex algorithms to estimate ex-ante the value of their
parameters for predictive purposes. For example, predicting the performance of ten stocks
traded on Wall Street would require historical information on those stocks, along with
similar ones, to generate any inference, as algorithms, such as Markowitz’s “mean-variance
portfolio” algorithm tell us. A limited data sample would be flawed by virtue of their having
a high variance. In these cases, therefore, basing one’s decision on data can be misleading.10
It is better to use the heuristic approach, and this is what Markowitz did when he had to
invest his severance package using the heuristic 1/N.

359
360

Riccardo Viale

In uncertain situations, the use of decisional heuristics can occur either automatically and intui-
tively or in a deliberate and reasoned way. How do you select which heuristic or heuristics?
Some of these, as animal ethology teaches us, are inbred and selected through evolution. The
spatial localization capacity of insects and birds is one such example. In humans, most heuristics
are, however, learned by individual learning and social imitation. Once these heuristics are
memorized, they become part of our knowledge and are retrieved implicitly and intuitively or
explicitly and deliberately depending on the task at hand. What are the fundamental building
blocks of heuristics? These are the expression of some cognitive characteristics even when the
psychological mechanisms at the basis of the heuristic decision-making are still unclear. Some
central inbred capacities certainly include recognition memory and frequency monitoring.
Additionally, there are more sensory properties like the visual tracking of an object (present
in the “gaze heuristic”) and social ones like social imitation and social rules (Gigerenzer &
Gaissmaier, 2011). To these acquired functions, others like aspiration levels, ordered search, and
the one-reason stopping rule (Hertwig & Herzog, 2009, p. 678) are added. An important role
is performed by emotion, especially in social-type heuristics. As we will see later, emotion plays
a key role in the ecological irrationality of psychiatric disorders.

Social rationality
The social dimension of human beings entails that most of their decisions are made in contexts
where others social actors are involved. This may happen in various ways and for each of these
ways, we can refer to social ecological rationality. Hertwig and Herzog (2009) identify four
such ways, of which I will discuss the two most relevant ones for the analysis of mental disease:

1. Social information in games against nature: Each one of us, when making a prediction
about a natural event (such as the risk of contracting malaria while traveling to India) or
solving a practical problem (buying a house) tends to use the knowledge accumulated
through personal experience, but primarily the knowledge handed down in society. It is
a knowledge of behaviors, preferences, decision-making modes, and problem solving of
others that guides our behavior. Depending on our trust in other people, our information
social circle may be small or large. Others may be a direct source of information through
direct questions, or simulated by putting oneself in the shoes of others when dealing with
a problem or a prediction to be made.
2. Social learning in games against nature: When dealing with the problems of everyday life,
we cannot count on our personal experience, but we need to base our decision on what
others are doing. Social learning leads to a reduction in the costs (time and errors) that
individual learning would require, and allows us to use behaviors that have been selected
because they are successful (Richerson & Boyd, 2005). This is also been explained at the
evolutionary level (Boyd & Richerson, 2005). What are the rules that allow us to use social
learning best? First of all, imitation or related advice-giving. That allows individuals to
learn about their environments without engaging in potentially hazardous learning trials
or wasting time and energy on exploration (Henrich & McElreath, 2003). According to
the “Imitate the successful” heuristic (Boyd & Richerson, 2005), the individual looks for
the most successful person in her reference group and imitates her behavior. This heuristic
is particularly effective for learning the cue orders for lexicographic strategies (Garcia-
Retamero, Takezawa, & Gigerenzer, 2006, 2007). In this type of heuristics (as in satisficing
and elimination by aspects), the critical point is the choice of the cue orders through
which to sequentially make the final selection of the option in a process of comparing

360
361

Psychopathological irrationality and BR

several alternatives in stages. Counting on the experience of the most successful can entail
a considerable decisional advantage. Depending on situational cues and task structures,
the imitation heuristics may be by majority11 or by the nearest individual. Another route
through which social learning can occur is by actively seeking the advice of others or by
interpreting institutional arrangements as implicit recommendations (Hertwig & Herzog,
2009).12 Social heuristics present the same building blocks as non-social ones with the add-
ition of social rules, social imitation, and emotion.

Ecological irrationality of psychiatric disorders

The articulation of psychiatric disorders is varied and subject to periodic changes in classifica-
tion. The Diagnostic and Statistical Manual of Mental Disorders (DSM) is now in its fifth edition
since 1952. In that time several changes have taken place. One significant example: in the
DSM V (American Psychiatric Association, 2013), all the paranoid and catatonic subcategories
have been removed from schizophrenia after having been the basis of all psychiatry manuals.
Even the expression “psychiatric condition” has been abandoned and replaced by “psychiatric
disorder.” When dealing with the rational aspect of these disorders, I will mostly refer to the
classification of the DSM IV (American Psychiatric Association, 2000) and particularly to per-
sonality disorders and bipolar, phobic, or obsessive-compulsive disorders.
As previously noted, most individuals are characterized by limitations of logical ration-
ality, which do not make them impeccable logicians, statisticians, and maximizers, but that
allow them to successfully adapt to the environment to solve problems and to learn from their
mistakes.13 From this standpoint, a behavior can be defined as pathological. This is not because
it deviates from normative standards, but because it is unable to successfully interact with the
social and physical environment, in other words, to have cognitive success. The irrationality
of the mental disorder is such that it differs from bounded ecological rationality because it
would not allow some of the following adaptive cognitive functions (depending on the mental
disorder): (1) learning from mistakes or environmental feedback, in particular, advice from
the social context; (2) using social imitation heuristics, like that of successful people or of
the crowd, that allow the individual to speed up the decision-making processes and to find
solutions that are readily available; (3) realistically representing the terms of a problem in the
context of a decision, without affective or emotional hyperpolarization, based only on some
salient variables and not on others that are more relevant to the decision; (4) having a correct
non-distorted perception of spatial and temporal coordinates in the context of a decision; and
(5) anticipating at the corporal level (e.g., somatic marker, Damasio, 1994) the affective effects
of a future choice in a way that corresponds to the reality of the phenomenon not in a distorted
or dissociated way (Viale, 2019). Any rigidity and impermeability to the signals from the envir-
onment and the lack of a realistic representation of the variables at work do not allow adequate
fitness of the structure of a decisional task to allow us to be successful in our choices. It is there-
fore from an ecological point of view rather than from a normative one that we can consider a
behavior irrational in the context of some mental disorders. We will now analyze some of these
points in more detail. This analysis will be entirely hypothetical, given the absence of empirical
data from the study of ecological rationality in individuals affected by mental disorder.

Impermeability to environmental feedbacks

One of the peculiar characteristics of ecological rationality is the possibility of learning spe-
cific decision-making options for a set of environmental tasks through decisional learning.

361
362

Riccardo Viale

This means remaining open and permeable to the reinforcement mechanisms coming from
the environment as responses to previous decisions, and adjusting the subsequent decision
to be more adaptive. In these conditions, it is possible, for example, to develop forms of
judgment and decisions that tend to converge toward the requirements of the Bayes theorem.
For example, it is well known that if information is presented as the outcome of learning from
experience, known as natural frequencies, and not as conditional probabilities, the proportion
of people reasoning by Bayes’ rule increases greatly (Gigerenzer & Hoffrage, 1995). Nudge
theory, too (Thaler & Sunstein, 2008), deals with the problem of helping citizens through
System 2 nudges that reinforce their decisional learning capacity (Viale, forthcoming). Among
them one of the most important is the creation of environments that are rich in feedback
that allows the citizen to adapt his decision-making process to previous experience.14 This
prerequisite of ecological rationality is almost never fulfilled in the case of mental disorders.
Patients with borderline personality disorder (BPD) tend to make risky choice and are unable
to improve their performance by experience (Schuermann et al., 2011). A reduced feedback
related negatively predicted an inability to learn from negative feedbacks (Mak & Lam, 2013).
As the following clinical examples will show, various emotional and cognitive disorders tend
to be impermeable to environmental responses. Delirium and hallucinations are experienced
as real and do not allow corrections. The schizophrenic really sees images and hears voices
that do not exist. If you suffer from paranoid delirium, you believe that threats to your life
are genuine and not falsifiable possibilities. Depression, phobias, and impulsive compulsive
disorders are cognitively correctable, but the emotional dimension is too strong to lead to a
real correction. The patient may acknowledge that the causes of the disorder are excessive
or not existent, but the affective and emotional state of polarization is predominant over this
judgment. Reinforcement learning through feedbacks, describing how the brain can choose
and value courses of actions according to their long-term future value, may account for how
depressive symptoms may result from aberrant valuations and prior beliefs about the loss of
agency (“helplessness”) (Adams et al., 2016). Lastly, personality disorders present a framework
of rigidity and impermeability, on an emotional basis, to forms of learning and improve-
ment of reasoning and decision-making procedures (Gangemi et al., 2013; Johnson-Laird
et al., 2006).

Disabled social learning

First of all, the structure of social environmental tasks is characterized by uncertainty more
strictly than non-social ones. The complexity and unpredictability of human decisions and
interaction between different social actors make social tasks primarily characterized by ambi-
guity and uncertainty. This characteristic of the task structure makes the use of heuristic
rules a necessity for adaptive success. In the case of mental disorders, there may be a distorted
opposite tendency to underestimate or overdramatize the uncertainty of the tasks, and
accordingly to approach the decision-making process in a non-adaptive way. Mental disorders
like narcissistic or bipolar personality disorder in the hypomaniac phase tend to represent the
external environment as characterized by predictable and limited options of which the indi-
viduals are aware and which are under their control. The illusion of control (Langer, 1975)
and overconfidence (Moore & Schatz, 2017) are phenomena in cognitive psychology that
have been studied in the normal adult and which are accentuated in the case of certain mental
disorders. On the contrary, patients suffering from hysterical or phobic personality have the
tendency to accentuate the possibility of danger and adversity. That makes the patient con-
stantly insecure with respect to future events and leads him to attach greater uncertainty to

362
363

Psychopathological irrationality and BR

tasks, even when the contexts present few variables and they are relatively closed to external
events. As noted above, social heuristics differ from non-social heuristics in their respective
building blocks: social rules, social imitation, and emotion. When talking about interaction
with the environment, the most relevant component in the evaluation of ecological ration-
ality in mental disorder is the social one. Mental disorders are above all disorders concerning
social relations, because it is at the social level that its consequences become most disabling.
Let us consider, according to the list of components of social rationality (Hertwig & Herzog,
2009), the various forms of ecological irrationality. First of all, there is a misguided use of
social information in the prediction of events and problem solving. This is particularly evi-
dent in schizophrenic patients, and autistic ones as well, but also in individuals suffering
from less crippling mental disorders like paranoid personality, phobic personality, or obsessive
personality. In all these disorders it is difficult for the individual to refer to other people as
sources of knowledge. This occurs both at the direct level and the simulated level. It is very
difficult for them to ask other people for information, including close relatives. In the case
of autism, even the simulation of what other people would do or think becomes impossible.
The informative social circle cannot be created and the individual can only rely on his/her
own idiosyncratic and limited knowledge. It is, however, primarily mostly in social learning
that the ecological rationality gap in mental disorder emerges. The competitive advantage
of resorting to imitation heuristics and social rules is absent in psychotic patients, but is also
lacking in many pathological and borderline personalities. An obsessive personality finds it
difficult to delegate to others due to lack of trust. A paranoid personality believes that others
are out to harm him/her and therefore is averse to any reference to social examples. The
antisocial personality is in conflict with social norms and rules. The narcissistic personality
tends to overestimate his or her individual capacities and to underestimate those of others.
Schizoid personality presents an absence of desire for close relationships with other human
beings and an emotional detachment from people around them and reality in general. In all
these cases, the individual is not inclined to rely on social intelligence to deal with everyday
decisions and finds himself experimenting instead and making decisions without relying on
other people’s experience. These decisions are uniquely personal, complex, and difficult
because of the variables at play and the difficulty in calculating the results of his decisions.
In particular, the identification of cue orders for a compared choice between alternatives is
suboptimal, because it relies entirely on individual learning instead of being based on social
imitation. The opposite situation can also be observed, however, in patients with a borderline
personality disorder (BPD) characterized by a strong emotional component. For example, in
the case of a hysterical personality, the greater emotional response increases mentalization,
defined as a form of interpretation of human behavior in terms of intentional mental states
(Sharp et al., 2011). A “hypermentalizing” tendency increases the cooperative answer in the
trust game. BPD patients were better at adjusting their investment to the fairness of their vir-
tual partner and assessing relevant emotional cues from unfair trustees, and at using objective
fairness of their counterparts to guide responses (Sharp et al., 2011). The “hypermentalizing”
tendency generates social reactions that may not always be appropriate and poorer social
perspective coordination, that is, the capacity to both differentiate from and integrate the
perspectives of self with those of others (Jennings et al., 2012). BPD patients had a biased per-
ception of participation. They more easily felt excluded even when included, in turn having
an increase in other-focused negative emotions (Staebler et al., 2011). This tendency reflects
on the use of social heuristics by exaggeratedly accentuating an acritical identification with
one’s choices and with other people’s cue orders, with no real selection of the source based
on merit and reliability.

363
364

Riccardo Viale

Distorted emotions as building blocks

Emotion is one of the two central specific building blocks of social heuristics (Hertwig &
Herzog, 2009). Its importance in guiding heuristic decision making is evident. Also evident is
the fact that its distortion, either in terms of hyperstimulation or hypostimulation, has negative
effects on the ecological rationality of individuals. According to some authors (Johnson-Laird
et al., 2006; Gangemi et al., 2013), psychological illnesses are disorders in emotion, not intel-
lect. In particular, hyper-basic emotions tend to occur at the onset of psychological illnesses.
Using the Multifaceted Empathy test, BPD patients were found to have significant impairments
in emotional empathy that could contribute to dysfunctional emotional and social responses
(Diziobek et al., 2011). They exhibit more overall errors in recognizing facial expressions15
on facial morph tasks, in particular, those with negative valence with fearful and surprised
expressions (Domes et al., 2008). This indicates an attentional bias for fearful faces characterized
by difficulty in disengaging attention from threatening stimuli. As noted, patients with BPD
tend to make riskier choices because they are unable to learn to adjust them based on nega-
tive feedback (Schuermann et al., 2011). According to the distorted emotional salience of
events, the tendency toward risk will change. For example, the phobic patient will tend to
attach greater weight to events that are not risky and, based on them, his behavior will become
risk-averse and very prudential. Conversely, the narcissistic or bipolar patient in their hypo-
manic phase will underestimate risk, will lean toward illusion control and will show a marked
propensity for risk. Lastly, emotion in psychotics reflects on their analysis of data relevant to
choices. People with psychosis make decisions on the basis of less evidence. They jump to the
conclusion (JTC). In a systematic review and meta-analysis to investigate the magnitude and
specificity of the JTC, they found that JTC is linked to emotional content as greater probability
of having delusions, and it generates a more extreme reasoning style than people with other
mental health conditions (Dudley et al., 2015). How does emotion reflect on the individual’s
ecological rationality? Emotion reflects on the salience of memorized information. When we
have an experience that has a strong emotional component, this information will become
easier to retrieve later. For example, if we have an emotional experience that is extremely posi-
tive concerning the teaching of philosophy, when deciding about your children’s education,
in the cue order generated, good philosophy teaching will be at the top of the ranking when
comparing lexicographically a number of schools. Another example concerns the fact that if
we admire a pair of parents for how they have educated their children and for their children’s
school performance, we will be inclined to imitate their choices in a selection of school or in
the selection of extracurricular activities, like the type of summer school. Emotion shapes the
dynamic or social heuristics and the associated social ecological rationality. The emotional label
makes salient who we consider to be a success in which social group to take as a reference for
the imitation of social rules. Under point 2 we noted the emotional component in some BPDs,
such as hysterical personality, generates an excessive mentalization that can have negative effects
on social heuristics and social coordination. A similar distortion in the role of emotion in heur-
istic decision making can also be observed in other mental disorders, such as bipolar syndrome,
where the emotional sphere is altered, for example, in hypomanic phases when overconfidence
and egocentrism prevail, with salient information concerning mostly groups or leaders, who
are in favor of one’s theories or designs or visions of the world. The hypomaniac will tend to
base his choices on his overestimated capacity to learn and evaluate instead of relying on the
people who are close to him, who, under different conditions, he would admire for the success
they have achieved. By contrast, in depressive phases, this individual has limited confidence in
himself, but at the same time is not very open to social learning. Unlike the hypomaniac, the

364
365

Psychopathological irrationality and BR

mildly depressed person is extremely critical and skeptical about other people’s capacities. His
emotional sphere, characterized by pessimism and negativity, leads him not to remain enthusi-
astic about external events, and consequently to attribute scarce salience to information that he
has memorized. Only very few events manage to overcome the threshold of positive attention
and salience. When this happens, the selection threshold is extremely high and leads to a good
ecological rationality in decision making. The situation is different with seriously depressed
individuals. In these cases, depression pervades all cognitive activity and every bit of informa-
tion tends to confirm pessimism and affective suffering. Social learning becomes impossible
because every bit of external information is distorted through the depressive filter. In other
mental disorders, like obsessive-compulsive personality or paranoid personality, hyperemotional
stimuli generate polarization in distortion of the emotional salience of information, leading
to a recall of distorted information that is not ecologically rational. The paranoid is focused
on information confirming the hostile attitude of other people. The obsessive concentrates
on information that confirms his fears. The role of information salience is central in the eco-
logical functioning of many heuristics and it has acquired growing importance in explaining
some psychiatric disorders. For example, Jim van Os and Shitij Kapur (2009) propose replacing
the term schizophrenia with “salience disorder.” Hallucinations are described as the expres-
sion of aberrant salience of visual and auditory information. In hallucinations, some sensorial
input acquires salience when the relative neural signal is amplified at the expense of others.
This attracts our attention in an aberrant way. According to Shitij Kapur (2003), hallucinations
reflect a direct experience of the aberrant salience of internal representations. Thus, a person
like Mr B––as described by Michael Gazzaniga (2018)––believes that the FBI is constantly
spying on him and any visual and auditory information becomes salient to prove his delirium.
He cuts social relations because he believes that the people around him are paid by the FBI
to inform on his actions. At home, he does not shower naked, because he thinks he is being
observed. What is interesting here is that Mr B’s delirium appears rational according to specific
hallucinations. Paradoxically these delusional, delirious beliefs are a rational cognitive effort to
give meaning to salient aberrant sensorial experiences. Consequently, Mr B is formally rational
in connection with the salient aberrant experiences, but he is not ecologically rational because
his hallucinations distort the salience of external information and decrease his cognitive success
in coping with the environment.

Some paradoxes of psychopathological irrationality

If the negative role of mental disorders on the individual’s ecological rationality appears evident,
a few last reflections can be made on some paradoxes. The following are examples thereof.

Greater logical rationality in psychiatric patients

Consider this clinical description of an impulsive-compulsive behavior (from Gangemi,
Mancini, & Johnson-Laird, 2013, p. 48):

The photographer must have been close to Rock Hudson because the photograph
was a “close up.” So, the photographer himself might have been contaminated. So,
when he developed the negative, he could have contaminated it. The negative was
in contact with the print of the photograph and so could have contaminated it. The
man in charge of printing the newspaper used the photograph, and so, he could have
passed its contamination on to the newspaper’s printer. The printing press could have

365
366

Riccardo Viale

passed the contamination on to the picture in every newspaper. So, when I touched
the newspaper, I may have been contaminated with the HIV virus.

The patient’s reasoning is impeccable. She constructs a long chain of logical inferences that aim
to demonstrate that she has contracted HIV. According to Gangemi et al. (2013), she realizes
that the conclusion is unlikely, yet typically for such patients, she cannot reject it. The chain of
logical inferences is based on a distorted knowledge of reality that she accepts arbitrarily and in
an ad hoc manner.
Logical rationality in several mental disorders can also be found in this simulation of con-
sistency of choice. As noted above, the norm of “independence of irrelevant alternatives”
presupposes this in the case of John and his choice to go to the club, when faced with two
alternatives A and B, he cannot prefer A to B. Subsequently, when a third alternative C is
added, a preference reversal occurs and John opts for B over A. John makes this inversion
because his reference is not only internal to preferences, but also external, that is to say, relating
to social norms. In this case, according to the norm, it would be correct not to go to a
club with the girlfriend of a friend. A different situation can be observed in the case of sev-
eral mental disorders, where the reference to social norms tends to dissolve. If John had an
antisocial or narcissistic personality, he would not have taken into account the social norm
of correctness. He would have chosen B, going to the club with Jane, in both the first and
second cases (with and without the alternative of meeting Carl at the club). By so doing, John
would not have broken the norm of “independence of irrelevant alternatives” and his behavior
would have been correct from the point of view of logical rationality. Paradoxically, this, like
other mental disorders, generates behaviors that are more coherent with the dictates of logical
rationality, unlike what was believed to be case in the past. This confirms that the criterion of
logical rationality is not adequate to account for the adaptive success and ecological rationality
of human behavior. Is the hypoemotion of mental disorders linked to economic rationality?
Above, we have seen that patients with autism were less responsive to the framing of monetary
outcomes as either losses or gains and, thus, exhibited more internally consistent behavior. Loss
aversion caused by heavier emotional content of loss vs. gain does not apply to autism and, as
shown by De Martino et al. (2008), the behavior is less subject to the framing effect. In short,
given the presence of fewer emotional states, the autistic individual has no greater propen-
sity to risk in frames of loss and no greater aversion to risk in frames of gain. Neurological or
mental disorders, reducing the activation of emotional mechanisms, generate behaviors that
are more coherent in the choice of a product, satisfy the criteria of maximizing social utility,
neutralize myopic loss aversion and the regret effect, are more rational in the ultimatum game,
and are less subject to fear in situations characterized by ambiguity and uncertainty (Hertwig
& Volz, 2013). In short, hypoemotionality makes behaviors more economically rational. These
phenomena confirm the gap between economic rationality and adaptive success in real social
environmental contexts. Human reason is characterized by the role of emotions and intuitions
in guiding decision-making of a heuristic type. The gut feeling (Gigerenzer, 2007) is often the
best option when making a decision in situations characterized by ambiguity, uncertainty, and
stringent time constraints. Cold rationality devoid of any emotional content would make the
subject unable to swiftly adapt to situations and would lead to adaptive failure.

Ecological rationality of the paranoid mind?

In the previous example concerning John’s behavior toward the decision of whether or not
to go to the club with Jane, it was concluded that, while it violated the rational norm of

366
367

Psychopathological irrationality and BR

“independence of irrelevant alternatives,” his behavior was certainly coherent with his strategy
and adaptive to the social context, and therefore ecologically rational. The coherence between
strategies and behavior is not only a characteristic of a normal subject. It could happen that in
cases of mental disorders, one’s behavior derives from a strategy that is coherent to one’s heur-
istic rules, but the result does not appear to be (at least at first glance) ecological compared
to the social context in which the decision is made. For example, a paranoid personality in
a company may behave according to his own heuristics (for example, the rule whereby “If
you see two colleagues who are talking and looking at you, then they are plotting something
against you”), his behavior may be coherent (for example, by trying to boycott their work in
the office), but presumably not adaptive to his work interests (by forcing his colleagues to mar-
ginalize him and demean him and generating an organizational malaise that would frame him
as the person responsible for the situation). The word “presumably” is used here because we
might wonder if these are the only adaptive criteria to be considered when judging his eco-
logical rationality. If our judgment results from an analysis of all external constraints and the
objectives of a paranoid personality on which to evaluate the success of his behavior, then those
that appear to us to be negative results (like work marginalization and hostility on the part of his
colleagues) may appear to the paranoid as unintentional objectives pursued in order to confirm
his hypothesis of persecution through a self-evident “hindsight bias” and “confirmation bias.”
The classic self-fulfilling prophecy would thus be: “I was right to suspect my colleagues. They
took their masks off and they are unfairly persecuting me.” This is particularly true if we con-
sider not only borderline patients, but paranoid schizophrenia, where hypotheses of persecution
take on proportions of true delirium. Depending on the objective and the external environ-
mental context, it could be ecologically rational for a paranoid patient to behave in a way that
allows him to achieve results that from our perspective are extremely negative. This could be
labeled irrational only if we assume a normative, benchmark for well-being and adaptive success
that is external in nature by which to evaluate the paranoid individual’s behavior. That is, we
would be forced to judge ecological rationality, not only based on the individual’s success in
achieving internal, endogenous objectives, but also in relation to external, exogenous, conven-
tional norms regarding social well-being or individual wellness. In other words, we would have
to take on a paternalistic evaluation approach. If the behavior was to be judged only on the basis
of subjective objectives and environmental constraints, then, paradoxically, the judgment would
be that the behavior is coherent and ecologically rational.

Conclusion: what role for the brain?

Unlike neurological diseases where the biological foundation of the alterations was never in
question, in psychiatry, this has happened in alternate phases. The main reason was empirical
evidence at the cellular level of many neurological diseases, especially those of the degenera-
tive and traumatic type, and the lack of psychiatric ones. Moreover, there was a strong cultural
resistance against considering psychiatric disorders of a cognitive, perceptive and emotional
type as diseases of the body and not of the mind. This is also because at the time there were
no instruments capable of detecting relevant cellular changes. In any case, the study of neuro-
logical diseases in the past has led to great advances in knowledge of both the central and
peripheral nervous system. Today, the situation has changed dramatically. As noted by Ferrer
(2018), innovation in the fields of imaging, CAT scans, and electroencephalography allows us
today to highlight even minimal changes in neuronal metabolism and electrical potential for
actions, something that was unthinkable only a short time ago. This has allowed us to study,
in completely new ways, the cognitive, perceptive, and emotional alterations in the fields of

367
368

Riccardo Viale

psychiatric diseases. This type of analysis has allowed us to define in a brand-new way the rela-
tionship between the environment, inheritance, and neuronal alteration, and thus to better
calibrate the possible therapies. Eric Kandel, in his book The Disordered Mind (2018), starts from
this relationship and asks the question: in light of the current knowledge of genetics in neuro-
science, is it possible to find a synthesis integrating the influence of the social environment, the
role of psychological therapies, and the biological dimension of mental alteration? To this end,
he offers autism as an example. Human beings are genetically social animals. Human mental
development and decision-making processes are constantly influenced by what other people
are doing. The use of social heuristics, that is to say decision-based rules of imitation of other
people’s behavior, gives us the advantage of the ability to solve many problems without having
to find a solution by ourselves. The capacity to imitate and interact socially is based on the cap-
acity to read and simulate the mind and the behavior of others (Viale, 2019). Leslie Brothers
(2002) identified a series of interconnected cerebral regions that elaborate social information
and which are relevant to autism. This “social brain” is made up of various parts that are respon-
sible for emotion (amygdala), the interpretation of biological motion (the upper temporal lobe),
the simulation of other people’s behavior (mirror neurons), the theory of the mind (the frontal-
temporal system), and facial recognition (inferior temporal cortex). Through fMRI (Functional
Magnetic Resonance Imaging), Stephen Gotts and colleagues (2012) recently discovered that in
autistic individuals, the neural circuit of this social brain is altered. In particular, this alteration
explains two important characteristics of the autistic mind: the difficulty in understanding the
meaning of biological movements like stretching out a hand or walking toward another person,
and the inability to understand our relationship between gaze and intention, and anticipating
and simulating other people’s behavior based on that input. Scholars like Bruno Bettelheim
argued that autism does not have a biological basis, but that it was the result of inadequate
maternal care (the regrettable and imputable theory of the “refrigerator mother”). Today
research has not only identified the neural mechanisms that are at the root of autism, but is also
discovering the related genetic mutations. This begins with one key observation: in homozy-
gous twins (twins with the same genetic code), if one is autistic, the probability that both are
autistic is 90 percent.
Looking at the origins of autism, the environmental influence does not appear to be rele-
vant, and this makes it difficult to identify potential treatment based on psychological therapy.
However, in other forms of mental disorders, the biological role of the social environment and
psychological therapy appears much more significant. By “biological role,” we refer to the cap-
acity to modify the synaptic connections responsible for mental alterations. This is a qualifying
point for Kandel, that allows him to discuss the synergy between pharmacology and psycho-
therapy. In pharmacological therapy, one can consider a biological-type treatment that acts on
the mediators and synaptic receptors, and the same is true also for cognitive behavioural psy-
chotherapy. This has been demonstrated through fMRI, which highlights the effects of psycho-
therapy on the activation of cerebral areas. One relevant example concerns depression. Today it
is possible with an fMRI to identify the patients who require psychotherapy and those who also
need pharmacological treatment. Some time ago, Helen Mayberg (2009) identified the neural
circuit of depression and discovered that when the basal activity of one of its components, the
right anterior insula (responsible for self-conscience and social experience) is below average,
people respond well to psychotherapy and not to anti-depressants, while when basal activity is
above average, the opposite is true. A relevant contribution to addressing the neurobiological
basis of mental illness and the rational characteristics of thought can come from the studies of
the emerging field of neurocomputational psychiatry.16 It might describe how computational

368
369

Psychopathological irrationality and BR

models of cognition can infer the current state of the environment and weigh up future actions.
It might allow us to understand how the abnormal perceptions, thoughts, and behaviors that are
currently used to define psychiatric disorders relate to normal functions and neural processes.
For example, reinforcement learning, describing how the brain can choose and value courses
of actions according to their long-term future value, may account for how depressive symptoms
may result from aberrant valuations and prior belief about the loss of agency (Adams et al.,
2016). Or the notion of the brain as a Bayesian inference machine may account for how several
cortical abnormalities in schizophrenia might reduce precision in predictive coding, biasing
Bayesian inference toward sensory data and away from prior beliefs (Adams et al., 2013). It
would be interesting to analyse from a neurocomputational standpoint the failure to satisfy the
ecological rationality in some mental disorders.
In conclusion, the ecological irrationality of mental disorders was discussed using a con-
ceptual hypothetical approach. The theme, however, deserves greater attention and empirical
study, which could favor the cognitive foundation of psychopathology based on ecological
rationality. This chapter aimed to provide ideas concerning a future research program using,
alongside the instruments of cognitive psychology, those of neurobiology and, in particular,
neurocomputational psychiatry, to provide tools to identify the causes of the limited ecological
adaptability of the psychiatric subject.

Acknowledgments
The author wishes to thank Ralph Hertwig and Gerd Gigerenzer for their invaluable and useful
suggestions.

Notes
1 Stated simply, probabilistic inference claims that whenever the two end-terms share the middle term in
common, they will be assumed to be related to each other. In other words, if the end-terms are related
positively, then a positive conclusion will be drawn, while if they are related negatively, then a negative
conclusion will be drawn (Evans et al. 1993, p. 239).
2 Conversion is a validating form of immediate inference for E (Negative Universal Proposition)—and
I (Affirmative Particular Proposition)-type categorical propositions. To convert such a proposition is to
switch the subject and predicate terms of the proposition, which is non-validating for the A (Affirmative
Universal Proposition)—and O (Negative Particular Proposition) type propositions. Hence, the fallacy
of illicit conversion is converting an A- or O-type proposition. For example, “We like the beautiful and
don’t like the ugly; therefore, what we like is beautiful, and what we don’t like is ugly...”
3 The Pepsi paradox refers to the observation that Pepsi is preferred to Coke in blind taste tests, despite
Coke being regarded as the more successful brand.
4 Suppose two urns each contain 100 balls. The risky urn contains 50 red and 50 black balls. The
ambiguous urn contains red and black balls in unknown proportions. When individuals who stand to
receive a prize for drawing a red ball are invited to choose between the two urns, they generally prefer
the risky to the ambiguous urn. When they are subsequently promised a prize for drawing a black ball,
their aversion to the ambiguous urn remains. This persistent preference, which leads humans to stray
from the axioms of standard decision theory, has been interpreted as ambiguity aversion. If individuals
prefer to draw a red ball from the risky rather than the ambiguous urn, then their subjective probability
of drawing a red ball from the ambiguous urn must be 0.5. Therefore, they should prefer to draw a black
ball from the ambiguous rather than the risky urn, but most do not. However, the “right” kind of brain
damage seems to cure humans of ambiguity aversion.
5 It is not only humans who violate this type of norms; animals do too. For example, honeybees violate
transitivity in their preferences, preferring flower A to B, B to C, and C to D, and yet preferring D to
A (Shafir, 1994). The violation of syntactical norms appears to be essential for survival.

369
370

Riccardo Viale

6 Heuristic refers here to a strategy ignoring part of the information for the purpose of making decisions
faster, more frugally and/or more accurately than when using more complex methods (Gigerenzer &
Gassmaier, 2011).
7 With a sequential study on several reasons, there are heuristics of the lexicographical and satisficing
type. In this case, too, the decision is made based on one reason without any compensatory analysis
of the trade-off between the weights of the various reasons. It is on the basis of these models that
the “fast-and-frugal trees” were conceived; these are decisional trees the branches of which present a
binary choice to which heuristics based on one reason are applied.
8 Decisions in conditions of uncertainty do not include risk situations with a high number of alternatives
in which heuristics can still be successfully applied. For example, when playing chess, the alternatives
are theoretically knowable, but in fact are so numerous as to produce a computational explosion of the
decisional problem. As Simon correctly points out, the only solution is the heuristic decision.
9 The Effective Annual Percentage Rate (EAPR) is the effective annual interest rate, including all con-
tract expenses.
10 In these cases, we are faced with the “bias-variance” dilemma. On one hand, statistical samples carry
the danger of variance. On the other, heuristics that ignore basic statistical information risk generating
bias. Consider, for example, the “base-rate fallacy” in the ex post evaluation of the phenomenon based
on new information. When the amount of data to be collected is enormous, the error derived from
small samples is higher than that derived from the information bias of the heuristics (Gigerenzer &
Gassmaier, 2011).
11 Nudge theory (Thaler & Sunstein, 2008) applies the imitation of social rules as an instrument to direct
the citizens toward behaviors related to their well-being, or the social or environmental good. In fact,
the imitative behavior that is encouraged refers to social imitation heuristics (Viale, forthcoming).
12 This last heuristic is also at the heart of the interpretation of the efficacy of policy defaults as implicit
recommendations in Nudge theory (Thaler & Sunstein, 2008; Viale, forthcoming).
13 This type of limited and ecological rationality is such because it builds on the “embodied” dimension
of cognition. Without this embodiment, limited individual rationality would not be ecological, in that
it would not possess the terminals to interact successfully with the environment (for an in-depth ana-
lysis of this theme, see Chapter 23 by Gallese, Mastrogiorgio, Petracca, & Viale in this volume).
14 Feedbacks are an important concept also in the theory of complexity and in particular in Complex
Adaptive Systems (CAS). When the interactions are not independent, feedback can enter the system
and alter its dynamics. When feedback is negative, changes get absorbed and the system tends toward
stability. When it is positive, changes get amplified, leading to instability (Miller and Page, 2007, p. 50).
Systems that settle into equilibrium tend to include negative feedback. On the contrary, systems that
generate complexity tend to include positive feedback. An example is the phenomenon of network
externality.
15 This test is relevant because facial emotion recognition may contribute to social cognitive responses.
16 Neurocomputational psychiatry aims to describe the relationship between the brain’s neurobiology, its
environment, and mental symptoms in computational mathematical terms (Adams et al., 2016).

References
Adams, A. R., Huys, Q., & Roiser, J. P. (2016). Computational psychiatry: towards a mathematic-
ally informed understanding of mental illness. Journal of Neurological and Neurosurgical Psychiatry,
87(1), 53–63.
Adams, A. R., Stephan, K. E., Brown, H. R., Frith, C. D., & Friston, K. J. (2013). The computational
anatomy of psychosis. Frontiers in Psychiatry, 4, 1–26.
Alos-Ferrer, C. (2018). Analisi critica di “Social Neuroscience”: la ricerca sul cervello sociale e l’economia
si influenzano a vicenda? Sistemi Intelligenti, XXX, 2, 229–274.
Alter, A.L., & Oppenheimer, D.M. (2006). Predicting short-term stock fluctuations by using processing
fluency., Proceedings of the National Academy of Sciences U.S.A., 103, 9369–9372.
American Psychiatric Association (2000). Diagnostic and statistical manual of mental disorders (DSM-IV-TR).
Washington, DC: APA.
American Psychiatric Association (2013). Diagnostic and statistical manual of mental disorders (DSM-V-TR).
Washington, DC: APA.
Arieti, S. (1963). Interpretazione della schizofrenia. Milan: Feltrinelli.

370
371

Psychopathological irrationality and BR

Arkes, H.R., Gigerenzer, G., & Hertwig, R. (2016). How bad is incoherence? Decision, 3(1), 20–39.
Bara, B. G. (Ed.) (2005). Nuovo manuale di psicoterapia cognitiva. Turin: Bollati Boringhieri.
Boyd, R., & Richerson, P. J. (2005). The origin and evolution of cultures. New York: Oxford University Press.
Brothers, L. (2002). The social brain: A project for integrate primate behaviour and neurophysiology in
anew domain. Concepts in Neuroscience, 1, 27–51.
Brunswik, E. (1952). The conceptual framework of psychology. In International encyclopedia of unified science
(Vol. 1, no. 10, pp. 4–102). Chicago: University of Chicago Press.
Camille N. et al. (2004). The involvement of the orbitofrontal cortex in the experience of regret. Science,
304, 1167–1170.
Chapman, L. J., & Chapman, J. P. (1959). Atmosphere effect re-examined. Journal of Experimental
Psychology, 58, 220–226.
Clark, A. (2016). Surfing uncertainty: Prediction, action, and the embodied mind. Oxford: Oxford University Press.
Damasio, A. (1994). Descartes’ error. New York: Avon.
De Martino, B., Harrison, N. A., Knafo, S., Bird, G., & Dolan, R. J. (2008). Explaining enhanced logical
consistency during decision making in autism. Journal of Neuroscience, 28, 10746–10750.
Domes, G., Czieschnek, D., Weidler, F., et al. (2008). Recognition of facial affect in borderline personality
disorder. Journal of Personality Disorders, 22, 135–147.
Dudley, R., Taylor, P., Wickman, S., & Hutton, P. (2015). Psychosis, delusions, and the ‘jumping to
conclusions’ reasoning bias: A systematic review and meta-analysis. Schizophrenia Bulletin, 42(3),
652–665.
Dziobek, I., Preissler, S., Grozdanovic, Z., et al. (2011) Neuronal correlates of altered empathy and social
cognition in borderline personality disorder. Neuroimage, 57, 539–548.
Evans, J. St. B., Over, D. E., & Manktelow, K. I. (1993). Reasoning, decision making and rationality,
Cognition, 49(1–2), 165–187.
Foucault, M. (1961). History of madness. London: Routledge.
Gangemi, A., Mancini, F., & Johnson-Laird, P. (2013). Emotion, reasoning, and psychopathology. In
I. Blanchette (Ed.), Emotion and reasoning. London: Psychology Press.
Garcia-Retamero, R., Takezawa, M., & Gigerenzer, G. (2006). How to learn good cue orders: When
social learning benefits simple heuristics. In R. Sun & N. Miyake (Eds.), Proceedings of the 28th annual
conference of the Cognitive Science Society (pp. 1352–1358). Mahwah, NJ: Erlbaum.
Gazzaniga, M. S. (2018). The consciousness instinct: Unraveling the mystery of the brain makes the mind.
New York: Farrar, Straus, and Giroux
Gigerenzer, G. (2007). Gut feeling. London: Penguin.
Gigerenzer, G., & Gaissmaier, (2011). Heuristic decision making. The Annual Review of Psychology, 62,
451–482.
Gigerenzer, G., & Hoffrage, U., (1995). How to improve Bayesian reasoning without instructions: Frequency
formats. Psychological Review, 102, 684–704.
Gigerenzer, G., Todd, P. M. & the ABC Research Group (Eds.) (1999). Simple heuristics that make us smart.
New York: Oxford University Press.
Goldstein, D. G., & Gigerenzer G. (2009). Fast and frugal forecasting. International Journal of Forecasting,
24, 760–77 2.
Gotts, S.J., Simmons, W. K., Milbury, L. A., Wallace, G. L., Cox, R. W., & Mertin, A. (2012). Fractional
of social brain circuits in autism spectrum disorders. Brain, 135, 9, 2711–2725.
Henrich, J., & McElreath, P. (2003). The evolution of cultural evolution. Evolutionary Anthropology: Issues,
News, and Reviews, 12(3), 123–135.
Hertwig, R., & Herzog, M. S. (2009). Fast and frugal heuristics: Tools of social rationality. 27(5),
661–698.
Hertwig, R., & Volz, K. G. (2013). Abnormality, rationality, and sanity. Trends in Cognitive Sciences, 17(1),
547–549.
Hsu, M., Bhatt, M., Adolphs, R., Tranel, D., & Camerer, C. F. (2005). Neural systems responding to
degrees of uncertainty in human decision-making. Science, 310, 1680–1683.
Jennings, T. C., Hulbert, C. A., Jackson, H. J., et al. (2012). Social perspective coordination in youth with
borderline personality pathology. Journal of Personality Disorders, 26, 126–140.
Johnson-Laird, P. N., Mancini, F., & Gangemi, A. (2006). A hyper-emotion theory of psychological
illnesses. Psychological Review, 113, 822–842.
Kahneman, D. (2011). Thinking, fast and slow. New York: Farrar, Strauss & Giroux
Kandell, E. (2018). La mente alterata. Milan: Cortina.

371
372

Riccardo Viale

Kapur, S. (2003). Psychosis as a state of aberrant salience: A framework linking biology, phenomenology,
and pharmacology in schizophrenia. American Journal of Psychiatry, 160, 13–23.
Knoch D. et al. (2006). Diminishing reciprocal fairness by disrupting the right prefrontal cortex. Science,
314, 829–832.
Koenigs, M., & Tranel, D. (2008) Prefrontal cortex damage abolishes brand-cued changes in cola prefer-
ence. Social Cognitive and Affective Neuroscience, 3, 1–6.
Koscik T.R., & Tranel D. (2013). Abnormal causal attribution leads to advantageous economic decision-
making: a neuropsychological approach. Journal of Cognitive Neuroscience, 25, 1372–1382.
Langer, E. J. (1975). The illusion of control. Journal of Personality and Social Psychology, 32(2), 311–328.
Mark, A., & Lam, L. (2013). Neurocognitive profiles of people with borderline personality disorder.
Current Opinion in Psychiatry, 26, 90–96.
Matte Blanco, I. (1981). L’inconscio come insiemi infiniti. Turin: Einaudi.
Mayberg, H. S. (2009). Targeted electrode-based modulation of neural circuits for depression. Journal of
Clinical Investigation, 119(4), 717–725.
Mendez, M. F. (2009). The neurobiology of moral behavior: Review and neuropsychiatric implications.
CNS Spectrums, 14, 608–620.
Miller, J.H., & Page, S.E. (2007). Complex adaptive systems. Princeton NJ: Princeton University Press.
Moore, D. A., & Schatz, D. (2017). The three faces of overconfidence. Social and Personality Psychology
Compass, 11(8).
Ortmann, A., Gigerenzer, G., Borges, B., & Goldstein, D.G. (2008). The recognition heuristic: A fast and
frugal way to investment choice? In C. R. Plott & V. L. Smith (Eds.), Handbook of experimental economics
results, vol. 1. Amsterdam: North Holland.
Richerson, P. J., & Boyd, R. (2005). Not by gene alone: How culture transformed human evolution.
Chicago: University of Chicago Press.
Samuelson, P. (1938). A note on the pure theory of consumers. Economica, 6, 61–71.
Schooler, L. J., & Hertwig, R. (2005). How forgetting aids heuristic inference. Psychological Review, 112,
610–628.
Schuermann, B., Kathmann, N., Stiglmayr, C., & Endrass, T. (2011) Impaired decision making and feed-
back evaluation in borderline personality disorder. Psychological Medicine, 41(9), 1917–1927.
Schurtz, G., & Hetwig, G. (2019). Cognitive success: A consequentialist account of rationality in cogni-
tion. Topics in Cognitive Science, 11, 7–36.
Sen, A. (1993). Internal consistency of choice. Econometrica, 61, 495–521.
Sen, A. (2002). Rationality and freedom. Cambridge, MA: Harvard University Press.
Serwe S., & Frings C. (2006). Who will win Wimbledon? The recognition heuristic in predicting sport
events. Journal of Behavioral Decision Making, 19, 321–332.
Shafir, E. (1994). Uncertainty and the difficulty of thinking through disjunctions. Cognition, 50(1–3),
403–430.
Shafir, S., & Smith, W. (2002) Context-dependent violations of rational choice in honeybees (Apis
mellifera) and gray jays (Perisoreus canadensis). Behavioral Ecology and Sociobiology, 51, 180–187.
Sharp, C., Pane, H., Ha, C., et al. (2011). Theory of mind and emotion regulation difficulties in adolescents
with borderline traits. Journal of the American Academy of Child and Adolescent Psychiatry; 50, 563–573.e1.
Shiv, B., Lowenstein, G., Bechara, A., Damasio, H., & Damasio, A. (2005). Investment behavior and the
negative side of emotion. Psychological Science 16, 435–439.
Simon, H. (1978). Rational decision making in business organizations. In A. Lindbeck (Ed.), From Nobel
lectures, economics 1969–1980. Singapore: World Scientific Publishing Co.
Simon, H. (1986). Decision making and problem solving. Washington, DC: National Academy Press.
Simon, H. (1990). Invariants of human behaviour. Annual Review of Psychology, 41, 1–19.
Staebler, K., Renneberg, B., Stopsack, M., et al. (2011). Facial emotional expression in reaction to social
exclusion in borderline personality disorder. Psychological Medicine, 41, 1929–1938.
Thaler, R., & Sunstein, C. (2008). Nudge. London: Penguin.
Todd, P. M., Gigerenzer, G., & the ABC Research Group (2012). Ecological rationality: Intelligence in the
world. New York: Oxford University Press.
Tversky, A., & Kahneman, D. (1974). Judgement under uncertainty: Heuristics and biases. Science, 185,
1124–1131.
Van Os, J., & Kapur, S. (2009). Schizophrenia. The Lancet, 374, 635–645.
Viale, R. (2012). Methodological cognitivism. Vol. I. Mind, rationality and society. Heidelberg: Springer.
Viale, R. (2013). Methodological cognitivism. Vol. II. Cognition, science and innovation. Heidelberg: Springer.

372
373

Psychopathological irrationality and BR

Viale, R. (2019). La razionalità limitata “embodied” alla base del cervello sociale ed economico. Sistemi
Intelligenti, 1.
Viale, R. (forthcoming). Nudging. Cambridge, MA: MIT Press.
Viale, R., Macchi, L., & Bagassi, M. (in preparation). Decision making as problem solving.
von Domarus, E. (1925). Ueber die Beziehung des normalen zum schizophrenen Denken. Archives of
Psychiatry, 74, 641–646.
von Domarus, E. (1944). The specific laws of logic in schizophrenia. Berkeley, CA: University of California Press.
Wason, P.C. (1966). Reasoning about a rule. Quarterly Journal of Experimental Psychology, 20(3), 273–281.
Wason, P. C., & Johnson-Laird, P. N. (1972). The psychology of reasoning. Cambridge, MA: Harvard
University Press.
Wübben, M. & Wangenheim F. (2008). Instant customer base analysis: managerial heuristics often “get it
right”. Journal of Marketing, 72, 82–93.
Young L., & Koenigs M. (2007). Investigating emotion in moral cognition: a review of evidence from
functional neuroimaging and neuropsychology. British Medical Bulletin, 84, 69–79.

373
374
375

PART IV

Embodied bounded rationality

376
377

23
EMBODIED BOUNDED
RATIONALITY
Vittorio Gallese, Antonio Mastrogiorgio, Enrico Petracca,
and Riccardo Viale

The root of bounded rationality in cognitive psychology and the

bounds of embodied cognition
There is little doubt that one of Simon’s key contributions throughout his scientific career––if
not the main one––was rooting the notion of bounded rationality in cognitive psychology
(Simon, 1976). He called his notion of bounded rationality in the cognitive psychology realm
“cognitivism” (Haugeland, 1978), an approach, also known as the “information-processing”
approach, which he contributed to affirming together with his colleague Allen Newell, starting
in the mid-1950s. According to cognitivism, cognition works through the internal (i.e. mental)
manipulation of representations of the external environment accomplished through referen-
tial “symbols” (e.g., Newell & Simon, 1972). Connecting Simon’s theory of cognition to his
theory of rationality is the notion that cognition works in a way that is necessary and sufficient for
intelligent behavior (what is known as the “physical symbol system hypothesis,” see Newell &
Simon, 1976). The idea that results from integrating Simon’s view of cognition with his view
of bounded rationality is that rationality is a “process and product of thought” (Simon, 1978), in
which the internal bounds of reason (Simon, 1955) adapt to the external bounds of the envir-
onment (Simon, 1956) in a disembodied fashion.
In this picture, in fact, there is no place for flesh and blood. Patokorpi (2008) has emphasized
that there is an inner tension in Simon’s thought, as, on the one hand, he was a strong advocate
for a realistic approach to the bounds of rationality, while, on the other, he represented them
through the “unbounded” power of digital computation and the metaphor of computers.
This inner tension, which did not simply concern Simon’s thought but an entire generation of
cognitive scientists, would have huge consequences in the history of cognitive psychology. In
the early 1990s, Newell and Simon’s physical symbol system hypothesis was questioned when
the “embodied robots” designed by Rodney Brooks proved able to simulate simple forms of
intelligent behavior by externalizing most of cognition onto the physical properties of envir-
onments, thus dispensing with abstract symbolic processing (Brooks, 1991). This is just one
instance from the recent history of cognitive science pointing to the fact that while bounded
rationality remains a pivotal notion in behavioral economics and economic psychology, new
theoretical views and massive experimental evidence in cognitive science have superseded

377
378

Vittorio Gallese et al.

cognitivism and its abstract representation of cognition (Wallace et al., 2007). Contemporary
cognitive psychology emphasizes that cognition is “embodied,” as it constitutively depends on
body states, on the morphological traits of the human body, and on the sensory-motor system
(see, e.g., Wilson, 2002). As such, it can be said to introduce another “bound” to human
cognition, able to integrate the internal bounds represented by cognitive limitations and the
external bounds of task environments: the human body. In this chapter, we argue that, in so far
as the human body represents a new bound for human cognition, it can also have an important
role in the re-conceptualization of bounded rationality. Since the new approach of embodied
cognition is so recent, it is still rather plural and variegated, and as such far from a stable syn-
thesis (for reviews on the issue of conceptual pluralism in embodied cognition, see the classic
Wilson, 2002; for more recent reviews, see Gallese & Lakoff, 2005; Clark, 2008; Kiverstein &
Clark, 2009). Without the claim of being exhaustive, here is a list of classic books on the idea
that the body is a constitutive part of cognition: Varela, Thompson, & Rosch (1991); Clancey
(1997); Clark (1997); Lakoff & Johnson (1999); Rowlands (1999); Shapiro (2004); Gallagher
(2005); and Pfeifer & Bongard (2006). As a matter of terminology, although different labels
have been used to identify and distinguish different views of embodiment, we will refer to
them all by means of the common synthetic label “embodied cognition” (Calvo & Gomila,
2008; Shapiro, 2014).

The morphology of the human body and the sensory-motor

system in cognition
Different arguments have been provided to support the idea that cognition is embodied. We
will consider here two main categories of arguments: the evolutionary arguments on the one
hand, and the developmental arguments on the other (Meier et al. 2012). An evolutionary
argument for embodied cognition consists in emphasizing that the brain is structured by evo-
lutionary layers, so that the new layers exploit and build upon the resources of older ones,
which were mostly devoted to ‘lower’ cognitive activities like sensory-motor control (this has
been called the “principle of neural exploitation,” see Gallese 2008, or “principle of neural
reuse,” see Anderson, 2010). Another evolutionary argument concerns the emergence of cog-
nitive artifacts from the morphological traits of the human body as is the case, for example,
with numeric systems. It is well known that numeric systems depend on the underlying
morphological traits used for counting (e.g., the 10-based numeral system uses the 10 fingers
of the hands, while other numeric systems stem from the use of other bodily resources; see
Gibbs, 2006). Numeric processing, as such, is not an abstract process (Cohen Kados & Walsh,
2009). In this regard, even the disembodied cognitive artifact par excellence, the entire apparatus
of mathematics, is rooted in the human body (Lakoff & Nuñez, 2000). On the other hand,
the developmental arguments for embodiment focus on childhood learning, emphasizing the
role of the child’s exploration of the surrounding space (and the related physical sensations)
in the development of cognitive faculties and structures in adulthood (e.g., Williams, Huang,
& Bargh, 2009). Further, developmental processes of imitation, based on “embodied simu-
lation” (Gallese 2003, 2007), lie at the root of intersubjectivity (e.g., Gallese & Goldman,
1998; Iacoboni, 2009; Gallese 2014). Far from being just an alternative hypothesis on the
foundations of cognition, the idea that cognition is embodied is evidenced by a number
of experiments connecting body states to judgment, decision-making, problem-solving,
attitude formation, etc. Experimental evidence shows that body variables decisively direct
and affect decision-making (see the classic study by Damasio, 1994; for a review of related

378
379

Embodied bounded rationality

neuroeconomics evidence, see Reimann & Bechara, 2010). Further, problem-solving is non-
trivially dependent on body correlates like, for instance, eye movement (e.g., Werner & Raab,
2014). Other various experimental evidence shows that people judge steepness depending on
the weight of their backpacks (Bhalla & Proffitt, 1999), that environmental temperature affects
social attitudes (e.g., Zhong & Leonardelli, 2008), that imagined food consumption makes
people satiated (Morewedge, Huh, & Vosgerau, 2010), or that physical weight induces the
perception of importance (Jostmann, Lakens & Schubert, 2009).

A more complex view of Simon’s scissors

Most of the experimental findings introduced above simply do not make sense in the cognitivist
paradigm. This is mainly because an entire conceptual locus, the human body, was missing in
the cognitivist picture of cognition. This is mostly visible if we consider the famous meta-
phor that Simon used to introduce bounded rationality: the scissors metaphor. As Newell and
Simon said:

Just as a scissors cannot cut paper without two blades, a theory of thinking and
problem solving cannot predict behavior unless it encompasses both an analysis of the
structure of task environments and an analysis of the limits of rational adaptation to
task requirements.
1972, p. 55

The existence of two “blades,” cognition and environment, is not able by itself, however,
to express the richness of human cognitive activity. In other words, human cognition is
underdetermined by the two loci of the scissors metaphor.
Before exploring the reasons and consequences of this underdetermination, it is important to
say something on the connection between Simon’s “bounded rationality” and Gerd Gigerenzer’s
“ecological rationality.” As is well known, Gigerenzer was deeply inspired by Simon’s scissors
metaphor (e.g., Gigerenzer & Goldstein, 1996). In line with Simon, Gigerenzer’s notion of
ecological rationality for heuristic judgment maintains that “A heuristic is ecologically rational
to the degree that it is adapted to the structure of the environment” (Gigerenzer et al., 1999,
p. 13). Furthermore, although much less committed than Simon to the cognitivist approach to
cognition, Gigerenzer admits that ecological rationality’s program is inspired by “Simon and
Newell’s emphasis on creating precise computational models” (Gigerenzer et al., 1999, p. 26;
see Petracca, 2017, for further points of contact between the two views). Undoubtedly, eco-
logical rationality moves a step forward with respect to Simon’s bounded rationality by empha-
sizing the crucial requirements of ecology for rational adaptation, but the ‘cognitive’ dimension
of ecology remains either Simonian (e.g., when emphasizing the modularity of human intelli-
gence, Gigerenzer, 1997) or not fully explored.
In a true ecological model of cognition, the role of the human body is pivotal (Hirose, 2002).
Although one may still maintain that adaptation is between “cognition” and “environments,”
this is not a good reason to rule out the role of the human body as a mediating evolutionary or
developmental interface. This is why we propose to enrich the traditional scissors metaphor by
identifying another conceptual locus, the human body, which would metaphorically take the
place of the ‘pivot’ of the scissors. In the next two sections, we will explore how heuristics and
representations, two notions that are central in Simon’s traditional view of bounded rationality,
can be reconceptualized in the light of embodied cognition.

379
380

Vittorio Gallese et al.

Rules of thumb: embodied heuristics

Embodied cognition can help to reconsider such a fundamental notion in bounded rationality
as heuristics. This may be accomplished, at a first level of approximation, by emphasizing that a
common name for heuristic is the “rule of thumb”: an expression emphasizing that heuristics
originate from body resources used for inferential purposes. The thumb can be considered as
a true cognitive resource, either when it is institutionally established as a measuring device or
when it is used on the fly as a tool to estimate approximate distance, length, etc. Less specu-
latively, the heuristic that better represents the trait d’union between ecological rationality and
the embodied cognition approach is the “gaze heuristic,” which is worth examining in some
detail. It is important to note, incidentally, that Gigerenzer discussed this well-known heur-
istic in his more ‘embodied’ book, when he explicitly emphasized the notion of “gut feelings”
(Gigerenzer, 2007). The gaze heuristic is a heuristic that is used to correct motion, on the fly,
to achieve a spatial goal (see Hamlin, 2017); as such, it is used in many real-world activities
like, for instance, catching a ball flying through the air, chasing prey, or landing an airplane.
Contrary to the idea that individuals unconsciously perform complicate computations in this
kind of task (Dawkins, 1976), the gaze heuristic shows that humans simply exploit the coup-
ling between the visual perceptual apparatus and the invariant (because of the laws of physics,
the relative invariance of human environments, etc.) properties of motor tasks. The gaze heur-
istic specifically reads: “Fix the gaze on the objective and adapt movement and running speed
so that the angle of gaze remains constant.” A wide array of motor tasks are possible through
this heuristic without the need to rely on complex computations, simply by focusing on one
variable: the object’s angle. The gaze heuristics shows that adaptation occurs not only between
cognition and environments, but it also crucially involves the structure of the perceptual and
motor apparatuses.

Affordances and heuristics

The gaze heuristic is based on a fundamental principle: perceptive estimations are driven by
pragmatic reasons, as perception aims to facilitate individuals’ adaption to different situations
(Mastrogiorgio & Petracca, 2018). The gaze mechanism is therefore a form of ecological ration-
ality based on perceptual processing (Viale, 2017). However, the perception of environmental
information (used by individuals to make decisions) is not a static and passive activity like taking
a photograph. It involves instead the active research and manipulation of the inputs coming
from eyes, head, and trunk movements. Moreover, an external object is perceived through
the signals that it sends to the individual, so as to trigger neuro-motor patterns functional to
this interaction. Objects ‘speak’ to the individual and ‘tell’ her/him what to do to allow inter-
action. This is the ‘affordance’ offered by the object, which the individual is invited to ‘follow.’
The concept of affordance, introduced by the psychologist of perception James Gibson (1950,
1979), concerns the dynamic relations that are established between an agent and a perceived
object. Any perceptual stimulus coming from objects not only represents a collection of proper-
ties, but also triggers a possibility for action coupled with the sensory and motor characteristics
of the perceiver. Our motor system is activated every time we observe an object that can be
grasped, and get ready to perform an action congruent with the physical features of the object
(Caruana & Borghi, 2016). For instance, a ball thrown in the air induces the action of picking
up the ball, and this is true for any other stimulus (the irregular slope of a mountain triggers
the affordance of climbing it, etc.). Evidence of this embodied mark of perception comes from
neuroscience. By observing the effect of damage to the inferotemporal and parietal regions,

380
381

Embodied bounded rationality

it was noted that during the observation of an object the information is sorted along two
different cerebral channels: a ventral pathway, which connects the visual to the temporal cortex
and oversees semantic recognition processes, and a dorsal pathway in the cortex, which reaches
the parietal lobe and processes the pragmatic aspects connected to visual-motor transformation
and to action (Caruana & Borghi, 2016). The parietal lobe, however, presents features that
are particularly interesting to explain the motor aspects involved in perception. In this region,
a dorso-dorsal pathway can be observed, as well as a ventro-dorsal one, and the intraparietal sulcus
(dividing the inferior parietal lobule from the superior one). The first pathway of visuomotor
transformation is aimed to enable the action of reaching and grabbing objects. The second
pathway concerns instead the storage of motor programs. The intraparietal sulcus is, however, the
element that presents the most significant characteristics in terms of visuomotor coordination.
The neuronal composition of its anterior part (AIP, Anterior Intraparietal Area) presents visual,
visuomotor, and motor neurons that are activated, selectively, every time we observe objects
that can be grasped. During the mere observation of graspable objects, this area automatically
instantiates a grasping action, regardless of the actual will to grasp them. The same happens
also in the area of F5 mirror neurons, to which the AIP is connected. According to Rob Ellis
and Mike Tucker (2000), the affordances concerning the automatic grasping of objects occur
because of the repeated co-occurrence of visual and action patterns. For example, the repeated
past association that has occurred between the ball falling and the action to pick it up results
in an affordance consisting in the automatic behavior to reach it, typical of the gaze heuristic.
Moreover, it would not seem possible for affordances to be triggered indiscriminately, without
any ‘top-down’ adjustment stemming from the context of outcomes and goals to be achieved.
Since we are surrounded by many objects and each of them has multiple affordances (e.g., a
cup may be grasped by the handle, the base, the lip, etc.), it is unlikely that our brain engages
in responding to this proliferation of affordances indiscriminately (Caruana & Borghi, 2016,
p. 48). Tipper and his team (2006) performed a clever experiment that seems to demonstrate
that the effect of affordances is task-driven and not automatic. For example, when facing a
book on a table, if the goal is grasping the book to put it on a shelf, a power grasp affordance
will be triggered. If, instead, the goal is to flip through the pages, a precision grasp one will be
triggered. Thus, when the ball is falling down, the grabbing affordance is activated when we
find ourselves in the context of playing and the goal is to grab the ball. This may be not the case
if we are taking our dog for a walk in a park and people are playing with a ball in the distance.
Almost all objects have conflicting and competing affordances, and it is the context and our
goals that make us select the coherent one. The context may be physical or social. For example,
when the grasping distance of an object makes it too far away in an extra-personal space (phys-
ical context effect), no affordance is triggered. If instead it is within the peri-personal space of
the individual (social context effect), the affordance is triggered. Our capacity to understand
other people’s intentions, through what is called embodied simulation (see below), leads us to treat
them as avatars with regard to the perception of objects and their affordances, as if we want
to act in their place. This effect seems to be generated by embodied simulation working as a
“mind-reading” activity (Viale, 2012).
These remarks on sensory-motor aspects of perception may provide the basis for a variety
of decision-making heuristics. As noted earlier, gaze heuristics can be explained according to a
sensory-motor model. Other heuristics, such as the ones based on recognition like the “take-
the-first” heuristic, can be explained within the same framework (Viale, 2017). If we think
of recognition as the first option coming to mind when playing a team game and having to
pass the ball to a teammate, it is clear that this process relies on sensory-motor memory. The
repeated co-occurrence of similar situations in previous games with the same teammates (or

381
382

Vittorio Gallese et al.

different players occupying similar positions) in the game dynamics leads to the creation of
mnemonic patterns (e.g., through the ventro-dorsal pathway for the storage of motor programs)
of a sensory-motor type that trigger automatic decisions about passing the ball. The player
passes the ball to the first player corresponding to the sensory-motor memory patterns. In other
words, it is the affordance of the teammate in a certain position on the field that triggers the
player’s sensory-motor memory and that makes him/her pass the ball in a certain way.

Heuristics and embodied emotions

People often talk about decisions based on a “gut feeling.” Particularly when a decision is
taken under time pressure, with limited information, and when the effect of this decision has
implications on one’s well-being, one does not rely on complex computations but on gut
feelings. Consider the required speed in financial trading decisions: there is no time for in-depth
reasoning, as decisions have to be made in just a few seconds. Only intuition and gut feeling
can work as decision makers. Gigerenzer (2007) presents numerous examples of decisions taken
in this way. What characterizes gut-based decisions? These are decisions where the emotional
component contained in choosing one option over another is expressed positively or negatively
at the bodily level (Damasio, 1994). When the taxi driver takes one glance at the customer and
feels a “knot in the stomach,” he or she suddenly decides the customer cannot be trusted and it
is not safe to take him on board.
The role of emotions is evident in many other aspects of decision making. When using
the recognition heuristic, it is the emotional element of familiarity that often leads us to pick
one option over another (Viale, 2017). Emotion plays an important role also in the case of
“one reason-based” heuristics (Gigerenzer & Gaissmaier, 2011). Let us consider the “take-
the-best” heuristic, according to which, when making a choice, an individual retrieves from
memory situations, facts, and reasons, and uses the first cue in order to evaluate alternatives.
The emotion related to specific episodes is important to strengthen memorization and the
subsequent retrieval. Heuristics based on recognition and on “one reason” have the capacity
to generate decisions that are more adaptive if compared to the algorithms of (neo)classical
rationality. They allow us to deal with an environment characterized by uncertainty and com-
plexity through rapid decisions that are simple, frugal, and effective in terms of prediction.
They are therefore the center of gravity of the ecological dimension of bounded rationality (Viale,
2017). This adaptive capacity can be explained precisely by the role that emotion plays in the
interaction between the individual and the environment. The mechanisms of emotion seem to
refer to the functions of the ventromedial prefrontal cortex, which includes the orbitofrontal cortex.
By virtue of its multiple incoming connections with sensory and enteroceptive areas and of its
outgoing connections with the autonomous nervous system, the ventromedial cortex serves to
read physical expressions and to assess the physiological state of the body. In this way, it reads
physical sensations connected to decision-making options and automatically leads the indi-
vidual toward those that are characterized by well-being and pleasure, avoiding negative ones.
Another important cortical structure involved in the emotional experience is the insula or Island
of Reil. Its posterior part functions as a primary enteroceptive cortex in its own right. It receives a
variety of stimuli from gustative representation to pain, tactile sensations, disgust, visceromotor
control, thermoception, etc. According to the neuroscientist Bud Craig (2011), the posterior
insula processes this information and constitutes a sort of cerebral map of corporal states. The
ventral part of the insula instead works for the subsequent processing of this map to generate a
sort of re-interpretation of the emotional experience. In short, the insula functions as an ‘emo-
tional eye’ of corporal states, interpreting them, and generating, together with the ventromedial

382
383

Embodied bounded rationality

prefrontal cortex, choices and decisions. (Neural imaging tests could verify the hypothesis that
recognition and one-reason-based heuristics rely on ventromedial prefrontal cortex and insula
activation, and are therefore driven by emotional salience-based retrieval.)

Embodied representations and simulations

Central notions in Simon’s approach to cognitive psychology and bounded rationality are
those of mental representation and simulation. Characteristically, in Simon’s framework,
mental representations and simulations would be constituted of abstract and amodal ‘symbols’
(Newell & Simon, 1972, 1976). Embodied cognition does not completely rule out the role
of representations and simulations in cognition (even if this is not true for all approaches to
embodied cognition, see Petracca, 2017), but definitely rejects the idea that they are abstract
and amodal.
One of the most exciting contributions of neuroscience to the debate about the nature of
the human mind and its functional mechanisms is the discovery of the cognitive role of the cor-
tical motor system. Empirical research has demonstrated, first in non-human primates and then
in humans, that the cortical motor system is functionally organized in terms of motor goals.
Many cortical motor neurons, both in the frontal and parietal lobes, do not discharge during
the execution of elementary movements, but are only active before and during purposive motor
acts, i.e., movements executed to accomplish specific motor outcomes, like grasping, tearing,
holding, or manipulating objects. The teleological dimension of behavior thus entirely belongs
to the functional properties of the motor system. A further element of novelty about the cog-
nitive role of the motor system is provided by the robust evidence of its involvement in per-
ception: premotor and parietal areas contain motor neurons that also perceptually respond to
visual, auditory, and somatosensory inputs (see Gallese & Cuccio, 2015).
The discovery––first in macaque monkeys, then in humans––of “mirror neurons” revealed
the cognitive role of the motor system in social cognition. Mirror neurons are motor neurons
that respond both when a given movement or action is performed and when it is observed
being performed by someone else. Mirror neurons reveal a new empirically founded notion
of intersubjectivity connoted first and foremost as intercorporeality: the mutual resonance of
intentionally meaningful sensorimotor behaviors. The ability to understand others as inten-
tional agents does not exclusively depend on propositional competence, but it is in the first
place dependent on the relational nature of action. According to this hypothesis, it is possible
to directly understand others’ basic actions by means of the motor equivalence between what
others do and what the observer can do. Thus, intercorporeality becomes the primordial source
of knowledge that we have of others.
These findings led to the “Motor Cognition” hypothesis (see Gallese et al., 2009): cogni-
tive abilities like the mapping of space and its perception, the perception of objects occupying
our visual landscape, the hierarchical representation of action with respect to a distal goal, the
detection of others’ motor goals, and the anticipation of their actions are possible because of the
peculiar functional architecture of the motor system, organized in terms of goal-directed motor
acts. The same motor circuits that control individuals’ behavior within their environment also
map distances, locations, and objects in that very same environment, thus defining and shaping
in motor terms their representational content. The way the visual world is represented by the
motor system incorporates agents’ idiosyncratic way to interact with it.
Empirical research demonstrated that the human brain is also endowed with mirror
mechanisms in the domain of emotions and sensations: the very same nervous structures
involved in the subjective experience of emotions and sensations are also active when such

383
384

Vittorio Gallese et al.

emotions and sensations are recognized in others. For example, witnessing someone expressing
a given emotion (e.g., disgust, pain, etc.) or undergoing a given sensation (e.g., touch) recruits
some of the viscero-motor (e.g., anterior insula) and sensori-motor (e.g., SII, ventral premotor
cortex) brain areas activated when one experiences the same emotion or sensation, respectively.
Other cortical regions, though, are exclusively recruited for one’s own and not for others’
emotions, or are activated for one’s own tactile sensation, but are actually deactivated when
observing someone else being touched (for review, see Gallese, 2014; Gallese & Cuccio, 2015).
Embodied simulation theory makes use of a notion of embodiment according to which
mental states or processes are embodied because of their bodily format. The bodily format
of a mental representation constrains what such mental representation can represent, because
of the bodily constraints posed by the specific nature of the human body. Similar constraints
apply both to the representations of one’s own actions, emotions, or sensations involved in
actually acting and experiencing, and also to the corresponding representations involved when
observing someone else performing a given action or experiencing a given emotion or sensa-
tion. These constraints are similar because the representations have a common bodily format.
Hence, embodied simulation is the reuse of mental states and processes involving representations
that have a bodily format. The nature and the range of what can be achieved with embodied
simulation are constrained by the bodily format of the representations involved.
To put it simply, the producer and repository of representational content are not the brain
per se, but the brain-body system, by means of its interactions with the world of which it is
part. The proper development of this functional architecture likely scaffolds more cognitively
sophisticated social cognitive abilities. As recently argued, embodied simulation as a component
of cognitive models, when recruited by the situated and contextualized process of meaning
construction, is an integral part of linguistic meaning, including conceptual knowledge (Cuccio
& Gallese, 2018).

The disembodied approach of current social neurosciences

What is the position of social neuroscience research today with respect to embodied cogni-
tion? Carlos Alós-Ferrer (2018), in line with Cacioppo et al. (2006) and Schutt et al. (2015),
attempts to survey the contribution of neuroscience to social sciences and economics. As
recently noted (Viale, 2019), he correctly emphasizes that social sciences can no longer
function without the empirical data provided by the research at the intersection between social
cognitive psychology and neuroscience. This need not only results from an epistemological
change in economics, but it is rooted in a deeper evolutionary argument: the bidirectionality
between brain and the social world. The brain and the early forms of social organization co-
evolved over time. While the role of the neurocognitive apparatus in the genesis of social phe-
nomena is now widely accepted, the importance of understanding social evolution for brain
studies is not. As Alós-Ferrer writes:

Understanding which characteristics this coevolution selected for and why they
provided an evolutionary advantage provides a far better understanding of the nature
and functioning of the brain than statistical studies on the relative proportion of
neurons and glial cells in each brain region.
2018, p. 235

And this integration between hereditary dimension, social environment, and biological com-
ponent explains, among other things, also the development and the treatment of mental illness

384
385

Embodied bounded rationality

(as highlighted in Eric Kandell’s recent book, Kandell, 2018). The representation made by
Alós-Ferrer of the relation between neuroscience and society suffers, however, from a neuro-
cognitivist bias, which is typical of neuroeconomics: forgetting the body as a fundamental
bridge between the brain and the environment. Accordingly, this bias also regards bounded
rationality as lacking conformity to the standards of economic rationality (thus making it a
synonym for irrationality) and not as a way to adapt to an uncertain and complex social world.
Alós-Ferrer’s approach adopts the epistemological stance of cognitivism with the sole diffe-
rence that behavior and decision-making processes are analyzed on the basis of neurocognitive
models, which are centered on the brain (Legrenzi & Umiltà, 2011; Gallagher, 2018a, 2018b).
What is lacking in this analysis is the ‘embodied’ dimension of cognition, that is, the integra-
tion of central nervous system with all the other visceral, sensory, and motor body parts. The
neurocognitivist bias of contemporary social neurosciences results in the polarization of the
analysis of the social brain mostly in terms of theory of the mind (Theory Theory of the Mind,
or TT). This focus on the theory of the mind as the central element of the social brain stems
from a distance from ecological and embodied arguments. Accordingly, this leads to overlooking
a fundamental component of that function altogether, namely, mind reading, originating
in “embodied simulation” produced by the mirror neurons system (Rizzolatti et al., 2001;
Goldman, 2006; Iacoboni, 2008; see above). According to the TT, the attribution of mental
states to others is possible only through the construction and the development of a theory on
these states (Premack & Woodruff, 1978). Understanding or predicting others’ actions means
engaging in a theoretical inference on propositional attitudes like beliefs, desires, and intentions
of others, through conscious and intentional representations of the other person’s mental states.
In this perspective, there is no room for automatic, unconscious forms of “mind reading” or
an empathetic interpretation of emotional states or “mind feeling” (Viale, 2011, 2012). TT is
typically the expression of the “disembodied cognition” of the cognitivist approach, and finds
its natural collocation in behavioral economics and neuroeconomics. According to Alós-Ferrer,
the theory of the mind is the cognitive equivalent of the game theory, which is fundamental to
understanding social interaction. As he writes: “Social neuroscience is to decision neuroscience
as game theory is to decision theory” (Alós-Ferrer, 2018, p. 259).1
As noted, the mutual understanding of social behaviors seems to occur mostly through an
embodied simulation mediated by the mirror neurons system. This type of study shows that
the social brain cannot be reduced to the fronto-temporal network of the theory of the mind.
On the contrary, it seems to point to the fact that conscious and intentional neurocognitive
activity is the tip of the iceberg of our social interaction. Most of our social life takes place
through processes of automatic, unconscious simulation that is often empathetic with other
people’s behavior, and, hence, embodied. The development and recognition of the self occur
through the imitation of others, and one’s personal identity seems shaped by a reflection onto
others. The newborn child seems to develop mirror neurons through the repeated imitation
of the expressions of the adults he or she interacts with. His/her brain, for example, associates
the image of his/her mother’s smiling face with the motor level intended to replicate it. Mirror
neurons would therefore serve as the basis of the reflection of our behavior in others. Through
mirror neurons, we see ourselves in others. Pfeiffer et al. (2008), for example, found a correl-
ation in children between emotional empathy, measured according to the Interpersonal Reactivity
Index, and the increase in mirror neurons’ activity while observing facial expressions with an
emotional component. The same increase was found in children who were more sociable, more
open, and with more relations than others. Other data seem to indicate that mirror neurons
are activated every time a social identification process occurs. This activation was observed, for
example, with advertisements and the identification with political figures (Iacoboni, 2008). In

385
386

Vittorio Gallese et al.

this way, they seem to provide the basis for understanding social identity, one of the fundamental
mechanisms at the origin of social interaction and the related decision-making processes.
In the emotional and empathetic dimension, mirror neurons also explain a series of emo-
tional social contagions. A case in point is contagious laughter, as mirror neurons explain why
we laugh more when we are in the company of others and why we laugh when others do. The
“audience effect” of empathic resonance also occurs for negative expressions, like pain, disgust,
anger, etc. While, on the one hand, we suffer more when we are in the presence of others, on
the other hand, we identify ourselves, empathically, with the suffering of others (Chartrand
& Bargh, 1999; Longo et al., 2008; Heyes, 2011). Lastly, the embodied simulation of mirror
neurons is clearly a useful decision-making tool in social contexts. When we are making rapid
and intuitive decisions, the capacity to simulate other people’s emotions and actions provides
important contextual elements to make adaptive choices. This type of simulation therefore
plays a heuristic role that fits into the toolkit of bounded and ecological rationality.

Conclusion
The re-conceptualization of human rationality in light of the embodied perspective is grad-
ually emerging in literature under the name of “embodied rationality” (see Spellman & Schnall,
2009; Mastrogiorgio & Petracca, 2015, 2016; Viale, 2017; Gallagher, 2018a, 2018b; see also
Oullier & Basso, 2010). This chapter aims to show that the adaptive and ecological dimension
of bounded rationality should be better analyzed by assuming an embodied cognition per-
spective. Ecological rationality and the functioning of simple heuristics, in particular, would
greatly benefit from inputs from neurobiological and embodied cognition studies. Until now,
few reflections about these opportunities have been made in the framework of the ecological
rationality approach. It remains mainly characterized by reference to information processing
psychology (see Petracca, 2017) and by a Marrian algorithmic level of analysis (Marr, 1982;
Brighton, Chapter 17 in this volume). In fact, something is changing among ecological ration-
ality scholars, in particular with reference to deeper analysis of the neurobiological dimension
of heuristics decision making. According to Nordli and Todd (Chapter 19 in this volume),
neurophysiological studies would promote a new theoretical framework for ecological ration-
ality. For instance, those studies may contribute to understanding strategy selection in decision
making, where strategy selection means the selection of a given heuristic for a particular con-
text. The selection is successful when the selected heuristic matches the structure of a given
situation. That may happen by developing a mapping between context structure and appro-
priate heuristics to use (Marewsky & Schooler, 2011). The mapping may be generated by a
“strategy selection learning” procedure as proposed by Rieskamp and Otto (2006), as strategies
are selected according to predictions regarding their expected results in particular tasks and
contexts. According Nordli and Todd (Chapter 19 in this volume), the reinforcement learning-
based mechanism of strategy selection

is consistent with accounts that tie reinforcement learning processes to recurrent

cortico-basal circuitry (e.g., Yin & Knowlton, 2006), which is the very same circuitry
that is critical to the exploitation of past behavior in the form of habits and fixed
action patterns.
Nordli & Todd, p. 320

In particular, the basal ganglia seem to lie behind most ecologically rational behavior. Basal gan-
glia seem to evaluate context and select actions based on either motor or cognitive past behavior

386
387

Embodied bounded rationality

(Graybel, 2008, quoted in Nordli & Todd, Chapter 19 in this volume). This neural mechanism
allows the strategy selection of adaptive heuristics based on evaluation of past behavior.2
In conclusion, as the arguments in this chapter and previous references to strategy selection
aim to show, embodied bounded rationality is the new framework to study adaptive decision-
making processes.

Notes
1 This means that game theory overrides the theory of decision making and finds its equivalent in the
whole of social neuroscience, which should rely on its formal rigor to prevent the fragmentation of
neuroscientific research on the theory of the mind. This desire extends to the point of wishing that
neuroscience, while thinking about thought, turns to game theory as a means to formalize the “thinking
of the thought thinking the thought thinking …” This seems to be a wish of neoclassical and cognitivist
inspiration––which is actually untenable, given the limited computation capacities of the human brain
and particularly of its operative memory.
2 There are other examples of neural modeling of adaptive behavior. One is proposed by Bhui (in
press). It is well known that the way we feel about an outcome depends on how we compare it to our
past and present experiences. However, it is not so clear what kind of comparisons we make, or even
why we make comparisons at all. Bhui refers to theoretical neuroscience to explain why this context
dependence is adaptive. Our sense of value may adapt to our experiences for the same computational
reason that our eyes adapt to light and dark: our neurons have a limited capacity to process informa-
tion, which is best spent distinguishing among the stimulus values we expect to encounter in our local
environment. Bhui shows that influential psychological theories of context-sensitive judgment can be
derived from the neurocomputational principle of efficient coding. This unites conflicting cognitive,
behavioral, and neural findings spanning both perceptual and value-based judgment across multiple
species into a single neurobiologically-grounded framework.

References
Alós-Ferrer, C. (2018). A review essay on social neuroscience: Can research on the social brain and eco-
nomics inform each other? Journal of Economic Literature, 56(1), 234–64.
Anderson, M. L. (2010). Neural reuse: A fundamental organizational principle of the brain. Behavioral and
Brain Sciences, 33(4), 245–266.
Bhalla, M., & Proffitt, D. R. (1999). Visual–motor recalibration in geographical slant perception. Journal
of Experimental Psychology: Human Perception and Performance, 25(4), 1076–1096.
Bhui, R. (in press). Testing optimal timing in value-linked decision making. Computational Brain &
Behavior.
Brooks, R. A. (1991). Intelligence without representation. Artificial Intelligence, 47(1–3), 139–159.
Cacioppo, J. T., Visser, P. S., & Pickett, C. L. (Eds.) (2006). Social neuroscience: people thinking about thinking
people. Cambridge, MA: MIT Press.
Calvo P., & Gomila T. (Eds.) (2008). Handbook of cognitive science: An embodied approach, San Diego,
CA: Elsevier.
Caruana, F., & Borghi, A. M. (2016). Il cervello in azione: Introduzione alle nuove scienze della mente.
Bologne: Il Mulino.
Chartrand, T. L., & Bargh, J. A. (1999). The chameleon effect: The perception-behavior link and social
interaction. Journal of Personality and Social Psychology, 76(6), 893–910.
Clancey, W. (1997). Situated cognition: On human knowledge and computer representations. Cambridge: Cambridge
University Press.
Clark, A. (1997). Being there: Putting brain, body, and world together again. Cambridge, MA: MIT Press.
Clark, A. (2008). Pressing the flesh: A tension in the study of the embodied, embedded mind?. Philosophical
and Phenomenological Research, 76(1), 37–59.
Cohen Kadosh, R. C., & Walsh, V. (2009). Numerical representation in the parietal lobes: Abstract or not
abstract? Behavioral and Brain Sciences, 32(3–4), 313–328.
Craig, A. D. (2011) Interoceptive cortex in the posterior insula: Comment on Garcia-Larrea et al. 2010
Brain 133, 2528. Brain, 134(4).

387
388

Vittorio Gallese et al.

Cuccio, V., & Gallese, V. (2018). A Peircean account of concepts: Grounding abstraction in phylogeny
through a comparative neuroscientific perspective. Philosophical Transactions of the Royal Society of London
B, 373, 20170128.
Damasio A. (1994). Descartes' error: Emotion, reason, and the human brain. New York: Putnam Publishing.
Dawkins, R. (1976). The selfish gene. New York: Oxford University Press.
Dorfman, H. M., Bhui, R., Hughes, B. L., & Gershman, S. J. (in press). Causal inference about good and
bad outcomes. Psychological Science.
Ellis, R., & Tucker, M. (2000). Micro-affordance: The potentiation of components of action by seen
objects. British Journal of Psychology, 91(4), 451–471.
Gallagher, S. (2005). How the body shapes the mind. New York: Oxford University Press.
Gallagher, S. (2018a). Embodied rationality. In G. Bronner & F. Di Iorio (Eds.), The mystery of ration-
ality: Mind, beliefs and social science (pp. 83–94). Berlin: Springer.
Gallagher, S. (2018b). Decentering the brain: Embodied cognition and the critique of neurocentrism and
narrow-minded philosophy of mind. Constructionist Foundations, 14(1), 101–134.
Gallese, V. (2003). The manifold nature of interpersonal relations: The quest for a common mechanism.
Philosophical Transactions of the Royal Society of London B, 358, 517–528.
Gallese V. (2007). Before and below ‘theory of mind’: Embodied simulation and the neural correlates of
social cognition. Philosophical Transactions of the Royal Society of London B, 362, 659–669.
Gallese, V. (2008). Mirror neurons and the social nature of language: The neural exploitation hypothesis.
Social Neuroscience, 3, 317–333.
Gallese V. (2014). Bodily selves in relation: Embodied simulation as second-person perspective on inter-
subjectivity. Philosophical Transactions of the Royal Society of London B, 369, 20130177.
Gallese V., & Cuccio, V. (2015) The paradigmatic body: Embodied simulation, intersubjectivity and the
bodily self. In T. Metzinger & J. M. Windt (Eds.), Open MIND (pp. 1–23). Frankfurt: MIND Group.
Gallese, V., Fadiga, L., Fogassi, L., & Rizzolatti, G. (1996). Action recognition in the premotor cortex.
Brain, 119, 593–609.
Gallese, V., & Goldman, A. (1998). Mirror neurons and the simulation theory of mind-reading. Trends in
Cognitive Sciences, 2(12), 493–501.
Gallese, V., & Lakoff, G. (2005). The brain’s concepts: The role of the sensory-motor system in reason and
language. Cognitive Neuropsychology, 22, 455–479.
Gallese, V., Rochat, M., Cossu, G., & Sinigaglia, C. (2009). Motor cognition and its role in the phylogeny
and ontogeny of intentional understanding. Developmental Psychology, 45, 103–113.
Gibbs, R. W. Jr. (2006). Embodiment and cognitive science. New York: Cambridge University Press
Gibson, J. J. (1950). The perception of the visual world. Boston: Houghton Mifflin.
Gibson, J. J. (1979). The ecological approach to visual perception. Boston: Houghton Mifflin.
Gigerenzer, G. (1997). The modularity of social intelligence. In A. Whiten & R. W. Byrne, (Eds.),
Machiavellian intelligence II: Extensions and evaluations (pp. 264–288). Cambridge: Cambridge
University Press.
Gigerenzer, G. (2007). Gut feelings: The intelligence of the unconscious. London: Penguin.
Gigerenzer, G., & Gaissmaier, W. (2011). Heuristic decision making. Annual Review of Psychology, 62,
451–482.
Gigerenzer, G., & Goldstein, D. G. (1996). Reasoning the fast and frugal way: models of bounded ration-
ality. Psychological Review, 103(4), 650–669.
Gigerenzer, G., Todd, P. M., & the ABC Research Group (Eds.) (1999). Simple heuristics that make us smart.
New York: Oxford University Press.
Goldman, A. (2006). Simulating mind. New York: Oxford University Press.
Graybiel, A. M. (2008). Habits, rituals, and the evaluative brain. The Annual Review of Neuroscience, 31,
359–387.
Hamlin, R. P. (2017). “The gaze heuristic”: Biography of an adaptively rational decision process. Topics in
Cognitive Science, 9(2), 264–288.
Haugeland, J. (1978). The nature and plausibility of cognitivism. Behavioral and Brain Sciences, 1(2),
215–226.
Heyes, C. (2011). Automatic imitation. Psychological Bulletin, 137(3), 463–483.
Hirose, N. (2002). An ecological approach to embodiment and cognition. Cognitive Systems Research, 3(3),
289–299.
Iacoboni, M. (2008). Mirroring people: The new science of how we connect with others. New York: Farrar, Strauss
and Giroux.

388
389

Embodied bounded rationality

Iacoboni, M. (2009). Imitation, empathy, and mirror neurons. Annual Review of Psychology, 60, 653–670.
Jostmann, N. B., Lakens, D., & Schubert, T. W. (2009). Weight as an embodiment of importance.
Psychological Science, 20(9), 1169–1174.
Kandell, E. (2018). The disordered mind. New York: Farrar, Strauss and Giroux
Kiverstein, J., & Clark, A. (2009). Introduction: Mind embodied, embedded, enacted: One church or
many? Topoi, 28(1), 1–7.
Lakoff, G., & Johnson, M. (1999). Philosophy in the flesh: The embodied mind and its challenge to Western
thought. New York: Basic Books.
Lakoff, G., & Nuñez, R. (2000). Where mathematics comes from: How the embodied mind brings mathematics
into being. New York: Basic Books.
Legrenzi, P., & Umiltà, C. (2011). Neuromania: On the limits of brain science. New York: Oxford
University Press.
Longo, M. R., Kosobud, A., & Bertenthal, B. I. (2008). Automatic imitation of biomechanically pos-
sible and impossible actions: Effects of priming movements versus goals. Journal of Experimental
Psychology: Human Perception and Performance, 34(2), 489–501.
Marewski, J. N., & Schooler, L. J. (2011). Cognitive niches: An ecological model of strategy selection.
Psychological Review, 118, 393–437.
Marr, D. (1982). Vision. San Francisco, CA: Freeman.
Mastrogiorgio, A., & Petracca, E. (2015). Razionalità incarnata. Sistemi Intelligenti, 27(3), 481–504.
Mastrogiorgio, A., & Petracca, E. (2016). Embodying rationality. In L. Magnani & C. Casadio (Eds.),
Model-based reasoning in science and technology (pp. 219–237), Berlin: Springer.
Mastrogiorgio, A., & Petracca, E. (2018). Satisficing as an alternative to optimality and suboptimality in
perceptual decision-making. Behavioral and Brain Sciences, 41, e235.
Meier, B. P., Schnall, S., Schwarz, N., & Bargh, J. A. (2012). Embodiment in social psychology. Topics in
Cognitive Science, 4(4), 705–716.
Morewedge, C. K., Huh, Y. E., & Vosgerau, J. (2010). Thought for food: Imagined consumption reduces
actual consumption. Science, 330(6010), 1530–1533.
Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall.
Newell, A., & Simon, H.A. (1976). Computer science as empirical inquiry: Symbols and search.
Communications of the ACM, 19(3), 113–126.
Oullier, O., & Basso, F. (2010). Embodied economics: How bodily information shapes the social coordin-
ation dynamics of decision-making. Philosophical Transactions of the Royal Society B, 365(1538), 291–301.
Patokorpi, E. (2008). Simon’s paradox: Bounded rationality and the computer metaphor of the mind.
Human Systems Management, 27(4), 285–294.
Petracca, E. (2017). A cognition paradigm clash: Simon, situated cognition and the interpretation of
bounded rationality. Journal of Economic Methodology, 24(1), 20–40.
Pfeifer, R., & Bongard, J. (2006). How the body shapes the way we think: A new view of intelligence. Cambridge,
MA: MIT Press.
Pfeiffer, J., Iacoboni, M., Mazziotta, J. C., & Dapretto, M. (2008). Mirroring other emotions relates to
empathy and interpersonal competence in children, Neuroimage, 39, 2076–2085.
Premack, D., & Woodruff, G. (1978). Does the chimpanzee have a theory of mind? The Behavioural and
Brain Sciences, 1, 515–526.
Reimann, M., & Bechara, A. (2010). The somatic marker framework as a neurological theory of decision-
making: Review, conceptual comparisons, and future neuroeconomics research. Journal of Economic
Psychology, 31(5), 767–776.
Rieskamp, J., & Otto, P. E. (2006). SSL: A theory of how people learn to select strategies. Journal of
Experimental Psychology General, 135(2), 207–236.
Rizzolatti, G., Fogassi, L., & Gallese, V. (2001). Neurophysiological mechanisms underlying the
understanding and imitation in action. Nature Review Neuroscience, 2, 661–670.
Rowlands, M. (1999). The body in mind: Understanding cognitive processes. Cambridge: Cambridge
University Press.
Schutt, R. K., Seidman, L. J., & Keshavan, M. S. (Eds.) (2015). Social neuroscience: Brain, mind, and society.
Cambridge, MA: Harvard University Press.
Shapiro, L. (2004). The mind incarnate. Cambridge, MA: MIT Press.
Shapiro, L. (Ed.) (2014). The Routledge handbook of embodied cognition. Abingdon: Routledge.
Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics, 69(1),
99–118.

389
390

Vittorio Gallese et al.

Simon, H. A. (1956). Rational choice and the structure of the environment. Psychological Review, 63(2),
129–138.
Simon, H. A. (1976). From substantive to procedural rationality. In S. J. Latsis (Ed.), Method and appraisal
in economics. Cambridge: Cambridge University Press.
Simon, H. A. (1978). Rationality as process and product of thought. American Economic Review, 68(2), 1–15.
Spellman, B., and Schnall, S. (2009). Embodied rationality. Queen’s Law Journal, 35(1), 117–164.
Tipper, S. P., Paul, M. A., & Hayes, A. E. (2006). Vision-for-action: The effects of object property dis-
crimination and action state on affordance compatibility. Psychonomic Bulletin & Review, 13, 493–498.
Varela, F., Thompson, E., & Rosch, E. (1991). The embodied mind: Cognitive science and human experience.
Cambridge, MA: MIT Press.
Viale R. (2011). Brain reading social action. International Journal of Economics, 58(4), 337–58.
Viale, R. (2012). Methodological cognitivism. Vol. 1: Mind, rationality and society. Heidelberg: Springer.
Viale, R. (2017). Corpo e razionalità. In A. Coricelli & D. Martelli (Eds.), Neurofinanza. Le basi neuronali
delle scelte finanziarie. Milan: Egea.
Viale, R. (2019) La razionalita limitata ‘embodied’ alla base del cervello sociale ed economic. Sistemi
Intelligenti, XXXI, 1.
Wallace, B., Ross, A., Davies, J. B., & Anderson T. (Eds.) (2007) The mind, the body and the world: Psychology
after cognitivism. London: Imprint Academic.
Werner K., & Raab M. (2014). Moving your eyes to solution: Effects of movements on the perception of
a problem-solving task, Quarterly Journal of Experimental Psychology, 67(8), 1571–1578.
Williams, L. E., Huang, J. Y., & Bargh, J. A. (2009). The scaffolded mind: Higher mental processes
are grounded in early experience of the physical world. European Journal of Social Psychology, 39(7),
1257–1267.
Wilson, M. (2002). Six views of embodied cognition. Psychonomic Bulletin and Review, 9, 625–636.
Yin, H. H., & Knowlton, B. J. (2006). The role of the basal ganglia in habit formation. Nature Reviews
Neuroscience, 7(6), 464–476.
Zhong, C. B., & Leonardelli, G. J. (2008). Cold and lonely: Does social exclusion literally feel cold?
Psychological Science, 19(9), 838–842.

390
391

24
EXTENDING THE BOUNDED
RATIONALITY FRAMEWORK
Bounded-resource models in biology

Christopher Cherniak

Perhaps the most fundamental psychological law is that we are finite organisms (Cherniak,
1986). Bounded-resource models of the agent characterize our rationality as falling in the
“Goldilocks” range, between nothing and perfection. This is not a mere quirk: The apercu
motivating the rationality critiques is conveyed by the recognition that standard idealizations
entail some deductive omniscience, for instance, triviality of portions of the deductive sciences.
Such ideal agent/logicians, if computational, would violate Church’s Theorem on undecid-
ability of first-order logic (Church, 1936) (see Figure 24.1). This insight in turn indicates
that NP-completeness (Garey & Johnson, 1979) is of corresponding interest: core compu-
tational intractability of a cosmos-consuming scale is a practical counterpart to traditional
absolute uncomputability––another layer of impossibility between the idealizations and reality
(Cherniak, 1984). This is some of the philosophical significance of computational complexity
(Aaronson, 2013; Dean: 2016). The formalism of scarcity of interconnections is combinatorial
network optimization theory, which emerged from the field of microcircuit design, and itself
includes many important intractable problems.
The research program here stems from a holistic rather than compartmentalized perspec-
tive: Philosophy and science are distinct but inextricably interconnected. For instance, the
classical paradoxes of semantics (e.g., the Liar Paradox (Tarski, [1935] 1983) and set theory
(e.g., Russell’s Paradox (Russell, 1903)) can be reexamined not as mere odd pathology, but
instead as deep indications of the use of “quick and dirty heuristics”––that is, the ultimate
speed-reliability tradeoffs of correctness and completeness for feasibility. Three disparate fields
thereby converge: (1) computational complexity theory; (2) empirical psychology of quick and
dirty heuristics; and (3) philosophical theory of bounded-resource rationality. In this way, the
bounded rationality models provide a foundation of current “behavioral economics” (Nobel
Prize Committee, 2017).

Brainwiring optimization
In addition, it is natural to explore extending the bounded-resource approach in this way down
from rationality to the brainwiring hardware level. Specifically, long-range connectivity is a
critically constrained neural resource, with great evolutionary pressure to employ efficiently.

391
392

Christopher Cherniak

|–(φ→ψ)?

“Yes” “No”
Halt Halt

Figure 24.1 Standard rationality idealizations of the agent require a deeply non-possible object: an Ideal
Logician. If its logical competence is represented in a finite algorithm, then it must violate Church’s
Theorem (1936): There is no “Yes/No” decision procedure for first-order predicate calculus. A proof
procedure exists––the left-hand half (“Yes” semi-algorithm) of the full algorithm––but that cannot
suffice for perfect rationality. Nor can any quick and dirty heuristic procedure.

Hence, “brain as microchip”: Connection minimization appears to be a first law of brain

tractography, an organizing principle driving neuroanatomy (Cherniak: 1994a), with hitherto
unreported precision.
Our laboratory has therefore been assessing how well wiring-optimization concepts from
computer microcircuit engineering apply to brain structure. “Save wire” turns out to be a
strongly predictive “best in a billion” model. Wiring minimization can be perceived at multiple
levels, e.g., placement of the entire brain, layout of its ganglia and/or cortex areas, subcellular
architecture of dendrite arbors, etc. Much of this biological structure appears to arise “for free,”
directly from physics.
Of course, if connectivity in a nervous system were unbounded or cost-free, there would be
no pressure for its economical deployment. However, much of higher central nervous systems
operates at signal propagation velocities not at light speed, but below the 60 mph speed limit.
Since connectivity is in limited supply, network optimization is quite valuable. As a result, this
scarce wiring is highly optimized, a prima facie paradoxical consequence.
For instance, a key specific wiring problem is component placement optimization: For a
set of interconnected components, what are the positionings of the components that min-
imize total interconnection costs (e.g., wire length)? Surprisingly, this idea seems to account
very precisely for aspects of neuroanatomy at multiple hierarchical levels. For example, the
nervous system of the nematode worm Caenorhabditis elegans (Wood, 1988) includes 11 gangli-
onic components, which have 11! (~40,000,000) alternative possible anteroposterior orderings.
In fact, “the actual is the ideal” here, in that the actual layout happens to require the min-
imum total possible wire length, a predictive success story (Cherniak, 1994b). However, such
problems are NP-complete: exact solutions generally appear to entail brute-force searches, with
exponentially exploding costs. Despite local minima traps, such neuroanatomy optimization is
approximated well by “mesh of springs” energy-minimization mechanisms (Cherniak et al.,
2002). (See discussion below of the “genomic bottleneck”.)

392
393

Bounded-resource models in biology

Figure 24.2 “Size Law” trend for macaque visual cortex areas––increasing optimality with increased
subsystem size. The system of components is 17 contiguous central visual areas of macaque cortex.
A layout is scored in terms of its violations of the Adjacency Rule. A series of nested compact subsets
of the set of visual areas is generated; each subset is compared with all possible alternative layouts of
that subset for adjacency-rule optimality. As subset size increases, optimality-ranking of the actual
layout consistently improves. Layout optimality rank for the complete visual system analyzed is in top
0.000000119 of all possible alternative layouts of the full 17 area set. For comparison, the corresponding
analysis for a layout of the 17 visual areas with their adjacencies randomly shuffled shows no trend
toward improving optimality. This analysis includes only 17 of the total 73 areas of macaque cortex
(Cherniak et al., 2004).

A corresponding approach can be explored for the placement of the interconnected func-
tional areas of the cerebral cortex. (For an overview of the anatomy of the human cortex
white-matter connections, see Carpenter and Sutin, 1983.) A first strategy is to use a sim-
pler connection cost measure, conformance of a cortex layout to a wire-saving Adjacency
Heuristic: “If components are connected, then they are placed adjacent to each other.”
Extensive sampling of all possible layouts is still required to verify the best ones. For example,
for 17 core visual areas of macaque cortex, the actual layout of this subsystem ranks in the top
10−6 layouts best minimizing their adjacency costing (Cherniak et al., 2004). Similar high opti-
mality rankings also hold for the core set of visual areas of cat cortex, and also for rat olfactory
cortex and amygdala (see Figure 24.2).
Furthermore, a “Size Law” appears to apply to systems with such local-global tradeoffs: “If
a complete system is in fact perfectly optimized, then the smaller the subset of it evaluated in
isolation, the poorer the optimization tends to appear.” (This seems to be a generalization of
a related perspective-dependent idea in theology, to account for the problem of evil (Tooley,
2015)––of apparent local imperfections in the universe with an omnipotent, benevolent deity.)
The Size Law applies well to each of the above brain systems, as well as elsewhere (e.g., for
microchip design).
We have also reported similar Size Law optimization results for C. elegans ganglia, rat amyg-
dala and olfactory cortex, cat visual cortex, and combined cat visual auditory and somatosen-
sory cortex. The human brain appears to be the most complex physical structure presently
known in the universe, an ultimate big dataset. With such a network optimization framework,
these “Save wire” results have been extended and replicated for the (1) complete (2) living and
(3) human cerebrum via fMRI, another predictive success (Lewis et al., 2012)––a “Mount
Everest” of neural optimization.

393
394

Christopher Cherniak

Figure 24.3 Actual versus optimal arbors: mouse thalamus extrinsic axon, ascending reticular
formation. The arbor best fits a volume-minimizing model. (A) Wireframe representation of 8-terminal
subtree of observed arbor. The actual tree, with actual topology in its actual embedding, appears in
broken lines. Optimal embedding of the actual topology with respect to volume minimization is
superimposed in solid lines. The cost in volume of the actual arbor exceeds that of the optimized
embedding of its topology by 2.20 percent. (B) “Best of all possible topologies” connecting the given
terminal loci: the optimal topology with respect to volume, optimally embedded. The volume cost of
the actual arbor exceeds that of the optimal topology by 2.47 percent. Only 10 of the 10,395 possible
alternative topologies here have lower total volume costs, when optimally embedded, than the actual
topology (Cherniak et al., 1999).

In addition, neuron dendrite and axon arbors also appear significantly to approximate a
generalization of minimum-cost Steiner trees (Cherniak et al., 1999). These optimal structures
are derivable via simple fluid dynamics. (“Instant arbors, just add water.”) Evolution gets a free
ride from the physics. This seems to constitute some of the most complex biostructure pres-
ently derivable from simple (non-DNA) physical processes (see Figure 24.3). Such a “Physics
suffices” account of biological morphogenesis constitutes “Nongenomic Nativism”––innateness
without DNA: The tabula rasa is pre-formatted. One rationale for the pre-biotic pervading the
biotic in this way is to cope with the “genomic bottleneck”: Like other organism systems, the
genome has limited capacities. The more neuroanatomy for free, directly from physics, the less
the genome information-carrying load.
A basic question arises: Neural wiring minimization is of course of value, but why should
it seem to have such a high priority––sometimes apparently near-maximal? The significance of
ultra-fine neural optimization remains an open question.
An additional methodological issue concerns how the intentional level of mind meshes
with the hardware level of the brain. Prima facie, that relationship appears in tension: In some
aspects, the brainwiring appears virtually perfectly optimized, yet the rationality has layers of
impossibility between it and perfection (Cherniak, 2009). Perhaps this is another manifestation
of irreducibility of the mental––the familiar loose fit of each domain with the other.

Genome as nanobrain
A next section of this research program: Similar computation theory concepts can be used for
understanding the structure and function of organism DNA. The Crick-Watson double-helix
model (Alberts et al., 2015) emerged at the same place and time as Alan Turing's final work, on
morphogenesis, namely, Cambridge around 1950. So an idea of DNA-as-Turing-machine-tape
has circulated for decades.

394
395

Bounded-resource models in biology

In particular, the genome itself can be treated like a “nano-brain” or pico-computer to see
whether similar connection minimization strategies also appear in gene networks. As outlined
above, for decades we have reported wiring optimization in the brain that begins to approach
some of the most precisely confirmed predictions in neuroscience.
Two meta-models of the genome compete today: One is a “genome as hairball” idea, of
the very vehicle of innateness itself effectively possessing minimal structure. (For example, our
genome is a mess. It is “in an alarming state of disarray” (Alberts et al., 2015).) The other pic-
ture, examined here, is structuralist––that the genome itself has large-scale global patterns.
We are now exploring a connection-minimization model for the human genome.
Information transmission may not be cost-free even within a cell, nucleus, or genome. For
instance, genes strongly expressed in particular tissues are not just randomly distributed in
the genome. Rather, the arrangement of such tissue-specific gene positions in the complete
chromosome set mirrors the antero-posterior, and dorso-ventral, configuration of the tissue-
locations in the body. A statistically significant supra-chromosomal “genome homunculus”––a
global, multi-dimensional, somatotopic mapping of the human body––appears to extend across
chromosome territories in the entire sperm cell nucleus (Cherniak & Rodriguez-Esteban,
2013) (see Figure 24.4). Such a mapping is a strategy for connection cost-minimization (e.g.,
cf. body maps reported in sensory and motor cortex since the nineteenth century). Also,
corresponding finer-scale somatotopic mappings seem to occur on each individual autosomal
chromosome (Cherniak & Rodriguez-Esteban, 2015).
Furthermore, the organelle sub-structure of the typical individual eukaryotic animal cell
also turns out to map similarly as a “cellunculus” onto the total genome, via organelle-specific
genes that express more strongly in particular organelle types (Cherniak & Rodriguez-Esteban,
2018) (see Figure 24.5). So, genome as palimpsest: multiple maps, at different scales, seem
superimposed upon the genome.

500 Pancreas
Ovary
400
~Tail
Nucleus: Tissue genes

300
head-tail gradient

Liver
Kidney
200
Testis
100
Heart Spleen
0
0 Brain 20 40 60 80 100
~Head

-100 Thymus

-200 Anterior Posterior

Body: Approximate antero-posterior locus of tissue

Figure 24.4 Antero-posterior “gradient of gradients” in nucleus. Tissue location in human body
correlates significantly with pattern of tissue genes’ positions in cell nucleus. (For 9 datapoints each
weighted by their own significance, r 2 = 0.62; p < 0.01, two-tailed.) That is, tissue location-in-body
relates to its genes’ distribution-gradient in the complete genome. The more forward-placed a tissue
in the body, the more forward-placed its genes on chromosomes in nucleus. The head of the genome
homunculus is at the head of the sperm cell nucleus. A corresponding body-genome mapping also holds
for the dorso-ventral body axis (Cherniak & Rodriguez-Esteban, 2013). In addition, the body similarly
maps onto individual chromosomes (Cherniak and Rodriguez-Esteban, 2015).

395
396

Christopher Cherniak

-0.3
Genome: Slope of Gene-Group PlasMem

Peripheral
Distribuon on CP Axis Nucleolus
-0.2

NucBod EndoRet
-0.1
Mitoch
0 2 Nucleolus 4 NucSpec 6 GolgiApp 8 10
0
Central

NucMem
0.1 NucFibCt
Nucleus
0.2
Central Peripheral
Cell: Radial Posion of Organelles on CP Axis

Figure 24.5 Isomorphism of cell microanatomy and large-scale human genome structure: Components
positioned more centrally in a cell tend to have their genes correspondingly concentrated on
chromosomes sited more toward the center of genome. In a plot of 10 organelles, this cell-genome
correlation is significant (r2 = 0.540; p < 0.015, two-tailed). Each of the datapoints is labeled with its
organelle-name (Cherniak & Rodriguez-Esteban, 2018).

Bounded Rat

Intractability Q&D Heuriscs

Network
Opmizaon

Neuro Anat Genome Anat

Steiner Tree Comp Plac Opt Body Map Cell Map ...

Dendrites, Axons Worm, Rat, Cat, A-P & D-V

Mac, H.sap., . . .
Total Genome & Individual Chromos

Figure 24.6 Bounded-resource models in biology: Concept map. For explanation of terms, refer to
preceding text.

Conclusion: The argument of this chapter starts with a synoptic view, from a “knowledge
without borders” perspective. The chapter aimed to extend the bounded rationality framework,
sketching some of the terrain of more realistic limited-resource models in the life sciences.
—Limits for growth: of mind, brain, genome, and other biological systems. Figure 24.6
summarizes the derivation of the research program. Further “Bounded Biology” results of this
type now also have support.

396
397

Bounded-resource models in biology

References
Aaronson, S. 2013. Why philosophers should care about computational complexity. In J. Copeland,
C., Posy, & O. Shagrir (Eds.), Computability: Turing, Gödel, Church, & Beyond (pp. 261–327). Cambridge,
MA: MIT Press.
Alberts, B., Johnson, A., Lewis J., Raff, M., Roberts, K., & Walter, P. 2015. Molecular biology of the cell (6th
ed.). New York: Garland Science, Chapter 4.
Carpenter M., & Sutin, J. 1983. Human neuroanatomy (8th ed.). Baltimore, MD: Williams & Wilkins,
Chapters 2, 19.
Cherniak, C. 1984. Computational complexity and the universal acceptance of logic. Journal of Philosophy,
81, 739–758.
Cherniak, C. 1986. Minimal rationality. Cambridge, MA: MIT Press.
Cherniak, C. 1994a. Philosophy and computational neuroanatomy. Philosophical Studies, 73, 89–107.
Cherniak, C. 1994b. Component placement optimization in the brain, Journal of Neuroscience, 14,
2418–2427.
Cherniak, C. 2009. Minimal rationality and optimal brain wiring. In C. Glymour, W. Wei, &
D. Westerstahl (Eds.), Logic, methodology and philosophy of science: Proceedings of 13th International
Congress (pp. 443–454). London: College Publications.
Cherniak, C., Changizi, M., & Kang, D. 1999. Large-scale optimization of neuron arbors. Physical Review
E, 59, 6001–6009.
Cherniak, C., Mokhtarzada Z., & Nodelman, U. 2002. Optimal-wiring models of neuroanatomy. In
G. Ascoli (Ed.), Computational neuroanatomy: Principles and methods (pp. 71–82). Berlin: Springer.
Cherniak, C., Mokhtarzada, Z., Rodriguez-Esteban, R., & Changizi, B. 2004. Global optimization of
cerebral cortex layout. Proceedings of the National Academy of Sciences of the U.S.A., 101, 1081–1086.
Cherniak C., & Rodriguez-Esteban, R. 2013. Body maps on the human genome. Molecular Cytogenetics,
6(1), 61.
Cherniak C., & Rodriguez-Esteban, R. 2015. Body maps on human chromosomes. Tech Report,
UMIACS-TR-2015–04. College Park, MD: University of Maryland Institute for Advanced Computer
Studies.
Cherniak C., & Rodriguez-Esteban, R. 2018. Cell maps on the human genome. Tech Report, 2018
UMIACS-TR-2018–01. College Park, MD: University of Maryland Institute for Advanced Computer
Studies,
Church, A. 1936. An unsolvable problem of elementary number theory. American Journal of Mathematics,
58 (2), 345–363.
Dean, S. 2016. Computational complexity theory. In E. Zalta (Ed.), Stanford encyclopedia of philosophy
(pp. 261–327). Available at: http://plato.stanford.edu
Garey, M., & Johnson, D. 1979. Computers and intractability: A guide to the theory of NP-Completeness. San
Francisco: W. H. Freeman.
Lewis, S., Christova, P., Jerde, T., & Georgopoulos, A. 2012. A compact and realistic cerebral cortical
layout derived from prewhitened resting-state fMRI time series: Cherniak’s adjacency rule, size law,
and metamodule grouping upheld. Frontiers in Neuroanatomy, 36.
Nobel Prize Committee 2017. The Prize in Economic Sciences 2017: Scientific background: Richard
H. Thaler: integrating economics with psychology. Stockholm. Available at: Nobelprize.org.
Russell, B. 1903. Principles of mathematics. Cambridge: Cambridge University Press.
Tarski, A. [1935] 1983. The concept of truth in formalized languages,. In J. Corcoran (Ed.), Logic, semantics,
metamathematics (2nd ed.), Indianapolis: Hackett.
Tooley, M. 2015. The problem of evil. In E. Zalta (Ed.), Stanford encyclopedia of philosophy. Available
at: https://plato.stanford.edu/archives/fall2015/entries/evil/
Wood, W. (Ed.) 1988. The Nematode Caenorhabditis elegans. Plainview, NY: Cold Spring Harbor
Laboratory.

397
398

25
HOW RATIONALITY IS
BOUNDED BY THE BRAIN
Paul Thagard

When Richard Thaler was asked how he planned to spend the million dollars from his 2017
Nobel Prize in economics, he replied: “I will try to spend it as irrationally as possible” (Politi,
2017). Thaler’s prize was awarded for decades of research showing that people are not nearly as
rational as economists have assumed.
Why are people frequently irrational? There are still many economists and philosophers
who believe that people are fundamentally rational, but a large accumulation of evidence from
psychology and behavioral economics shows that people often fall short of rational standards.
The evidence is descriptive, showing that people make thinking errors in systematic ways. But
these descriptions leave open the question of why people do not think in ways that support their
long-term interests.
Cognitive science usually explains thinking in terms of mental representations and processes,
but advances in neuroscience make it increasingly feasible to explain many mental processes as
brain mechanisms. So the question becomes: What are the brain mechanisms that lead people
to be irrational?
The term “bounded rationality” originated in the 1950s in the writings of Herbert Simon
(e.g., Simon, 1972), but appreciation that people are often limited in their rationality goes back
to Aristotle and Francis Bacon. The human brain is marvelous in many of its accomplishments,
but I will describe its inherent limitations in size, speed, and cognitive-emotional functioning.
Herbert Simon (2000, p. 25) characterized bounded rationality as follows:

Bounded rationality is simply the idea that the choices people make are determined
not only by some consistent overall goal and the properties of the external world,
but also by the knowledge that decision makers do and don't have of the world,
their ability or inability to evoke that knowledge when it is relevant, to work out
the consequences of their actions, to conjure up possible courses of action, to cope
with uncertainty (including uncertainty deriving from the possible responses of other
actors), and to adjudicate among their many competing wants. Rationality is bounded
because these abilities are severely limited.

I show how these mental limitations derive in part from brain limitations.

398
399

How rationality is bounded by the brain

Table 25.1 3-analysis of rational

Exemplars Logical deduction, probabilistic reasoning and statistical inference, making

decisions based on maximizing expected utility.
Typical features Following rules of reason, careful, conscious.
Explanations Explains: why people get true beliefs; why people make good decisions.
Explained by: ability of humans to follow appropriate rules of reasoning.

Table 25.2 3-analysis of irrational

Exemplars Fallacious deductive inferences such as affirming the consequent; defective

probabilistic inferences such as judging a conjunction to be more probable
than either of its conjuncts; making bad decisions such as emphasizing sunk
costs rather than future expectations.
Typical features Violation of normative rules, careless succumbing to fallacies and biases.
Explanations Explains: why people arrive at dumb beliefs and make bad decisions.
Explained by: susceptibility to psychological processes that interfere with the
application of good rules.

What is rationality?
The question “what is rationality?” sounds like a request for a definition, but decades of work
on the psychology of concepts show that concepts outside of mathematics are rarely suscep-
tible to clear definitions in terms of necessary and sufficient conditions. Rather, concepts are
better characterized in terms of the exemplars that provide standard examples of them, typical
conditions for the concept, and explanations that the concepts can be used to provide (Murphy,
2002). A new neural theory of concepts shows how to combine all three of these aspects of
concepts in a unified neural model (Blouw, Solodkin, Thagard, & Eliasmith, 2016).
Accordingly, we can characterize concepts by identifying these three dimensions of
exemplars, typical features, and explanations, a method of conceptual analysis that Thagard
(2019a) calls “3-analysis.” Table 25.1 provides a 3-analysis of the concept rational.
According to currently dominant traditions, standard examples of rationality include using
formal logic to deduce truths from truths, using probability theory to perform inductive reasoning,
and using maximization of expected utility to decide what to do by combining probabilities and
utilities to choose actions. The typical features of being rational include carefully and consciously
applying such normative rules. Such applications explain why people sometimes succeed in
acquiring true beliefs and making good decisions. That people are rational is explained by their
ability to follow rules. This concept of rationality is complemented by the prevalent concept of
irrationality, which can also be captured by a 3-analysis, as shown in Table 25.2.

The recognition of irrationality

Formal logic begin with Aristotle’s doctrine of the syllogism, but Aristotle also recognized
that people often fall short of good syllogistic reasoning. His Sophistical Refutations describes

399
400

Paul Thagard

numerous fallacies that people are prone to commit, such as equivocating on the meanings of
ambiguous words (Aristotle, 1984). The study of fallacious reasoning has a long history in phil-
osophy and survives today in the discipline of informal logic (Hanson, 2015).
In the seventeenth century, Francis Bacon (1960) provided a sophisticated discussion of
inductive reasoning in Novum Organon. In addition to giving good advice about how to go
from observations to generalizations, Bacon generated a list of mistakes that people often make
in inductive reasoning. He called them “idols,” which he colorfully described as idols of the
tribe (due to human nature), idols of the cave (due to what a particular human cares about),
idols of the marketplace (due to communication), and idols of the theater (due to philosoph-
ical prejudices). Bacon’s idols capture some of the errors recognized by twentieth-century
psychologists, such as the availability heuristic and motivated inference.
In the 1950s, Herbert Simon recognized the limited extent of human rationality, but he
did not systematically investigate the ways in which people fall short of good decision making
and inductive inference. Beginning in the 1970s, psychologists such as Daniel Kahneman and
Amos Tversky and economists, such as Richard Thaler, conducted experiments that identify
many ways in which human thinking falls short of good reasoning (Kahneman & Tversky,
2000; Thaler, 2015).
I use the term “error tendencies” to cover all the fallacies, idols, heuristics, biases, and typ-
ical mistakes that have been identified by philosophers, psychologists, and economists, and have
compiled more than 50 of them (Thagard, 2011). Why are there so many? Why did evolution
by natural selection fail to optimize people’s ability to reason well about what to believe and
what to do?
The biological answer is that brains have evolved to be only somewhat effective at
performing deductive, inductive, and practical inferences. Optimization is constrained by bio-
logical mechanisms and the difficulties of survival and reproduction in changing environments.
I will describe how these constraints have produced brain mechanisms that often work well
but are limited by size, speed, cognitive-emotional functions, and limitations of attention and
consciousness.

Brain size and speed

The argument that people must be rational because of optimization through natural selection
has two flaws. First, it gets evolution wrong, because natural selection does not optimize (Gould
& Lewontin, 1979). Rather, nature selects for organisms that are somewhat better at surviving
and reproducing than organisms with different genes.
Second, the optimization argument forgets that the current standards of rationality are rela-
tively recent cultural innovations. Humans have been around for at least 100,000 years, but
formal logic only began with Aristotle around 2,500 years ago, and sophisticated understanding
of deduction only began with the work of Gottlob Frege and Charles Peirce in the late nine-
teenth century. Probability and utility theories are products of mathematical thinking that began
in the seventeenth and eighteenth centuries, and the elegant version now used by economists
was only developed in the 1940s. For some purposes, these tools are useful, but there is no
reason to suppose that they are built into the human brain by evolution.
Human brains have numerous strengths that have enabled people to spread all over the planet
and increase in population to more than 7 billion. The most impressive features of human brains
are not the special-purpose adaptations touted by evolutionary psychologists, but rather the
general adaptability furnished by the flexible ways in which humans can learn from experience
(Quartz & Sejnowski, 1997).

400
401

How rationality is bounded by the brain

Nevertheless, the brain has numerous limitations that forestall optimal rationality. Our
brain’s assemblage of 86 billion neurons provides a lot of computing power, but elephants have
three times as many. In order to have more neurons, people would need to have bigger brains
that require bigger heads, but childbirth is already often a difficult procedure. Human brain size
reflects a trade-off between the benefits of more processing power and ease of delivery through
a pelvis that also must function for bipedal locomotion.
Another constraint on human brain size concerns energy. Even though the roughly 1.4 kg of
the human brain take up less than 3 percent of the average human weight, the brain uses up to
20 percent of the energy available to the body. Larger brains would require more energy, which
either requires less energy available for other functions such as metabolism and reproduction,
or greater sources of food. The evolution of human brains requires a trade-off between size and
energy efficiency, as well as between size and birth delivery.
These limitations on the size of human brains place important constraints on rationality
because there is a limited amount of information that people can store. Some philosophers
have maintained that it is rational to believe all the logical consequences of one’s beliefs. An
infinite set cannot be stored in any human brain or even in all the computers run by Amazon
and Google.
Moreover, even for a finite number of beliefs, the size limitation of the brain places sharp
constraints on the combinations of beliefs that can be considered. Educated people have a
vocabulary of around 30,000 words, so with 10 beliefs for each word they would have 300,000
beliefs. Understanding is growing of how such beliefs can be stored in the human brain through
distributed representations (Eliasmith, 2013). But brains cannot accommodate belief revisions
that require consideration of combinations of 2300,000 subsets of these beliefs. Such subsets are
required for considering whether human belief systems are consistent. So it is unreasonable to
expect that people should be consistent in their beliefs.
Human brains also come with limitations in speed of processing. Billions of neurons allow
for massively parallel operation, but the neurons themselves are slow. A typical neuron fires up
to 200 times per second, whereas current computers have operations at the rate of trillions of
times per second.
Why are neurons so slow? Most neural connections are chemical, requiring the movement of
neurotransmitters such as glutamate and gamma aminobutyric acid (GABA) from one neuron to
another. This chemical transmission is slower than purely electrical signaling, which occurs rarely
in brains, but it allows for flexibility in timing and the development of different kinds of pathways.
In principle, the brain could operate with only two neurotransmitters, one for enabling one
neuron to excite another (increasing its rate of firing), and another for enabling one neuron to
inhibit another (decreasing its rate of firing). Glutamate and GABA play these excitatory and
inhibitory roles, respectively. But there also dozens of neurotransmitters that operate in human
brains, carrying out diverse functions at different time scales. For example, circuitry involving
the neurotransmitter dopamine is important for motor control and learning about rewards,
while circuitry involving serotonin influences perception and emotion.
If brains were faster, they still would not be able to do an infinite amount of processing,
but they would be able to better approximate some of the standards required for the rational
norms of deductive logic and probability and utility theory. The brain lacks the speed to be
able to do all of the calculations that are required for absolute standards of rationality. I estimate
that speed and size limitations contribute to more than 30 of the 53 error tendencies listed in
Thagard (2011). For example, people would be less prone to representativeness (the tendency
to use assessments of similarity in causal reasoning) if they had the cognitive capacity to do fuller
statistical calculations.

401
402

Paul Thagard

On the other hand, brains are appropriately efficient at carrying out computations that are
important for the survival and reproduction of organisms. Perception, inference, and decision-
making can all be modeled as processes of parallel constraint satisfaction, in which a brain or
computer considers a range of possible interpretations of a conflict situation and comes up with
a good but not necessarily optimal solution (Rumelhart & McClelland, 1986; Thagard, 2019a).
For example, recognizing a moving object as an instance of prey or predator should consider
alternative hypotheses about the animal, constrained by perceptual and environmental informa-
tion. Parallel constraint satisfaction is efficiently computed by neural networks that implement
constraints by excitatory and inhibitory links (Thagard, 2000).
Therefore, the brain can be understood as an engine of coherence rather than deduction or
calculation. Coherence requires satisfying multiple constraints in parallel, not making zillions
of calculations. In general, coherence is computationally intractable (Thagard and Verbeurgt,
1998), but in practice is efficient as long as the number of inhibitory connections is small
compared to the number of excitatory ones, as is true of the brain (van Rooij, Blokpoel,
Kwishout, & Wareham, 2018).
In sum, human rationality is bounded by the limited size and speed of the brain, but it can
often function well by making coherence judgments rather than overextending itself by making
too many deductions and calculations. In exciting work in progress, Dan Simon and Stephen
Read show that a large number of cognitive biases can be explained by thinking of the mind
as a coherence process. My aim is less specific, to show why the alternative methods of rational
calculation are at odds with the size and speed of the brain.

Brain integration of cognition and emotion

Rationality is also bounded by more specific aspects of how brains function, including the
integration of cognition and emotion and the limited role of consciousness. An obsolete view
of the brain takes it as combining a recent cognitive system consisting of areas such as the pre-
frontal cortex built on top of an ancient limbic system consisting of primitive areas such as
the amygdala. This view was exploded by findings that the most high-level areas are intensely
interconnected with the emotional systems (Damasio, 1994; Pessoa, 2013).
The integration of cognition and emotion in the brain is generally a feature rather than a
bug. Accounts of deduction and probability assume that the brain is largely a syntactic engine,
with semantics (meaning) and pragmatics (context and purpose) only playing peripheral roles.
But syntax alone cannot explain numerous aspects of human thinking, such as performance
in the selection task of Wason (1966) and the complexity of analogical inference (Holyoak &
Thagard, 1995). Standard computer programs are fabulous at rapid syntactic manipulations, but
are much less successful in the semantics of connections to the world and the pragmatics of
accomplishing important goals in particular contexts.
In contrast, neural representations mingle syntax and semantics by means of representations
that can simultaneously handle relational structure and connections to the world. The best
current account of this mingling is Chris Eliasmith’s (2013) Semantic Pointer Architecture,
which shows how populations of neurons can retain information gained from the world by sen-
sory and motor operations, but also build up syntactically-rich structures. It thereby provides a
biologically plausible synthesis of syntax and semantics.
This architecture extends to explain emotions as brain representations that integrate infor-
mation drawn from (1) physiological changes in the body; (2) cognitive appraisals in the brain;
and (3) contextual information including the use of language in humans (Thagard & Schröder,
2014; Thagard 2019a, 2019b; Kajić, Schröder, Stewart, & Thagard, 2019). Emotions that

402
403

How rationality is bounded by the brain

pervade cognition show how goals and purposes can fundamentally influence the operation of
the neural system, making it a pragmatic as well as a syntactic and semantic addition. Hence
adding emotions to the Semantic Pointer Architecture shows how brains accomplish a synthesis
of syntax, semantics, and pragmatics. This integration helps to ensure that human thinking is
not just idle deduction or calculation, but operates effectively in the world and accomplishes
human goals and purposes.
Nevertheless, the evolutionary feature of integration of cognition and emotion comes
with bugs. In the brain, cognitions and emotions are not independent of each other, unlike
the calculations in expected utility theory where probabilities and utilities are distinct before
being mathematically combined. Under different circumstances, the brain might have evolved
with separate modules for probability calculations, utility calculations, and their integration in
calculations of expected utility. But these are cultural developments that came late in the his-
tory of our species.
Some of the error tendencies (bugs) that arise from illicit mingling of cognitions and
emotions are shown in Table 25.3. Normatively, calculations of probabilities and utilities should
be independent of each other, but independence fails in the brain. Motivated inference is the
well-known tendency of people to base their beliefs not just on relevant evidence, but on what
they want to believe (Kunda, 1990). This thinking is more complicated than wishful thinking,
because it requires interactions of goals, memory, and inference making. The brain has no
firewall between cognition and emotion, so it is not surprising that people often adopt beliefs
that they find emotionally appealing, in domains that range from politics to relationships. For
example, people who like particular politicians find it hard to believe that they have misbehaved.
Surprisingly, however, people do not always believe what makes them happy, but instead
believe things because they scare them, which is fear-driven inference. The classic example
is Shakespeare’s Othello, who has only scanty evidence that his wife is unfaithful, but cannot
block the inference of infidelity because fear keeps him thinking about it. The emotional sig-
nificance of the hypothesis of unfaithfulness is so strong that it hijacks attention while Othello
ignores contrary evidence and alternative hypotheses. High disutility prompts an estimation of
high probability.
Another illicit interaction of probability and utility is sour grapes, where the fox concludes
that the grapes are sour because he cannot reach them. Logically, low utility should be

Table 25.3 Error tendencies resulting from confusions of probability and utility caused by cognition-
emotion interactions

Error tendency Example Illicit interaction

Motivated inference I really want that job, so I’m sure high utility → high probability
to get it.
Fear-driven inference I’m terrified that this mole is high disutility → high probability
cancerous, so it must be.
Sour grapes I can’t afford a BMW, but they’re low probability → low utility
too unreliable anyway.
Rationalization I have to go to Moose Jaw, so it high probability → high utility
should be a fun city.

Source: Based on Thagard (2019c).

403
404

Paul Thagard

independent of low probability, but the brain again mixes cognition and emotion to yield the
dubious conclusion.
Another mixture that people succumb to is rationalization, where the high probability of
an occurrence makes us think that it is at least okay, with more utility than we would other-
wise judge it to have. After Donald Trump was elected president, many pundits erroneously
argued that it could not really be as bad as initially feared because of political constraints on the
presidency. For emotional rather than cognitive reasons, people have a tendency to accept their
current situations.
Other familiar error tendencies in decision making result from inappropriate interconnections
of cognition and emotion. People are prone to the fallacy of sunk costs, making their decisions
based on past results rather than on future expectations. For example, people sometimes stay in
careers and romantic relationships too long, because of past investments rather than consider-
ations of future prospects. Instead of calculating the expected utility of continuing the career or
the relationship, people are driven to avoid the emotion of regret that they would feel if they
bailed out, along with the possibility of other negative social emotions such as embarrassment,
guilt, and shame. Such emotions get in the way of rational calculations about what to do in the
future rather than focusing on the unchangeable past.
Similarly, emotional effects on cognition explain the well-known tendency of people to
be unduly influenced by immediate context rather than by long-term effects. People have a
tendency to go for short-term small gains in neglect of long-term large gains that they prefer
on reflection. For example, people tend to buy things immediately rather than save for retire-
ment. A famous example of this time discounting is the psychological study in which most
children choose to eat a marshmallow immediately rather than wait a short time to get two
marshmallows.
The neural explanation of time discounting is that different brain areas are involved (McClure
et al., 2007). Decisions about what to do immediately engage emotion-related areas such as the
ventral striatum and orbitofrontal cortex, whereas decisions about what to do in the long run
are not so emotionally engaging and therefore can be done by calculations in prefrontal and
parietal areas. Faced with immediate rewards, people do not perform calculations of long-term
expected utilities, with emotion overwhelming cognition.
Other well-known incursions of emotion into decision making include risk aversion and
the tendency to frame losses as more salient than gains. Neurobiological investigations find that
loss aversion correlates with activity in the amygdala, suggesting that dealing with losses has
greater emotional effect than dealing with gains (De Martino, Kumaran, Seymour, & Dolan,
2006). I estimate that more than half of the 53 error tendencies listed in Thagard (2011) have a
substantial emotional component. Hence the integration of cognition and emotion in the brain
contributes as much as size and speed limitations to the boundedness of rationality.

Brain limitations on attention and consciousness

In his best-selling book, Thinking, Fast and Slow, Daniel Kahneman (2011) uses dual processes
theories of mind to explain why people are so prone to the many thinking biases that he and
Amos Tversky identified. Like many other psychologists, he distinguishes between System 1,
which is fast, automatic, involuntary, and unconscious, and System 2 which is slow, deliberate,
voluntary, and conscious. The distinction explains numerous thinking biases as resulting from
the unreflective operations of System 1, whereas System 2 allows people to apply appropriate
formal rules such as theories of probability and utility.

404
405

How rationality is bounded by the brain

The problem with this explanation is that there is no specification of how the two systems
actually operate. Suggestions have been made about how the two systems might map onto
different areas of the brain, but these proposals have not stood up to empirical scrutiny (Spunt,
2015). Dual process theory is a useful, descriptive, way of classifying different kinds of thinking,
but provides no explanation of them, because it does not specify the mechanisms that underlie
either of the processes.
Thagard (2019a, Chapter 8) uses the Semantic Pointer Architecture to describe mechanisms
that produce the differences between thinking fast and slow. In the slow mode, inferences
and actions take place because of interactions among brain areas that interpret sensory inputs,
evaluate sensory inputs, and generate new neural representations. The slow mode is much rarer
and occurs when competition among brain representations generates conscious awareness via
a small subset of them.
Since George Miller's (1956) landmark paper on the magical number seven, psychologists
have been aware that conscious, working memory is severely limited. The reasons for this
limitation are not clear: it may be an adaptive feature designed to focus mental resources on
potential actions, which have to be serial rather than parallel; or it may just be a side effect of
the large amount of neural resources required to produce bindings of representations accessible
to consciousness.
Either way, consciousness is limited in ways that block people's awareness of their failures
to follow normative principles rather than committing the errors so far. For example, when
people persist in relationships or businesses because of the sunk cost error tendency, they may
not be aware that they are failing to do a good calculation of the expected consequences of their
actions because of emotions such as regret and embarrassment. On the other hand, the emo-
tional importance of personal relationships and careers should encourage people to consider
what past events reveal about the satisfaction of their basic goals, and thereby discourage them
from tossing away what worked in the past based on a superficial calculation of future gains. For
example, if you have spent years in a romantic relationship, it is worthwhile reflecting on why
there were some good times before making inferences about future consequences.
Most inferences in decision-making occur unconsciously, without people being aware of
what they were doing. The limits of conscious attention exacerbate the limits of size and speed
discussed earlier. Even if the brain has the resources to carry out complex inferences, it often
does not have the capacity to become aware of how it is performing. Hence, people cannot
consciously check whether they are following appropriating normative rules or just sliding into
error tendencies.
Without completely reengineering the brain in a way that is not evolutionarily available,
there is no way to enable people to make more of their thinking consciously evaluable by nor-
mative principles. People can use external memory such as paper and spreadsheets to write
down relevant considerations, overcoming the limitations of working memory and attention.

Conclusion: helping brains to be more rational

Faced with the perplexing plethora of error tendencies, we can ask how people can be helped
to think more rationally. The standard pedagogical method is to teach critical thinking courses
in which students are made aware of fallacies, biases, and other error tendencies, in the hope
that this awareness will reduce mistakes. That is the practice that I followed when I taught crit-
ical thinking, and students said they thought that the class helped. But careful studies of the
effectiveness of critical thinking instruction are rare (Cotter & Tally, 2009).

405
406

Paul Thagard

A deeper strategy would be to look critically at what counts as rationality. My 3-analysis

of rationality in Table 25.1 assumes that people should aim to meet the standards of deductive
logic, probability theory, and utility theory, but these have psychological and philosophical
limitations. Outside mathematics, logical deduction is rare, so it does not provide much of a
standard for rational inference. Far more common is abductive inference, where hypotheses
are accepted because they provide the best explanation of the available evidence. Such infer-
ence is better understood as based on coherence rather than on formal principles (Thagard,
1989, 2000).
Probability theory is immensely useful when frequencies are known, for example, in games
of chance and statistically rich sciences. But the applicability of probabilities understood as
subjective degrees of belief is much more contentious, because their potential objectivity and
psychological reality are suspect (Thagard, 2019c). When we know statistical probabilities, we
should use them, but wildly guessing about them does not further the aims of making good
inferences.
Similarly, the theory of expected utility theory assumed by economists rests on psychologic-
ally dubious foundations. Whereas nineteenth-century theorists such as Bentham viewed utility
as a psychological quantity, twentieth-century economists tried to reconstruct utility from
preferences. The result was mathematically elegant, but got the causal explanation backward.
People have the preferences that they do because of estimations of value, not utilities because of
preferences. Hence it is not clear that the theory of expected utility provides the desired nor-
mative standard for decision-making. An alternative is to view decisions as inferences to the best
plan, where assessment of actions is based on coherence with emotion-laden goals (Thagard,
2006, 2019c).
Accordingly, we should consider replacing the exemplars of rationality in Table 25.1 with
some of the following: coherence-based abductive inference, probabilistic inference in appro-
priate domains, and decisions based on inference to the best plan (Thagard, 2000). The other
dimensions of the 3-analysis survive, but take on a less formal and more realistic aim to help
people improve their acquisition of beliefs and their choices of actions. People still qualify as
frequently irrational, for example, in motivated inference, but at least we have a better sense of
why they succumb because of close ties between normative practices and irrational deviations.
It is an open question whether understanding the roots of bounded rationality in the brain
can help people to avoid error tendencies. There need to be controlled experiments that evaluate
the effectiveness in improving inferences about beliefs and actions under these conditions:

1. Make people aware of error tendencies with vivid examples, as is currently done in critical
thinking courses.
2. Make people aware of error tendencies along with psychological explanations of why
people are prone to them, such as dual process theories of cognition and social influences
(Nisbett, 2015).
3. Make people aware of error tendencies along with neural explanations of why people
are prone to them, including the brain limitations with respect to size, speed, cognition-
emotion interactions, and limitations of consciousness.

My conjecture is that understanding why people are so prone to irrational inferences might
help students of critical thinking to be more rational.
Regardless of pedagogic effectiveness, understanding how rationality is bounded by the
brain should allow psychologists, economists, and philosophers to have a better theoretical
understanding of why people are so frequently irrational.

406
407

How rationality is bounded by the brain

Acknowledgments
The author thanks Riccardo Viale for helpful comments.

References
Aristotle. (1984). The complete works of Aristotle. Princeton,, NJ: Princeton University Press.
Bacon, F. (1960). The New Organon and related writings. Indianapolis: Bobbs-Merrill.
Blouw, P., Solodkin, E., Thagard, P., & Eliasmith, C. (2016). Concepts as semantic pointers: A framework
and computational model. Cognitive Science, 40, 1128–1162.
Cotter, E. M., & Tally, C. S. (2009). Do critical thinking exercises improve critical thinking skills?
Educational Research Quarterly, 33(2), 3.
Damasio, A. R. (1994). Descartes' error: Emotion, reason, and the human brain. New York: G. P. Putnam's Sons.
De Martino, B., Kumaran, D., Seymour, B., & Dolan, R. J. (2006). Frames, biases, and rational decision-
making in the human brain. Science, 313(5787), 684–687.
Eliasmith, C. (2013). How to build a brain: A neural architecture for biological cognition. Oxford: Oxford
University Press.
Gould, S. J., & Lewontin, R. C. (1979). The spandrels of San Marco and the Panglossian paradigm: A cri-
tique of the adaptationist programme. Proceedings of the Royal Society of London B, 205(1161), 581–598.
Hanson, H. (2015). Fallacies. Stanford encyclopedia of philosophy. Available at: https://plato.stanford.edu/
entries/fallacies/
Holyoak, K. J., & Thagard, P. (1995). Mental leaps: Analogy in creative thought. Cambridge, MA: MIT Press/
Bradford Books.
Kahnemann, D. (2011). Thinking, fast and slow. Toronto: Doubleday.
Kahneman, D., & Tversky, A. (Eds.) (2000). Choices, values, and frames. Cambridge: Cambridge
University Press.
Kajić, I., Schröder, T., Stewart, T. C., & Thagard, P. (2019). The semantic pointer theory of emotions.
Cognitive Systems Research, 58: 35–53.
Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108, 480–498.
McClure, S. M., Ericson, K. M., Laibson, D. I., Loewenstein, G., & Cohen, J. D. (2007). Time discounting
for primary rewards. Journal of Neuroscience, 27(21), 5796–5804.
Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for pro-
cessing information. Psychological Review, 63, 81–97.
Murphy, G. L. (2002). The big book of concepts. Cambridge, MA: MIT Press.
Nisbett, R. E. (2015). Mindware: Tools for smart thinking. New York: Farrar, Straus, and Giroux.
Pessoa, L. (2013). The cognitive-emotional brain: From interactions to integration. Cambridge, MA: MIT Press.
Politi, D. (2017). Richard Thaler wins economics Nobel for recognizing people are irrational. Slate.
Available at: www.slate.com/blogs/the_slatest/2017/10/09/richard_thaler_wins_economics_nobel_
for_recognizing_that_people_are_irrational.html.
Quartz, S. R., & Sejnowski, T. J. (1997). The neural basis of cognitive development: A constructivist
manifesto. Behavioral and Brain Sciences, 20, 537–556.
Rumelhart, D. E., & McClelland, J. L. (Eds.) (1986). Parallel distributed processing: Explorations in the micro-
structure of cognition. Cambridge, MA: MIT Press/Bradford Books.
Simon, H. A. (1972). Theories of bounded rationality. Decision and Organization, 1(1), 161–176.
Simon, H. A. (2000). Bounded rationality in social science: Today and tomorrow. Mind & Society,
1(1), 25–39.
Spunt, R. P. (2015). Dual-process theories in social cognitive neuroscience. In A. Toga & M. D. Lieberman
(Eds.), Brain mapping: An encyclopedic reference (pp. 211–215). Amsterdam: Elsevier.
Thagard, P. (1989). Explanatory coherence. Behavioral and Brain Sciences, 12, 435–467.
Thagard, P. (2000). Coherence in thought and action. Cambridge, MA: MIT Press.
Thagard, P. (2006). Hot thought: Mechanisms and applications of emotional cognition. Cambridge, MA: MIT Press.
Thagard, P. (2011). Critical thinking and informal logic: Neuropsychological perspectives. Informal Logic,
31, 152–170.
Thagard, P. (2019a). Brain-mind: From neurons to consciousness and creativity. Oxford: Oxford University
Press.
Thagard, P. (2019b). Mind-society: From brains to social sciences and professions. Oxford: Oxford University
Press.

407
408

Paul Thagard

Thagard, P. (2019c). Natural philosophy: From social brains to knowledge, reality, morality, and beauty.
Oxford: Oxford University Press.
Thagard, P., & Schröder, T. (2014). Emotions as semantic pointers: Constructive neural mechanisms.
In L. F. Barrett & J. A. Russell (Eds.), The psychological construction of emotions (pp. 144–167).
New York: Guilford.
Thagard, P., & Verbeurgt, K. (1998). Coherence as constraint satisfaction. Cognitive Science, 22, 1–24.
Thaler, R. H. (2015). Misbehaving: How economics became behavioural. New York: Norton.
van Rooij, I., Blokpoel, M., Kwisthout, J., & Wareham, T. (2018). Cognition and intractability: A guide to
classical and parameterized complexity analysis. Cambridge: Cambridge University Press.
Wason, P. C. (1966). Reasoning. In B. M. Foss (Ed.), New horizons in psychology (pp. 1–43). Harmondsworth:
Penguin.

408
409

26
BUILDING A NEW
RATIONALITY FROM THE NEW
COGNITIVE NEUROSCIENCE
Colin H. McCubbins, Mathew D. McCubbins, and Mark Turner

The theory of mind within the theory of games

Game-theoretic models are used to explain human interactions across a wide range of activities,
such as allocation of security forces, allocation of health care services, and the design of polit-
ical, legal, social, and market institutions (Roth, 1990; Fudenberg and Tirole, 1991; Kagel and
Roth, 1997). Despite the widespread use of game-theoretic models to explain human behavior,
we often observe behavior, from voting to the divergence of political parties’ platforms to
market bubbles and crashes that do not easily accord with game-theoretic predictions (Camerer,
2003, 2008). Further, it is a common finding that experimental subjects, in tightly controlled
settings, do not make choices that comport with Nash equilibrium strategies, or indeed, von
Neumann-Morgenstern utility maximization (Plott, 1967; Smith, 2010).
To begin, we need to define our terms and identify game theory’s flaws. A game is defined
by identifying the items in PAISPOE:

• Players;
• Actions available to players;
• Information they have about the game; the timing of access to such information; including
the knowledge the players have about what other players know or will know and when
those other players will know it;
• Strategies define the action rules that specify what actions to take under every circumstance;
• Payoffs associated with each outcome;
• Outcomes which define the consequences of the game;
• and, finally, Equilibria. A Nash Equilibrium (NE) results when, given their strategies, no
player can do better by changing her strategy while all others remain unchanged. An
equilibrium concept defines which strategies are allowed and which are precluded.

Noncooperative games, at least in their classical format, assume that subjects share common
knowledge. Classical game theory requires players to have correct and consistent beliefs. That
is, individuals are perfectly able to predict the actions of others and those others are, likewise,
perfectly able to predict that individual’s actions. Thus, the equilibrium path in any game
is predictable by assumption. To have “correct beliefs” is to regard other players as following

409
410

Colin H. McCubbins et al.

NE strategies and to predict that they follow classical equilibrium strategies. Indeed, it is also
required that players know that other players know that all players, including themselves, are
following NE strategies, and so on, ad infinitum. As Lupia, Levine, and Zharinova (2010)
note, the condition is even stronger, in that a Nash Equilibrium “requires shared conjectures.
… Common Nash refinements ... continue to require that actors share identical conjectures of
other players’ strategies” (p. 106). This is part of what economists assume when they accept that
the players in a game share ‘common knowledge.’ As Smith (2000, p. 9) writes, “The common
knowledge assumption underlies all of game theory and much of economic theory … Without
such common knowledge, people would fail to reason their way to the solution arrived at cog-
nitively by the theorist.”
Unfortunately, the natural, biological, or cognitive means of achieving such knowledge
are not specified in game theory; rather, it is merely assumed that, given enough time and
effort, players can all learn what behaviors to adopt and when to adopt them, or, barring
this possibility, that societies will adopt rules, laws, or norms to restrict and channel behavior
to more efficient forms. Prior work on subjects’ beliefs in experimental settings suggest that
subjects possess non-equilibrium beliefs (Croson, 2007) and that, in at least some settings, their
behavior can be reasonable, given such beliefs, although this is not always true (Camerer, Ho,
and Chong, 2004).
The laboratory and market mispredictions of game theory have led to the establishment of
behavioral game theory. The main approach of behavioral game theory has been to propose
systematic deviations from the predictions of rationality in classical game theory, deviations that
arise from, for example, character type, cognitive overloads or other reasoning, constraints and
conditions of memory, or informational limitations. The fundamental idea of behavioral game
theory is that, if we know the deviations, then we can correct our predictions accordingly, and
so get it right. Indeed, if we know these deviations, we can anticipate them and we can nudge
people to act more in accord with the equilibrium predictions of classical game theory, as modi-
fied by behavioral game theory (Sunstein, 2014).
There are two problems with this approach, however, and each is fatal. (1) For the chooser
to contemplate the range of possible deviations actually makes it exponentially harder for
the chooser to figure out a path to an equilibrium, since there are many dozens of possible
deviations. This array of possible deviations makes the theoretical models useless for modeling
human thought or human behavior in general (Roth, 1990). Modeling deviations is helpful
only if the deviations are consistent, so that scientists (and indeed decision-makers) can make
predictions about future choices on the basis of past choices. They must be able to gener-
alize from the particular deviation. But as we have shown (Lucas, McCubbins, and Turner
2015; McCubbins, Turner, and Weller 2012), the deviations are not consistent. In general,
deviations from classical models are not consistent for any individual from one task to the
next or between individuals for the same task. In addition, people’s beliefs are in general not
consistent with their choices. Accordingly, all hope is hollow that we can construct a general
behavioral game theory.

Building a new rationality from the new cognitive neuroscience

To address the discrepancy between predicted and actual behavior, we must build a better
theory of human behavior. To do this, we must start with an appreciation of how we actually
reason. As cognitive science has shown, intuitive notions of how the mind works (vision, lan-
guage, memory, etc.) may be very useful for humans to hold as scaffolding for consciousness,
but they are comprehensively wrong and simplistic. Intuitive notions of how we reason are

410
411

Building a new rationality

not a basis for science. How we reason must be discovered, not assumed, and certainly not
borrowed from intuition.
The first step in designing a new model of human choice is to determine why game theory
failed. We propose that the failure of game theory derives from its mistaken theory of mind. In
this theory of mind,

• People have unitary selves and fixed preferences about outcomes.

• People think about hypothetical paths that lead to outcomes.
• People have cognitive models about other people, including
• those other people have unitary selves and consistent preferences about outcomes;
• those other people think about hypothetical paths that lead to outcomes;
• those other people have cognitive models about other people.
• An actor considers everyone’s fixed preferences (to the extent that they are known),
the choices available to each actor at each moment in the rest of the decision tree, the
information known to each actor at each moment, and the ultimate payoffs to the
actors of ending up at each of the specific possible final outcomes.
• An actor chooses, at a moment of decision, the available choice that can lead ration-
ally to the expected best outcome available for the actor. “Rationally” in this pic-
ture means not only that the actor will choose in this way but also that the actor
understands that the other actors will choose in this way.
• Accordingly, actors make choices by backward induction, looking ahead to the final
outcomes, and making choices at each node along the way so as to create a path to the
expected best outcome for the actor at the end. In this picture, an actor’s self is fixed
throughout the course of complicated action, however many moments of choice are
required, and indeed fixed across all games.

Flexibility and blending

In contrast, the world’s cultures have unhesitatingly recognized human flexibility. Even
conditioned and trained humans in utterly confining conditions do not comply reliably to
expectations of fixity. To succeed at all, game theory must control the activity of actors by
imposing extreme, artificial constraints—as in chess, tic-tac-toe, or common-value auctions.
A thorough theory of human decision-making will subsume game theory as a limiting case.
The reverse subsumption is impossible, because modeling the vast majority of human choosing
requires eliminating the constraints on flexibility imposed by game theory.
Many researchers outside game theory have proposed theories, according to which any
human being contains multiple selves (Angyal, 1965; Baumeister, 1998; Berne, 1961; Elster,
1985; Lester, 2010; Mischel, 1968; Rowan, 1990). We acknowledge this tradition and take the
view that an individual assembles, on the fly, very many versions of self, one in each moment
of choosing. Indeed, a self assembled in such a brief moment might never be assembled by that
individual human being again.
If the self is not fixed, how do we model choice by a self? Turner (2014, Chapter 4), reviews
how selves are assembled via conceptual blending. A conceptual blend can have many inputs,
including inputs that conflict strongly with each other. Structure can be projected selectively
from each of these inputs to create a coherent blend with emergent structure of its own.
Consider Bill and Peter, brothers-in-law, each happy. Bill is a mathematically-talented professor
in the Eastern time zone who likes investing and San Francisco. Peter is a stockbroker in San
Francisco. Bill wonders, analogically, should he move to San Francisco and be a stockbroker and

411
412

Colin H. McCubbins et al.

get a huge raise? No: Bill is a night owl but Peter must arise at 5:30am Pacific Time to deal with
the stock market’s opening at 9:30am Eastern Time. Mentally, Bill has created a new, blended
person—Bill-as-Peter, who is miserable arising at 5:30am every morning to do his stockbroker
job, even though misery is not in any of the inputs. This kind of blending, depending on what
is active in the mind as an input, what is selectively projected to the blend, and what new struc-
ture arises in the blend, is standard for human beings despite its utter incompatibility with the
conception of selves in game theory.
Possible inputs to a conceptual blend are very many, including memories of previous selves,
ideas of other people’s selves, ideas of selves which we encountered through reports, news,
books, fiction, songs, movies, fairytales, or generic conceptions of character. Activation of ideas
is highly variable in the brain; the potential inputs vary moment-to-moment. Projection to the
blend is highly selective. Quite variable ranges of emergent meaning can develop in the blend.
There are analogies and disanalogies across all the different selves one has been, in reality
or imagination. In constructing a self by blending, one may select some inputs and com-
press the analogies connecting them to an identity in the blend, and compress the disanalogies
into change for that identity. For example, there are astonishing disanalogies between a person
before marrying and the “identical” person after marrying. The analogies are compressed to
an identity—the person—and the disanalogies are compressed to change for that identity: Mary
got married, or Mary became a wife. This common compression pattern—analogies to identity,
disanalogies to change—is extremely common in mental blending. It is the same pattern we
use for “dinosaurs turned into birds,” “the fences get taller as you drive west across the United
States,” and “his girlfriend gets younger every year.” The resulting blend of a stable self is a useful
illusion. Fauconnier and Turner (2008) explore constraints on blending and argue that nearly all
attempts at blending take place outside of consciousness and fail almost immediately. Very few
survive to be embraced by a community.
Adam Smith masterfully analyzed the creation of variable selves through blending (Turner,
2014). However, classical economics now pays no attention to the role of the dynamic and
highly variable construction of a self over time. This failing is at the root of game theory’s mis-
taken view of mind.

Selves and choices in wayfinding

A person in a moment needs a self that serves for collective action but not the straitjacket
internal consistency assumed by game theory. Person A encountering person B needs to be
a self in the moment of encounter, but there might be many serviceable selves, and the brain
may be running many of those incompatible alternatives in parallel. One will precipitate in
the moment of action, but that does not indicate that the brain was bent exclusively on that
approach. It means only that one self precipitates to serve the moment of action. A different
precipitation might occur in the same circumstances next time. A human being about to act
needs a sense of a stable self, one that will make a choice leading to the next choice point,
where the self will be different but not randomly so. Stability is important; uniqueness, stasis,
rigidity are not.
Or, a small group of colleagues, ignorant of the town, meet and decide to go out. Each
participant needs to know how to choose at any given choice point. The chooser needs a
wayfinding marker: You Are Here Now. Or perhaps several such markers, only one to precipi-
tate at the moment of choice. The chooser needs to be a wayfinder in this scene. The chooser
does not need to try to model this as a game.

412
413

Building a new rationality

Choosers are wayfinders. At each choice point, the self may differ, in a different story. The
possibility space of these alternative selves in stories is large. All that is needed is for one self to
precipitate for each choice.

Collective action in the wild

The design of actual choice situations in the wild supports our view of choosing as wayfinding.
Consider a restaurant. The owner does not seek equilibria by saying to customers, “Tell us what
you want, and we will see whether we can make it.” Instead, the restaurant imposes constraints
on the customer by its name, style, design, menu, suggestions (“for those who prefer something
light”), presentations (dessert cart, case of grilled vegetables), safety valves (corkage), delegates
(the bartender or mother who chooses for you).

Selves and choices in cognitive neuroscience

The view of people as wayfinders, inventors of multiple selves in multiple stories, accords with
suggestions in recent cognitive neuroscience.
What is the brain doing? Older views of the brain as driven by tasks prompted by external
stimuli have been cast into doubt by evidence suggesting that most of the brain’s work is
intrinsic. When human beings are unengaged in tasks, many brain networks are highly active,
but what they might be doing is hotly debated in cognitive neuroscience. One of the main
proposals is that the brain is largely engaged in composing matrices of stories and selves. We
clarify that the engagement imagined is of course not during moments of focused decision
making, which are task-oriented, but instead in the temporal background to such moments,
preparing resources for such brief moments of choosing when they arise. Michael Anderson
(2014, p. 113) writes:

Over the past few years there has been growing interest in something called “resting
state functional MRI,” a technique for seeing what your brain is doing when you
aren’t doing much of anything at all. It turns out that brains at rest are pretty restless,
consuming far more energy than they do when doing. More interesting, “resting”
activity is not random but highly coherent, consistent, and predictable. The discovery
of the brain’s characteristic resting behavior led some years ago to the postulation of a
“default network” for the brain—a set of regions that consistently cooperate to do ...
well, what, exactly, we don’t know. But surely it must be something interesting. Your
brain would hardly waste all of that energy dancing to the beat of its inner drummer
if it didn’t serve some function, right?
Our ignorance regarding the function of all that fluctuation isn’t for lack of trying.
The discovery of the brain’s default network has led to hundreds of studies relating the
default network to the brain’s anatomical structure as well as to mood disorders, such
as depression, developmental problems, such as autism, and degenerative diseases, such
as Alzheimer’s. It has even been suggested that resting-state activity holds the key (cue
deep voice and echo effect) to understanding consciousness itself (e.g., Raichle, 2010).
Now, when neuroscientists start brandishing the “c” word, there are two predictable
reactions: increased public interest and attention and increased scientific scrutiny and
criticism. Both have happened here, generating a cadre of enthusiastic adherents and an
equally committed group of critics who question whether we should continue wasting

413
414

Colin H. McCubbins et al.

our energy figuring out why the brain appears to be wasting its energy. Or, as one
prominent neuroscientist put it to me recently, “It’s just such a fad. I kind of hate it.”

Hyperbolic allusions to cracking the mystery of consciousness excepted, I don’t think any-
body should hate it. But to see why we should, if not love, then at least care about the brain’s
intrinsic activity requires us to think about brain function in a new and unfamiliar way.
In earlier notions of the computational brain, it was easy to view brains as having program-
ming that remained fixed and that responded to external stimuli by taking inputs and com-
puting actions in response. The fit between the theory of the computational brain and theory
of games was pat: the chooser has preferences and rationality and, given knowledge of external
circumstances, computes a choice and enacts it. But, today’s cognitive neuroscience is more
compatible with the view that action is driven largely by intrinsic and imaginative construction
of selves and stories in a process of wayfinding through life.

The brain as an imagination engine for selves and stories

The Neuron Doctrine was compatible with a computational view of the brain as producing 1s
and 0s in response to input much the way a theoretical Turing machine operates by receiving
input—one symbol on a tape at a time—and changing the static state of the machine according
to algorithms for responding to the symbol. Instead, recent work paints a picture of a brain
engaged in largely intrinsic work, often connected to constructing selves and narratives. As
Kaplan et al. (2017, p. 5) write:

In attempting to characterize the kind of psychological operations that appear to engage

this (i.e. the default) network, researchers have described them as related to social cog-
nition (Mars et al. 2012), internally directed processing (Immordino-Yang et al. 2012),
mental time travel (Ostby et al. 2012), or self-related processing (Northoff and Qin
2011). Interestingly, all of these operations are either involved in the processing of
narratives or rely on a narrative organization of information. For example, the majority
of studies that have shown cortical midline activations for self-related processing have
focused on aspects of the autobiographical self, such as personality trait judgment,
rather than transient present-moment aspects of the self (Northoff et al. 2006). The
autobiographical self is, in essence, a process of generating fragmentary narratives of our
personal lives built from a multitude of recorded experiences (Damasio 1998). These
same midline structures are activated just as much or more when we think about the
biographies of other people (Araujo et al. 2013, 2014), suggesting that the processing of
narratives may be more important for activating these structures than self-relatedness.

The search for neuroscientifically relevant human

psychological factors (NRPs)
Contemporary cognitive neuroscience is reconsidering what might be the “neuroscientifically
relevant human psychological factors.” Back-of-the-envelope notions of the mind as a system of
passive perception, active response, fear, greed, desire, executive function, inference, memory,
etc., were an unfortunate place for psychology and economics to begin (see Anderson, 2014,
for a review). A view of higher-order human cognition as largely intrinsic and constantly
blending even strongly incompatible inputs in imaginative ways to produce innovations—in
selves, stories, and possibilities—was largely unavailable in these previous folk psychological

414
415

Building a new rationality

notions. Even less available was an idea of mental activity as generating simultaneously very
many of these blending networks, in parallel, often with strong conflicts between them, and of
action—including enacted choice—as the momentary precipitation of just one of these par-
allel imaginative lines into an attractor basin in the behavioral landscape. Such ideas of mental
activity are being actively explored in the new cognitive neuroscience.

Dynamical cognition
Cognitive neuroscientists have proposed that the human brain is constantly scanning over a
range of often-conflicting alternatives and collapsing that range to an action only in the moment
of decision (Spivey, 2008). Consider someone who wants to pick up a coffee cup. There are
many ways to do so successfully. The brain may explore many of those approaches simultan-
eously. The neurobiological basis of action can be varied, with simultaneous but conflicting
lines, each with some probability of being given, at the final instant, control over skeletal and
motor programs. One of those possibilities will precipitate in the moment of action. Our
enacting only one action suite does not mean that the brain was exclusively focused on that
one suite; it means only that, in the moment of action, one coherent action was executed, as
others were forsaken.
The analysis for picking up the coffee cup applies to any moment of choice. The human
being requires great flexibility. Stability of self is not repetition or fixity but instead a coherent
migration path. Perhaps the next moment of picking up the coffee cup will have a self who is
inviting, or dismissive, or finalizing the communicative turn, or trying to elongate the moment,
or oblivious to the coffee because she is paying attention to other things, or styled as uncaring,
or attentive, or sensitive, or concerned with the warmth of the coffee, or turning from that sip
of coffee to more important or less important things, or simply bored with the usual ways of
picking up a coffee cup.
Action appears to be unitary and specific: we must pick up the cup just one way, not
lots of incompatible ways. Misled by the cause-effect isomorphism fallacy, we think the cause
must have the attributes of the effect. Given that performed actions in the world are unitary
and specific and form a nice chain, it is easy to imagine that the mind works something like
that. Instead, as Spivey (2008) explains, it is better to think of multiple pathways of thinking
happening in parallel and of one of them precipitating into an attractor basin for the purpose of
performing a specific action. But the neural activity could drop into a different attractor basin.

Towards a new model of rationality

From the observations of the brain’s intrinsic work, we can develop a new model of rationality.
The main assumption of this model is what we call the Principle of Least Cognitive Effort—that is
to say, people will most likely choose the path of action that requires the least cognitive effort.
Often they will not, but most often they will.
It is noteworthy that this is not a simple reformulation of the Cognitive Miser Hypothesis,
which states that the brain actively and unconsciously avoids spending computational effort on
tasks if it can be avoided. It would seem that the model of the brain as a limited memory com-
puter is being slowly proven flawed, as neuroscience research has demonstrated that brains have
much larger computational ability than we had previously hypothesized or observed. Instead,
we argue here that the Principle of Least Cognitive Effort is a conscious choice that individuals
are making: that is, this is not the result of the brain somehow trying to save itself from its own
limitations but rather the individual trying to save what limited time they have in the day. As

415
416

Colin H. McCubbins et al.

such, the limiting factor here is not brain computational ability but rather individual time and
energy.
This model of rationality requires that we make assumptions similar to those in PAISPOE.
We will, however, depart in significant ways. Instead of modeling individual actions as binary,
consistent responses derived from external stimuli, we can develop a probabilistic model of
choice that is more compatible with observations in cognitive neuroscience and game theoretic
experimentation. Individual behavior in the moment is unique and it is conditional on experi-
ence. Given a game, an individual will respond by choosing one of many potential selves to act
with in this setting. The choice of self determines the payoffs for any particular outcome and
it affects the information conditions. A person’s type will determine how the game is played.
To this model of decision, we add that behavior is also conditional on context: the who,
what, where, when, and why. Context is a parameter that changes the probability space over
which an individual is making a decision; it forces that individual to reconsider the con-
struction of the theory of self and the theory of mind that they are using in a specific game.
Context is a multidimensional construct that is also dependent on time. Consider the scenario
in which an individual is visiting a restaurant for the second time in two days. The who, what,
where, and why are the same, but the when is different. Thus, we may expect that individual
to order something completely different even though the choice situation is the same as it was
the day before except that it is now the following day. Thus, context, itself, is conditional on
previous context, which creates path dependence in action as well as the strategies and equi-
libria that follow.
Suppose that each individual has a set of types. For any specific context and game combin-
ation, there is a subset of types that have a non-zero probability of being chosen by that indi-
vidual for use in that situation. Thus, in considering the choice of strategy, in other words an
action plan, the choice space over types and the type chosen greatly affect the equilibria.
These subsets of types for individuals define the probability space of actions that an indi-
vidual can choose from, given a particular context and game. This probability space of actions is
narrowed to a single action at time t when a decision is necessary. We can define the probability
function of actions Θ as a function of an individual’s type T, context C, and the game Γ. This
probability function has a density corresponding to the probability of a specific action being
taken. We assume that the highest peak of this density function corresponds to an individual’s
most commonly completed action.
This probability space of actions defines the cognitive sum over actions much in the same
way that quantum path integral formulation defines the sum over histories of a particle: there is
a continuum of possible actions that an individual can take but a much narrower band of likely
actions. The most likely of these, given the assumption above, is the action that requires the
least cognitive effort. Because expertise reduces the cognitive effort of taking an action, we
would thus infer that the probability density would become unimodal and narrower, such that
the potential range of actions for the expert would be much less varied and more structured.
Thus, expertise would conduce to action path consistency: we would expect an individual to
have a similar response to the same game and context over repeated trials. However, the exact
opposite would occur in the absence of expertise: the probability density would become much
wider and possibly have multiple modes, which would lead an individual’s responses to be
inconsistent. This model, then, has the potential to corroborate laboratory experimental results.
Simon long ago studied the decision-making speed and accuracy of experienced chess
players as an example of the impact that expertise has on cognition:

416
417

Building a new rationality

We have seen that a major component of expertise is the ability to recognize a very
large number of specific relevant cues when they are present in any situation and
then to retrieve from memory information about what to do when those particular
cues are noticed. Because of this knowledge and recognition capability, experts can
respond to new situations very rapidly – and usually with considerable accuracy.
Simon, 1957

Chase and Simon’s (1973) experimental studies on chess expertise exemplify the type of
cognitive action this model supposes. While it has historically been the case that laboratory
experiments that expose subjects to game theoretic situations tend to produce results far
afield from what would be expected, given the mathematical models they purport to study,
it is also likely that the subjects being studied are just not experts in participating in these
types of situations in a laboratory setting. Thus, we would expect responses to these situ-
ations to be inconsistent and slow, just like the chess novice’s responses to chess scenarios in
Chase and Simon (1973). Having an understanding of how expertise shapes the cognitive
space that individuals are working in seems key to moving forward with a broader theory
of cognition and can help to design better laboratory experiments that have greater levels of
external validity.

Conclusion
As Engel (2005) argues, human beings are by nature flexible and variable, so much so that
cultures must invent institutions to generate enough predictability in human performance to
make interactions even possible.1 Cultures exploit the great flexibility of the human imagination
to invent institutions for the purpose of generating sufficient regularity and predictability in the
face of otherwise uncontrollably wild behavior. In the last 50,000 years or so—a blink of the eye
in evolutionary time—cultures have invented contracts, classrooms, courts, constitutions, retail
counters, certified public accountants, conjugal arrangements, … One of their central purposes
is to enforce patterns that constrain possibilities and generate predictability. It is impossible to
model human collective action unconstrained by cultural institutions, because the possibilities
for human collective action are too many and too variable. As Wittgenstein writes and Geertz
quotes (Geertz, 1973, p. 13):

We … say of some people that they are transparent to us. It is, however, important as
regards this observation that one human being can be a complete enigma to another.
We learn this when we come into a strange country with entirely strange traditions;
and, what is more, even given a mastery of the country's language. We do not under-
stand the people. (And not because of not knowing what they are saying to them-
selves.) We cannot find our feet with them. —Ludgwig Wittgenstein.

Geertz himself comments:

[C]ulture is best seen not as complexes of concrete behavior patterns—customs,

usages, traditions, habit clusters—as has, by and large, been the case up to now, but
as a set of control mechanisms—plans, recipes, rules, instructions (what computer
engineers call “programs”)—for the governing of behavior.
Geertz 1973, p. 44

417
418

Colin H. McCubbins et al.

and “One of the most significant facts about us may finally be that we all begin with the natural
equipment to live a thousand kinds of life but end in the end having lived only one” (Geertz,
1973, p. 45).
Game theory represents the utter extreme of eliminating personal flexibility, by creating dra-
conian governing conditions under which we might have expected to be able to predict both
behavior and an understanding of that behavior. However, as has been demonstrated in decades
of experimentation, people retain impressive flexibility even under these conditions.
We have proposed that, in the wild, people use flexible but orderly mental processes, such
as conceptual blending and narrative imagination, to operate as wayfinders, constructing selves
as they go along to make choices. We have argued that this is consistent with modern cogni-
tive neuroscience. Our prescription for finding our way forward in creating a new rationality
is to explore a fruitful alliance between institutionalist views of collective action and cognitive
neuroscientific views of multiple, variable selves in multiple, various stories and their precipita-
tion in some moments into individual coherent actions.

Note
1 This kind of unpredictability (and the resulting lack of equilibrium) was discussed by Quine (2013) in
terms of the indeterminacy of translation. Quine argues that we do not have an identity criterion for
most of our utterances and thus there is no way for people to be sure that their translation of a sentence
is the correct one.

References
Anderson, M. 2014. After phrenology: Neural reuse and the interactive brain. Cambridge, MA: MIT Press.
Angyal, A. 1965. Neurosis and treatment. New York: Wiley.
Baumeister, R. F. 1998. The self. In D. T. Gilbert, S. T. Fiske, and G. Lindzey (Eds.), The handbook of social
psychology. Vol. 1. (pp. 680–740). Boston: McGraw-Hill.
Berne, E. 1961. Transactional analysis in psychotherapy. New York: Grove Press.
Camerer, C. F. 2003. Behavioral game theory. New York: Russell Sage Foundation/ Princeton, NJ: Princeton
University Press.
Camerer, C. 2008. The Potential of Neuroeconomics. Economics and Philosophy, 24(3), 369–379.
doi:10.1017/S0266267108002022.
Camerer, C. F., Ho T. H., and Chong. K. 2004. A cognitive hierarchy model of games. Quarterly Journal
of Economics, 119(3), 861–898.
Chase, W. G., and Simon, H. A. 1973. Perception in chess. Cognitive Psychology, 4, 55–81.
Croson, R. 2007. Theories of commitment, altruism and reciprocity: Evidence from linear public goods
games. Economic Inquiry, 45, 199–216.
Denzau, A., and North, D. C. 2000. Shared mental models: Ideologies and institutions. In A., Lupia,
M., McCubbins, and S. Popkin (Eds.), Elements of reason: Cognition, choice and the bounds of rationality.
Cambridge: Cambridge University Press.
Elster, J. (Ed.) 1985. The multiple self. Cambridge: Cambridge University Press.
Engel, C. 2005. Generating predictability: Institutional analysis and institutional design. Cambridge: Cambridge
University Press.
Fauconnier, G., and Turner, M. 2008. The way we think: Conceptual blending and the mind's hidden complex-
ities. New York: Basic Books.
Fudenberg, D., and Tirole, J. 1991. Game theory. Cambridge, MA: MIT Press.
Geertz, C. 1973. The interpretation of cultures. New York: Basic Books.
Gigerenzer, G. 2000. Adaptive thinking: Rationality in the real world. New York: Oxford University Press.
Kagel, J. H., and Roth, A. E. (Eds.) 1997. The handbook of experimental economics. Princeton, NJ: Princeton
University Press.

418
419

Building a new rationality

Kaplan, J. T., Gimbel, S. I., Dehghani, M., Immordino-Yang, M. H., Sagae, K., ... Damasio, A. 2017.
Processing narratives concerning protected values: A cross-cultural investigation of neural correlates.
Cerebral Cortex, 27(2), 1428–1438.
Lester, D. 2010. A multiple self theory of personality. Hauppauge, NY: Nova Science.
Lucas, G. M., McCubbins M. D., and Turner M. 2015. Against game theory. In R. A. Scott and
S. M. Kosslyn (Eds.), Emerging trends in the social and behavioral sciences: An interdisciplinary, searchable, and
linkable resource. Hoboken, NJ: Wiley.
Lupia, A., Levine, A. S., and Zharinova, N. 2010. Should political scientists use the self confirming equi-
librium concept? Benefits, costs and an application to the jury theorem. Political Analysis, 18, 103–123.
McCubbins, M. D., and Turner, M., 2012. Going cognitive: Tools for rebuilding the social sciences. In
R. Sun, P. Thagard, N. Ross, B. Shore, and S. Feldman (Eds.), Grounding Social Sciences in Cognitive
Sciences, Cambridge, MA: MIT Press.
McCubbins, M. D., Turner, M., and Weller, N. 2012. The mythology of game theory. In International
Conference on Social Computing, Behavioral-Cultural Modeling, and Prediction (pp. 27–34). Berlin: Springer.
McKelvey, R., and Page, R. T. 1990. Public and private information: An experimental study of informa-
tion pooling. Econometrica, 58, 1321–1339.
Mischel, W. 1968 Personality and assessment. New York: Wiley.
Plott, C. R. 1967. A notion of equilibrium and its possibility under majority rule. The American Economic
Review, 57(4), 787–806.
Quine, W. O. 2013. Translation and meaning. In W. O. Quine, Word and object (new ed.), (pp. 23–72).
Cambridge, MA: MIT Press.
Raichle,, M. 2010. The brain’s dark energy. Scientific American, 302(3), 44–49.
Rasmusen, E. 2006. Games and information (4th ed.). Oxford: Blackwell Publishers.
Roth, A. E. 1990. New physicians: A natural experiment in market organization. Science, 250, 1524–1528.
Rowan, J. 1990. Subpersonalities. London: Routledge.
Simon, H. 1957. A behavioral model of rational choice. In H. Simon, Models of man, social and
rational: Mathematical essays on rational human behavior in a social setting. New York: Wiley.
Smith, A. 2010. The theory of moral sentiments. London: Penguin.
Smith, V. 2000. Rational choice: The contrast between economics and psychology. In Bargaining and
Market Behavior: Essays in Experimental Economics (pp. 7–24). New York: Cambridge University Press.
Spivey, M. 2008. The continuity of mind. Oxford: Oxford University Press.
Sunstein, C. R. 2014. Why nudge? The politics of libertarian paternalism. New Haven, CT: Yale University Press.
Turner, M. 2009. The scope of human thought. Available at: http://onthehuman.org/humannature/.
Turner, M. 2014. The origin of ideas: Blending, creativity, and the human spark. New York: Oxford University
Press.

419
420
421

PART V

Homo Oeconomicus Bundatus

422
423

27
MODELING BOUNDED
RATIONALITY IN ECONOMIC
THEORY
Four examples

Ariel Rubinstein

Introduction
The terms bounded rationality and economic theory mean different things to different people.
For me (see Rubinstein (2012)), economic theory is a collection of stories, usually expressed
in formal language, about human interactions that involve joint and conflicting interests.
Economic theory is not meant to provide predictions of the future. At most, it can clarify
concepts and provide non-exclusive explanations of economic phenomena. In many respects, a
model in economic theory is no different than a story. Both a story and a model are linked to
reality in an associative manner. Both the storyteller and the economic theorist have in mind a
real-life situation but do not consider the story or the model to be a full description of reality.
Both leave it to the reader to draw their own conclusions, if any.
For me, models of bounded rationality are (see Rubinstein (1998)) models that include
explicit references to procedural aspects of decision making. A common critique of bounded
rationality models is that they are more specific, less general, and more arbitrary than models
in the mainstream areas of economic theory, such as general equilibrium or game theory in
which full rationality is assumed. In response, I would claim that every model makes very
(very) special assumptions. Without strong assumptions, there would be no conclusions. It
is true that rationality is a special assumption since it is viewed by many as being normative,
whereas models of bounded rationality are viewed as dealing with deviations from normative
behavior. Returning to the analogy of a story: What story is more interesting—one about
normative people who behave “according to the book” or one about people who deviate
from normative behavior?
Not every model that is inconsistent with some aspect of rationality is a model of bounded
rationality. A model in which rational agents ignore some aspect of rationality is a bad model
rather than a model of bounded rationality. A good model of bounded rationality should
include a procedure of reasoning that “makes sense” and is somewhat related to what we
observe in real life.

423
424

Ariel Rubinstein

I am not a big fan of abstract methodological discussions. I prefer to demonstrate an approach

by discussing examples. Accordingly, the chapter discusses four models, in which economic
agents are assumed to reason in systematic ways using a well-defined procedure that is outside
the standard scope of rational behavior. My choice of models is totally subjective. I am par-
tial to these models because I was involved in constructing and analyzing them over the last
35 years. Rubinstein (1998) surveyed the field at the time, while Spiegler (2011) surveys more
recent models in which aspects of bounded rationality have been inserted into classic models
of industrial economics.

Bounded rationality and mechanism design

Story
The director of a prestigious MBA program has been persuaded by Choi, Kariv, Muller, and
Silverman (2014) that transitivity is strongly correlated with success in life (as measured by
wealth). Thus, he decides to accept only those candidates who hold transitive preferences.
Accordingly, the director designs a simple test. He presents three alternatives a, b and c to
each candidate and asks them to respond to a questionnaire consisting of three questions Q(a,
b), Q(b, c) and Q(a, c) where Q(x, y) is the quiz question:

Do you prefer x to y or y to x?
I prefer x to y.
I prefer y to x.

For each of the three questions a candidate must respond by clicking on one and only one of
the two possible answers. Thus, the questionnaire has eight possible sets of responses.
The director is obligated by law to inform candidates about the conditions that will gain
them admission to the program. The director has specified the following two conditions:

R1: If you prefer a to b and b to c then you must prefer a to c.

R2: If you prefer c to b and b to a then you must prefer c to a.

The candidates are reminded that an “if ” proposition is violated only if its antecedent (the “if ”
part) is satisfied and its consequent (the “then” part) is not.
The director was hoping that the candidates would feel obliged to report the truth and thus
the simple questionnaire should separate perfectly between “good” candidates who hold transi-
tive preferences and “bad” candidates who do not. In order to encourage instinctive responses,
the director also sets a short time limit for completing the questionnaire.
A disappointment: the director learns that all the candidates have been admitted and
concludes that the candidates with cyclical preferences had gamed the system. The frustrated
director opens an investigation. Researchers are rushed to the scene. They interview candidates
to reveal how they answered the questionnaire so “successfully”. Apparently, candidates treated
the questionnaire as a puzzle to be solved in order to be admitted to the program. The time
limit made solving the “puzzle” a challenging task. The following procedure was identified:

Step 1: Examine whether your honest set of answers satisfies all the conditions. If it does, then
happily submit those answers. If not, go to step 2.

424
425

Modeling BR in economic theory

Step 2: Find a condition that is violated by your honest answers (that is, your true answers sat-
isfy the antecedent but not the consequent of the condition). Try modifying your answers
with respect only to the consequent. If the modified set of answers satisfies all conditions,
then submit them. If not, iterate step 2 (starting with your honest set of answers) until it is
exhausted. Then, proceed to Step 3.
Step 3: Give up. Submit the honest answers (and be rejected).

Faced with the questionnaire and the admission conditions, a “good” candidate (who has tran-
sitive preferences) reports the truth and is admitted. What about a “bad” candidate with cyclical
preferences a  b  c  a (or c  b  a  c )? An honest response satisfies R2 (the antecedent
is false) but violates R1. The candidate remains with the true answers to Q(a, b) and Q(b, c)
and modifies his answer to Q(a, c). Thus, he responds as if he holds the preferences a  b  c and
is admitted. In this way, all the candidates are admitted.
Realizing that people are prepared to cheat (when they have to) by using the procedure to
find a persuasive set of responses, the director tries to come up with a modified set of admission
conditions so that all the good candidates will pass the test while the bad candidates will not. He
adds two new admission conditions, so that the new set of admission conditions is as follows:

R1: If you prefer a to b and b to c, then you must prefer a to c.

R2: If you prefer c to b and b to a, then you must prefer c to a.
R3: If you prefer a to b and a to c, then you must prefer c to b.
R4: If you prefer c to a and c to b, then you must prefer a to b.

This set of conditions seems a bit odd. How would the candidates respond to it?
First, consider a “good” candidate who holds the transitive relation b  a  c (or one of the
following three other preferences : b  c  a, a  c  b or c  a  b ). None of the antecedents
of the four conditions are satisfied and responding truthfully leads to acceptance, as desired by
both the candidates and the manager.
What about a “good” candidate who holds the preferences a  b  c (or c  b  a )? If he
responds truthfully, then he violates R3 and is rejected. However, he is induced by the violation
of R3 to modify his answer to Q(b, c); this leads him to respond as if he holds the preferences
a  c  b and thus he cheats successfully.
On the other hand, a “bad” candidate with cyclical preferences a  b  c  a, (or their
counterpart) violates only R1 and is led to try only the set of answers corresponding to the
preferences a  b  c which do not satisfy the set of conditions. Thus, such a candidate fails the
test, as desired by the program director.
A happy ending, at least for the program director. Interestingly, he does not mind that some
candidates cheat. The questionnaire “works” in the sense that it separates between the good
and bad candidates. Actually, the director also finds that there is no alternative set of conditions
which will separate between the two types of candidates without some good candidates having
to cheat. He concludes that “white lies” are sometimes necessary.

Discussion
This section is based on Glazer and Rubinstein (2012). The model comes under the rubric
of Bounded Rationality since it explicitly specifies a process of reasoning used by the individ-
uals. The paper characterizes the circumstances under which the designer of the mechanism is

425
426

Ariel Rubinstein

able to implement his target given that the individuals use the discussed procedure. Conditions
under which implementation does not require that some “good” candidates need to cheat are
established as well.
No one claims that most people use this exact procedure. Nonetheless, Glazer and Rubinstein
(2012) provide experimental evidence that the two central ingredients of the procedure are
present in many people’s minds: (1) The truth is the anchor for cheating. When one invents
a false set of answers, he starts from the truth and modifies it to look better. (2) Given that an
“if ” sentence is violated, an individual who wishes to cheat successfully will modify his true set
of answers by reversing his answer to fit in the consequent of the condition. Note that without
(1) separation between the candidates would be impossible.
In the current story, the bounded rationality of the candidates is an advantage for the
designer. If all candidates were fully rational, as is commonly assumed in the mechanism design
literature, all of them would be able to game the system if necessary. It is the cognitive imper-
fection of the individuals which opens a door to obtaining a desirable outcome.

Bibliographic notes
1 Glazer and Rubinstein (2014) study a related model in which a candidate responds to a
questionnaire without being notified of the acceptance conditions though he has access to
the data about the set of responses that achieve admission and tries to make sense of things.
There is a bound on the complexity of the regularities that the candidates can detect in
the data. It is shown that whatever this bound is, the director can construct a sufficiently
complex questionnaire such that agents who respond honestly to the questionnaire will be
treated optimally and the probability that a dishonest agent will cheat successfully is “very
small”.
2 de Clippel (2014) investigates Maskin’s classical implementation question in environments
in which individuals follow systematic procedures of choice that are not consistent with
rational behavior.
3 Li (2017) is motivated by an observation that bidders in an ascending bid auction tend
to wait until the bid reaches their reservation value whereas a majority of bidders don’t
report the true value in the “equivalent” second-bid auction. An obviously dominant strategy
is one in which the decision at each decision node is “obvious” in the sense that it does
not depend on conjecture about the other players’ actions. Li identifies several targets that
are implementable using an extensive game in which the desirable outcome is obtained
through obviously dominant strategies. Glazer and Rubinstein (1996) is an earlier paper
that suggests a different criterion for simple implementation.

Elections and sampling equilibrium

Story
Three candidates are running for office. A large number of voters will participate in the elections
and each will cast one vote in favor of one of the candidates. The issues on the agenda are “hot”
and all of the voters are expected to cast their ballots. The winner will be the candidate with
the largest number of votes (even if he does not achieve an absolute majority).
The candidates are labeled as Left, Center, and Right. The population of voters is split into
three classes Leftist, Centrist, and Rightist with the proportions A, x and p:

426
427

Modeling BR in economic theory

Leftist: Individuals with the preferences L  C  R .

Centrist: Individuals with the preferences C ≻ L ∼ R .
Rightist: Individuals with the preferences R  C  L .

It is known that the Leftist and Rightist groups are more or less of equal size and that the
Centrist group is smaller but not negligible (1/3 < λ = ρ < 3/7). Some observers predict that C
will win since some of the leftists and rightists will vote for C in order to prevent the election of
the candidate on the other extreme. Some view such a result as a desirable compromise. Others
find it unacceptable that the least popular candidate might win the election.
Voters by nature tend to vote sincerely, but they will nonetheless vote for another candi-
date if they think that their candidate will certainly lose and the race between the other two
candidates is close. In such an event, they vote for their second choice.
Surveys are forbidden and voters base their prediction of the election results on occasional
discussions with casual acquaintances who express their voting intentions (without specifying
their first choice). As is often the case, people say that they “talked with several people” but
actually mean that they talked with just two. Thus, based on a sample of size two, people decide
who to vote for. Therefore, they may change their mind during the campaign according to the
sample results.
Some game theorists and political scientists analyze complicated models in which voters
think strategically, in the sense that they put themselves in the shoes of other voters, calculate
correctly the distribution over states conditional on the infinitesimal probability event that their
personal vote will be pivotal, and vote optimally given that event. No traces of such behavior
can be found in our story (nor in real life). No standard game-theoretical considerations are
used. At each moment in the campaign, each voter constructs his prediction and decides who
to vote for based on the small sample he has drawn.
Accordingly, if a voter samples two people who intend to vote for the same candidate, the
voter treats the election as already decided, concludes that his vote will not make any difference
and votes sincerely. If the two sampled individuals intend to vote for two different candidates,
he views himself as pivotal and votes for the candidate he prefers out of these two (who is not
necessarily his favorite). Thus, a C supporter intends to vote sincerely, regardless of the sample
findings. An L supporter will vote sincerely unless he draws a sample of “C and R” and then
votes for C. An R supporter will vote for C if his sample is “C and L” and sincerely otherwise.
Thus, although voters are stubborn in their political views and never change their basic
position, they may change their intention during the campaign on the basis of the sample. The
proportions of voters voting for each candidate will fluctuate until eventually stabilizing around
a distribution of votes which we call equilibrium.
Given that all C followers vote C, L supporters will never vote R and R supporters will
never vote L, a distribution of votes is characterized by two numbers: a, the proportion of the
L camp that votes C, and b, the proportion of the R camp that votes C. In equilibrium, those
proportions are stable. Notice that even when equilibrium is reached, members of the L and R
camps may still change their intentions; however, in equilibrium, the changes are “balanced”
and the distribution of votes remains constant.
Election day arrives. The ballots are closed. Will C win? Let’s examine the equilibrium.
Recall that the distribution of the L,C,R groups is (λ, 1 — 2 λ, λ). If at some point in time
the proportions of manipulative voters are a and b, then the proportions of voters’ intentions
(following the above procedure) is (λ (1 — a), λ a + (1 — 2 λ) + λb,ρ (1 — b)). The propor-
tion of L supporters who will sample “C and R” is 2[λa + (1 — 2λ) + λb][ λ (1 — b)]. The

427
428

Ariel Rubinstein

distribution is stable if this proportion is equal to a (and analogously for R). Thus, equilibrium
is characterized by two equations:
a = 2[λa + (1 – 2λ) + λb][λ(1 – b)]
and
b = 2[λa + (1 – 2λ) + ρb][λ(1 – a))].
This set of two equations has a unique solution which can easily be calculated and accordingly
the minority candidate C will win! Incidentally, when the size of C’s base drops below 1/7,
candidate C would lose, at least according to the story we have told.

Discussion
This section is based on Osborne and Rubinstein (2003). The bounded rationality component
of the model is the procedure used by each voter to estimate the results. The paper analyses
some variants of the model. An example of a result: R will not be elected whenever his base is
smaller than that of L, regardless of C’s base.
As mentioned before, the procedure of choice assumed here is in sharp contrast to what is
assumed in standard game-theoretical voting models, in which a voter is assumed to fully cal-
culate the conditions under which his vote will be pivotal given the correct expectations about
the votes of all other voters.

Bibliographic notes
1 The model was inspired by Osborne and Rubinstein (1998) who proposed the concept of
S-1 equilibrium and applied it to symmetric two-player games. A population of players is
pair-wise randomly matched to play a game. A newcomer to the population samples each
of the potential actions once by interviewing one person whose experience depends on the
behavior of the randomly matched player he played against. Then, the newcomer chooses
the action which, according to the sample, yields the best outcome. An equilibrium is a
stable distribution of choices in a population, in the sense that the probability that a player
chooses a particular strategy is equal to the frequency of the strategy in the distribution.
This equilibrium concept differs fundamentally from the standard game-theoretical ana-
lysis and has special properties. For example, a dominated strategy (which is never the best
response against any belief about the other player’s behavior) might appear with positive
probability in support of the equilibrium when the symmetric game has at least three
strategies.
2 Rani Spiegler constructed several economic models in which individuals follow an S-1 type
of procedure (see Spiegler 2011, Chapters 6 and 7). Spiegler (2006a) presents and analyses
a model of a market for quacks. The market consists of several healers whose success rates
are identical to that of “non-treatment.” The healers compete on price. Patients rely on
“anecdotal” evidence regarding the healers’ success and also sample the free non-treatment.
Spiegler (2006b) presents a model of competition between providers of a service in which
each chooses a distribution of “price” in the range (-∞, 1). Consumers sample each of
the providers and choose the one with the lowest price (which might be negative). The
uniqueness of the symmetric Nash equilibrium with expected price of 1/2 is proven.
Increasing the number of competitors leads to a mean-preserving spread in the equilibrium
price distribution.

428
429

Modeling BR in economic theory

Long interactions and finite automata

Story
Two players are involved in a repeated Prisoner’s Dilemma interaction. Each of the players
chooses one of two modes of behavior each day: C (cooperative) or D (noncooperative).
A player’s daily payoff if he chooses the action corresponding to a row while his partner chooses
the action associated with a column is given by the corresponding entry in the following
matrix:

C D

C 3 1

D 4 0

Each period ends with one of four outcomes (C, C), (C, D), (D, C), or (D, D) and the four
pairs of payoffs that can be obtained each day are (3,3), (1,4), (4,1), or (1,1). The infinitely long
interaction yields an infinite sequence of outcomes and two streams of payoffs, one for each
player. A player cares only about his long-term average payoff.
In the long-term interaction (i.e., a repeated game), each player chooses a strategy, i.e. a plan
that specifies whether to play C or D after every possible sequence of his opponent’s actions.
A strategy can be complex since it specifies an action for an infinite number of contingencies.
A player’s “language” in formulating a strategy is a finite automaton (machine). This is not an
actual machine used to play the strategy but rather an abstract description of the mental pro-
cess used by players to determine an action at each point in time after any history he might
encounter.
A finite automaton consists of:

1 a finite set of states (of mind);

2 an initial state from which the machine starts operating;
3 an output function that indicates, for each state, one action (C or D) which the player will
play whenever his machine reaches that state.
4 a transition function which determines the next state of the machine for each state and each
possible observation (C or D) of the action taken by the other player.

The two machines are operated in the expected way: The two initial states and the two output
functions determine the first-period pair of actions. Each player observes the other player’s
action and his transition function determines the next state of his machine. The process con-
tinues recursively as if each pair of states is the initial one.
Any weighted average of the four one-shot payoff pairs that equal at least 1 (a payoff each
player can guarantee by playing D) is a long-term average payoff vector of some stable pair of
strategies (stable in the Nash equilibrium sense). In other words, no player can increase his
average payoff by using a different strategy. To obtain such a pair of average payoffs (π1, π2),
players can play in a cycle of length nD,D+nD,C+nC,D+nC,C so that the outcome (i, j) will appear
ni,j times in the cycle such that

n1,1 (1,1) + n4,0 (4, 0 ) + n0,4 (0, 4 ) + n3,3 (3, 3)

= ( π1 , π 2 )
n1,1 + n4,0 + n0,4 + n3,3
.

429
430

Ariel Rubinstein

Players will support this path of play by threatening that any deviation will trigger moving to
the non-cooperative behavior forever.
However, in our story and unlike the standard repeated game model, a player cares not
only about his stream of payoffs but also about the complexity of the machine he employs,
which is measured by the number of states in the machine (without taking into account the
complexity of the transition function). Players view the complexity as a secondary consid-
eration after the desire to increase the average payoff. Thus, a pair of machines will not be
stable if one of the players can replace his machine with another that either yields a higher
average payoff or the same payoff but less complexity. In other words, each player has lexico-
graphic preferences with the first priority being to increase average payoff and the second to
reduce complexity.
Initially, the two players use a naïve strategy—the one-state machine MC (see Figure 27.1)
which plays C independently of what is observed. This situation is not stable. One of the players
modifies his machine to MD, which plays D independently of what it observes. Such a deviation
improves his payoff each period without increasing complexity.

C, D C, D C C, D

C D C D

Start Start Start D

Friendly Punishment

MC MD Mgrim

Figure 27.1

The two players then adopt the machine Mgrim. This machine has two states: friendly and
punishment. The initial state is friendly. The output function assigns C to friendly and D to pun-
ishment. In the case that D is observed when the machine is at friendly, it moves to punishment,
which is a terminal state. A player who plays against Mgrim cannot increase his long-term
average payoff whatever machine he chooses. However, eventually one of the players notices
that he can reduce the complexity of his machine without losing payoff by dropping the state
punishment which does not lead to any loss in terms of payoff.
The situation has some happier endings. Following are two examples:

Give and Take

Player 1 chooses the two-state machine M1 (see Figure 27.2) with initial state Take, in which he
plays D and moves to the second state Give after the other player plays C. If M1 reaches Give it
plays C and moves to Take independently of what it observes. Note that “giving” means playing
C and tolerating the other player playing D. “Taking” means playing D and expecting the other
player to play C. Player 2 adopts the machine M2 which is identical to M1 except that its initial
state is Take.

430
431

Modeling BR in economic theory

C, D C

D D

D C C D

Start C Start C, D
Take Give Give Take

M1 M2

Figure 27.2

Adopting this pair of machines results in stability. Players expect to alternate between giving
and taking. A player’s threat against the other player “not giving” is to avoid moving to Give, a
state in which he is supposed to give until the other player gives. As long as player 1 follows M1
and player 2 follows M2, the players will alternate turns between giving and taking and each
player obtains an average payoff of 2.5. No player can save on states without losing on payoff
and no player can increase his average payoff no matter what he does.

The cooperative equilibrium

Each player uses the machine M* (see Figure 27.3) which has two states: Showoff and Cooperative.
In Showoff, a player plays D and moves to Cooperative only if the other player plays D. In
Cooperative, the machine plays C and moves to Showoff only if the other player played D.
The machine’s initial state is Showoff. Thus, players start by proving their ability to punish one
another and then and only then move to the cooperative mode.

C C

Start D
Showoff Cooperative

D C
M*

Figure 27.3

Each player obtains an average payoff of 3. A player cannot increase his payoff whatever he
does. Reducing the number of states would cause a payoff loss (the one-state machines MD and
MC achieve an average of only 1 or 0 against M*).

Discussion
This section is based on Rubinstein (1986) and Abreu and Rubinstein (1988). The bounded
rationality component of the model is the complexity of the strategy and the desire of players
to minimize it as long as the repeated game payoff is not reduced.

431
432

Ariel Rubinstein

The two equilibria described above demonstrate the logic of two different types of social
arrangements. In the first, players alternate between giving and taking; the threat not to give
tomorrow deters the other player from not giving today. The second arrangement achieves
ideal cooperation but on the way players must demonstrate their ability to punish if necessary.
Abreu and Rubinstein (1988) characterize the Nash equilibria of the infinitely repeated
game with finite automata for (1) general one-shot two-player games; (2) discounting evalu-
ation criterion; and (3) any preferences that are increasing in the payoffs and decreasing in
complexity. The payoff vectors that can be obtained are those on the “cross” in Figure 27.4:

−1 1 2 3

−1

Figure 27.4

A few structural results were proven, such as that the two machines must have the same
number of states in equilibrium and that there is a one-to-one correspondence between the
occurrence of states in the two machines (namely, in equilibrium a machine “knows” the state
of the other player’s machine). An improved presentation of the results, following Piccione
(1992), appears in Rubinstein (1998, Chapter 8).
In a previous paper, Rubinstein (1986) showed that the only equilibria of the machine game
with the additional constraint that all states are revisited infinitely often (otherwise they would
eventually be dropped) yield either the non-cooperative outcome or cyclical combinations of
(C, D) and (D, C) only. Cooperation cannot be obtained in such an equilibrium.

Bibliographic notes
1 Neyman (1985) and Ben Porath (1993) investigate a repeated game with finite automata in
which the number of states a player can use in his machine is exogenously bounded. Neyman

432
433

Modeling BR in economic theory

(1985) shows that in the finite-horizon Prisoner’s Dilemma, cooperation can be achieved in
equilibrium by applying the idea that the machines waste their limited resources by following
an initial string of actions (“a password to heaven”) that prevents them from identifying the
point in time—toward the end of the game—at which it is profitable for them to deviate to
the non-cooperative mode of behavior without being punished in the future. Ben Porath
(1993) shows that in an infinitely repeated zero-sum game, if players use the “limit of the
means” criterion and are limited in the size of the automaton they can use, then as long as the
number of states of one machine is not much larger (in a well-defined sense explained there),
the equilibrium payoff vector will be close to the equilibrium payoff of the basic game.
2 Examples of other models of repeated games with bounds on the strategies: Lehrer (1988)
studies the model of repeated games with bounded recall and Megiddo and Wigderson
(1986) and Chen, Tang and Wang (2017) study the model of repeated games in which
players use Turing machines instead of finite automata.
3 Eliaz (2003) and Spiegler (2004) study models of repeated games and bargaining where
the simplicity considerations were applied to the player’s belief regarding the opponent’s
strategy, rather than his own.

Agents with different models in mind

Story
Each of two agents, 1 and 2, wishes to purchase a particular service each day. The service is
provided by P (a provider) but can also be obtained from a local source at a price of 6 by agent
1 and at the price of 4 by agent 2.
All involved parties interact daily for a “very long period of time.” Every morning, without
knowing the price charged by P, each agent decides whether to purchase the service from the
local source or to go to P. A cost of 3 is associated with each daily trip to P.
While the agents are making their daily decisions, P posts a price at which he commits to
sell the service to whoever asks for it. The provider P can supply the service at no cost to one
or both agents. An agent who comes to P is informed of the price offer and decides whether to
buy the service from P or return to the local price. At the end of the period, the price becomes
common knowledge to all individuals.
The provider cares only about his average long-run profits. He commits to a sequence of
prices. An agent behaves myopically in every period. He makes the trip to P only if his expected
costs (given his beliefs) do not exceed the local charge. He bases his belief on the regularities he
detects in this sequence. The ability to detect a regularity is given by a non-negative integer k
(referred to as the order) which expresses a player’s “cognitive depth”: an agent of order k detects
all regularities of the type “after a particular string of k prices the price will be ...” but cannot
detect more complicated regularities. Thus, an agent of order 0, for example, only learns the
frequency at which different prices appear in the price sequence. An agent of order 1 learns
the frequency of each price given the last price. If the real sequence is (0,0,1,1,0,0,1,1...) an
agent of order 0 or 1 will always believe that the next element in the sequence is 0 or 1 with
equal probability. An agent of order 2 (or higher) will accurately predict the next element in
the sequence since the last two periods fully “encode” the next element.
Our two agents differ in their ability to recognize patterns. While agent 1 can detect a
pattern that determines the next price based on the last three periods (k1 = 3), agent 2 is more
sophisticated (which aligns with the assumption that his outside option is better than agent
1’s) and can detect patterns on the basis of the last 4 periods (k2 = 4). Nothing essential would

433
434

Ariel Rubinstein

change in the story if we replace these bounds with any other non-negative integers k1 and k2,
as long as k1 < k2.
The provider understands that if he always charges a price higher than 3, then he will not attract
any customers; if he constantly charges a price between 1 and 3, then only agent 1 will make the
trip; and if he always charges a price of 1 or below, then both agents will buy his service but his
profits will be at most 2. Thus, by charging any constant price P his daily profits cannot exceed 3.
Furthermore, the provider realizes that if both agents have the same cognitive limit, then
using random price sequences will not help him since both agents will eventually compare their
local price to the expected price plus 3 and he would not make more than what he would have
with a constant price.
Finally, it comes to the provider’s attention that he can obtain higher profits if he wisely
exploits the differences in the agents’ cognitive ability and the correlation between an agent’s
order and the agent’s outside option, namely, that the more sophisticated agent has a better
outside option.
He learns about the existence of “DeBruijn sequences of order k” (for every k). To under-
stand this concept, consider the following infinite sequence of prices which has a cycle of
length 16 and is an example of a a DeBruijn sequence of order 4:

(5,5,5,5,1,5,5,1,1,5,1,5,1,1,1,1,....).

An agent with k = 16 would, of course, be able to detect a rule that creates this sequence and
to correctly predict the next period’s price. In fact, any agent with k ≥ 4 would also be able to
do so. The sequence is built so that all 16 combinations of 4 prices (1 and 5) appear in the cycle
exactly once and thus the last four prices predict the next element in the sequence. On the other
hand, the last three values in the sequence predict that it is equally likely that the next price will
be 1 or 5. Thus, an agent with k = 4 would be able to predict that a low price follows four high
prices, that after a sequence of three high prices and one low price always comes a high price,
etc. An agent with k = 3 will always maintain the belief that the next price will be either 1 or 5.
Thus, agent 2 will know when the low price will be posted and will approach the provider
only on mornings when he expects the price to be 1. Agent 1 is not able to find any useful
regularity. After the appearance of any 3 numbers, he will believe that there is an equal chance
of observing a high or a low price. His expected price is 3, so he every morning he approaches
the provider and buys the service.
Thus, the provider has found a non-constant sequence, which is complicated enough that
agent 1 is left confused and regular enough so that agent 2 is able to predict fairly well. The
provider’s expected profit is 7/2 per period, which is more than he could obtain using any con-
stant price sequence.

Discussion
This section is based on Piccione and Rubinstein (2003). The bounded rationality element
in this model is the limited ability of agents to recognize patterns. Once there is a correlation
between this ability and other economic factors, sophisticated manipulators can try to use the
correlation as a way to sort out agents to their benefit. Conditions for this manipulation are
analyzed in the paper.
From an economic perspective, this line of research is related to Spence (1973). However,
whereas in Spence (1973), the correlation is between two materialistic factors (productivity
and ease of performing a worthless task), here the separation between agents is based on the

434
435

Modeling BR in economic theory

correlation between a materialistic factor (the willingness to pay) and cognitive ability (to rec-
ognize patterns in a time series). Note that the separation can also emerge in market equilib-
rium without the explicit interference of an interested party.

Bibiliographic notes
1 Rubinstein (1993) is an earlier paper demonstrating the ability of a monopolist to use the
correlation between cognitive differences and willingness to pay in order to increase his
profits. The modeling of agents’ limits in understanding a multi-part price offer makes use
of the formal concept of a perceptron.
2 Eyster and Piccione (2013) is a model of competitive markets, in which riskneutral traders
trade a one-period bond against an infinitely lived asset in each period. Traders lack struc-
tural knowledge of the situation and use various incomplete theories of the type discussed
in this section in order to form statistically correct beliefs about the long-term asset price
in the next period. One of the results is that the price of the long-term asset is affected by
the diversity in the agents’ cognitive levels.
3 In a series of papers starting with Jehiel (2005), Philippe Jehiel developed the concept
of analogy-based equilibrium. The basic idea is related to the way we play games like chess.
When planning a move, we evaluate the board positions that it can lead to. In analogy-
based equilibrium, those evaluations are “correct on average.” Players differ in their par-
tition of the set of situations they bundle together, where finer partitions reflect a better
understanding of the situation.
4 Spiegler (2011, Chapter 8.) includes a pedagogical exposition of the model and links it to
other concepts like Jehiel’s.
5 Rani Spiegler suggests a general framework for studying economic agents who have imper-
fect understanding of correlation structures and causal relations (Spiegler, 2016). Spiegler
(2018) identifies a condition under which the wrong model does not allow an outsider to
systematically fool the agent.

Acknowledgments
I would like to thank Jacob Glazer, Martin Osborne, Michele Piccione and Rani Spiegler for
their comments.

References
Abreu, Dilip and Ariel Rubinstein. 1988. “The structure of Nash equilibrium in repeated games with
finite automata.” Econometrica, 56, 1259–1282.
Ben-Porath, Elhanan. 1993. “Repeated games with finite automata.” Journal of Economic Theory,
59, 17–32.
Chen, Lijie, Pingzhong, Tang, and Ruosong, Wang. 2017. “Bounded rationality of restricted Turing
machines.” In Proceedings of Thirty-First AAAI Conference on Artificial Intelligence, AAAI,
pp. 444–450.
Choi, Syngjoo, Shachar Kariv, Wieland Muller, and Dan Silverman. 2014. “Who is (more) rational?”
American Economic Review, 104, 1518–1550.
de Clippel, Geoffroy. 2014. “Behavioral implementation.” The American Economic Review, 104,
2975–3002.
Eliaz, Kfir. 2003. “Nash equilibrium when players account for the complexity of their forecasts?.”
Games Economic Behavior, 44, 286–310.

435
436

Ariel Rubinstein

Eyster, Erik and Michele Piccione. 2013. An approach to asset-pricing under incomplete and diverse
perceptions,” Econometrica, 81, 1483–1506.
Glazer, Jacob and Ariel Rubinstein. 1996. “An extensive game as a guide for solving a normal game, with
Jacob Glazer.” Journal of Economic Theory, 70, 32–42.
Glazer, Jacob and Ariel Rubinstein. 2012. “A model of persuasion with a boundedly rational agent.”
Journal of Political Economy, 120,1057–1082.
Glazer, Jacob and Ariel Rubinstein. 2014. “Complex questionnaires.” Econometrica, 82, 1529–1541.
Jehiel, Philippe. 2005. “Analogy-based expectation equilibrium.” Journal of Economic Theory, 123, 81–104.
Lehrer, Ehud. 1988. “Repeated games with stationary bounded recall strategies.” Journal of Economic
Theory, 46,130–144.
Li, Shengwu. 2017. “Obviously strategy-proof mechanisms.” American Economic Review, 107, 3257–3287.
Megiddo, Nimrod and Avi Wigderson. 1986. “On play by means of computing machines.” in Theoretical
Aspects of Reasoning About Knowledge Proceedings of the 1986 Conference, 259–274.
Osborne, Martin and Ariel Rubinstein. 1998. “Games with procedurally rational players.” American
Economic Review, 88, 834–847.
Osborne, Martin and Ariel Rubinstein. 2003. “Sampling equilibrium with an application to strategic
voting.” Games and Economic Behavior, 45, 434–441.
Piccione, Michele 1992. “Finite automata equilibria with discounting.” Journal of Economic Theory,
56,180–193.
Piccione, Michele and Ariel Rubinstein. 2003. “Modeling the economic interaction of agents with
diverse abilities to recognize equilibrium patterns.” Journal of European Economic Association, 1, 212–223.
Rubinstein, Ariel. 1986. “Finite automata play the repeated Prisoner’s Dilemma.” Journal of Economic
Theory, 39, 83–96.
Rubinstein, Ariel. 1993. “On price recognition and computational complexity in a monopolistic model.”
Journal of Political Economy, 101,473–484.
Rubinstein, Ariel. 1998. Modeling Bounded Rationality. Cambridge, MA: MIT Press.
Rubinstein, Ariel. 2012. Economic Fables. Cambridge: Open Book Publishers.
Spence, Michael. 1973. “Job market signaling.” Quarterly Journal of Economics, 87, 355–374.
Spiegler, Ran. 2004. “Simplicity of beliefs and delay tactics in a concession game.” Games and Economic
Behavior, 47, 200–220.
Spiegler, Ran. 2006a. “The market for quacks.” Review of Economic Studies, 73,11131131.
Spiegler, Ran. 2006b. “Competition over agents with boundedly rational expectations.” Theoretical
Economics, 1, 207–231.
Spiegler, Ran. 2011. Bounded Rationality and Industrial Organization. Oxford: Oxford University Press.
Spiegler, Ran. 2016. “Bayesian networks and boundedly rational expectations.” Quarterly Journal of
Economics, 131, 1243–1290.
Spiegler, Ran. 2018. “Can agents with causal misperceptions be systematically fooled?” mimeo. Available
at: papers.ssrn.com/sol3/papers.cfm?abstract_id=2807830.

436
437

28
BOUNDED RATIONALITY,
SATISFICING AND THE
EVOLUTION OF
ECONOMIC THOUGHT
Diverse concepts

Clement A. Tisdell

Introduction
Neoclassical economic models are based on the absence of any constraints on the exercise of
rationality and most rely heavily on optimization goals in order to predict economic outcomes.
This is particularly evident in traditional microeconomic theory. Most theories presume that
both consumers and producers are omniscient and not hindered in any way in making decisions
needed to achieve their optimization goals. As is well known, these assumptions are too strin-
gent to reflect reality and have probably become more so as the economic world has evolved
to become more complex.
This does not mean that neoclassical economic theories have no value for understanding the
operation of economic systems, especially market systems. Many do have predictive value, even if it
is sometimes only of a qualitative nature, e.g., prediction of price changes as a result of alterations in
the market supply or demand conditions for commodities. Furthermore, the assumption of unre-
stricted rationality optimization is stronger than is required for perfect decision making of the type
assumed in neoclassical economics (Tisdell, 1975). Nevertheless, there are many economic situ-
ations in which the presence of bounded rationality is a significant influence on economic behav-
iour, and is consequential for the evaluation of this behaviour and economic valuation.
In this chapter, an initial sketch will first be provided of the influence of the concept of
bounded rationality on the evolution of economic thought. It is then argued that it is impera-
tive to clarify the various meanings of the concepts of rationality and bounded rationality.
Attention is paid to the evolution of economic thought taking into account the diversity of these
concepts. Subsequently, the analysis considers satisficing behaviour as a reaction to bounded
rationality. Simon (1957, 1961) stressed the importance of this type of behaviour. His emphasis
on satisficing behaviour contrasts strongly with the central assumption of unbounded opti-
mizing behaviour in neoclassical economic theories. Before concluding, an overall assessment
of the place of bounded rationality in economics is provided. Note that the coverage of this
chapter is very selective because of limited space.

437
438

Clement A. Tisdell

An initial sketch of bounded rationality in economic thought

Some questioning of the applicability of the neoclassical model of economic behaviour had
begun already in the 1930s. For example, in relation to macroeconomic theory, Keynes (1936)
emphasized the importance of “animal spirits” as an influence on the behaviour of investors
and the effect of this on the level of economic activity. Hall and Hitch (1939) came to the
conclusion that cost-plus pricing was prevalent in several sectors of the economy. This was
attributed to two different possible causes: it could be a result of oligopolistic market behav-
iour. Alternatively, it might be employed as a rule of thumb because many firms lack the
capacity to determine their profit-maximizing level of pricing. Hall and Hitch (1939) found
the latter to be very important. This publication subsequently sparked debate about whether
this cost-plus procedure might, in fact, maximize a firm’s profit and whether all firms are
profit-maximizers.
Another important development in the evolution of economic thought about bounded
rationality was the publication of the theory of Games and Economic Behavior (von Neumann and
Morgenstern, 1944). It highlighted limits to the exercise of unrestricted individual rationality as
a means for providing solutions to group behaviour involving conflict. In addition, it helped to
explain failures to achieve socially optimal group outcomes, such as Pareto optimality (Tisdell,
1996). For example, it indicated that under omniscient conditions for decision making, economic
behaviour cannot be precisely determined, as in the case of zero-sum games where the solution
relies on mixed strategies and in empty-core games involving the possibility of transactionless
coalition formation. This development of the theory of games prompted Simon (1955) to doubt
the applicability of the unrestricted neoclassical rationality assumption as a determinant of eco-
nomic behaviour, and he further developed his concept of bounded rationality as being more
relevant. It is, however, pertinent to note that Morgenstern was aware of limits to the applicability
to economics of the neoclassical concept of unlimited rationality (Morgenstern, 1964).
As an alternative approach to economic behaviour, Herbert Simon proposed a satisficing
theory of decision making and economic behaviour. The cost of obtaining information, of
retaining it, and of reasoning, was seen as an important restriction on the ability of individuals
to make absolutely optimal (perfect) decisions of the type assumed in neoclassical economics.
The assumption of satisficing behaviour was subsequently applied to consumer behaviour and
to the theory of the firm, particularly to the latter.
Another significant but embryonic development was the publication of an article by Baumol
and Quandt (1964) outlining a theory of optimally imperfect decisions. This provided insights
into how much information gathering, the amount of its retention, and the extent of reasoning
in decision making are likely to be economically optimal. They pointed out that the use of
some rules of thumb (such as those employed by some firms in adopting cost-plus pricing)
could be rational from an economic optimization perspective.
The type of modelling of Baumol and Quandt (1964) belongs to the class of modelling of
bounded rationality sometimes described as optimization under constraint, that is, optimiza-
tion which takes into account the cost of decision-making. An earlier example of this was
the stopping rule of Stigler (1961) for searching for the purchase of a used car, namely, stop
searching when the extra cost of searching equals the extra expected benefits. Similar sorts of
stopping rules have been adopted for quality control involving serial sampling by producers
and are relevant to other types of sampling. However, Gigerenzer and Selten (2002, pp. 4–5)
question whether this type of optimization modelling captures the essence of the occurrence of
bounded rationality. This issue will be considered in the discussion section.

438
439

BR, satisficing and economic thought

Tisdell (1963; 1968; 1996; 2013) emphasized that, as a result of bounded rationality (and
other factors), economic behaviours can be expected to be diverse and that this diversity had
been neglected in neoclassical theory. Nevertheless, diverse behaviours have predictable eco-
nomic consequences. Empirical investigations are needed to determine the extent of that diver-
sity and its other attributes such as its variation with the passage of time. Studies in behavioural
economics, experimental economics and psychological economics are all relevant to exploring
this aspect of economics.
Another area of economic thought (which has advanced as a result of giving attention to the
occurrence of bounded rationality) is whether it is more desirable to follow rules rather than
discretion in decision making, that is, the benefits of engaging in flexible or less flexible types of
decision making. The more flexible type of decision making may, for example, involve adjusting
controlled variables based on short-term predictions of uncontrolled variables in an attempt to
achieve a particular objective. The size and nature of the divergence between the predicted
and actual values of controlled variables may be such that greater benefit can be obtained by
ignoring short-term predictions and acting on long-term central tendency predictions, for
example, predicted central values of the uncontrolled variables (or approximations to these)
(Tisdell, 1971).
Friedman (1968) pointed out that following rules rather than engaging in discretionary
zig-zag or fine-tuning behaviour could result in a more desirable type of monetary policy.
Tisdell (1971, 1974), in criticizing Muth’s theory of rational expectations (Muth, 1961),
came to a similar conclusion and pointed out that this also applied to other areas of eco-
nomics as well.
Rapid development of other areas of economics associated with the concept of bounded
rationality occurred in the 1970s. Considerable attention was paid to how the transac-
tion costs involved in economic organization resulted in participants in economic activity
having incomplete knowledge (see, for example, Williamson, 1975). Issues such as the
following were highlighted: principal-agent problems, the incompleteness of contracts and
the importance of trust in exchange (Williamson, 1975, 1979) and the possibility of market
collapse or the impeded operations of markets due to the asymmetry of information of
market participants (Akerlof, 1970). These phenomena were shown to be a source of poten-
tial economic loss.
Neoclassical economic theories of behaviour were mostly based on introspection but also
obtained empirical support from observations of the operations of markets. This was partly
because the assumption of unrestricted rationality is stronger than necessary for qualitative
predictions about how many markets work (Tisdell, 1975). Nevertheless, it became increas-
ingly clear that not all economic phenomena could be understood or predicted by relying on
the assumption of unrestricted (unbounded) rationality. This led to an upsurge in the devel-
opment of behavioural economics and psychological economics and to increasing attention
being paid to experimental economics. However, questions have been raised about whether
many of the advances made in these fields of inquiry are consistent with the theory of bounded
rationality. For example, Gigerenzer and Selten (2002, p. 4) claim they are not, and they also
argue that optimizing theories based on decision making under constraints do not reflect the
essence of bounded rationality. In order to help clarify this problem, it is helpful to consider
the different meanings of the word “rational” as specified in English dictionaries. This will
provide scope for some discussion of additional theories of economic behaviour which have
been associated with bounded rationality as well as consideration of the concept of ecological
rationality.

439
440

Clement A. Tisdell

Different meanings of rationality and the further development of economic

thought, including the concept of ecological rationality
The word “rationality” can have several different meanings in English, according to English
dictionaries. Two relevant different meanings, as stated in the Australian Macquarie Dictionary
(Delbridge, 1981) are:

A decision or behaviour is rational if it involves reasoning (Type 1 rationality).1

A decision or behaviour can be judged to be rational if it is reasonable or sensible (Type 2
rationality).2

The New English Dictionary and Thesaurus (Anon, 1994) gives the same interpretations of the
word “rationality”. Given the first interpretation, a behaviour that does not involve reasoning
would not be a rational form of behaviour. Nevertheless, it could be effective for particular
purposes, such as instinctive behaviour is in particular circumstances and is judged to be rational
in the second sense of the word. The second meaning involves judgement about whether a
behaviour is reasonable. Whether or not a behaviour is judged to be reasonable or sensible can
depend upon the circumstances surrounding the behaviour (that is, the environment). Given
the second meaning, behaviour that does not involve forethought or reasoning can be rational.
Moreover, given this meaning, “excessive” forethought or reasoning in decision-making is not
reasonable and therefore is irrational.
Not all decisions and behaviours which are effective in achieving a desired outcome are
based on reasoning. For example, some instinctive and emotional behaviours are effective in
particular circumstances for achieving desired ends. They are not a result of type 1 rationality,
but may satisfy type 2 rationality.
Given the two meanings of rationality outlined above, it is clear that the extent to which
rationality is present in decision making and behaviour can vary in degrees. Moreover, the
presence of rationality in the second sense outlined above is subject to personal judgement. For
example, much of the study of ecological rationality is focused on non-optimizing behaviours
which are sensible (given bounded rationality) and behaviours that are effective for some par-
ticular purpose given the surrounds in which they occur. Gerd Gigerenzer is a prominent advo-
cate of this approach which concentrates on type 2 rationality.
Gigerenzer and Selten (2002, p. 38) describe ecological rationality as “the match between
heuristics and environmental structures” and indicate that this requires paying particular
attention to satisficing behaviours as part of search and decision making and the adoption of fast
and frugal heuristics, for example, involving the use of cues in making decisions. However, this
ecological approach is even wider than this because it judges some behaviours to be rational
which do not involve heuristics or rules of thumb. These can include some forms of intuitive
and emotional behaviour as well as various social norms. These behaviours are considered to be
rational if they are effective in achieving a desired end or purpose.3 Also, the Gigerenzer group
of ecological rationalists is aware that the amount of thought it is rational to give to a decision
depends on the time constraint faced by decision makers. However, this group rejects the rele-
vance of economic models of optimization based on constrained decision making and even
more strongly rejects the neoclassical model of unbounded rationality (Selten, 2002). They also
cast doubts on the relevance to bounded rationality of behavioural psychological studies, such
as those associated with Kahneman (2003) (Gigerenzer and Selten, 2002, p. 4).
An overlapping but narrower view of ecological rationality is adopted by Vernon Smith
(2003). He describes ecological rationality as “an emergent order based on trial-and-error

440
441

BR, satisficing and economic thought

cultural and biological evolutionary processes” (Smith, 2003, pp. 499–500). His primary con-
cern is with group rationality in economics. He rejects the relevance of constructivist ration-
ality, namely, that social mechanisms are as a rule thoughtfully created to serve a perceived
intended purpose (Smith, 2003, p. 470). In general, he believes that evolutionary processes
and trial-and-error processes are effective in developing optimal social rules of behaviour and
social norms that are beneficial in promoting desirable social ends. However, his view is too
sweeping. In the past, some societies developed social norms and religious beliefs which did
not promote desirable social ends. Examples of this have been proposed by Diamond (2011).
These include the deforestation of Easter Island (Rapa Nui in the Pacific Ocean) by its ori-
ginal inhabitants and the Mayan collapse. Several other societies have engaged in persistent
irrational behaviours – some as a result of their religious beliefs or their adoption of forms of
unsustainable economic development – which eventually proved to be catastrophic from their
point of view, for example, early producers of copper and bronze in central Europe (Tisdell and
Svizzero, 2018). Today there are concerns that we may not be able to establish effective norms
and behaviours to restrict global warming “adequately”. Theoretically, there is no guarantee
that selective evolutionary processes will result in the prevalence of “optimal” decisions or even
socially satisfactory ones (Tisdell, 2013).4
Another approach to considering the consequences of bounded rationality in economics
is based on the unearthing biases in economic behaviour. This type of approach to bounded
rationality was initially developed by Amos Tversky and Daniel Kahneman. Kahneman (2003,
p. 1449) explains that this type of research about bounded rationality explores “the systematic
biases that separate the beliefs that people have and the choices they make from optimal beliefs
and choices assumed in rational-agent choice models”. These biases are mainly identified by
relying on experiments but may also be discovered by considering observations from non-
experimental situations (Camerer et al., 2003).
One of the significant results from this line of enquiry, which has extended the findings of
Thaler (1980) demonstrates the importance of loss aversion, endowment or status quo effect
as an influence on several types of economic decision-making. Kahneman (2003, p. 1457)
explains that this effect is present when “the value of a good to an individual appears to be
higher when the good is viewed as something that could be lost or given up when the same
good is evaluated as a potential gain”. Examples are given by Kahneman (1990), Tversky and
Kahneman (1991) and Kahneman et al. (1991). Bandara and Tisdell (2005) found evidence of
the importance of this effect in relation to willingness to pay for the conservation of elephants
in Sri Lanka. The status quo effect is not allowed for in traditional economic theory. There
can be several reasons for the occurrence of this effect, for example, a psychological desire to
keep valued items which one already has (possessiveness), transaction cost considerations, and
the possibility that the consumption or enjoyment of the commodity alters the taste of the
possessor. These aspects require further investigation.
Another aspect of bounded rationality which has been paid much attention by contributors
to behavioural economics is the importance of frames in shaping decisions (Kahneman, 2003).
The emphasis, in this case, is on how individuals perceive alternative possible states of nature,
or more generally, possibilities. Both the selective nature of perceptions and their distortions are
studied. These aspects of perception are relevant for predicting economic behaviours and also
for assessing the worth and limitations of economic valuation studies, particularly those valuing
alterations in the supply of public and quasi-public goods, especially environmental goods.
Results from these investigations (and other types of studies) demonstrate that economic valu-
ation (reliant on the assumptions of neoclassical economics) of the demand for changes in the
supply of public goods, particularly environmental goods, can be problematic.

441
442

Clement A. Tisdell

Satisficing and bounded rationality

Simon (1957; 1961) stressed that, because of the presence of bounded rationality, individ-
uals and organizations often adopt satisficing behaviours or rules rather than optimizing. In
principle, satisficing behaviours which occur because of bounded rationality can take sev-
eral different forms. Some such behaviours may be based on aspiration levels or targets for
performance. These are usually not optimal in the neoclassical sense. The degree to which
these targets are adaptive is liable to vary. Other behaviours of a satisficing type may rely
on rules of thumb or heuristics and limited cues about states of nature. If they give satisfac-
tory benefits, their use may persist. However, this is not always so. Searches may continue
for more effective rules of thumb, heuristics and cues, especially if the environments in
which decisions are to be made alter. The adaptive rationality of heuristic decision making
is important.
Not all satisficing behaviour is a result of bounded rationality, even though some forms are at
odds with the underlying assumptions of neoclassical economic theory. (Bendor, 2015, p. 774)
states, for example:

Bounded rationality should not be confused with a theory (e. g., of satisficing),
much less with a specific formal model (e.g., Simon, 1957). It is best considered a
research program: a sequence of theories with overlapping sets of assumptions, aimed
at solving similar problems … In principle, the program’s empirical domain is vast—it
is as imperialistic as the rational choice program—and so its set of possible theories is
also very large.5

Baumol’s theory of behaviour of an imperfectly competitive corporation assumes that the

company tries to maximize the value of its sales subject to ensuring that its shareholders receive
a satisfactory level of return on their capital (Baumol, 1959).6 In this instance, bounded ration-
ality is not involved in this constrained optimization theory. Sahlins’ theory of the affluence
of some ancient societies supposes that members of these societies were completely satisfied
with their low level of consumption of material goods (Sahlins, 1972). It does not rely, per se,
on any assumptions about bounded rationality (Tisdell, 2018). Some theories also exist which
suppose that the utility obtained by individuals is a function of the difference between the
level of the income to which they aspire (or some other economic variables) and the levels
achieved. This is, for example, a component of Weckstein’s model (Weckstein, 1962). Such
constructs may but need not involve bounded rationality.7 These types of models are of par-
ticular interest because they raise questions about how aspiration levels are determined and
adjusted.
Another relevant rationality aspect of satisficing goals is the extent to which possible failure
to achieve these goals should be foreshadowed and when and in what depth plans ought to
be drawn up to address this possibility? To what extent is delay in decision making a result
of bounded rationality or due to other factors such as social embedding? These two types of
behaviours are, for example, highlighted by Tisdell and Svizzero (2017) in their discussion of
the transition of ancient societies from hunting and gathering to agriculture.

Discussion
Gigerenzer and Selten (2002) suggest that two sets of models that are often discussed under the
banner of bounded rationality have not been appropriately classified. They state:

442
443

BR, satisficing and economic thought

Bounded rationality is neither optimization nor irrationality. Nevertheless, a class of

models known as optimization under constraints is referred to in the literature as ‘bounded
rationality’, and the class of empirical demonstrations of ‘so-called’ errors and fallacies
in judgment and decision-making has been labeled ‘bounded rationality’. The fact
that these two classes have little if anything in common reveals the distortion that the
concept of bounded rationality has suffered.
Gigerenzer and Selten, 2002, p. 4

However, the first set of models is relevant to the study of bounded rationality because they
highlight limits to the neoclassical vision of unrestricted rationality. While their knowledge
and rationality assumptions are still too strong, they can help to identify factors that ought to
influence behaviours under conditions of restricted rationality. As for the second class of models
(which include behavioural ones), most (but not all) identify limits to perceptions of states of
nature and common faults in reasoning, both of which can be considered to be a consequence
of bounded rationality. These classes of models (mostly behavioural economic ones) do demon-
strate some of the limits to unrestricted rationality and can have predictive value.
One of the important consequences of bounded rationality is that it gives rise to variations
or differences in the behaviours of individuals and groups. Individuals differ in the perception
of states of nature, in the estimates of probabilities and risk, and in their willingness to take
risks. This aspect has been stressed by Tisdell (1963, 1968) and in some of his later publications.
These variations have predictable economic consequences but they have not been given enough
attention in the economic literature. Bendor (2015, p. 774) mentions that behavioural (eco-
nomic) theories do not pay enough attention to differences in human behaviour. Differences
and changes in behaviours all have important economic consequences.
The time available for decision making limits the scope for the gathering of informa-
tion and reasoning, as stressed by Selten (2002). Sometimes, there is a definite end-point by
which a decision must be made and action taken. In extreme cases, there may be little or no
time available for rational decision making involving data collection and thought. In these
cases, action may be dictated by instinct or learned reactions. In other cases, a final decision
may be delayed, resulting in both benefits and costs. Some of the factors that influence the
optimality of delayed decisions have been examined by Tisdell (1970; 1996, Chapter 5).
Where decisions may have to be made at short notice, it can be rational to prepare for these
in advance of their possible occurrence. However, the amount of rational preparation can be
expected to vary.
The value of the strategy of delaying decisions to gain extra information depends on the
environmental scope for responding to this information, as does the ability of decision makers
to take advantage of changes in economic information (Tisdell, 1970, 1996, Chapter 5).
Therefore, apart from collecting more information, a rational response to bounded rationality
can be to alter the environment in which decisions can have effect, for example, it may be pos-
sible to change existing environments to allow greater flexibility for responding to decisions.
Examples of this include the adoption of production processes (techniques) that exhibit greater
adaptability than otherwise in the production of different commodities (Tisdell, 1963, 1968)
and an increasing liquidity of assets to take advantage of varying investment opportunities
which are subject to uncertainty. However, changing economic environments in this way usu-
ally comes at a cost. Therefore, analysis and judgement are necessary to decide whether acting
in this way is worthwhile.
A related concept in environmental economics is the precautionary principle (Tisdell, 2010,
2015). If unrestricted rationality occurred, this principle would be irrelevant. Because the

443
444

Clement A. Tisdell

environmental future is uncertain, it becomes relevant. One of the manifestations of the prin-
ciple is that in view of uncertainty, it is often desirable to keep options open, for example, con-
serve biodiversity. This permits advantage to be taken of new information which may become
available in the future. Once again, environmental variation may be made in order to provide
greater flexibility in available decision-making strategies. This may come at a cost, and how
sensible it is depends on attitudes to the bearing of risk or uncertainty and the anticipated net
benefits from increased flexibility.
Gigerenzer and others have emphasized the importance of fast and frugal heuristics and
the use of selective cues as a guide to behaviour. Presumably, the usefulness of these as rea-
sonable guides to behaviour depends on the economic situation that is being responded to.
Considerable use of these types of heuristics appears to be made in trading in financial markets,
for example, trading on the stock exchange. Identifying the types of cues that traders use for
exchange in these markets (and other markets) is also important for predicting and evaluating
their consequences. This opens up a large area for empirical economic research. This is par-
ticularly so because different individuals and groups may employ different rules of thumb and
cues. Mixed behaviours of market participants as well as alterations in the diversity of these
behaviours can be important for the operation of markets (Lasselle et al., 2005; Tisdell, 2013).

Conclusion
Neoclassical economic theory pays no attention to the costs and other restrictions on rational
decision making and has therefore developed optimizing models of economic behaviour which
assume unrestricted rationality. These models are, in fact, special cases. Nevertheless, they
do have some predictive value because their assumptions are stronger than is necessary for
forecasting or explaining some types of economic behaviour and for providing a guide to how
some markets work. On the other hand, it is a mistake to assume that all economic behaviour
reasonably accords with that assumed in neoclassical economic theory. The realization of this
has resulted in substantial progress in economic thought in recent decades and has created a new
academic environment in which further progress is being facilitated, for example, as a result of
joint contributions by psychologists, economists and others.
In this short chapter, it has not been possible to consider all the advances in economic
thought which have stemmed from research on bounded rationality. Much of this research is
based on examining particular situations. We are now challenged to determine whether general
principles can be distilled from these studies. It is also important that greater attention be paid
analytically to determine how reasonable or sensible decisions made and behaviours observed
under conditions of bounded rationality are, that is to go beyond the empirical determination
of the impact of bounded rationality on behaviours and decision-making. For example, to what
extent can the rules of thumb used for economic decision-making be improved or replaced
by ones that are more effective in achieving desired goals? To what extent are decision-makers
cognisant of the factors which ought to guide their decisions when they are acting under
bounded rationality and is there scope for them to improve their decisions by paying greater
attention to such factors?
The presence of bounded rationality has also created complications for methods derived
from neoclassical economics of valuing public goods and experiential goods, especially environ-
mental commodities. Results from the application of these methods (both revealed and elicited
preference methods) need to be treated with caution given the presence of bounded rationality
and the occurrence of biases in observed behaviours and in responses to elicitation of values
(see, for example,Tisdell, 2017, Chapter 4, especially Table 4.1; as well as Tisdell, 2014). The

444
445

BR, satisficing and economic thought

challenge now is to determine what use can sensibly be made of these results for the purpose
of social economic valuation.

Notes
1 A problem with this meaning is that not all decision making and behaviour involving reasoning would
normally be regarded as rational. For example, illogical reasoning, failure to assess the environment
appropriately, excessive or insufficient reasoning may be sources of irrationality. Also, this meaning
seems to imply conscience deduction but, as mentioned in the text, some behaviours that do not
involve reasoning could (in particular circumstances) be rational.
2 This meaning enables a wider range of behaviours to be regarded as rational than the previous one. It
leaves open the question of who is to judge whether a decision or behaviour is sensible or reasonable.
A reviewer of the draft of this chapter suggested that this should be mainly judged by the decision
maker. However, in most societies, others are mainly called on to judge the rationality of decisions or
behaviours of individuals or groups. For instance, a clinically insane person may believe that he/she is
asking rationally but others may judge otherwise.
3 A problem here is that some of these behaviours may only be effective for a limited period of time and
they trap individuals in a long-term situation which is unsatisfactory. Furthermore, some behaviours
could be effective by chance for a limited period of time, e.g., forms of gambling or risk-taking. Even
though they may be effective for a while, they are not rational from a long-term point of view.
4 Social embedding of various kinds can result in societies being unable to take advantage of the mistakes
or errors of earlier societies or groups. See, for example, Tisdell (2017, Chapter 5). Also, as soci-
eties evolve and become larger, their proneness to prisoner dilemma-type problems may increase, as is
illustrated by a number of sustainability issues including those associated with human-induced climate
change.
5 While satisficing behaviours are only one of the many ways of responding to bounded rationality, not
all satisficing behaviours are a consequence of bounded rationality. Satisficing behaviour, for example,
occurs if there are thresholds in preference functions and uncertainties are absent.
6 The level of this satisfactory return from the point of view of shareholders is the prevailing rate of return
on capital in the capital market.
7 Weckstein does not include uncertainty and bounded rationality in his modelling of decision-making.
For further discussion of Weckstein’s model, see Tisdell (1983) or Chapter 14 in Tisdell (1996).
Although Weckstein does not incorporate bounded rationality into his type of modelling, it can be
allowed for in these types of models. For instance, the difference between what is aspired to and what
can be achieved is often uncertain. Consequently, a dynamic interplay is frequently observed between
these two behavioural components with adjustment being made to aspirations in the light of learning
about what it is possible to achieve.

References
Akerlof, G. (1970). ‘The market for lemons: quality and the market mechanism’, The Quarterly Journal of
Economics, 84(3), 488–500. doi:https://doi.org/10.2307/1879431.
Anon (1994). New English dictionary and thesaurus, New Lanark: Geddes and Grosset.
Bandara, R., and Tisdell, C. (2005). ‘Changing abundance of elephants and willingness to pay for
their conservation’, Journal of Environmental Management, 76(1), 47–59. doi:10.1016/j.jenvman.
2005.01.007.
Baumol, W. J. (1959). Business behavior, value and growth, New York: Macmillan.
Baumol, W. J., and Quandt, R. E. (1964). ‘Rules of thumb and optimally imperfect decisions’, American
Economic Review, 54, 23–46.
Bendor, J. (2015). ‘Bounded rationality’, in International encyclopedia of the social & behavioral sciences
(pp. 773–776). Oxford: Elsevier.
Camerer, C., Loewenstein, G., and Rabin, M. (eds) (2003). Advances in behavioral economics.
New York: Russell Sage Foundation/Princeton, NJ: Princeton University Press.
Delbridge, A. (1981). The Macquarie dictionary, St Leonards, NSW, Australia: Macquarie Library.
Diamond, J. (2011). Collapse: How societies choose to fail or succeed, Revised ed. New York: Penguin.
Friedman, M. (1968). ‘The role of monetary policy’, The American Economic Review, 58(1), 1–17.

445
446

Clement A. Tisdell

Gigerenzer, G., and Selten, R. (eds) (2002). Bounded rationality: The adaptive toolbox, Cambridge, MA:
MIT Press.
Hall, R., and Hitch, C. (1939). ‘Price theory and business behaviour’, Oxford Economic Papers, 2, 12–45.
Kahneman, D. (1990). ‘Experimental tests of the endowment effect and the Coase theorem’, Journal of
Political Economy, 98(6). 1325–1348. doi:10.1086/261737.
Kahneman, D. (2003). ‘Maps of bounded rationality: Psychology for behavioral economics’, American
Economic Review, 93(5). 1449–1475. doi:10.1257/000282803322655392.
Kahneman, D., Knetsch, J. L., and Thaler, R. H. (1991). ‘The endowment effect, loss aversion and status
quo bias’, Journal of Economic Perspectives, 5, 193–206.
Keynes, J. M. (1936). The general theory of employment, interest and money, London: Macmillan.
Lasselle, L., Svizzero, S., and Tisdell, C. (2005). ‘Stability and cycles in a cobweb model with heteroge-
neous expectations’, Macroeconomic Dynamics, 9(5), 630–650. doi:10.1017/s1365100505050017.
Morgenstern, O. (1964). ‘Pareto optimum and economic organization’, in W. Krelle and H. Müller (eds),
Systeme und Methoden in den Wirtschafts- und Sozialwissenschaften, Tübingen: J.C.B. Mohn.
Muth, J. F. (1961). ‘Rational expectations and the theory of price movements’, Econometrica, 29, 315–335.
Sahlins, M. (1972). Stone Age economics, Chicago: Aldine de Gruyter.
Selten, R. (2002). ‘What is bounded rationality?’, in G. Gigerenzer and R. Selten (eds), Bounded ration-
ality: The adaptive toolbox (pp. 13–36). Cambridge, MA: MIT Press.
Simon, H. (1955). ‘A behavioral model of rational choice’, The Quarterly Journal of Economics, 5, 99–118.
Simon, H. (1957). Models of man, New York: John Wiley & Sons, Ltd.
Simon, H. (1961). Administrative behavior, New York: The Macmillan Company,
Smith, V. L. (2003). ‘Constructivist and ecological rationality in economics’, American Economic Review,
93(3), 465–508. doi:10.1257/000282803322156954.
Stigler, G. J. (1961). ‘The economics of information’, Journal of Political Economy, 69(3), 213–225.
doi:10.1086/258464.
Thaler, R. (1980). ‘Toward a positive theory of consumer choice’, Journal of Economic Behavior and
Organization, 1(1), 39–60. doi:10.1016/0167–2681(80)90051–7.
Tisdell, C. A. (1963). ‘Price uncertainty, production and profit’, PhD thesis, Australian National University,
Canberra.
Tisdell, C. A. (1968). The theory of price uncertainty, production and profit, Princeton, NJ: Princeton
University Press,
Tisdell, C. A. (1970). ‘Implications of learning for economic planning’, Economic Planning, 10(3), 177–192.
Tisdell, C. A. (1971). ‘Economic policy, forecasting and flexibility’, Weltwirtschaftliches Archiv-Review of
World Economics, 106(1), 34–54.
Tisdell, C. A. (1974). ‘Comments on Muth’s note on economic policy, forecasting and flexibility’, Review
of World Economics, 110(1), 176–177.
Tisdell, C. A. (1975). ‘Concepts of rationality in economics’, Philosophy of the Social Sciences, 5(3), 259–272.
Tisdell, C. A. (1983). ‘Dissent from value, preference and choice theory in economics’, International Journal
of Social Economics, 10(2), 32–43.
Tisdell, C. A. (1996). Bounded rationality and economic evolution, Cheltenham: Edward Elgar.
Tisdell, C. A. (2010). ‘The precautionary principle revisited: Its interpretations and their conservation
consequences’, Singapore Economic Review, 55, 335–352.
Tisdell, C. A. (2013). Competition, diversity and economic performance: Processes, complexities, and ecological simi-
larities, Cheltenham: Edward Elgar.
Tisdell, C. A. (2014). Human values and biodiversity conservation, Cheltenham: Edward Elgar.
Tisdell, C. A. (2015). Sustaining biodiversity and ecosystem functions: Economic issues, Cheltenham: Edward Elgar.
Tisdell, C. A. (2017). Economics and environmental change: The challenges we face, Cheltenham: Edward Elgar.
Tisdell, C. A. (2018). ‘The sustainability and desirability of the traditional economies of Australian
Aborigines: Controversial issues’, Economic Analysis and Policy, 57, 1–8. doi:org/10.1016/j.eap.
2017.11.001.
Tisdell, C. A. and Svizzero, S. (2017). ‘Optimization theories of the transition from foraging to agricul-
ture: A critical assessment and proposed alternatives’, Social Evolution and History: Studies in the
Evolution of Human Societies, 16(1), 3–30.
Tisdell, C. A., and Svizzero, S. (2018). ‘The economic rise and fall of the Silesian Únětice cultural
populations: A case of ecologically unsustainable development?’, Anthropologie, 56(1), 21–38.
Tversky, A., and Kahneman, D. (1991). ‘Loss aversion in riskless choice: A reference-dependent model’,
The Quarterly Journal of Economics, 106(4), 1039–1061.

446
447

BR, satisficing and economic thought

von Neumann, J., and Morgenstern, O. (1944). Theory of games and economic behavior, Princeton,
NJ: Princeton University Press.
Weckstein, R. S. (1962). ‘Welfare criteria and changing tastes’, The American Economic Review, 52(1),
133–153.
Williamson, O. E. (1975). Markets and hierarchies: Analysis and anti-trust implications, New York: Free Press.
Williamson, O. E. (1979). ‘Transaction-cost economics: The governance of contractual relations’, The
Journal of Law and Economics, 12(2), 233–262.

447
448

29
BEYOND ECONOMISTS’
ARMCHAIRS
The rise of procedural economics

Shabnam Mousavi and Nicolaus Tideman

Outcome vs. process in modeling choice

The paradigm of mainstream economics is a coherent one built logically on a substantive notion
of rationality, with expected utility theory as its crowning achievement. As Gary Becker (1962)
neatly summarized, “now, everyone more or less agrees that rational behavior simply implies
consistent maximization of a well-ordered function, such as utility or profit.” In this frame-
work, an actor seeks the best or optimal outcome, and specifying the criteria for its existence
and (preferably) uniqueness occupies theorists and modelers, who rely mainly on deductive
methods. Issues of how to collect data for testing the theory, or building models to achieve
concrete real-world goals are usually not of primary concern. The search for information is also
assumed to be optimal. That is, the rational agent uses an optimal stopping rule, continuing to
calculate marginal costs and marginal benefits of further search, and stops when they are equal.
This is acknowledged to be an artificial search process, with no claim to represent the actual
search process. Understanding reality is thus pursued through “stylized facts”1 (Abad & Khalifa,
2015) instead of through observation. Clever stylized facts, “as-if ” locutions (Lehtinen, 2013),
and coherent axioms are the bread and butter of the smooth operationalization of concepts in
the realm of general equilibrium economic models and subjective expected utility, wherein
polished artefacts are used to account for the behavior of utility-maximizing individuals and
profit-maximizing firms. Game theory from mathematics (von Neumann & Morgenstern,
1944), equilibrium analysis akin to physics (Arrow & Debreu, 1954), and the axioms of statis-
tical decision theory (Savage, 1954) allowed economics to claim the status of being a science.2
The social legitimacy of this claim was sealed when, in 1968, Riksbank, the Swedish Central
Bank, established an endowment in perpetuity with the Nobel Foundation, to award a presti-
gious prize in Economic Sciences. Interestingly, the prize was awarded not only to economists
and mathematicians, constructors of the rational core of the economics discipline, e.g., Arrow
in 1972, Debreu in 1983, Becker in 1992, Nash in 1994, but also to scientists who took alter-
native views on the topic, the most famous of them arguably being Herbert Alexander Simon
in 1978 (after receiving the Turing Award in 1975, together with Newell).
In his Nobel Prize lecture, Simon, (1978b) propounded: “There are no direct observations
that individuals or firms do actually equate marginal costs and revenues.” That is, real-world
individuals (firms) do not show indications of utility (profit) maximization with unbounded

448
449

The rise of procedural economics

rationality. The actual form of observed rationality exerted by individuals and firms has a
bounded nature.
The initial thrust of bounded rationality was to question the relevance of (subjective)
expected utility calculations for observed human behavior. The general features of a boundedly
rational agent, sans the term itself, are described as administrative behavior in Simon (1955).
The term “principle of bounded rationality” first appeared in the Models of Man (Simon, 1957,
p. 200), and later in a chapter (Simon, 1972) entitled “Theories of bounded rationality.” Simon
viewed economists’ detachment from observed behavior as a serious flaw in their methodology.
He advocated the tradition in psychology, of starting from evidence for studying, modeling,
and theorizing about human choice behavior (Barros, 2010). In an interview with Challenge in
1986, he elaborated:

They [economists] don’t talk about evidence at all. You read the pages where Lucas
talks about why businessmen can’t figure out what’s happened to prices and it is just
what he feels as he sits there smoking his cigarettes in his armchair. I don’t know what
Keynes smoked, but when you look at the pages where he talks about labor’s money
illusion, no evidence is cited. So the real differences in economics, as compared with
psychology, is that almost everybody operates within the theoretical logic of utility-
maximization in the neoclassical model.
When economists want to explain particular phenomena in the real world, they
have to introduce new assumptions. The distinctive change in the behavior of the
economic actors comes from a change in the behavioral assumptions. No empirical
evidence is given to support those changes. They emerge from the mind of the eco-
nomic theorist sitting in his armchair.
The Swedish economist Assar Lindbeck recently did a survey of some 30 business-
cycle theories. When I read it, I saw a pattern emerging as one theorist after another
attempted to explain how unemployment developed. Each introduced into his
model some particular behavioral departure from perfect economic rationality. I'm sure each
departure led to a journal article on business cycles, and in response, other economists
soon joined the debate. All of this theory and debate was over assumptions, but not
one was bolstered by evidence from empirical observation. Economists treat behavior
as if it were in fact the right behavior for the actual circumstances: that the world is
out there, people see that world accurately, understand that world, and adapt perfectly
to it. That is what I call “substantively rational behavior.”
Challenge, 1986, pp. 22–23

Background and focus

This chapter discusses the trend of the integration of bounded rationality concepts into the
economists’ toolkit. In the early 2000s, entertaining the possibility of axiomatizing bounded
rationality as a primary framework—as opposed to formalizing it as partial deviations from full
rationality—produced a doctoral dissertation (Mousavi, 2002) for one author of this chapter
under the supervision of the other author. The core argument of that dissertation was that
parts of the initial idea of bounded rationality do not lend themselves to the tools that popu-
late the paradigm of full or substantive rationality, namely, optimization techniques and the
focus on existence (and uniqueness) of answers. Alternatively, in an imaginative new para-
digm where goals not only can be multiple and incommensurable, but also do not need to
be predetermined, bounded rationality can be formalized. One suggested means of doing so

449
450

Shabnam Mousavi and Nicolaus Tideman

entails turning the focus from cognitive limitations to the nature and structure of goals pursued
by real-world boundedly rational agents.
Another central argument arose from the observation that fundamental uncertainties of situ-
ations, the ones that cannot be reduced to risk calculation, do not paralyze decision makers.
This fact seems to be lost on economic modelers, who focus intently on reducing uncer-
tainty in formal representations of choice behavior. But people make simple and hard choices
every day without reducing all aspects of their situations to ones with probabilities assigned to
them. In other words, decisions are made all the time under persisting, irreducible uncertainty.
Mousavi (2002) argued that this characteristic of actual choice behavior calls for moving away
from quantifiable uncertainty and systematic processes for reducing it. A first step in this direc-
tion is to allow goals to remain less than fully specified during the process of inquiry, in order
to permit an ontic notion of uncertainty (see also Mousavi & Garrison, 2003; Mousavi, 2018).
Technical requirements for such formalization were scarcely developed at the time, and not-
able attempts in this direction, e.g., Bewley (1986, 1987) on individual choice behavior under
Knightian uncertainty, and March and Simon (1958) on goal generation (as opposed to goal
orientation) in firms, did not get into the mainstream of economic modeling. Much has been
achieved since.
In what follows, we draw on two themes from Mousavi (2002). The first is to position
bounded rationality in the larger framework of social sciences by employing Simon’s view3
on the satisficing nature of search and stopping rules. The second is to juxtapose the substan-
tive rationality of neoclassical economics with procedural bounded rationality, going on to
point out less discussed commonalities between these two views of rationality, which we find
to be rooted in a Humian notion of uncertainty (Hume, 1739 [1985]). Hume’s empiricism
and epistemic conception of uncertainty father modern economic modeling methodologic-
ally (Binmore, 2011). Finally, we recount the basics of the cognitive revolution, which sets the
context for the Simonian approach to the study of human behavior. The arguments from 2002
are brought up to date through additions, modifications, and afterthoughts. The main message
is that over the decades since Simon called on economists to stroll away from their armchairs,
they have kept central notions of full rationality near and dear, while branching out into other
fields in a series of ongoing methodical efforts to tie down bounded rationality. Experimental
economists play a pronounced role in this movement. This rising procedural metropolis of eco-
nomics has now successfully acquired and built many behavioral and cognitive science suburbs.

The satisficer: the poster child of bounded rationality

Satisficing is a theory of search (Bendor, 2010). The term is a combination of “satisfying”
and “sufficing,” which can be juxtaposed, respectively, to “necessary” and “sufficient” as
requirements for mathematical completeness. Thus, replacing “necessary” with “satisfying”
sums up what bounded rationality changes in rational choice theory, and “sufficiency”—not
having been replaced by an alternative—is the part that remains intact and is thus the common
component of modeling both fully and boundedly rational behavior. To elaborate further on
this parallel connection, note that instead of requiring all criteria for a substantively rational
optimum of an objective function of many variables, a problem can be solved in a boundedly
rational way by satisfying target variables, say, through a sequential scheme. The magnitude of
available computational resources and attainable information provides a satisfactory solution
of a boundedly rational problem, when the potential limitlessness of search and calculation
necessitated by substantive rationality can defy solvability. (Think of solvability in familiar terms
of uniqueness and existence.)

450
451

The rise of procedural economics

The most famous of all boundedly rational agents is the satisficer, who divides the known
options into two categories, satisficing and non-satisficing, then chooses an available satisficing
option (Newell & Simon, 1972)—a process also known as achieving a “good-enough” choice.
The satisficer needs only one criterion, a threshold of acceptability called an aspiration level.
If a choice cannot be made in a unique way or does not exist, then the agent will attempt to
adjust the aspiration level (up or down) or to expand/contract the set of alternatives, or both
(Simon, 1955, 1957). If finding satisficing options are easy and low cost, the agent will raise the
aspiration level and can simultaneously look for more options, and vice versa. Notice, however,
that these two activities, adjusting the level of aspiration versus the set of choice options, are not
the same and can be triggered by different underlying conditions. For instance, a higher level
of persistence is associated with finding more options rather than adjusting the aspiration level.
Also, persistence corresponds to the effort of information gathering.
The satisficer is best understood in contrast to the rational representative agent (Kirman,
1992) or the “economic man.” A major operational property of rational formalization is its
independence of the situation. That is, the rational agent acts the same way across situations,
and thus the maximization of utility remains a process that yields the optimal outcome from a
full exploration of information, configuration of relevant distributions, and a comprehensive
cost-benefit calculation (Güth & Kliemt, 2000). Simon’s “administrative men,” in contrast,
are sensitive to the structure of the environment, e.g., the organization, within which they
search for, obtain, and analyze information to make their choices and solve the problems they
face. Their process of problem solving cannot be configured independently of their situation.
Whereas the cross-situation mobility of the rational economic man brings about an elegant,
tractable, coherent formalism, the “muddling through” of the administrative man defies clean
formalization. The representative agent configuration allows for seamless scaling, whereas the
complication of formalizing boundedly rational processes escalates quickly as the situation of
choice acquires more dimensions, becomes more than pairwise comparisons, or does not lend
itself to a binary structure that can generate thresholds for setting aspiration levels. At a more
fundamental level, this complication is verbalized by Viale (2012, pp. 161–162):

Insofar as Simon’s bounded and procedural rationality seeks to be a realistic represen-

tation of human cognitive activity, it finds its raison d’être in the hypothesis and know-
ledge of psychology. The first question we have to ask is: “what type of psychology?”.
There are almost as many types of psychology as there are psychologists! … [even] the
sea of cognitive psychology too has grown increasingly stormy.

An important challenge facing modelers from both camps, substantive and procedural ration-
ality, is to account for the observed data collected on choice processes. Competition in this
domain becomes a matter of fluency with techniques. Rational choice theory can expand and
extend preference structures to account for “deviations” from full rationality. Artful executions
of this method have generated marvels, such as Gary Becker’s seminal work that analyzed, in the
framework of rational choice theory, problems from other disciplines, such as discrimination
(Becker, 1971) and marriage (Becker, 1973, 1974a) from sociology and criminology (Becker,
1974b), and from law. In Simon’s words:

Becker follows Milton Friedman’s advice that economists should focus on testing the
conclusions of a theory, not testing the factual assumptions that underlie it. In his
Treatise on the Family, Becker [1981] looks at the steady rise in the labor force par-
ticipation of married women since World War II. The major cause for this change,

451
452

Shabnam Mousavi and Nicolaus Tideman

he believes, is the growth in the productivity of women as the American economy

developed. … He could also have asked if the increase in women’s weekly earnings
might have resulted from an increase in average hours worked; that is a form rather
than a consequence of greater labor force participation. But he explains it all in terms
of an unexplained shift in the demand curve for women’s labor. He doesn’t say why
this event should have taken place at this particular moment in American history. Was
it an unexpected shock, or a continuing development?
Challenge, 1986, p. 20

In another place, Simon (1978a) makes the connection between his idea and Becker’s clear,
“What Becker [1962: Irrational behavior and economic theory] calls irrationality in his art-
icle would be called bounded rationality here” (p. 3). Interestingly, the formally “weaker” and
functionally “broader” definition of rationality allowed economics to expand its domain from
the early days of being “the science which treats of those social phenomena that are due to
wealth-getting and wealth-using activities of man” (Ely, 1930, p. 4) to the Samuelsonian “the
study of the allocation of scarce resources among unlimited and competing uses” (Encyclopedia
of the Social Sciences, vol. 4, p. 472, cited by Simon, 1978a). While the consequences of the
bounded conception of rationality range all over social sciences (Simon, 2000), our attention
in the current chapter is limited to its implications for economics, on which we elaborate
further next.

Economics: thesis, antithesis, synthesis

Interestingly, in the 1930s, Ely described economics as “a branch of sociology. Next to language
it is the best developed of them all, and is by far the best introduction to the larger group.” He
urged broadening the definition of economics to

the science (1) which treats of those social phenomena due to the wealth-getting and
wealth-using activity of man, and (2) which deals with all other branches of his life in
so far as they affect his social activity in this respect.

All the while, Ely keeps economics at the center of social sciences.4
Whereas Simon seems to view economics as expanding its domain from Ely to Samuelson,
it seems to us that the Samuelsonian definition is merely more technical, context-free, and is
thus actually narrower in scope. That is, economics went from a broad verbal social science
(following Ely) to a mathematically oriented technique-centered quest (following Samuelson),
then returned to apply those refined tools to a wider scope of problems (such as bounded
rationality). Along this path, economics exceeded other branches of social sciences in formal-
ization and is now an apparatus for turning observations from experiments and field studies
into discrete models, all with respect to one central theory of rational choice. Triumphantly,
the economists’ toolkit has generated reliable mechanisms for capturing observations in elab-
orate descriptive accounts—observations that were originally suggestive of a paradigm change,
as they appeared to violate rationality’s characteristic assumptions.
Gary Becker, trained in sociology (recall that Ely viewed economics as a branch of soci-
ology), spent his professional career developing rigorous accounts of social topics by expanding
rational choice theory. As he put it, “The purpose … is to show how the important theorems
of modern economics result from a general principle which not only includes rational behavior
and survivor arguments as special cases, but also much irrational behavior” (Becker, 1962, p. 2).

452
453

The rise of procedural economics

It is tempting to imagine that accounting for the context equates to realistic modeling. That
does not follow, however, since modelers are only as good as their tools and the formalization
of observed phenomena has to be channeled through abstraction. For instance, a skillful mod-
eler in the framework of bounded rationality can extend the definition and scope of the target
to accommodate deviations from a pre-specified form of bounded rationality such as satisficing
behavior—technically, in the same way that rational choice theory users extend preferences
to account for deviations from full rationality. Linking empirical data to formal statements
remains a matter of fluency in mathematical techniques, whereas the claim to capture the
actual process at play is at best speculative. Besides psychological conditions, neurological and
physiological markers have been studied and documented in association with certain behavior
patterns, shedding light on even more aspects involved in the generation of humans’ observed
choice behaviors. As more aspects of the problem unravel, so emerge more and deeper puzzles
to grapple with. One of the most plausible data-driven perspectives we have encountered so
far is that of accepting the multiplicity of processes that people employ in face of the same
problem (Harrison 2018).5 Added to this is the possibility that the same process of choice can
be employed in different situations by the same person or by different people. In other words, a
fundamental flaw in the neoclassical methodology is the tireless search for one-to-one and even
onto mappings that can generate functional forms for formalizing choice behavior (Mousavi
& Gigerenzer, 2017). On the other hand, a main guiding insight from taking the bounded
rationality stance lies in liberation from the constraint of looking for a one-size-fits-all structure
to make scientific sense of human choice behavior. The admission of the bounded rationality
approach into the methodology of modeling choice behavior implies multiplicity of the ways
in which the same problem can be tackled by different agents, say, a group of satisficers, each
holding a different aspiration level, or giving their individual limited attention to different
choice sets.
There is more to Simon’s method. He criticizes armchair economists who “don't seem to
hesitate to extrapolate from themselves to the population. But they shouldn’t” (Challenge, 1986,
p. 25). At the same time, he warns against seeking the answer in extensive data collection (a cri-
terion for enhancing the accuracy of statistical inference). Simon shares a very important insight
that multiplicity is not extensive:

Take a mechanical puzzle, like this one on my desk, and give it to ten people. You
will get quite a range of individual behavior in trying to solve it. But you won’t learn
very much more from the next ten people you give it to. So the idea that we must
have huge samples in order to know how a system works is not necessarily so. Human
beings aren’t that variable.
Challenge, 1986, p. 25

This vision calls for rethinking the value of big data in exploring beliefs and decision making
tools, notwithstanding its value for profiling individual preferences. This passage also provides a
Simonian valuation of the limits on the knowledge and insight that can result from the study of
multiplicity determinants such as analytic vs. intuitive as distinctive modes of cognition.
The satisficer as the poster child for bounded rationality, both in its static version of fixed
aspiration level and the two dynamic dimensions of adjusting the choice set or the aspiration
level, is a gate to investigating human choice behavior in many ways that depart from rational
choice theory. In practice, however, the elegance and convenience of the tools and techniques
from rational choice theory have largely resulted in the transfer of them to the realm of bounded
rationality—so much so that the many emerging forms of theorizing about bounded rationality

453
454

Shabnam Mousavi and Nicolaus Tideman

are still presented with respect to the rational choice benchmarks, and not as primarily self-
containing structures. Bounded rationality was born from revisiting human choice behavior
through the lens of cognitive psychology. A closer look at cognitive science can shed light on
why some old ways are actively lingering in the new frameworks developed since the rise of the
concept of bounded rationality.

What substantive and procedural rationality (don’t) share

To appreciate the form of bounded rationality that Simon upholds as the alternative to the
rationality of the economic man, take a step back to the origin of the idea. Howard Gardner’s
(1985) acclaimed history of the cognitive revolution, The Mind’s New Science, acknowledges
Herbert Simon as one of the founders of the field, for which Gardner identifies the five
following tenets:

[1] the belief that, in talking about human cognitive activities, it is necessary to speak
about mental representations and to point to a level of analysis wholly separate
from the biological or neurological, on the one hand, and the sociological or
cultural, on the other.
[2] the computer as the most viable model of how the human mind functions.
[3] the deliberate decision to de-emphasize certain factors which may be important
for cognitive functioning but whose inclusion at this point would unnecessarily
complicate the cognitive-scientific enterprise. These factors include the influ-
ence of affective factors or emotions, the contribution of historical and cultural
factors, and the role of the background context in which particular actions or
thoughts occur.
[4] cognitive scientists harbor the faith that much is to be gained from interdiscip-
linary studies.
[5] a key ingredient in contemporary cognitive science is the agenda of issues, and
set of concerns, which have long exercised epistemologists in the Western philo-
sophical tradition.
Gardner, 1985, pp. 6–7

Simon’s cognitive psychology, in adherence to these five tenets, depicts the path from sub-
stantive to procedural (bounded) rationality (Simon, 1979), whereby he lays out the following
four defining aspects of the psychology of choice (or deliberation), and declares the fourth one
outside his domain of study. (By and large, the biological and neurological conditions of risky
choice do not specifically interest Simon.)

1 The study of cognitive processes involves the elements of action in problem situations and
consists of three strands: learning, problem solving, and concept attainment. Learning refers
to extraction of information from one situation and the consequent storing and use of this
information in similar problem situations. Problem solving involves gathering information
from various sources in different manners and combining the components to arrive at a
course of action for solving the problem at hand. In problem solving research, trial-and-
error procedures are complemented by insights. Concept attainment refers to generaliza-
tion and extraction of rules from existing problem situations to predict patterns and attain
solutions to future problems.

454
455

The rise of procedural economics

2 Computational efficiency refers to the time and effort required to solve a problem, for which
Simon retains the normative status of mathematical computation: “computational effi-
ciency is a search for procedural rationality, and computational mathematics is a normative
theory of such rationality” (Simon, 1979, p. 69).
3 Computation: risky decisions. Viewing the human mind as a computer that deals with risky
decisions by computations that are constrained by the limits of cognitive functioning,
stemming from and affected by biological, neurological, cultural, and sociological
conditions. This is a systems view of the human organism, which is a pillar of cognitive
science.
4 Man’s computational efficiency is only of interest to Simon in association with thinking, not
acting. He clarifies, “In my comparison of computer and Man, I am leaving out of account
the greater sophistication of Man’s input and output system, and the parallel processing
capabilities of his senses and his limbs. I will be primarily concerned here with thinking,
secondarily with perceiving, and not at all with sensing or acting” (Simon, 1979, p. 72, fn).

On this ground, the difference between Simon and neoclassical economics in the study of
human choice behavior can be understood as interest in and focus on different aspects of the
phenomenon. Simon is focusing on the cognitive processes involved in gathering, storing, and
interpreting information for solving present or future problems. Rational choice theory in
economics, on the other hand, has largely left the middle processing stage in a black box and
focused on reconstructing reasoning schemas (Smith, 2007) that map a given situation to the
final action, grounded in the fundamental assumption that best actions deliver optimal payoff as
defined with respect to a specifiable preference ordering. Simon’s focus on the process opens an
exciting gate into exploration of the economists’ black box, while leaving biological and emo-
tional mechanisms for other fields.

Conclusion
The visionary polymath Simon foresaw that for economics to survive and thrive, substantive
rationality must be augmented with procedural components and that this requires acquiring
tools from other disciplines (Simon, 1978a, pp. 2–3):

Economics has largely been preoccupied with results of rational choice rather than the
process of choice. Yet as economic analysis acquires a broader concern with dynamics
of choice under uncertainty, it will become more and more essential to consider
choice processes. In the past twenty years [1960s–1970s], there have been important
advances in our understanding of procedural rationality, particularly as a result of
research in artificial intelligence and cognitive psychology. The importation of these
theories of the processes of choice into economics could provide immense help in
deepening our understanding of the dynamics of rationality, and of the influences
upon choice of the institutional structure within which it takes place.

Today, economics have absorbed the knowledge and techniques from operations research for
achieving targets. Also, the behavioralization of economics has resulted in the absorption of
psychology to a great extent. This trend of enriching economics with tools from other disciplines
is an ongoing one: In the footsteps of cognitive psychology, evolutionary theory, bioeconomics,
and neuroeconomics have gradually joined the family of economic science––with more yet

455
456

Shabnam Mousavi and Nicolaus Tideman

to join the club. All the while, the supremacy of rational choice theory, both as the norma-
tive benchmark and as the central primary formalization with respect to which the many
versions of boundedness of rationality are being operationalized, remain strongly intact. Today,
economics—as a whole, including its behavioral branch—has successfully acquired the alterna-
tive rival of process modeling by systematically following a simple technical recipe: Add pro-
cedural to substantive as needed to incorporate bounded rationality.
Simon presented bounded rationality in contrast to substantive rationality that is fixated
on outcomes and oblivious to procedures. He invited economists to leave their armchairs
and observe the many ways in which actual human behavior defies the calculus of costs and
benefits. This chapter recounted some early Simonian cognitive-based notions that initiated
departure from the deductive economics methodology of rational choice theory. Today,
the adoption of procedural rationality notions from other fields has effectively customized
bounded rationality in the economists’ arsenal. This is a development that we propose to
call procedural economics in reference to the decades-long efforts of formalizing observed
(boundedly rational) choice behavior processes with respect to and as a result of extensions
of rational choice theory.

Notes
1 For a brief overview, see the helpful slides available at the link below, starting with this defin-
ition: “Stylized facts refer to generalizations that hold approximately, but not exactly.” /www.albany.
edu/~bd445/Economics_301_Intermediate_Macroeconomics_Slides_Spring_2014/Stylized_Facts_
of_Economic_Growth.pdf.
2 The use of the term “Economic Science(s)” is not rare nowadays. Economics’ status as a science, how-
ever, has been continuously debated for over a century: from Thorstein Veblen’s (1898) Why is economics
not an evolutionary science? to Alfred Eichner’s (1983) Why economics is not yet a science. The longing to be
a science among economists has been also dubbed “physics envy.” (Do an online search of “physics envy
+ economics” for a flow of opinions on all sides of the matter.)
3 For a critical review of the Simonian view of expectations and beliefs, valuation, cognitive abilities, and
Turing machines, see Mousavi and Garrison (2003).
4 Ely concludes his suggestion for a broader definition of economics as follows: “Of course the other
social sciences require a similar extension, and so they all are dependent upon economics.”
5 Glenn Harrison (1990) has worked with several collaborators (e.g., Harrison, List & Towe, 2007) on the
development of a field that he calls behavioral econometrics, where he applies rigorous econometric
methods to the data while keeping an open mind about the human behavioral phenomena at work and
trying to account for them by allowing different strategies to surface in solving one problem by several
subjects, say, by clustering them into groups that can be modeled as users of satisficing, maximization,
minimax, etc.

References
Abad, L. A., & Khalifa, K. (2015). What are stylized facts? Journal of Economic Methodology 22(2): 143–156.
doi:10.1080/1350178X.2015.1024878.
Arrow, K. J., & Debreu, G. (1954). Existence of an equilibrium for a competitive economy. Econometrica
22(3): 265–290. doi:10.2307/1907353.
Barros, G. (2010). Herbert A. Simon and the concept of rationality: Boundaries and procedures. Brazilian
Journal of Political Economy 30(3): 455–472.
Becker, G. S. (1962). Irrational behavior and economic theory. Journal of Political Economy 70: 1–13.
Becker, G. S. (1971). The economics of discrimination. Chicago: University of Chicago Press.
Becker, G. S. (1973). A theory of marriage: part I. Journal of Political Economy 81(4): 813–46. doi:10.1086/
260084.
Becker, G. S. (1974a). A theory of marriage: part II. Journal of Political Economy, Special Issue: Marriage,
Family Human Capital, and Fertility (part 2) 82 (2): s11–s26. doi:10.1086/260287.

456
457

The rise of procedural economics

Becker, G. S. (1974b). Essays in the economics of crime and punishment. New York: National Bureau of
Economic Research, distributed by Columbia University Press.
Becker, G. S. (1981). A treatise on the family. Cambridge, MA: Harvard University Press.
Bendor, J. (2010). Bounded rationality and politics. Oakland, CA: University of California Press.
Bewley, T. F. (1986). Knightian decision theory: Part I. Cowles Foundation Discussion Papers 807. Cowles
Foundation for Research in Economics, Yale University.
Bewley, T. F. (1987). Knightian decision theory: Part II. Intertemporal problems. Cowles Foundation
Discussion Papers 835. Cowles Foundation for Research in Economics, Yale University.
Binmore, K. (2011). David Hume: Grandfather of modern economics? Working Paper.
Challenge (1986). Interview with Herbert A. Simon: The failure of armchair economics. Available
at: http://digitalcollections.library.cmu.edu/awweb/awarchive?type=file&item=34037.
Eichner, A. S. (1983). Why economics is not yet a science. Journal of Economic Issues 17(2): 507–520.
doi:10.1080/00213624.1983.11504135.
Ely, R. T. (1930). Outlines of economics (rev. ed.). New York: Macmillan.
Harrison, G. W. (1990). Risk attitudes in first price auction experiments: A Bayesian analysis. Review of
Economics and Statistics 72, 541–546.
Harrison, G. W., List, J. A., & Towe, C. (2007). Naturally occurring preferences and exogenous laboratory
experiments: A case study of risk aversion. Econometrica 75, 433–458.
Hume, D. (1739 [1985]). A treatise of human nature. London: Penguin.
Gardner, H. (1985) The mind’s new science: A history of the cognitive revolution. New York: Basic Books.
Güth, W., & Kliemt, H. (2000). From full to bounded rationality: The limits of unlimited rationality. ZiF –
Mitteilungen special 2000, Zentrum für interdisziplinäre Forschung der Universität Bielefeld. 1–15.
Harrison, G. W. (2018). The methodologies of behavioral econometrics. In M. Nagatsu & A. Ruzzene
(Eds.), Philosophy and social science: An interdisciplinary dialogue. London: Bloomsbury. Available at: https://
cear.gsu.edu/files/2017/03/WP_2017_04_The-Methodologies-of-Behavioral-Econometrics_2018_
0409.pdf
Kirman, A. P. (1992). Whom or what does the representative individual represent? Journal of Economic
Perspectives 6(2): 117–136.
Lehtinen, A. (2013). Three kinds of ‘as-if ’ claims. Journal of Economic Methodology 20(2): 184–205.
doi:10.1080/1350178X.2013.801560.
March, J. G. & Simon, H. A. (1958) Organizations. New York: John Wiley & Sons.
Mousavi, S. (2002) Methodological foundations for bounded rationality as a primary framework. Doctoral
dissertation. Available at: http://theses.lib.vt.edu/theses/available/etd-12222002–183717/
Mousavi, S. (2018). Behavioral policymaking with bounded rationality. In R. Viale, S. Mousavi, B.
Alemanni, & U. Filotto (Eds.), The behavioral finance revolution: A new approach to financial policies and
regulations. Cheltenham: Edward Elgar.
Mousavi, S., & Garrison, J. (2003). Toward a transactional theory of decision making: Creative rationality
as functional coordination in context. Journal of Economic Methodology 10(2): 131–156.
Mousavi, S., & Gigerenzer, G. (2017). Heuristics are tools for uncertainty. Journal of Homo Oeconomicus
34: 361–379.
Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall.
Savage, L. J. (1954). The foundations of statistics. New York: John Wiley and Sons.
Simon, H. A. (1955). A behavioral model of rational choice. Quarterly Journal of Economics 69(1): 99–118.
Simon, H. A. (1957). Models of man: Social and rational. Mathematical essays on rational human behavior in a
social setting. New York: John Wiley & Sons, Inc.
Simon, H. A. (1972). Theories of bounded rationality. In C. B. McGuire & R. Radner (Eds.), Decision and
organization (pp. 161–176). Amsterdam: North-Holland.
Simon, H. A. (1978a). Rationality as process and as product of thought. American Economics Review
68(2): 1–16.
Simon, H. A. (1978b) Nobel Prize Lecture: Rational decision-making in business organizations. Available
at: www.nobelprize.org/nobel_prizes/economic-sciences/laureates/1978/simon-lecture.html
Simon, H. A. (1979). From substantive to procedural rationality. In F. Hahn & M. Hollis (Eds.), Philosophy
and economic theory, Essay 5. Oxford: Oxford University Press.
Simon, H. A. (2000). Bounded rationality in social science: Today and tomorrow. Mind and Society
1(1): 25–39. https://doi.org/10.1007/BF02512227
Smith, V. X. (2007). Rationality in economics: Constructivist and ecological forms. New York: Cambridge
University Press.

457
458

Shabnam Mousavi and Nicolaus Tideman

Veblen, T. (1898). Why is economics not an evolutionary science? The Quarterly Journal of Economics
12(4): 373–397.
Viale, R. (2012). Methodological cognitivism. vol. 1: Mind, rationality, and society. Berlin: Springer.
Von Neumann, J. & Morgenstern O. (1944), Theory of games and economic behavior, Princeton, NJ: Princeton
University Press.

458
459

30
BOUNDED RATIONALITY
AND EXPECTATIONS
IN ECONOMICS
Ignazio Visco and Giordano Zevi

Introduction
In his contribution to the Scandinavian Journal of Economics on the occasion of Herbert Simon’s
Nobel Prize award in 1979, Albert Ando identified a consistent theme running through Simon’s
vast contribution to economics: his attempt “to construct a comprehensive framework for
modelling and analysing the behaviour of man”, acknowledging the implicit “limitations of his
ability to comprehend, describe and analyse” the complex environment he inhabits.1 In essence,
this is a rich and, to a large extent, exhaustive description of Simon’s work, and it provides the
foundations for an operational definition of “bounded rationality”, resulting, with the benefit
of hindsight, even more far-reaching than was Ando’s original intention.
At the time, bounded rationality in economics was generally meant to indicate the subject
of Ando’s companion paper in the same journal, in which William Baumol compared the
“satisficing” criterion followed by Simon’s economic agents in their attempts to reach the best
decisions with the traditional microeconomic optimisation problem, which essentially consists
in the constrained maximisation of a utility function of some sort. The crucial difference,
Baumol noted, was that maximisation required a process of comparison of all the available
alternatives, while Simon’s satisficing criterion was aimed at pinning down “the first decision
encountered which passes the acceptability test”,2 among those decisions which are subjectively
considered to be feasible.3
In Simon’s view, this satisficing criterion was not intended to contrast with rationality. On
the contrary, this kind of behaviour was plausibly motivated at the very least by some signifi-
cant informational deficiency and, more in general, by the natural constraints on the economic
agents’ ability to gain, store and process the information they receive. In a nutshell, Simon
pointed out that since in most cases identifying and comparing all the possible alternatives is,
in fact, a very costly option computationally, if at all feasible, it should generally be excluded
by agents who are rational. Therefore, they will simplify the choice problem they face by only
concentrating their efforts on a subset of all the possible choices,4 and on devising the best strat-
egies to delimit such a subset.
With the passing of time, however, bounded rationality in economics came to be defined
not only as the procedural rationality implicit in the satisficing criterion, but also as the broader
“consistent theme” Ando had identified in Simon’s work, the studying of the behaviour of man

459
460

Ignazio Visco and Giordano Zevi

when faced with economic choices. Moreover, the renewed attention that scholars and policy
makers have progressively accorded to the limits of economic agents’ knowledge, and to the
recognition of their more common salient psychological traits, has even led some researchers
to consider it as being the study of the deep consequences of including the actors’ cognitive
restrictions within a standard maximisation framework. The unifying theme of these two ori-
ginally separate streams in Simon’s thinking, as it was progressively received by the economics
profession, is the recognition that economic agents have to devise behavioural strategies in a
space where rationality (in the neo-classical utility-maximising sense) is itself a scarce resource,
subject to the law of diminishing returns. Exercising rationality requires effort. This holds true
for both the procedurally rational economic agents of theoretical models and those whose real-
life decisions are the subject of empirical work. Consequences are pervasive, especially when
studying organisations and institutions.5
When rationality is costly, then, there is scope to cut these costs along a number of
dimensions (time, computational power, etc.). In his survey on the advances of behavioural
economics, DellaVigna frames Simon’s original contribution as one that was made to the “non-
standard decision making” stream, as opposed to other lines of research dedicated to non-
standard preferences and non-standard beliefs.6 Somewhat in line with, or stemming from,
Simon’s contribution are other works investigating the “limited attention” (or rational inatten-
tion, in macroeconomics parlance7) paid by agents to aspects of the choice problems they face
that are not deemed salient, with salience (a thoroughly subjective criterion) being, therefore,
the constraint that determines the previously quoted “acceptability” region. The deviation from
the choices that would be made under the same circumstances by neoclassical agents, measured
by the welfare losses that they would incur if they had to identify all the alternatives that are
possible in principle, could then be meaningfully used to measure the savings made thanks to
the procedural rationality implemented by Simon’s agents.
In this environment, heuristics (i.e. simple decision-making rules based on repeated experi-
ence, observation, intuition or common wisdom) are rational strategies and can be a very effi-
cient tool for making choices when the cost for acquiring and processing information is high.
According to some views,8 other perturbations of the choice process, such as framing (i.e.,
the impact of changes in the external context on the final outcome of a problem of choice),
endowment effects (i.e., the influence of any given initial condition on preferences, even when
these conditions could, in principle, be swapped at no cost) and sunk-cost fallacies (i.e., in a
choice problem, non-zero weights given to sunk costs that, theoretically, should be deemed
irrelevant) are disturbances of little or no importance, as they are scarcely able to modify the
terms of the choice problem for procedurally rational agents.9

Bounded rationality and expectations

We will briefly consider the possible links between Simon’s bounded rationality ideas and the
different approaches adopted by economists in dealing with household and business expectations
on the evolution of economic variables. By connecting the choices made in the present to
possible future scenarios, expectations drive the evolution of the economy over time. How
expectations are formed also changes through time, as economic systems evolve and adjust con-
tinuously. Fittingly with some procedural notions of rationality, households and businesses need
to learn how best to make decisions in an environment of continuous change and adaptation.10
Adapting to new circumstances is indeed a common experience. It requires the selection
of implicit or deliberate strategies, based both on free as well as on costly information and on
the more or less sophisticated methods and tools needed to process this information. Lessons

460
461

BR and expectations in economics

learned in the past, for example, on how best to collect and treat the available information, may
provide skills that could be applied valuably to decrypting the present and plan for the future.
However, change is often itself a non-stationary process, in a statistical sense. Deep recessions,
technological revolutions, institutional transformations and sudden changes in social habits can
all make the set of abilities possessed by the economic subjects abruptly obsolete, including
those related to the comprehension of the neighbouring environment. Crucially, they can also
induce a rapid decay in the usefulness of routines followed by economic agents in the forma-
tion of their expectations, something that necessarily leaves them in need of re-learning and
defining new routines.
Such a description does not appear to be controversial. Indeed, in the history of economics,
the relevance of expectations for any intertemporal decisions (i.e., almost all relevant economic
decisions, except for the simplest ones in static environments) has been repeatedly underlined.11
However, the economists’ degree of attention to how expectations are actually formed and to
how they interact with the economic “observables” has followed high and low cycles.12
One “high” part of the cycle coincided with the prevalence of Keynesian theories, deeply
informed by the recognition of the radical difference between risky and uncertain events13
and of the wide-ranging consequences that the latter could exert – mediated by their impact
on expectations – on the functioning of the economy.14 On the contrary, a “low” cycle began
when Bob Lucas famously criticised the way expectations were treated by macroeconomists at
the time, rightly pointing out that they should not, in general, be invariant to changes in “the
structure of series relevant to the decision maker”, and in particular to changes in economic
policy.15
In a few years, his critique had been fully taken on board by the discipline. However, Lucas
did not suggest improving either the use of expectations in macroeconomic models, or the
tools used to measure them, in order to identify the revision processes that businesses and indi-
viduals implement when the underlying reality changes. Rather, he ignored the early attempts
of empirical research on expectations and instead introduced the powerful construct of the
rational expectations hypothesis, based on Jack Muth’s earlier work.16 Muth, who had previ-
ously shared ground-breaking empirical work on expectations with Simon,17 described the
hypothesis of rational expectations as a situation in which the agents’ average subjective prob-
abilities on the distribution of the relevant variables coincide with the conditional probabilities
in the “true” model of the economy, making “optimal” use of all the information available in
the economy and deviating from perfect foresight only by some random noise.
Although Muth’s contribution was originally intended for some specific and somewhat
narrow circumstances, Lucas extended it by assuming it to be a necessary consistency condi-
tion in macroeconomic models, and the sheer theoretical power of the Muth-Lucas construct
came to be seen as a revolution. In a few years the “rational” expectations hypothesis spread
through the practice of the profession of macroeconomics, most of the time in conjunction
with the notion of a “natural” level of unemployment, in a fundamentally stable environment.18
It is perhaps somewhat ironic that in the (general) equilibrium representation of the economy
associated with the rational expectations hypothesis, there is no role for the state of (subjective)
expectations.
Three complementary conditions coincided to make the success complete: (1) rational
expectations allowed leading macroeconomic models to be closed in a straightforward way,
spurring great waves of new research in macroeconomics; (2) rational expectations were seen
as a natural benchmark for the comparison of the deviations posed by other formulations of
expectations;19 and (3) the findings of models that included rational expectations did not appear
to be obviously contradicted by the empirics of the economy, at least in advanced economies

461
462

Ignazio Visco and Giordano Zevi

(to which most of those models were applied) and chiefly in the US, in particular during the
period of the so-called “Great Moderation”.
Indeed, in the relatively stable economic environment that prevailed during this period, it
could also be argued that, in forming their expectations about the future paths of aggregate
variables, the agents who populate the macroeconomic models did not have to be entirely
rational in the fullest neo-classical sense of the word (i.e., able to consider all available possibil-
ities, compare them at no cost and subsequently select the best one). The hypothesis of rational
expectations could be unobtrusively introduced by assuming that economic agents lived in an
environment that they had come to know sufficiently well over time, following a readily avail-
able learning procedure.20
It should also be mentioned that a number of critiques were raised early on. Davidson
pointed to the fact that if the economic processes move in time, so that the data-generating
processes are non-time invariant and the economic world these processes describe is not erg-
odic, Muth’s rational expectations hypothesis cannot hold, as “calculable probability statements
may have no relation to future events”.21 Pesaran observed that rational expectations could be
plausible only in contexts where all uncertainty was exogenous to the subjects formulating the
expectations, meaning that, subjectively, the actions of those agents were not relevant in shaping
the aggregate economic outcomes.22 Outside the realm of rational expectations, there were
both non-stationary economic environments and also circumstances like the “beauty contest”
famously described by Keynes (i.e. states of the world where the collective outcomes came from
a general second-guessing of other agents’ behaviour).23 The latter is a good description of the
conditions of endogenous uncertainty.
A more nuanced view on expectations has started to prevail again only in the last decade
or so. Dynamic stochastic general equilibrium (DSGE) models, which have been the work-
horse of the new classical (and neo-Keynesian) macroeconomics originating from the rational
expectations revolution, began increasingly to allow for the presence of rigidities and free
parameters directly linked to intuitions on the actual workings of the economies that predated
the rational expectation revolution.24 Models retained their tendency to gravitate towards a
long-run equilibrium, but they became much slower to converge, aiming to allow in their the-
oretical framework for the complexities of the economies that have challenged policy makers
and academics alike in the wake of the so-called Great Recession.25 In such more complex
theoretic environments, expectations are re-gaining the central position that was attributed to
them by the original Keynes’ contribution.
The cycle thus appears to have now turned towards the “high” region again, with new
characteristics, in particular, a stronger focus on micro-data. Recent work has promoted a
renewed interest in empirical inquiries on how expectations are actually formed. The larger
and ever increasing availability of survey-based data on subjective expectations, collected
from businesses and individuals, represents a crucial motivation of this current research.26
The fact that the empirical expectations recovered in the surveys repeatedly deviate from
implied rational expectations reinforces this motivation (and is not surprising at a time of
great disruptions in the global economy).27 It has also been suggested that the progressive
marginalisation of empirical studies on expectations formation from the mainstream eco-
nomic research has come to the great detriment of the discipline.28 In monetary policy
making, while ample use of survey data, especially on savings and income, has been made for
a long time by central banks,29 it is mostly in the last decade that special attention has been
paid to measures of inflation expectations. In this respect, the risks of de-anchoring have
been thoroughly assessed, making extensive reference to household, business and professional
forecasters’ survey data.30

462
463

BR and expectations in economics

The extent to which expectations collected directly from households and firms help to
better forecast their future behaviour remains, of course, an open empirical question. However,
careful extraction of the information they contain is undoubtedly useful in order to empirically
discriminate among alternative hypotheses. Over time, both before as well as after Muth’s pro-
posal of the rational expectations hypothesis and its exponential adoption in macroeconomic
literature following works by Lucas, Thomas Sargent, Edward Prescott and many other prom-
inent economists not limited to the neoclassical camp, various suggestions on how economic
agents may form their expectations have been advanced, following extrapolative, adaptive,
regressive, error-learning or return-to-normality specifications.31 In a recent review by Manski
about the increasingly relevant field of probabilistic expectations collection in micro data sets,
the evolution of thinking about the formation of expectations in policy analysis is discussed in
depth.32 Manski’s conclusion is that the ongoing progressive demise in policy analyses of the
rational expectation assumptions (a welcome development) should not be accompanied by a
proliferation of alternative ad hoc models of expectations formation (a confusing consequence).
Careful empirical work would be needed in order to discriminate among these alternative
models. In addition, while up to now most of the work on eliciting expectations from indi-
vidual agents has been pursued by microeconomists, Manski advocates a more direct involve-
ment of macroeconomists even at this early stage, given the crucial impact of expectations on
the macroeconomic aggregates. Surveys can shed light not only on the state of expectations, but
also on the main drivers of their revisions, when the economic reality changes.
These arguments had long been made by Herbert Simon himself, when he discussed
objections to behavioural criticism of the rational expectation hypothesis, in particular those
calling attention to its overall lack of coherence. He acknowledged these objections, accepting
that simply highlighting the general shortcomings of a theory does not guarantee its demise,
unless a newly formed alternative theory has been developed, which is also Mark Blaug’s
familiar argument that “you cannot beat something with nothing”.33 However, in Simon’s
words, Behaviouralism could be built only through extensive empirical research, and this had
to be done even in times when it did not yet provide a fully-fledged alternative to mainstream
theory. Consequently, the dispute around expectations could be settled only by “painstaking
microeconomic empirical study of human decision making and problem solving”.34
Interestingly, this analysis had also to be done in order to pin down and estimate the particular
deviation from full rationality that was needed to complement Lucas’s rational expectations-
based models in order to generate a meaningful business cycle theory: the inability of Lucas’s
agents to fully discriminate between shocks to own prices and shocks on the general level of
prices. The failure of the Lucas islands model to confirm its results when the stochastic envir-
onment in which the agents act is not stationary was also pointed out by Albert Ando, Simon’s
former student and co-author, even if with a different focus.35 Ando suggested that there was a
lack of coherence between the professed general value of agents’ expectations in Lucas’s model
and the strict (but not explicit) hypotheses under which they acted. As the fortune of Lucas’s
contribution was partly linked to his critique of previous models’ ad hoc assumptions, rather
than to new empirical findings disproving such models, Ando’s assessment was particularly
sharp. Equilibrium cycle models, in the Lucas and Sargent formulations, were unable, in his
view, to say anything significant about the origins of business cycles.36 Adding to Ando’s and
Benjamin Friedman’s view, Simon pointed out that Muth’s expectations were more a special
case than “a paradigm for rational behaviour under uncertainty”37 and that their implied deci-
sion rule, rather than rational, should be labelled a “consistent expectations” rule.38
Simon also addressed the fundamental non-stationarity of economic systems and the
existence of states of the world where endogenous uncertainty is pervasive. For Simon’s

463
464

Ignazio Visco and Giordano Zevi

rationally-satisficing agents, the “acceptability region”, in which the first satisficing choice must
lie, is historically and institutionally determined. Inherently unstable economies will tend to
gravitate towards the path set by institutions that have been devised to contrast such perceived
instabilities.39 This will produce conventions and habits as well as simple decision rules which
individual agents will then be able to use in their choice-making, helping them to solve their
particular beauty contest-like problems while saving on costly rationality. History is of the
essence, and this renders any attempt to approximate the mechanisms of expectation formation
without specific knowledge of the local circumstances driving the economy relatively useless.40
Welcome discipline on theorising about expectations must therefore come from the empirical
analysis and its detection of regularities.

Cognitive limitations and learning

The comparison between the economic agents’ theoretical cognitive abilities and the outcomes
derived from empirically observed data is particularly relevant in policy making. A rich research
agenda is associated with the need to understand how agents learn from past mistakes and from
the availability of new information on the state of the world they are in. “Learning” as iden-
tified in the influential contribution by George Evans and Seppo Honkapohja, implies that a
rational expectations equilibrium is only one of the possible outcomes when agents continu-
ously update their expectations based on the comparison between past expectations and actual
realisations.41
The roots of this stream of literature can be found in simpler models, such as the cobweb
described by Nerlove, that showed how the extrapolation of specific adaptive expectations
functions could lead either to equilibria (not necessarily unique) or to explosive outcomes.42
Error-learning mechanisms are, in fact, very general, and can rationalise sophisticated optimal
forecasting rules.43 Muth himself advanced his own rational expectations hypothesis in order to
estimate the parameters of a model of adaptive expectations (where the change in the expected
price level depends only on the forecast error just observed). Learning by boundedly rational
agents is central also to Sargent’s successive research on expectations in macroeconomics,44
where the limitations of the straightforward rational expectation hypotheses are contrasted with
a more nuanced view of rationality, one that weakens the strong informational assumptions
implicit in Lucas’s and Sargent’s original contributions.45
The application of these tools in monetary policy has far-reaching policy implications. For
example, Ferrero describes an environment where boundedly rational agents face the issue of
formulating expectations on the future rate of inflation.46 In this environment, economic agents
learn from their past mistaken predictions by combining old and new information to form
new beliefs; the policy maker (specifically in this case the monetary policy authority) can exert
some influence on the agents’ learning process. The model is able to explain why policies that
would be optimal under rational expectations could instead perform poorly when knowledge is
imperfect. It also highlights that, under some conditions, converging to a rational expectations
equilibrium could take an extremely long time, making it ill-suited to represent an anchor for
rational agents. Welfare consequences are pervasive, especially when the policy maker is willing
to influence the public learning rules.47 In recent work, Busetti et al. investigate the effects of
a sequence of deflationary shocks on expected and actual inflation and output in the context
of a New Keynesian model where agents have incomplete information about the workings of
the economy and form expectations through adaptive learning processes.48 They show that the
learning process could imply a de-anchoring of inflation expectations from the central bank’s
target that is entirely data-driven, through the workings and feedback of the learning process.

464
465

BR and expectations in economics

More complex agent-based models have also been employed to explore the impact of het-
erogeneous expectations in environments where there could not exist any rational expectations
equilibrium towards which agents can coordinate.49 In line with Simon’s (1955) seminal con-
tribution, later developed with March, agents use expectations based on simple heuristics as a
device to willingly ignore part of the available information in order to reach local optima which,
in this particular setting, beat the outcomes of fully rational choices.50 This strategy proves to be
superior to more sophisticated ways of forming expectations, grounded, for example, in recur-
sive least squares as in mainstream learning literature, due to the highly unstable and uncertain,
in the Knightian sense, environment. Rather than making the case for a general superiority
of simple versus sophisticated rules, this stream of works provides a counterexample to the
opposite claim, even in the event that information is not costly per se. There are conditions
under which less is clearly more.
Again, rules of thumb, traditions and common received wisdom can be somewhat effective
tools in helping economic agents to efficiently reach their goals. It should be noted that also
the results that come from the learning behaviour of boundedly rational (though not neces-
sarily à la Simon) agents could and should be confronted with the “direct observation of human
behaviour in the market and in the firm” available from survey-based data on expectations.51
Indeed, the strategy that Simon recommended to economists in the early 1980s was that of
securing “new kinds of data at the micro level, data that will provide direct evidence about the
behaviour of economic agents and the ways in which they go about making their decisions”.52
The massive amount of empirical work at the level of individual decision makers, that was
only envisioned more than 30 years ago when Simon was advancing his criticism of rational
expectations, has now been made possible by the availability of data and computational power.

Conclusion
Individual agents’ expectations on the state and evolution of the economy have a great influ-
ence on macroeconomic variables. Macroeconomists have oscillated between giving prominent
importance to survey-based expectations data (as in the first analyses of business cycles in the
immediate aftermath of the Second World War) and ignoring empirical data on expectations
altogether, relying instead on model-consistent rational expectations. In recent years, the
increasing availability of individual survey data and the failings of models based on purely
rational (representative) agents have prompted renewed interest in inquiries into the direct
measurement of individual expectations and empirical studies of their formation. Herbert
Simon’s legacy provides a useful guide for both these activities.
Over the past decades, the success by policy makers in stabilising expectations could have
generated the falsely reassuring conclusion that expectations are of limited consequence and can
be assumed away. In fact, the unbiasedness that stems from Muth’s original rational expectations
hypothesis could hold both when agents behave as neo-classically rational agents as well as
when they inhabit a sufficiently stable economic environment and follow a boundedly-rational
learning procedure. Conclusions, however, differ when the economy is hit by sudden changes
such as deep recessions, technological revolutions, institutional transformations, and rapid
modifications in social habits. In such cases, which are by far the most common, possibly
excluding only the 1960s and the couple of decades of the so-called Great Moderation around
the turn of the last century, in a world populated by neoclassical agents, economies would be
stabilised only by a quick, widespread agreement on the new general economic equilibrium
(or its deep parameters), or if agents are closer to boundedly-rational ones, with the help of the
actions of institutions and other focal actors. In the latter case, stories and narratives, as pointed

465
466

Ignazio Visco and Giordano Zevi

out by Akerlof and Shiller, as well as norms and conventions (in the spirit of Keynes and Simon)
may become prominent macroeconomic forces.53
Sound new research on expectations is warranted. It should be both empirical and the-
oretical. On the empirical side, better and larger data collecting is already ongoing. Big data
techniques could possibly complement more traditional survey-based methods. With regard to
theory, research on learning and information has been a promising avenue and should regain
importance. Discipline in relation to a model internal consistency provided by the rational
expectation hypothesis could be meaningfully substituted by strong, repeated empirical verifi-
cation in order to avoid the proliferation of model-specific settings of expectations.
Finally, institutions should pay particular attention to current research on agents’ cogni-
tive limits. The reason is twofold. It would help, on the one hand, to perfect their policy
tools, by relying on models better apt to gauge the reactions of businesses and households
to policy interventions, and it would improve, on the other hand, their ability to drive the
economy in times of great disruption, by better focusing on salient communication to the gen-
eral public. In following Simon, economists and social scientists in general, not only those of
strict behaviourist observance, could also benefit from this heightened institutional attention,
as policy actions could provide the replicability they need in order to discriminate between
model-specific local behavioural hypotheses and more general behavioural traits to be possibly
included in macroeconomic models.

Notes
1 Ando (1979, p. 83).
2 Baumol (1979, p. 76).
3 That is, decisions that lie in a region of the space defined by a set of acceptability constraints.
4 Simon (1955).
5 Simon (1978).
6 DellaVigna (2009, p. 348).
7 Sims (2003).
8 See, among others, Gigerenzer (1996).
9 For an alternative view, that deems these perturbations as fundamental for modeling human choices,
see Tversky and Kahneman (1974, 1981) and Kahneman, Knetsch and Thaler (1991).
10 See also Visco (2013).
11 The seminal works of Keynes (1936) and Hicks (1939) are well known examples.
12 The heterogeneous relevance of expectations in economic models that have been progressively
considered mainstream has also made it difficult to develop a “theory of expectations”, one that would
form a basis for a well-defined field in the economic discipline dedicated to expectations. For an
extensive analysis of this issue, see the Introduction to Visco (1985) and Visco (2009a).
13 Keynes (1921), Knight (1921).
14 Keynes (1936).
15 Lucas (1976, p. 41).
16 Muth (1961).
17 Holt, Modigliani, Muth and Simon (1960). In their contribution the authors derived a decision
rule for firms that found the optimal level of production, given the state of inventories and the
expectations on future sales. The cooperation, in Simon’s words, of two Keynesian economists, one
rational expectationist and a behaviouralist, gave origins to contributions that came to be prominent
but opposing: the rational expectation hypothesis and the bounded rationality streams of research. An
analysis is given in Egidi (2017).
18 As Simon (1984, p. 49) fittingly observed, the special virtue of these terms was largely due to their
ability to “win instantly by taking the breath away from would-be disputants, whose very skepticism
now accuses them of ‘unnaturalness’ or ‘irrationality’, as the case may be.”
19 Woodford (2013, p. 304).

466
467

BR and expectations in economics

20 See Evans and Honkapohja (2001). In a stable environment a learning procedure based on a set com-
prising only the past realisations of the variable to be expected may lead to an unbiased though possibly
inefficient expectation of its future path.
21 Davidson (1982, p. 190).
22 Pesaran (1988, Chapter 2).
23 Keynes (1936, p. 156).
24 Visco (2009a).
25 Visco (2009b).
26 For a survey on recent contributions see Coibion, Gorodnichenko and Kamdar (2018). See also Visco
(1984) for an extensive discussion of results from early surveys of price expectations, as well as an
empirical analysis of the formation of expectations in Italy from the early 1950s through the late 1970s.
27 Similar results hold for the years following the 1973–74 oil crisis, see Visco (1984, Chapters 4 and 5).
See also Cukierman (1986).
28 Gennaioli, Ma and Shleifer (2016, p. 380).
29 For example, Banca d’Italia has carried out a Survey on Household Income and Wealth (SHIW)
since the mid-1960s, complemented over the years by a number of surveys on firms’ intentions
and expectations. Similar information is gauged for instance by the Survey of Consumer Finances
(SCF) and by several business surveys produced by the Federal Reserve System (e.g. the Empire State
Manufacturing Survey by the New York Fed). In addition, both the Philadelphia Federal Reserve
Bank (since 1968) and the European Central Bank (since 2002) release quarterly surveys on profes-
sional forecasters’ expectations about real and nominal macroeconomic variables (SPF).
30 Recent work by Aruoba (2016) and Doh and Oksol (2018) use direct observations on agents’
expectations to estimate the term structure of inflation expectations in the United States and evaluate
the degree to which inflation expectations have been anchored over time. Buono and Formai (2018)
use expectations recovered from the Consensus Economics survey to compare inflation anchoring
in the euro area versus other major economies. Łyziak and Paloviita (2017) show some signs of
de-anchoring in the euro area inflation expectations in one analysis based on the ECB Survey of
Professional Forecasters and on the European Commission Consumer survey.
31 See, for a succinct overview of some of the proposals advanced in the literature, Visco (1984,
Chapter 6).
32 Manski (2018) reviews and extend a previous contribution on the same issue (Manski, 2004) where
the use of survey-based expectations data in macroeconomic modelling was strongly suggested. Based
on new empirical evidence, directly eliciting probabilistic expectations from households and businesses
in surveys is found to be particularly effective.
33 Simon (1984, p. 52) summarising Blaug (1980, p. 186).
34 Simon (1984, p. 35).
35 Ando (1983).
36 See Lucas (1972, 1980) and Sargent (1976).
37 Simon (1979, p. 505).
38 Simon (1978, p. 10); see also Friedman (1979) and Ando (1981).
39 On this, see Simon (1958) and for similar arguments Keynes (1936, Chapter 12), even if it should be
acknowledged that, according to Simon, “Keynes’ modes of reasoning in the General Theory are only
locally heretical. His general form of argumentation is the one that is standard in economics: what
might be called ‘what would I do if I were a rational man’ argumentation” (Simon, 1984, p. 36).
40 A more radical approach addressing the interaction between historical evolutions, market and non-
market behaviour and economic theorizing is found in Polanyi (1944).
41 Evans and Honkapohja (2001).
42 Nerlove (1958).
43 See, for a perceptive analysis, Rose (1972).
44 Sargent (1993).
45 According to Sent (1997), the link between Sargent’s and Simon’s bounded rationalities is still, after all,
rather weak. This does not reduce the influence exerted by Simon, but while Sargent advanced in a
world of adaptively learning (based on parallel computing) standard utility-maximising agents, Simon
kept a clear distance from neoclassical theory, maintaining the prominence of “satisficing” behaviour
and dismissing the adoption of parallel adaptive computing systems while continuing to prefer the
recourse to more traditional serial symbol processing procedures.
46 Ferrero (2007).

467
468

Ignazio Visco and Giordano Zevi

47 “A policy-maker who considers his role in determining the dynamics of the agents’ learning process
could choose a policy rule that induces agents to learn at a given speed, affecting the welfare of society
along the transition.” (Ferrero 2007, p. 3034).
48 Busetti, Ferrero, Gerali and Locarno (2014) and Busetti, Delle Monache, Gerali and Locarno (2017).
49 See, for a recent contribution, Dosi, Napoletano, Roventini, Stiglitz and Treibich (2017).
50 See Simon (1955) and March and Simon (1993).
51 Simon (1984, p. 52).
52 Simon (1984, p. 40), where he also observes, somewhat surprised, that a similar suggestion had also
been advanced by Lucas in his book on the business cycle (Lucas, 1980, pp. 288–289).
53 Akerlof and Shiller (2009).

References
Akerlof, G. A. and R. J. Shiller, 2009. Animal spirits, Princeton, NJ: Princeton University Press.
Ando, A., 1979. “On the contributions of Herbert A. Simon to economics”, The Scandinavian Journal of
Economics, 81(1), 83–93.
Ando, A., 1981. “On a theoretical and empirical basis of macroeconomic models”, in J. Kmenta
and J. B. Ramsey (eds), Large-scale macro-econometric models: Theory and practice,( pp. 329–367),
Amsterdam: North-Holland.
Ando, A., 1983. “Equilibrium business-cycle models: an appraisal”, in F. G. Adams and B. G.
Hickman (eds), Global econometrics: Essays in honor of Lawrence R. Klein (pp. 39–67), Cambridge,
MA: MIT Press.
Aruoba, S. B, 2016. “Term structures of inflation expectations and real interest rates”, Federal Reserve
Bank of Philadelphia, Working Paper 16–09, December.
Baumol, W. J., 1979. “On the contributions of Herbert A. Simon to economics”, The Scandinavian Journal
of Economics, 81(1), 74–82.
Blaug, M., 1980. The methodology of economics, Cambridge: Cambridge University Press.
Busetti, F., G. Ferrero, A. Gerali and A. Locarno, 2014. “Deflationary shocks and de-anchoring of infla-
tion expectations”, Banca d’Italia, Occasional Papers 252.
Busetti, F., D. Delle Monache, A. Gerali and A. Locarno, 2017. “Trust, but verify. de-anchoring
of inflation expectations under learning and heterogeneity”, European Central Bank, Working
Paper 1994.
Buono, I. and S. Formai, 2018. “New evidence on the evolution of the anchoring of inflation expectations”,
Journal of Macroeconomics, 57, 39–54.
Coibion, O., Y. Gorodnichenko and R. Kamdar, 2018. “The formation of expectations, inflation and the
Phillips curve”, Journal of Economic Literature, 56(4).
Cukierman, A. 1986. “Measuring inflationary expectations. A review essay”, Journal of Monetary Economics,
17, 315–324.
Davidson, P., 1982. “Rational expectations: A fallacious foundation for studying crucial decision-making
processes”, Journal of Post-Keynesian Economics, 5(2), 182–198.
DellaVigna, S., 2009. “Psychology and economics: Evidence from the field”, Journal of Economic Literature,
47(2), 315–372.
Doh, T. and A. Oksol, 2018. “Has the anchoring of inflation expectations changed in the United States
during the past decade? Federal Reserve Bank of Kansas City”, Economic Review, 103(1), 31–58.
Dosi G., M. Napoletano, A. Roventini, J. E. Stiglitz, and T. Treibich, 2017. “Rational heuristics?
Expectations and behaviors in evolving economies with heterogeneous interacting agents”, Scuola
Superiore Sant’Anna, Pisa, LEM Working Paper 31.
Egidi M., 2017. “Paths in contemporary economics and sciences of artificial intelligence that originate
from Simon’s bounded rationality approach”, PSL Quarterly Review, 70(279), 7–33.
Evans, G. W. and S. Honkapohja, 2001. Learning and expectations in macroeconomics, Princeton, NJ: Princeton
University Press.
Ferrero, G., 2007. “Monetary policy, learning and the speed of convergence”, Journal of Economic Dynamics
& Control, 31(9), 3006–3041.
Friedman, B. M., 1979. “Optimal expectations and the extreme information assumptions of ‘rational
expectations’ macromodels”, Journal of Monetary Economics, 5(1), 23–41.

468
469

BR and expectations in economics

Gennaioli, N., Y. Ma and A. Shleifer, 2016. “Expectations and investment”, NBER Macroeconomics
Annual, 30, 379–442.
Gigerenzer, G., 1996. “On narrow norms and vague heuristics: A reply to Kahneman and Tversky”,
Psychological Review, 103(3), 592–596.
Hicks, J. R., 1939. Value and capital, Oxford: Clarendon Press.
Holt, C. C., F. Modigliani, J. F. Muth and H. A. Simon, 1960. Planning production, inventories and work force,
Englewood Cliffs, NJ: Prentice-Hall.
Kahneman, D., J. L. Knetsch and R. H. Thaler, 1991. “Anomalies: The endowment effect, loss aversion,
and status quo bias”, Journal of Economic Perspectives, 5(1), 193–206.
Keynes, J. M., 1921. A treatise on probability, London: Macmillan.
Keynes, J. M., 1936. The general theory of employment, interest and money, London: Macmillan.
Knight, F. H., 1921. Risk, uncertainty and profit, New York: Houghton Mifflin.
Łyziak, T. and Paloviita, M., 2017. “Anchoring of inflation expectations in the euro area: Recent evidence
based on survey data”, European Journal of Political Economy, 46, 52–73.
Lucas, R. E., 1972. “Expectations and the neutrality of money”, Journal of Economic Theory, 4(2),
103–124.
Lucas, R. E., 1976. “Econometric policy evaluation: a critique”, Carnegie-Rochester Conference Series
on Public Policy, 1(1), 19–46.
Lucas, R. E., 1980. Methods and problems in business cycle theory, Cambridge, MA: The MIT Press.
Manski, C. F., 2004. “Measuring expectations”, Econometrica, 72(5), 1329–1376.
Manski, C .F., 2018. “Survey measurement of probabilistic macroeconomic expectations: Progress and
promise”, NBER Macroeconomics Annual, 32(1), 411–471.
March, J. G. and H. A. Simon, 1993. “Organizations revised”, Industrial and Corporate Change, 2(3),
299–316.
Muth, J. F., 1961. “Rational expectations and the theory of price movements”, Econometrica, 29(3),
315–335.
Nerlove, M., 1958. “Adaptive expectations and cobweb phenomena”, Quarterly Journal of Economics, 72(2),
227–240.
Pesaran, H., 1988. The limits to rational expectations, Oxford: Basil Blackwell.
Polanyi, K., 1944. The great transformation, New York: Farrar & Rinehart.
Rose, D. E., 1972. “A general error-learning model of expectations formation”, University of Manchester,
Inflation Workshop Discussion Paper.
Sargent, T. J., 1976. “A classical macroeconomic model for the United States”, Journal of Political Economy,
84(2), 207–237.
Sargent, T. J., 1993. Bounded rationality in macroeconomics, Oxford: Oxford University Press.
Sent, E. M., 1997. “Sargent versus Simon: Bounded rationality unbound”, Cambridge Journal of Economics,
21, 323–338.
Simon, H. A., 1955. “A behavioral model of rational choice”, Quarterly Journal of Economics, 69(1),
99–118.
Simon, H. A., 1958. “The role of expectations in an adaptive or behavioristic model”, in M. J Bowman,
(ed.), Expectations, uncertainty and business behavior (pp. 49–58), New York: Social Science Research
Council,
Simon, H. A., 1978. “Rationality as process and as product of thought”, American Economic Review,
68(2), 1–16.
Simon, H. A., 1979. “Rational decision making in business organizations”, American Economic Review,
69(4), 493–513.
Simon, H. A., 1984. “On the behavioral and rational foundations of economic dynamics”, Journal of
Economic Behavior and Organization, 5(1), 35–55.
Sims, C. A., 2003. “Implications of rational inattention”, Journal of Monetary Economics, 50(3), 665–690.
Tversky, A. and D. Kahneman, 1974. “Judgement under uncertainty: Heuristics and biases”,
Science, 185.
Tversky, A. and D. Kahneman, 1981. “The framing of decisions and the psychology of choice”, Science,
211 (4481).
Visco, I., 1984. Price expectations in rising inflation, Amsterdam: North Holland.
Visco, I., 1985. Le aspettative nell’analisi economica, Bologna: il Mulino.

469
470

Ignazio Visco and Giordano Zevi

Visco, I., 2009a. “On the role of expectations in Keynesian and today’s economics (and economies)”,
paper presented at Accademia Nazionale dei Lincei, Rome, 11–12 March (available at the Banca
d’Italia’s website)
Visco, I., 2009b. “The financial crisis and economists’ forecasts”, BIS Review, 49.
Visco, I., 2013. “The aftermath of the crisis: Regulation, supervision and the role of central banks”,
CEPR Policy Insight, 68.
Woodford, M., 2013. “Macroeconomic analysis without the rational expectations hypothesis”, Annual
Review of Economics, 5, 303–346.

470
471

31
LESS IS MORE
FOR BAYESIANS, TOO
Gregory Wheeler

Lore has it that a fundamental principle of Bayesian rationality is for decision makers to never
turn down the offer of free information. Cost-free information can only help you, never hurt
you, and in the worst case will leave you at status quo ante. Purported exceptions to this prin-
ciple are no exceptions at all, but instead involve a hidden cost to learning. Make those costs
plain and the problem you face is one of balancing the quality of a choice against the costs to
you of carrying it out, a trade-off that Bayesian methods are ideally suited to solve.
This piece of Bayesian lore, that rationality compels you to never turn down free infor-
mation, is sometimes called Good’s Principle, after I. J. Good’s concise formalization of the
reasoning behind it (Good, 1967). But the argument goes back to the beginning of modern
Bayesian probability theory, with remarks by Ramsey (1931), an argument by Savage (1972),
the formalization of a key piece of it by Raiffa and Schlaiffer (1961), followed thereafter by
assertions in textbooks, starting with Lindley (1965). Put a bit more carefully, Good’s prin-
ciple recommends to delay making a terminal decision between alternative courses of action
if there is an opportunity to learn, at zero cost, the outcome of an experiment relevant to
the decision (Pedersen and Wheeler, 2015). This will be put more carefully still later in this
chapter.
Objections to Good’s principle have surfaced in the last half-century, some of which are well
known by now but others less so, forming part of a rich discussion of the value of information
to rational decision making (Wakker, 1988; Machina, 1989; Seidenfeld, 1994; Grünwald and
Halpern, 2004; Gigerenzer and Brighton, 2009; Siniscalchi, 2011; Hill, 2013; Pedersen and
Wheeler, 2015). Since then, a picture has emerged about the value of information that is more
restricted and more nuanced than Good’s principle states, suggesting a revision to Bayesian
lore. For even in highly idealized settings, ignorance can be a virtue. Sometimes less is more
for Bayesians, too.

Asymmetric information in strategic games

The first dent to this folklore comes from the theory of games, where some strategic interactions
can result in a player being better off having less information. George Akerlof ’s study of market
failures created by asymmetric information is a classic example (Akerlof, 1970). Adverse selection

471
472

Gregory Wheeler

occurs when one side of a trade has less information than the other side and withdraws from
trading from fear of being unfairly taken advantage of by the more informed party. Akerlof
offered the used car market as an example where adverse selection occurs, a particularly apt
example in 1970. A used car salesman will know which cars on the lot are bad and which are
good, knowledge an ordinary consumer will not have. But the consumer will know that the
dealer knows which car is of which quality type and recognize the upper hand the dealer has in
any trade. Afraid of paying a good-car price and driving home in a bad one, the customer may
choose not to buy any car at all. The reasoning for this idealized single transaction generalizes,
resulting in a market failure for used cars where nobody is willing to pay more than the going
rate for a bad car.
Used car dealers have overcome their adverse selection problem by certifying the quality of
used cars, and backing those claims with a warranty, thereby leveling the information playing
field between dealers and customers by letting customers in on what dealers know about the
quality of the cars they sell. (Making better cars has helped, too.) Yet, since the problem here
is asymmetrical information, this isn’t the only way to restore the market. Rather than making
consumers as informed as dealers, another option is to make dealers as ignorant as consumers.
The following example, due to Martin Osborne, illustrates the ignorance option (Osborne
2003, 9.3). Imagine there are two states of the world, ω1 and ω2—which could be understood
to correspond to the state in which a car is more likely to be good than bad and vice versa,
for instance. Suppose there are two Bayesians, Player 1 and Player 2, but neither player knows
which state of the world they are in. Both are ignorant, so they both assign a probability of
one-half to ω1 and one-half to ω2. Figure 31.1 gives the payoff tables for Player 1 and Player
2, where the material difference to each player from their uncertainty over which state they
are in, ω1 or ω2, is reflected in the last two columns of the respective payoff tables. Given this
setup, with both players ignorant of the state, the strategy L is Player 2’s unique best response to
every strategy of Player 1, which yields Player 2 the expected payoff of 2–2(1–ε)p, whereas M
and R both yield –32 – –32 (1-ε)p where p is the probability Player 1 assigns to T. Player 1’s unique
best response to L is B. Therefore, (B,L) is the unique Nash equilibrium of the game, yielding
each player a payoff of 2.
Now imagine that instead of both players being ignorant of which state they are in, exactly
one of the players is informed of the state. Specifically, suppose Player 2 is informed of the state

Player 2
L M R
T (1, 2ε) (1, 0) (1, 3ε)
Player 1
B (2, 2) (0, 0) (0, 3)
State ω 1

Player 2
L M R
T (1, 2ε) (1, 3ε) (1, 0)
Player 1
B (2, 2) (0, 3) (0, 0)
State ω 2

Figure 31.1 Payoffs to Players 1 and 2 in states ω1 and ω2 with 0 ≤ ϵ ≤ ½

Source: Osborne (2003).

472
473

Less is more for Bayesians, too

whereas Player 1 remains ignorant but nevertheless knows that Player 2 is informed. In this
game (T, (R, M)) is the unique Nash equilibrium yielding to her at most a payoff of 1.5. Why?
Choosing R is Player 2’s best response in state ω1 and her worst response in ω2. Similarly, M is
Player 2’s best response in ω2 and worst in ω1. Player 1 knows this too, knows that Player 2 is
informed of the state, thus knows that Player 2 will never choose L. With column L removed
from consideration, Player 1’s best response to (M,R) is T.
Despite her information advantage over Player 1, Player 2’s payoff in this second game is 3ϵ
in each state, which is at most 1.5. Thus, Player 2 is worse off learning the state than remaining
ignorant. Given the choice between the original game, where both players are ignorant, and
the second game in which Player 1 remains ignorant but Player 2 is informed, it is rational for
Player 2 to choose to remain ignorant, even if the information about the state is offered to her
for free.
Akerlof ’s and Osborne’s examples are part of a broader collection of counter-intuitive results
that can arise when the rational choice of one player changes the probability assessments of
another about which state will occur. In this case, the negative value of information stems from
act-state dependence of Player 1’s strategic response to Player 2’s informed choice. Osborne seems
to think that the prospect of information having negative value appears only in games, not in
decision problems:

A decision-maker in a single-person decision problem cannot be worse off if she has

more information: if she wishes, she can ignore the information. In a game the same
is not true: if a player has more information and the other players know that she has
more information, then she may be worse off.
Osborne, 2003, p. 281

This position that Osborne expresses, that Good’s principle governs single-person decision
making but not strategic decision making (i.e., games), remains something of a received view
on the possibility of negative-valued information. Over the last half-century, decision and
game theorists have become keenly aware of the crucial role that act-state independence plays
in standard decision theory, aided by a slew of puzzles and aberrant behavior in examples
that are found upon close inspection to depend on violations of this independence condition
(Kadane, Seidenfeld, and Schervish, 2008). Act-state independence is the first thing to go in
the theory of games, however, as the whole point of strategic decision making is to factor in
the consequences to you from the rational acts of others. So, one might conclude, to avoid the
specter of negative-valued information, restrict the scope of Good’s principle to single-person
decision problems. That, in a nutshell, is the received view on Good’s principle. The received
view is wrong, however.

Good by Savage
To see why single-person decision-making is not immune to negative-valued information, let
us consider more carefully Savage’s argument that it is immune. To be clear at the outset, our
analysis will not uncover a mathematical mistake or faulty theorem. Rather, our aim is to draw
attention to another important type of qualification to Good’s principle.
Good’s principle appears in Savage’s discussion in Foundations of Statistics of the differences
between a basic decision problem and a derived decision problem. A basic decision problem is
one in which an agent is to choose a basic action from among a collection he judges to be avail-
able for choice. A derived decision problem is one in which the agent is to choose from the same

473
474

Gregory Wheeler

collection of basic actions but only after considering the associated conditional expected util-
ities for a basic action given each possible outcome of some experiment. Given the assumption
that you wish to maximize your expected utility, why should you prefer a derived decision
problem over a basic decision problem? Because you cannot be made worse off in expectation
and may well come out better. “It is almost obvious,” Savage remarks,

that the value of a derived problem cannot be less, and typically is greater, than the
value of the basic problem from which it is derived. After all, any basic act is among
the derived acts, so that any expected utility that can be attained by deciding on a
basic act can be attained by deciding on the same basic act considered as a derived act.
In short, the person is free to ignore the observation. That obvious fact is the theory’s
expression of the commonplace that knowledge is not disadvantageous.
Savage, 1972, p. 107

Good later showed that Carnap’s principle of total evidence (Carnap, 1947) follows as a con-
sequence of the principle to maximize expected utility, so long as the costs of acquiring infor-
mation are negligible.

[I]n expectation, it pays to take into account further evidence, provided that the cost
of collecting and using this evidence, although positive, can be ignored. In particular,
we should use all the evidence already available, provided that the cost of doing so is
negligible. With this proviso then, the principle of total evidence follows from the
principle of rationality.
Good, 1967, p. 319

Our discussion in the next two sections will be helped along by introducing a bit of formalism
now to set up Savage’s version of Good’s principle.
Following (Pedersen and Wheeler, 2015), consider an illustration of Good’s principle in
Figure 31.2. Suppose that at some time t1 you are to face a choice, A, among two courses of
action, a1 or a2. Prior to this choice you face a decision, O, at some time t0 prior to t1, between
o1, the basic decision of choosing a1 or a2 at time t1, and o2, the derived decision of choosing a1 or
a2 at some later time t2 after you have observed, at no cost, the outcome of an experiment ε,
with outcomes e1 or e2.1

σ (a1 , ωi ) σ (a2 , ωi ) σ (a1 , ωi ) σ (a2 , ωi )

A A t2
e1 e2
σ (a1 , ωi ) σ (a2 , ωi ) E

a1 a2
A t1
o2
o1
O t0

Figure 31.2 Illustration of Good’s principle

474
475

Less is more for Bayesians, too

Choice, being governed here by dominance reasoning, comes after ruling out those options
for choice that are worse than all others. Those acts that survive the cull are admissible for
choice. Suppose that your judgments of admissibility can be represented in terms of subjective
expected utility maximization with respect to a real-valued expectation p[⋅] agreeing with a
real-valued probability function p defined on a Boolean algebra A over the set of states Ω and
a real-valued utility function u defined over the set of consequences.2 Then, at time t0 you
confront a decision problem O = {o1,o2}. If you implement option o1 at time t0, then at time t1
you will face a decision problem A = {a1,,a2} without observing the outcome of experiment ε.
If you implement option o2 at time t0, then at time t2 you will face the same decision problem
A after observing the outcome of experiment ε.
Abusing notation, let ‘o1’ also stand for the event of facing the decision problem A after
implementing option o1 at t0. (Context should make clear which use of ‘o1’ we intend.) In
a similar manner, let ‘o2, ε = ei’ stand for the event of facing the decision problem A after
implementing o2 and observing the outcome ei of experiment ε. The choice set c of admissible
options from A for choice given each alternative, written c ( A|o1 ) and c ( A| o2 , ε = ei ) for
options o1 and o2, respectively, may be defined by Equations (31.1) and (31.2):

c ( A| o1 ) = arg max  p(|⋅ o1 ) (u  σ )(a, p(d ω | o1 )) (31.1)

a ∈A

c ( A| o2 , ε = ei ) = arg max  p (⋅| o2 , ε = ei ) (u  σ ) (a, p(d ω | o2 , ε = ei ) (31.2)

a ∈A

where ∘ denotes functional composition.

Good’s principle assumes that at t0 you are certain, regardless of whether or not you choose
to observe the outcome of experiment ε, that you will choose an option a ∈ A that maximizes
your expected utility. This assumption is codified in how admissible choices are determined for
each option o1 and o2 in Equations (31.1) and (31.2), respectively. A second assumption is that
your preferences over consequences remain unchanged. A third assumption is that your beliefs
given hypotheses accord with Bayesian conditionalization. With all of this in place, Good’s
principle states that your expectation of (1) your maximum conditional expected utility of choosing
from A under option o1 is less than or equal to (2) your maximum conditional expected utility under
o2 of choosing from A given experiment ε. Your expectation of (2) is strictly greater than (1) unless
there is an action from A that maximizes conditional expected utility from A regardless of the
experimental outcome of ε. In other words, unless the experiment is irrelevant (i.e., probabil-
istically independent), then c(O) = {o2}.

Uncertainty and imprecision

According to the canonical theory of synchronic decision making under risk, a perfectly rational
person is one whose comparative assessments of a set of consequences satisfies the recommen-
dation to maximize expected utility. What underpins this claim is the assumption that a person’s
qualitative comparative judgments of those consequences (aka, preferences) are structured in a
particular way (satisfy specific axioms) to admit a mathematical representation in terms of inequal-
ities of mathematical expectations, ordered from worst to best on the real number line. This
structuring of preference through qualitative axioms to admit a numerical representation is the
subject of expected utility theory (Wheeler, 2018, §1.1).

475
476

Gregory Wheeler

Savage’s theory tells us how to represent preference in terms of some pair of numer-
ical probability and numerical utility functions, an ingenious extension of prior work that
showed how to quantify each piece separately, principally von Neumann and Morgenstern’s
numerical representation of utility, which presupposes a numerical probability function (von
Neumann and Morgenstern, 1944); and de Finetti’s numerical representation of probability,
which presupposes a cardinal utility function (de Finetti, 1974). Let’s focus here on probability
assessments, which for Bayesians are understood as a person’s partial beliefs.
Consider what it means for a person to have a partial belief in the proposition, L, expressing
that a particular car is a lemon. What does it mean for a person to have a partial of belief of
0.40 that L is true? According to the Ramsey-de Finetti conception of partial belief, this
means the person is indifferent between two sorts of hypothetical transactions. The first hypo-
thetical transaction calls on him to buy a contract for €0.40 that pays him €1 if the car is a
lemon, whereas the second hypothetical transaction calls on him to sell such a contract for the
same price. Put differently, the first type requires the person to surrender a sure 40 cents for
the promise of 1 Euro on the event of A occurring and risk receiving 0 if L does not occur.
The second type of transaction requires the person to accept payment of the sure reward of
40 cents in exchange for agreeing to risk paying back 1 Euro on the event of L occurring and
paying out nothing—in terms of this contract—otherwise. The choice of price is up to the
person, the utility of Euros is assumed to be linear, and the stakes are presumed to be small
enough to not bankrupt the person yet large enough for him to care. (De Finetti was a thor-
oughly pragmatic fellow, a point sometimes lost on his critics.) The price of 40 cents is fair
to this person just in case he is indifferent between buying and selling contracts on L at 40
cents. A person is rational just in case there is no possible way to put together a finite set of
buy and sell positions on that person’s announced fair prices to cause him a sure loss, a return
to that person of a value less than zero no matter how the uncertain events in those contracts
are resolved.
We rehearse this canonical account in order to introduce a slight generalization. Airport
currency exchange counters post different prices for buying and selling trades between a
pair of currencies. While they do so primarily to turn a profit, the same idea can be used to
express your uncertainty about the event, or events, controlling the payoffs in the contract.
So, rather than require decision makers to post the same number for buying as for selling a
contract, we wish to allow for the possibility that they post different numbers. Put differ-
ently, rather than oblige an agent to give a single two-sided price for betting on and against
the event L, written P(L), we instead oblige the agent to give two one-sided numbers:
(1) a one-sided lower probability denoting the maximum buying price for a bet on L, written
P(L); and (2) a one-sided upper probability denoting the minimum selling price for a bet on
L, written P̅(L).
Notice that for someone whose fair price is P(L), he will judge any price α < P(L) to buy a
bet on L (to bet on L) as desirable. Similarly, prices to sell a bet on L (to bet against L) that are
strictly greater than P(L) will likewise be judged desirable. It is only the fair price, the single
numerical value of P(L), that marks the agent’s indifference. Similar reasoning applies to the
one-sided lower probability P(L). Any price α < P(L) will be judged a desirable price to bet
on L, and any α > P̅(L) will be judged a desirable price to bet against L. The difference is that
there are (possibly) two price points where the agent expresses indifference between a sure
award and risky reward in the same currency, namely, when the buying price for a bet on L is
P(L) and when the selling price for bets on L is P̅(L). Only when they are the same value is the
agent committed to a fair price. Since 0 ≤ P(L) ≤ P̅ (L) ≤ 1, one consequence is that any price

476
477

Less is more for Bayesians, too

p offered between the agent’s lower and upper probabilities for L, any α such that P(L) < α <
P̅(L), the agent is neither obligated to sell nor to buy contracts on L.
It is a commonplace to distinguish between risk and uncertainty, an idea that both Knight
and Keynes put forward a century ago (Keynes, 1921; Knight, 1921). The notion that it is some-
times sensible to permit a bounded range of probability values rather than to insist on numeric-
ally determinate probability values is an even older idea, dating back at least to (Bernoulli, 1713)
and (Boole, 1854). But the rich mathematical and philosophical consequences from working
out these ideas have only begun to come into focus more recently (Walley, 1991; Augustin,
Coolen, de Cooman, and Troffaes, 2014; Troffaes and de Cooman, 2014).
The lower probability model presented above is very basic and supplied with a behavioral
interpretation that is very close to the original, canonical model: instead of one number to
describe two attitudes, we allow each attitude to have its own number. This slight change, how-
ever, from a fair-price model to a buying and selling price model of belief, is enough to put
another dent in Good’s principle. We turn to see how, next.

Dilating probabilities
What does Knightian uncertainty look like in our bare-bones lower probability model? The
short answer is that we have the means to distinguish between indifference and incomparability,
and to do so behaviorally in the same simple terms of the canonical Bayesian model. For a
longer answer and a consequence, an example.
Suppose there is a ticket that pays to its owner 100 euros on the event of G, Germany wins
the next World Cup. If you owned such a ticket, how much would you demand to part with it?
100 euros would make you whole, so you should at least be indifferent between receiving a sure
100 for the promise of 100 on the event of G being true.3 Similarly, if you are sure they would
lose, the ticket would be worthless to you, so you would find any (positive) price a desirable
selling price. Conversely, how much would you pay to buy such a ticket? Here again if you
were maximally uncertain (but otherwise abided by the setup for the model), you might not be
willing to pay anything for such a ticket. In such a case, your lower probability would be 0. If
instead you were certain they will win, you would find any price less than 100 euros desirable
and be indifferent to owning the ticket and having a 100 euro note in your pocket: for you,
being certain of the outcome G, those two rewards are equivalent.
For my part, I would not know how to give a fair price for G. This does not rule out being
bullied by a Bayesian into announcing one, but then again that would be a different decision
problem. Hypothetically, I would pay up to 10 euros for a chance to win 100 if Germany won
the next World Cup. They’ve done it before, I reckon, so there is some chance they could do
it next time. On the other hand, if I had such a ticket, what price would I accept to relinquish
my chance at 100 euros if they win? Here I might accept nothing less than 90 euros. So, any
price between 10 and 90 euros I would neither buy nor sell a 100 euro contract on G. These
prices don’t have to be symmetric, nor need they be calibrated to a statistical model. This is still
a subjective probability model and these are my attitudes toward buying and selling hypothetical
100 euro contracts on G.
Let us introduce some notation to reason with attitudes like the one I have toward G.
A lower probability space is a quadruple (Ω, A, , P) such that Ω is a set of states, A is an algebra
over Ω,  is a nonempty set of probability functions on A, and P is a lower probability function on
A with respect to —that is, P(F ) = inf {p(F) : p ∈ } for each F ∈ A. The value P(F) is called
the lower probability of F. The upper probability function P̅ is then defined by stipulating that P̅(F) =

477
478

Gregory Wheeler

1– P(Fc) for each F ∈ A; the value P̅(F) is called the upper probability of F. If P(H) > 0, then
conditional lower and upper probabilities are defined as P(E∣H) = inf {p(F∣H) : p ∈ } and P̅
(F∣H) = sup{p(F∣H) : p ∈ }, respectively.
Now return to the highly uncertain event, G, that Germany wins the next World Cup. The
upper probability of G is close to 1, P̅(G) = .9, and its lower probability is close to 0, P(G) = .1,
such that

P̅(G) – P(G) = 0.8. (31.3)

Next, imagine a fair coin toss, whose outcomes are heads (H) and tails (Hc). The outcomes
of this normal coin flip form a partition, ε = {H,Hc}, and the same is true of this future
championship title, G = {G,Gc}. With these preliminaries in place, we rehearse an example
from (Seidenfeld, 1994) in which a probability estimate of an event becomes less precise upon
receiving information about how the tossed coin lands, regardless of whether it lands heads or
lands tails.
Since we judge the coin flip to be fair, our expectation of the coin landing heads is the same
as our expectation of it landing tails.

1
P=
( H ) P=
( H=
) P=
( H c ) P ( H c ).
2 (31.4)
Equation (31.4) is what a fair price looks like in a lower probability model. We also assume
that the outcome of this coin toss landing heads is independent of Germany winning this future
championship. If any pair of events are probabilistically independent, surely the events heads and
Germany wins! are. So, for each p ∈ , we have

p(G )
p(G ∩ H ) = p(G ) p( H ) = . (31.5)
2
Lastly, let F be the event of either G and H both occurring or both failing to occur, namely,
F := (G∩H)∪(Gc∩Hc). Given our setup, it follows that the probability of F is determinate: that
is, p(F ) = 12 , for all p ∈ .
Proof. For each p ∈ , observe that

p( F ) = p(G , H ) + p(G c , H c )
p(G ) 1 − p(G )
= + [by (31.5)]
2 2
p(G ) + 1 − p(G ) (31.6)
=
2
1
= .
2
Figure 31.3 may help to fix intuitions as to why p(F ) = 12 , for all p ∈ , is so by visualizing
three probability mass functions that differ with respect to the probability that G. Note that the
counter-diagonal is the complement of F, Fc, which is the event that Germany wins if and only
if the coin lands tails. Put differently, if Germany winning and the coin landing heads are each
coded as “success” and Germany losing and tails are coded as “failure,” the event F says that the
coin and Germany both succeed or both fail, whereas the complement event Fc that exactly
one of the two succeeds.

478
479

Less is more for Bayesians, too

H Hc H Hc H Hc

G F
F
F
F Gc

(a) (b) (c)

Figure 31.3 Tables for an uncertain event (row) G = {G,Gc}, a fair coin randomizer (column) ε =
{H,Hc}, and the pivotal event (diagonal) F denoting G if and only if H. (a) illustrates when p(G) = 9∕10
and p(Gc) = 1∕10, (b) when p(G) = p(Gc) = ½, and (c) when p(G) = 1∕10 and p(Gc) = 9∕10, for p ∈ . For
each (a), (b), and (c), p(F) = ½.

Let ε be a positive measurable partition of Ω. We say that ε dilates F just in case for
each e ∈ ε:

P(F | ε = e ) < P( F ) ≤ P( F ) < P( F | ε = e ).

In other words, ε dilates F just in case the closed interval P( F ), P( F ) is contained in the open
interval ( P ( E |ε = e ), P ( E |ε = 3)) for each e ∈ ε (Walley, 1991, Seidenfeld and Wasserman,
1993, Pedersen and Wheeler, 2014). What is remarkable about dilation is the specter of turning
a more precise estimate of F into a less precise estimate, no matter what event from the parti-
tion occurs.
Observe that in our World Cup example, F is dilated by the coin toss ε = {H,Hc}: although
the initial estimate of F is precisely one-half, learning the outcome of the coin toss, whether
heads or tails, dilates the probability estimate of F from one-half to [.1,.9].
Proof. We show that 0.1 = P(F∣H) < P(F ) = 1∕2.

P ( F | H ) = inf { p( F | H ) : p ∈ }
 p([(G ∩ H ) ∪ (G c ∩ H c )] ∩ H ) 
= inf  : p ∈ 
 p( H ) 
 p(G ∩ H ) 
= inf  : p ∈ 
 p( H ) 
 p(G ) p( H ) 
= inf  : p ∈ 
 p( H ) 
= 0.1

A similar argument establishes 9∕10 = P̅ (F∣H) > 1∕2, and the same argument holds if instead the coin
lands tails, i.e., P(F∣Hc) = 1∕10 and P̅(F∣Hc) = 9∕10. Thus, F is dilated by the coin toss, ε = {H,Hc}.
Here again Figure 31.3 may help fix intuitions about this result. Notice that the observa-
tion of the coin landing heads (H) effectively restricts attention to the first column. Since we
learn that H has occurred, the possibilities in the second column associated with tails (Hc) are
ruled out. But, the probability mass assigned to the event F in the first column varies widely
in Figures 31.3(a), 31.3(b), and 31.3(c). Only in Figure 31.3(b) does the F have the value 1∕2;
Figures 31.3(a) and 31.3(c) reveal that the range of uncertainty for F given H is precisely the

479
480

Gregory Wheeler

uncertainty for G displayed in Equation (31.3). The same argument applies if the coin instead
landed tails. As these two outcomes exhaust the possible outcomes of the coin toss, being
told that the coin was tossed is enough for a Bayesian to dilate his probability assessment of F.
For a discussion of the philosophical and mathematical features of dilation, see (Pedersen and
Wheeler 2014, 2015).

Good’s principle and dilation

Recall the illustration of Good’s principle in Figure 31.2. Following the presentation in
(Pedersen and Wheeler, 2015) of an example due to (Seidenfeld 1994), suppose that at t0 you
face a decision problem O = {o1,o2} where, as before, option o1 is a basic decision problem A in
which you are to choose at t1 between two acts: a1, which pays you €1 if E occurs and ‘pays’
you -€1 if Ec, i.e., σ(a1,F ) = €1 and σ(a1,Fc) = -€1;4 or the act a2 which ‘pays’ you a constant
-€0.50. Assume that your utility is linear in euro amounts with u(€x) = x. Figure 31.4 fills in
these details.
In this basic decision problem A, which is the result of implementing option o1, the sub-
jective expected utility of a1 is €0 and the subjective expected utility of a2 is -€0.50. So, a1 is
uniquely admissible from A: receiving nothing is better than paying 50 cents.
Turn now to option o2, whereby at t2 you face a derived decision problem conditional on
the outcome of experiment ε. Here you are confronted with the same decision problem A at
t2 after learning (only) that H obtains or Hc obtains at t1. In the derived decision problem act
a1 is inadmissible against a2. Why? Because in the basic decision problem p(F ) = 1∕2, but in the
derived decision problem F is dilated by ε to 0.1 and 0.9: whether the outcome of the fair coin
toss is heads or tails, F conditional on that outcome is highly uncertain. Thus, in the derived
decision problem, there are probability mass functions p ∈  whereby p(Fc) is .9, in which case
the minimum expected utility of a1 is -€0.80. So, in the derived decision problem, by Savage’s
maxmin decision rule, a2 has a higher minimum expected value (-€0.50) than a1 (-€0.80) regardless
of the outcome of the experiment, ε.
Assume that a decision maker is certain that she will not change her preferences, will update
her belief state by Generalized Bayesian conditionalization (Walley 1991), and that she will
choose to maximize her minimal expected utility. Then, in a pairwise choice between a1 of

1 if F 1 if F
− 0.50 − 0.50
− 1 if F c − 1 if F c

a1 a2 a2 a1
A A
1 if F
− 0.50
− 1 if F c
H Hc
a1 a2
A E
o1 o2
O

Figure 31.4 A sequential decision example

480
481

Less is more for Bayesians, too

the basic decision problem determined by option o1, which has an expected value of zero, and
a2 of the derived decision problem determined by option o2, which has an expected value of -
$0.50, observing cost-free information at t1, i.e., learning the outcome of the fair coin toss €,
is devalued. Here, under the conditions for Good’s principle slightly adapted to a lower prob-
ability model, we have a case where the decision maker would strictly prefer not to receive
cost-free information!

Conclusion
Let’s review. The informal version of Bayesian lore has it that it is irrational to turn down cost-
free information, since the worst case—when the information is irrelevant to your decision
at hand—will leave you at status quo ante. The first restriction to Good’s principle is that it
does not apply to strategic decision problems, where strategic considerations may disadvan-
tage a player with more information than her opponent. The problem of adverse selection
is the classic example, and we discussed Akerlof ’s market for lemons example and Osborne’s
formalization. This limitation is fairly well known, however, which is why Good’s principle
is usually formulated to govern single-person decisions. We then formalized Savage’s version
of Good’s principle in terms of his distinction between a basic and derived decision problem,
and where the role maxmin reasoning plays is clear. But, in what may be less widely known,
we appealed to the phenomenon of dilation to argue that there are exceptions to Good’s prin-
ciple even for single-person decision problems. Specifically, if one introduces an upper and
lower probability model to accommodate a modest form of “Knightian uncertainty,” then a
probability assessment can become less precise after learning the outcome of an experiment,
no matter how that experiment turns out. Finally, we returned to our discussion of basic
and derived decision problems in Savage’s framework to show that this dilation example can
be plugged into Savage’s original formulation of Good’s principle to show that, by applying
Savage’s maxmin principle, the decision maker would rationally choose to forgo the offer to
receive cost-free information about the coin flip experiment ε. Thus, for imprecise probabil-
ities, the “commonplace that knowledge is not disadvantageous” is false, even when the costs
of obtaining the information is zero. The upshot is that the scope of Good’s principle is far
narrower than originally conceived and narrower still than many current decision theorists
maintain.
The role that Bayesian methods ought to play in models of bounded rationality remains
controversial, and there are some good reasons. Models of bounded rationality typically
focus on procedures, algorithms, or psychological processes involved in reaching a deci-
sion, securing a goal, or making a judgment, yet these details are ignored in the canon-
ical model. Another branch of bounded rationality focuses on adaptive behavior, and
coherent comparative judgments are not, directly at least, the most obvious way to frame
this problem.
But it would be incautious to dismiss all of the tools of statistical decision theory, and unwise
to ignore the developments in the field over the last half-century. It is hoped that a wider
awareness of better outcomes with less information results in decision theory—even under the strict
adherence to the highly idealized conditions of those mathematical models—will plant a seed
of future progress in psychology, where concrete examples are well known. From studying
axiomatic departures from the canonical Bayesian theory, it is hoped that the grip of Bayesian

481
482

Gregory Wheeler

dogma will loosen to expand the range of new, creative possibilities for applying a set of prac-
tical and powerful mathematical methods to the study of bounded rationality (Wheeler, 2018).

Coda: blinded by omniscience

We end with a short remark on logical omniscience. Most formal models of judgment and
decision making entail logical omniscience, the presumption that agents have complete know-
ledge of all that logically follows from their commitments combined together with any and all
set of options that are admissible to them for choice. This is as psychologically unrealistic as it is
difficult, technically, to remove from formal models. The problem is especially troublesome to
Bayesian decision theory, making it difficult to apply the theory to uncertainty about matters of
logic and mathematics. Savage, ever prescient, saw the problem that logical omniscience poses
to the subjective theory of probability:

The analysis should be careful not to prove too much; for some departures from
theory are inevitable, and some even laudable. For example, a person required to risk
money on a remote digit of π would, in order to comply fully with the theory, have
to compute the digit, though this would really be wasteful if the cost of computation
were more than the prize involved. For the postulates of the theory imply that you
should behave in accordance with the logical implications of all that you know. Is it
possible to improve the theory in this respect, making allowances within it for the
cost of thinking, or would that entail paradox, as I am inclined to believe but unable
to demonstrate?
Savage, 1967, excerpted from Savage’s prepublished draft.
See notes in Seidenfeld et al., 2012

Notes
1 Good’s principle is a synchronic rationality principle, governing here the synchronic choice at t0
between options o1 and o2. Our informal discussion of choices taken at future times ought to be viewed
as all hypothetical choices entertained at t0. Put differently, in choosing between o1 and o2, we are com-
paring at t0 the consequences from engaging in two lines of suppositional reasoning.
2 Often a uniqueness result for probabilities and utilities accompanies the representation result (asserting,
for example, that the probability function is unique and that the utility function is unique up to a posi-
tive affine transformation).
3 Assume a euro today is worth the same euro in the future, or that values are so-adjusted.
4 Here we abuse our notation by writing σ(a,F) = €1, for instance, to express that σ(a,⋅) is a constant
€1 on F.

References
Akerlof, G. A. (1970). The market for “lemons”: Quality uncertainty and the market mechanism. The
Quarterly Journal of Economics 84(3), 488–500.
Augustin, T., F. P. A., Coolen, G., de Cooman, and Troffaes, M. C. M. (2014). Introduction to Imprecise
Probabilities. Chichester: Wiley.
Bernoulli, J. (1713). Ars Conjectandi. Basel: Thurnisius.
Boole, G. (1854). An Investigation of the Laws of Thought. New York: Dover.
Carnap, R. (1947). On the application of inductive logic. Philosophy and Phenomenological Research 8,
133–148.
de Finetti, B. (1974). Theory of Probability: A Critical Introductory Treatment, Vols 1 and 2. Chichester: John
Wiley & Sons, Ltd.

482
483

Less is more for Bayesians, too

Gigerenzer, G., and Brighton, H. (2009). Homo heuristicus: Why biased minds make better inferences.
Topics in Cognitive Science 1(1), 107–143.
Good, I. J. (1967). On the principle of total evidence. The British Journal for the Philosophy of Science 17(4),
319–321.
Grünwald, P., and Halpern, J. Y. (2004). When ignorance is bliss. In J. Y. Halpern (Ed.), Proceedings
of the 20th Conference on Uncertainty in Artificial Intelligence (UAI ’04) (pp. 226–234). Arlington,
VA: AUAI Press.
Hill, B. (2013). Dynamic consistency and ambiguity: A reappraisal. Technical Report ECO/SCD-2013–
983. Paris: HEC Paris,
Kadane, J. B., Seidenfeld, T., and Schervish M. J. (2008). Is ignorance bliss? Journal of Philosophy
105(1), 5–36.
Keynes, J. M. (1921). A Treatise on Probability. London: Macmillan.
Knight, F. H. (1921). Risk, Uncertainty and Profit. Boston: Houghton Mifflin.
Lindley, D. V. (1965). Introduction to Probability and Statistics. Cambridge: Cambridge University Press.
Machina, M. J. (1989). Dynamic consistency and non-expected utility models of choice under uncer-
tainty. Journal of Economic Literature 27(4), 1622–1668.
Osborne, M. J. (2003). An Introduction to Game Theory. Oxford: Oxford University Press.
Pedersen, A. P., and Wheeler, G. (2014). Demystifying dilation. Erkenntnis 79(6), 1305–1342.
Pedersen, A. P., and Wheeler, G. (2015). Dilation, disintegrations, and delayed decisions. In Proceedings of
the 9th Symposium on Imprecise Probabilities and Their Applications (ISIPTA), Pescara, Italy, pp. 227–236.
Raiffa, H., and Schlaiffer, R. (1961). Applied Statistical Decision Theory, Vol. 1 of Studies in Managerial
Economics. Boston: Harvard Business School Publications.
Ramsey, F. P. (1931). The Foundations of Mathematics and Other Essays, vol. 1. New York: Humanities Press.
Savage, L. J. (1967). Difficulties in the theory of personal probability. Philosophy of Science 34(4), 311–325.
Savage, L. J. (1972). Foundations of Statistics (2nd ed.). New York: Dover.
Seidenfeld, T. (1994). When normal and extensive form decisions differ. In D. Prawitz, B. Skyrms, and
D. Westerstahl (Eds.), Logic, Methodology and Philosophy of Science. Oxford: Elsevier Science B. V.
Seidenfeld, T., Schervish, M. J., and Kadane, J. B. (2012). What kind of uncertainty is that? Using
personal probability for expressing one’s thinking about logical and mathematical propositions. Journal
of Philosophy 109(8–9), 516–533.
Seidenfeld, T., and Wasserman, L. (1993). Dilation for sets of probabilities. The Annals of Statistics 21(9),
1139–1154.
Siniscalchi, M. (2011). Dynamic choice under ambiguity. Theoretical Economics 6, 379–421.
Troffaes, M. C. M., and de Cooman, G. (2014). Lower Previsions. Chichester: Wiley.
von Neumann, J., and Morgenstern, O. (1944). Theory of Games and Economic Behavior. Princeton,
NJ: Princeton University Press.
Wakker, P. (1988). Nonexpected utility as aversion of information. Journal of Behavioral Decision Making
1, 169–175.
Walley, P. (1991). Statistical Reasoning with Imprecise Probabilities. London: Chapman and Hall.
Wheeler, G. (2018). Bounded rationality. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy
(Winter 2018 ed.). Stanford, CA: Metaphysics Research Lab, Stanford University.

483
484

32
BOUNDED RATIONALITY
AS THE COGNITIVE BASIS
FOR EVOLUTIONARY
ECONOMICS
Richard R. Nelson

A foundational feature of modern evolutionary economics1 is its commitment both to a

Schumpeterian perspective on modern capitalist economies as dynamic systems,2 always
evolving, with change being driven largely by innovation, and to the proposition developed by
Herbert Simon and his colleagues that the behaviors of human and organizational actors should
be understood as boundedly rational.3 The way the presumption that economic actors are
boundedly rational is employed in evolutionary economics is consistent in a broad sense with
the way that concept has been used in other arenas of research and writing, but the contexts
and modes of behavior treated go significantly beyond the more conventional orientation. The
objective of this chapter is to explain why and how the presumption of individual and organ-
izational bounded rationality has been used in evolutionary economics.

Innovation, continuing unpredictable change, and bounded rationality

The evolutionary economics described in this chapter is based on a perspective regarding what is
going on in the economy that is very different from that presented in today’s standard textbooks.
The overarching difference is that evolutionary economists see continuing change, largely
driven by innovation, as a central characteristic of modern capitalist economies. Evolutionary
economists are Schumpeterian in that basic sense. This means that innovation, and operation
in contexts where innovation-driven change is going on, are fundamental aspects of what
economic actors do and the contexts in which they operate. Of course, economic sectors and
activities differ in the pace and character of change. In many parts of the economy, innovation
is rapid and continuing and the context for economic action taking is almost always shifting
and providing new challenges and opportunities. And while in some activities and sectors, the
rate of innovation is more limited, attempts at doing something new are going on almost every-
where in the economy, and so too change that can make obsolete old ways of doing things. This
is a very different perspective on the nature and context for economic action taking than that
described in the standard economics textbooks.
Proponents of the neoclassical economic theory that provides the intellectual scaffolding for
the conventional view of what economic activity is all about never explain or rationalize how

484
485

BR: basis for evolutionary economics

economic actors come to be able to choose courses of action that are optimum for them, given
their objectives and the context in which they are operating. It is highly relevant to our discus-
sion here, however, that almost no formulation of that theory of behavior mentions attempts
by economic actors to do something that has not been done before, or the difficulties they face
when they are operating in contexts that they have not previously experienced. While in evo-
lutionary economics not all situations are of these sorts, many of the most important are. And
this has led us to develop a version of Simon’s bounded rationality as our basic assumption about
the cognitive strengths and limits of economic actors, and our theory of how they behave.
The perspective on economic behavior as mostly “boundedly rational” has the particular
attractiveness for economists of being consonant with the traditional economic theory of
behavior, going back to the days of Adam Smith, that sees economic actors doing what they do
with purposes in mind and in many contexts at least a rough understanding of the consequences
of following various courses of action. I believe that, treated with care, and recognizing human
fallibility, this broad theoretical perspective has shown considerable explanatory and predictive
power. The problem with the full-blown rational behavior theory of neoclassical economics is
that it does not recognize these caveats, and they are particularly germane in contexts important
to evolutionary economics.
The proposition that human rationality is bounded highlights that there are limits to the
reasoning power of the human mind as well as of the knowledge and information actors can
master and work with. As Simon has stressed, the contexts for human action very often are too
complicated or subtle for actors to understand and take into account adequately the wide range
of factors bearing on what they should be doing. I would like to add that this general argument
is quite open to recognition of significant differences across contexts in the strength of human
understanding.
However, I would propose that, to address the range of phenomena of particular interest to
Schumpeterian economists, several distinctions and factors need to be highlighted much more
than they have been to date in the literature on bounded rationality.
First, it is important to distinguish between choice contexts which are familiar to the eco-
nomic actor and who responds to them more or less automatically by taking actions that have
sufficed before in this kind of context, and contexts that induce the actor to engage in serious
contemplation of alternatives. And where action taking is preceded by conscious deliberation,
it is important to distinguish between contexts where the actor’s attention is focused on courses
of action the actor has followed before, perhaps in another context, and perhaps others well
known in his community, and those that involve trying to do something new. Schumpeterian
and evolutionary economists of course have a special interest in the latter – that is what innov-
ation is all about – but innovation only can be understood in juxtaposition to more routine
behavior, and more generally action taking that involves doing the familiar.
Second, particularly for understanding the kinds of phenomena that most interest
Schumpeterian economists, it is important to recognize that actors differ in the capabilities
that they bring to various choice contexts. They differ in their knowledge and experience,
and in the skills they possess. For these reasons they may differ significantly in what they do in
contexts that, to an outside observer, may look basically the same. And some will do better than
others will. I note that this aspect of behavior – differences in capabilities – has received little
attention from either main line economists, or scholars working with the theory that economic
actors are boundedly rational. Yet differences in capabilities obviously are of central interest to
Schumpeterian and evolutionary economists.
A third important limitation of most of the writings on economic behavior, including those
oriented by the assumption of bounded rationality, that needs to be remedied is the failure to

485
486

Richard R. Nelson

relate the perceptions of individual actors about the contexts they face, the courses of action
that they understand and are competent to employ, and their judgments about which of these
actions are appropriate and likely to be effective, to the beliefs and understandings and know-
how of the broader community of which the actor is a part. This can and has been raised as a
criticism of modern psychology in general. And it is hardly recognized in behavioral economics.
I want to argue that it is especially important that Schumpeterian economists clearly recog-
nize the social and cultural context of action taking. We are centrally interested in how modern
capitalist economies have become so productive, and the sources and mechanisms of future
progress. When one observes powerful and complex methods being used by individuals and
organizations to achieve their ends, it is almost a sure thing that the heart of the knowledge
base of what they are doing is shared by their professional peers, and is acquired by individ-
uals only as they are part of this broader community. And almost always powerful knowledge,
common to professionals in a field of activity, has been achieved through a lengthy cultural
learning process.
To return to the general theme, under the perspective on economic behavior and cognition
I am describing here, economic actors are assumed to be boundedly rational. When in contexts
that call for them to do something, they proceed with some notions about the outcomes they
would like to see happen, a perception of at least some actions they might take that seem plaus-
ible, and some thoughts on which of these might be most appropriate. But the contexts they
face differ widely, and they go about generating the actions they actually take in different ways
in different kinds of contexts.
In contexts where change is relatively slow, actors are likely to respond to the requirement to
do something by following patterns of behavior that they have used successfully before. In other
situations, the context is different from what the actor has faced before, or while the context
may be familiar, for various reasons the actor may want to consider a range of options before
doing anything. Simon himself made this distinction in a number of his analyses, and this also
is a distinction made by Daniel Kahneman (2011).4 Sidney Winter has reminded me that John
Dewey ([1922] 2002) presented a similar view of behavior, with perhaps more emphasis on the
role played by emotion and anxiety in some contexts.
I note that Kahneman puts less weight than do I on the argument that routine behavior often
is highly effective. Also, he seems to presume to a greater extent than do I that active deliber-
ation is highly likely to come up with an effective course of action. It may or may not. I would
propose that the chances are better that it will when the actor’s own experience or the know-
ledge the actor has of what others have done effectively includes actions that will be effective
in this context. Where dealing with the problem induces or requires a quite new attack on it,
the chances of speedy success are slim. With continued effort and cogitation, success may be
achieved thought trial and error learning. But it may not.
Innovating is a rational activity, in that those who try have particular objectives in mind and
draw as best they can on the knowledge they have. But that rationality surely is bounded.

Routines
The range of actions that need to be taken even over a short period of time by an economic
actor often is far too great for that actor to be able to think carefully before taking each required
action. However, where the environment for action has been relatively tranquil, actors gen-
erally have had time to learn the kinds of action that works in that context and what doesn’t.
The argument above is that most of the actions one observes in such contexts should be under-
stood as actors following routines that have in the past yielded satisfactory outcomes, and are

486
487

BR: basis for evolutionary economics

triggered relatively automatically by circumstances under which action along these lines seems
appropriate.
I suggest that individual or household shopping for the kinds of items bought relatively
regularly largely involves following routines. In my recent paper with Davide Consoli (Nelson
and Consoli, 2010) we propose that much of household behavior can be understood in terms
of the routines they use. And of course a quite extensive empirical and theoretical literature
exists, arguing that firm behavior largely involves the following of established routines.5 In
our earlier work, Sidney Winter and I used the term “routine” to characterize these aspects of
firm behavior. Here I am using it to denote the relatively automatic behavior patterns of any
economic actor.
The fact that little conscious thought is involved in the invoking and execution of a rou-
tine does not imply that routines are crude ways of doing things. The routines a store has for
reordering stock and for setting prices may be quite elaborate, even though once in place they
are carried out routinely. The operation of highly sophisticated technologies largely involves
the use of routines. Many of the routines used by economic actors are very powerful and highly
effective in meeting their objectives.
Also, routines need not be rigid. Indeed, viable routines generally have a reasonable amount
of flexibility built into them to enable them to adjust to the kind of variable circumstances that
are to be expected in the broad context where they are operative. Household shopping routines
need to be sensitive to what is and is not available at the store, and to some degree to prices.
Firm pricing routines need to take costs into account. But my argument is that in established
shopping routines, these adjustments generally are made relatively routinely. There may be
some conscious consideration of alternatives, but so long as the context remains in the normal
range, wide search and intensive deliberation are highly unlikely. Similarly, the pricing routines
of firms almost always are sensitive to costs, with much of that sensitivity, if not necessarily all,
built into a formula used relatively routinely.
Elsewhere (Nelson, 2013), I have used the term “adaptively responsive” to denote the sen-
sitivity of routines to broadly experienced and thus anticipated variation in the details of the
context that invokes their use. My proposal is that most routines that are used for a significant
time are adaptively responsive.
Economists of a neoclassical persuasion would be inclined to argue that routines persistently
employed by an economic actor must be, in some sense, optimal. Proponents of the view that
the rationality of economic actors is bounded would point out that the fact that behavior is
reasonably effective, given the actors goals, and adaptively responsive to common variations in
context for action taking, do not imply optimality. However, that an actor continues to use a
particular routine indicates that the results are “satisfactory” in the sense that doing things in a
significantly different way is not being actively considered.6 On the other hand, of course, some
of the actions that are carried out routinely by some actors are clearly clumsy, and some likely
even counterproductive, given the objectives they aim to reach. An important challenge for
evolutionary economics is to illuminate the conditions under which routines are effective, and
those where they often are not.
From one point of view, to explain or predict what an economic actor does in a domain of
activity marked by the use of routines, it is sufficient to identify and analyze the routines that
are in use. And this is exactly what is done in studies like those reported in the classic book by
Cyert and March (1963), A Behavioral Theory of the Firm.
But for the theory of behavior to have depth, it is important to understand why the routines
in use are what they are. I have argued that the neoclassical mode of answering that question –
to propose that they are optimal – is not convincing if one holds to a theory of bounded

487
488

Richard R. Nelson

rationality, and wants an explanation, not simply a purported characterization, of observed

behavior. Under evolutionary theory such an explanation needs to be posed in terms of
learning and selection processes.

Deliberating, problem solving, choosing

The proposition that much of the economic behavior one observes at any time should be
understood as actors following routines is not meant to play down the role of deliberation,
problem solving, and often creativity in the generation of economic activity. These more active
cognitive processes are brought into play when economic actors face contexts with which they
are not familiar and for this or other reasons no established response seems appropriate, or more
generally where the actor for whatever reason wants to do something new. And, of course, in
many cases they are involved in the genesis of prevailing routines in the first place.
This perspective is, of course, very Schumpeterian. Chapter 1 of his Theory of Economic
Development (Schumpeter, 1934) is all about routine activity in an economic steady state. In the
actual economic world as we know it, no context is as constant as the context for economic
action Schumpeter depicts in Chapter 1, or is laid out in general equilibrium theory. However,
evolutionary economists would argue that at any time a good portion of economic activity
proceeds in contexts that are regular enough so that behavior that follows an established rou-
tine can suffice to meet the actor’s objectives, at least if the routine used has a certain amount
of built in flexibility.
In Chapter 2, Schumpeter describes a very different kind of economic behavior: innovation.
Innovation is creative by desire or necessity, uncertain as to success, often failing, sometimes
winning big. But involving thinking and problem solving in an essential way.
In recent years cognitive scientists have significantly improved our understanding of how the
cognitive capabilities and practices of human beings differ from those of other higher animals;
the most interesting comparisons have been with other primates.7 There would appear to be
two basic capabilities that humans have that other primates do not. One is built in biologically.
The other, while based on this, is essentially cultural.
Other animals share with humans the ability to solve problems by doing different things
until they find something that works, and then carrying over what has been learned to subse-
quent experiences with situations like that. But humans have the ability, that even other pri-
mates have to a far lesser degree, to in effect reflect on a context or a way of doing something
(perhaps something they have observed others doing) even when that context is not present or
that action not being actually implemented, in effect, anticipating future situations and actions.8
Thus, the kind of deliberation we are considering here would seem to be a capability that is
largely unique to humans.
And humans are unique in having the capacity for cumulative collective learning. While the
cutting edge of progress generally has been discovery or trying out of a new method by an indi-
vidual actor, major advances over time have depended on the spread across the community of
what has been learned, and the further building on that by others.9 The development of shared
language has been essential for this to happen to any major extent. There is no question that
the ability of humans to reflect and gather and process relevant knowledge prior to action is an
important capability in its own right. However, I would argue that, in the absence of strong
cultural know-how that has been developed over time through collective learning on which
that capability can draw, what human reflection can achieve on its own is modest.
In my view, Schumpeter draws too sharp a line between innovating, and the imita-
tive responses by followers to the innovations of others. The latter also requires ability to

488
489

BR: basis for evolutionary economics

conceptualize a way of doing something that is new for the particular actor, and often involves
considerable uncertainty.
However, what is a new situation or new activity for a particular actor will tend to be
conceived very differently if that actor knows about and can draw on the experience of other
actors, than if the actor is all alone, as it were. Much of what actors do that is new to them is
invoked by their knowledge of the experience of others. The abandonment by an actor of an
old routine and the adoption of a new one may be induced simply by knowledge that others are
doing something different and doing well, as contrasted with any compelling evidence that the
old routine is not yielding satisfactory results. While direct imitation often is not easy, and the
efforts of one economic actor to do what another is doing may achieve something somewhat
or widely different, at any time a shared body of know-how provides the basis for the range
of activities used in a field, and is the reason why one generally observes a certain amount of
similarity in what the various actors are doing.

Innovation and the advance of know-how

While I believe the lines are blurred not sharp, the term “innovation” as contrasted with “imi-
tation” connotes an endeavor by an actor to do something new not only to that actor, but to
the community of actors doing roughly similar things. Empirical research shows clearly that
innovators, like imitators, almost always draw heavily on know-how, and more general know-
ledge, possessed by their peer community. And a large share of innovation is based on and aims
to improve artifacts and processes that are in use, often used by the innovator. But innovators
are reaching beyond what has been done before. And if they are successful, what they have
achieved sooner or later becomes part of the knowledge base shared by that community. That
is, know-how in an area of economic activity advances over time through an evolutionary pro-
cess driven largely by the innovation going on.
The principal difference between the orientation of evolutionary and Schumpeterian
economists, and that of today’s more orthodox orientation to the study of economics, is our
focus on innovation. Our argument is that what makes economic activity today so effective in
meeting a wide variety of human wants is that the means we have available to achieve our ends
have become so powerful as the result of cumulative innovation. It is not because economic
decision makers are so effective. Human economic decision making remains, as it always has
been, often mechanical, sometimes creative, but in these cases often mistake-ridden. Human
rationality is bounded.
While there has been considerable research over the years by scholars of management on
what makes firms successful, there is little evidence that firm managers today are more effective
than firm managers were a half century or a century ago. The failure rates of new firms, and of
new ventures by established firms, remains high. Business management remains an art, in which
luck is an important factor determining success.
The situation is similar regarding household purchases and other decisions regarding how
they spend their money. It is not for naught that we have in place a number of regulatory
agencies justified explicitly by the proposition that households often have limited understanding
of what they are buying. Wesley Clair Mitchell’s The Backward Art of Spending Money, published
in 1912, rings as true today as it was then.
However, boundedly rational human actors can achieve remarkably good outcomes, if the
know-how they have to work with, the means they know how to use, are good enough.
And where powerful know-how is available, and one observes highly effective human action
going on, the principal reason is not so much that someone or some organization has effectively

489
490

Richard R. Nelson

thought through the background problem and surmised, or calculated, a good way of doing
things in that context, but rather that there has been a lot of collective learning going on gener-
ally over a considerable period of time that, cumulatively, has led to the development of ways of
doing things that work reasonably, or even extraordinarily, well. Thus, a key part of the theory
of behavior and cognition that we need is a theory of how collective learning occurs.
Attempts at innovation clearly are the key driving force. However, a key premise of evo-
lutionary economics, amply supported by empirical evidence, is that the efforts of economic
actors to venture beyond established practice almost always are associated with uncertain
outcomes (see Dosi and Nelson, 2010, for a broad review). While in areas where knowledge
is reasonably strong, innovative efforts are far from blind, nonetheless all areas of innovative
activity are marked by failures as well as successes, and even the most knowledgeable experts
sometimes turn out to be wrong. A fundamental consequence is that, while economic progress
certainly depends on the creative efforts of individual inventors and innovators, it depends at
least as much on the existence of a number of potential innovators holding somewhat different
perceptions of the most promising routes to advance, with competition in ex-post practice
being a large part of the selection process determining the winners. And continuing progress
depends on the essence of what has been achieved in one round of innovative effort becoming
part of the collective knowledge base for the next round.
Put more generally, the remarkable increases in human knowhow that have been achieved
over the years have been the result of the work of boundedly rational human actors, operating
in a dynamic evolutionary context in which at any time effective new ways of doing things
are separated from the not-so-good, and brought into wider practice. And in turn this sets the
stage for the next round of efforts to advance the state of the art, which in turn are subject to
selection mechanisms.

A brief summing up
Earlier I noted that, since the times of Adam Smith, economists observing the behavior of eco-
nomic actors in the contexts in which they had a central interest have assessed these behaviors
as largely reasonably rational, given the actors’ apparent objectives, and the range of options
they faced.
But over the past half-century, modern neoclassical economics has transformed what had
been a quite flexible view of what “rationality” means into the much narrower notion that eco-
nomic actors “optimized.” Evolutionary economists have not been alone in arguing that this
has been a very unfortunate development. Our position is that the presumption that economic
actors mostly behave rationally is a powerful and useful theoretical position to take so long as
that rationality is understood as bounded. The conception of rational behavior must have room
for both creativity and habit, for both insightful understanding of the situation, and biased or
simply ignorant views of what is going on.
I believe that the kind of perspective on economic behavior and cognition that I have
sketched here, based on the presumption that economic behavior is “boundedly rational” and
recognizing important differences associated with different kinds of contexts and conditions,
has the promise of doing this. It provides a much better and richer characterization of economic
behavior that is for the most part purposeful and functional than the theory of full-blown opti-
mization that neoclassical theory is stuck with. It is applicable across a much wider spectrum of
conditions. And for those who care about such matters, it provides an explanation for much of
economic behavior that one actually can believe.

490
491

BR: basis for evolutionary economics

Notes
1 Key references here are Nelson and Winter (1982), and Nelson and colleagues (2018).
2 This perspective is first laid out in Schumpeter (1934).
3 The most relevant references for our discussion here are Simon (1957) and Cyert and March (1963).
4 These two different modes of action taking were built into most of the models developed in Nelson and
Winter (1982).
5 For a fine review of the literature on organizational routines, see Becker (2004).
6 This is Herbert Simon’s concept of “satisficing.”
7 Donald (1991) provides a splendid discussion of these and related matters.
8 There is some evidence that certain other animals have this capability, but to a very limited degree.
9 Other species have the capability of spreading the effective behaviors learned by one individual to others
in the community, but not of building further and cumulatively from that.

References
Becker, M. (2004). Organizational routines: A review of the literature, Industrial and Corporate Change,
13(4), 643–678.
Cyert, R., and March, J. (1963). A Behavioral Theory of the Firm, Englewood Cliffs, NJ: Prentice Hall.
Dewey, J. (2002 [1922]). Human Nature and Conduct, Minola, NY: Henry Holt.
Donald, M. (1991). Origins of the Modern Mind, Cambridge, MA: Harvard University Press.
Dosi, G., and Nelson, R. (2010). Technical change and industrial dynamics as evolutionary processes, in
B. Hall and N. Rosenberg (Eds.), Economics of Innovation, Amsterdam Elsevier.
Kahneman, D. (2011). Thinking, Fast and Slow, New York Macmillan.
Mitchell, W. C. (1912). The backward art of spending money, American Economic Review, June, 269–281.
Nelson, R. (2013). Demand, supply, and their interaction on markets as seen from the perspective of evo-
lutionary economic theory, Journal of Evolutionary Economics, 23(1), 17–38.
Nelson, R., and Consoli, D. (2010). An evolutionary theory of household behavior, Journal of Evolutionary
Economics, 20(5), 665–687.
Nelson, R., Dosi, G., Helfat, C., Pyka, A., Saviotti, P., ... Dopfer, K. (2018) An Overview of Modern
Evolutionary Economics, Cambridge: Cambridge University Press.
Nelson, R., and Winter, S. (1982). An Evolutionary Theory of Economic Change, Cambridge, MA: Harvard
University Press.
Schumpeter, J. (1934), The Theory of Economic Development, Cambridge, MA: Harvard University Press,
Simon, H. (1957), Models of Man, New York: John Wiley & Sons, Inc.

491
492

33
BEYOND “BOUNDED
RATIONALITY”
Behaviours and learning in complex
evolving worlds

Giovanni Dosi, Marco Faillo, and Luigi Marengo

Introduction
Let us begin by the very notion of bounded rationality. Should we assume that there is an
“unbounded” rationality as a benchmark? Should one start, in order to describe and interpret
human behaviour from a model which assumes that we, human beings, have complete and
well-defined knowledge of our preferences, all possible states of the world, all possible actions
(our “technologies”), and the mappings among them?1
Savage was extremely careful in limiting his choice-theoretic exercise to the normative
domain and to “small worlds”, i.e., stationary and isolated portions of the world wherein
decision makers know the full set of possible events and can attribute probabilities to them
(Savage, 1954).
Jumping from this normative small world domain to the descriptive framework where one
builds theory of human behaviour in a complex world, characterized by radical uncertainty, a
multiplicity of interactive agents, and persistent endogenous innovation – we suggest – is deeply
misleading. Having “Olympic rationality” as a benchmark is like starting from the thermo-
dynamic equilibrium death with full entropy in order to interpret the biological world!
Rather, the question should be: how do human agents and organizations thereof behave in
complex and changing environments? Answering this question, we suggest, entails also a sig-
nificant departure from what is now accepted as behavioural economics, often meant as the
analysis of more or less significant deviations – called “biases” – from the “Olympic rationality”.
On the contrary, we suggest, human beings and human organizations behave quite distinctively
from the prescriptive model derived from the axioms of rationality.
As is well known, the standard decision-theoretic model depicts agency (and, in primis,
economic agency) as a problem of choice where rational actors select, among a set of alterna-
tive courses of action, the one which will produce (in expectation) the maximum outcome as
measured against some utility yardstick. In that, agents are postulated to know the entire set
of possible events of “nature”, all possible actions which are open to them, and all notional
outcomes of the mapping between actions and events (or at least come to know them after some
learning process). Clearly, these are quite demanding assumptions on knowledge embodied into
or accessible to the agents – which hardly apply to complex and changing environments. In

492
493

Beyond “bounded rationality”

fact, they cannot apply almost by definition in all environments where innovations of some
kind are allowed to occur – irrespective of whether they relate to technologies, behavioural
repertoires or organizational arrangements. If an innovation is truly an innovation, it could not
have been in the set of events that all agents were able to contemplate before the innovation
actually occurred.
Moreover, equally demanding are the implicit assumptions concerning the procedural ration-
ality involved in the decision process.
As a paradigmatic illustration, take the usual decision-theoretic sequence leading from
(1) representation/“understanding” of the environment (conditional on whatever avail-
able “information”), to (2) evaluation/judgement; (3) choice; (4) actions, and, ultimately,
(5) consequences – determined, e.g., by the stochastic pairing of actions and “events of nature”
and/or actions by other agents.
In order for this “rationalist” view to hold, at least two assumptions are crucial. First, the
linearity of that sequence must strictly hold. That is, one must rule out the possibility of
reversing, so to speak, the procedural sequence. For example, one cannot have preferences
and representations which adapt to an action that has already been undertaken. and, like-
wise, one must assume that consequences do not influence preferences (i.e. preferences are not
endogenous).
Second, at each step of the process agents must be endowed with, or able to build, the
appropriate algorithm in order to tackle the task at hand – whether it is representing the envir-
onment, evaluating alternatives or choosing courses of action, etc.
There are, indeed, a few rather compelling reasons why these assumptions might be a
misleading starting point for any positive theory of learning and choice.
Human agents tackle every day, with varying degrees of success, highly complex and “hard”
(in the sense of computability theory) problems with their highly limited computational capabil-
ities. Cognitive sciences have made impressive progress in the recent decades in understanding
how we do that. We are bad in processing information, we cannot handle more than a very
limited number of the overwhelming number of interdependencies the characterize our world,
but nevertheless we go along, sometimes decently well, with simple but useful representations
and simple but effective heuristics. As suggested by Gigerenzer and his group, such heuristics
are not the outcome of our biases, although they may sometimes produce them (Gigerenzer
et al., 1999). On the contrary it is their very simplicity which makes them “smart”, and gen-
erally well adapted to the complex and fast-changing world in which we live. They require
simple representations and neglect part of the available information, that is, they radically depart
from that model of rationality which assumes correct representation and unlimited informa-
tion processing capabilities, but, on the contrary, excel in simplicity, frugality, adaptability, i.e.,
features with are not even considered in the rational choice framework. “Olympic rationality”,
in fact, implies the availability of some inferential machinery able to extract the “correct” infor-
mation from environmental signals, Bayes rule being one of them, and possibly also the most
demanding in terms of what the agents must know from the start about alternative hypotheses
on what the world “really is”. But, again, such an inferential machinery cannot be innocently
postulated. Indeed, outside the rather special domain of “small worlds” whose structure is
known ex ante to the agents, a few impossibility theorems from computation theory tell us
that a generic inferential procedure does not and cannot exist (more on this point in Binmore,
1990; Dosi and Egidi, 1991; Dosi et al., 1994). This applies even more to so-called “Rational
Expectations”. It has repeatedly been shown that agents cannot generically learn even in simple
stationary environment, and less so in complex evolving ones. More than that: under the
latter circumstances trying sophisticated forms of learning is bad for the agents – in terms of

493
494

Giovanni Dosi, Marco Faillo, and Luigi Marengo

prediction and performance – and is bad for the system – in terms of its growth and stability
(on both points, see the discussion and the results from Dosi, Napoletano, Roventini, Stiglitz,
and Trebich, 2017).2
Complexity arguments also imply a radical critique to the idea that “rationality” – however
defined – rather than being an approximation to the empirical behaviours of purposeful, cog-
nitively quite sophisticated, agents, could be, so to speak, an “objective” property of behaviours
in equilibrium. Add the presumption that (most) observed behaviours are indeed equilibrium
ones. And finally postulate some dynamics of individual adaptation or intra-population selec-
tion leading there. What one gets is some version of the famous “as...if ” hypothesis, suggested
by Milton Friedman (1953) and rejuvenated in different fashions by more recent efforts to
formalize learning/adaptation processes whose outcome is precisely the “rationality” assumed
from the start (archetypical examples of this faith can be found in Sargent, 1993, and McGrattan
and Marimon, 1995).
A thorough, critical, discussion of the “as...if ” epistemology has been put forward by Sidney
Winter, in various essays (e.g., Winter, 1971) to which we refer the interested reader (and see
also Silverberg, 1988; Andersen, 1994 and Hodgson, 1988).
For our purposes here, let us just note the following:

1 Any “as...if ” hypothesis on rationality, taken seriously, is bound to involve quite a few
restrictions similar to those briefly overviewed earlier with reference to more “constructive”
notions of rational behaviours, simply transposed into a more “ecological” dimension –
whether it is the “ecology” of minds, ideas, organisations, populations, etc. That is, canon-
ical rationality, stricto sensu, postulates that one decides and acts by purposefully using the
appropriate procedures (or by learning them in purposeful, procedurally coherent, ways).
“As...if ”s of any kind apparently relax the demands on what agents must consciously know
about the environment, their goals, the process of achieving them, but at the same time
must assume some background mechanism that generates the available alternatives – which
must include the “correct” ones. However, without any further knowledge of the specific
mechanisms, such a possibility remains a very dubious shortcut. And it is utterly unlikely
when there are infinite alternatives which ought to be scanned.
2 While “realistic” interpretations of rationality put most of the burden of explanation upon
the power of inbuilt cognition, “as...if ” accounts shift it to selection dynamics – no matter
whether driven by behavioural reinforcements alike salivating Pavlovian dogs, or by differ-
ential reproduction of traits within populations.3 But, then, supporters of the view ought to
show, at the very least, robust convergence properties of some empirically justifiable selection
dynamics. As it stands, in our view, nothing like that is in sight. On the contrary, except for
very special set-ups, negative results are abundant in e.g., evolutionary games or other forms
of decentralized interactions – no matter whether applied to biology or economics – path-
dependency cannot easily be disposed of; cyclical limit behaviours might occur (cf. Posch,
1997, Marengo and Pasquali (2011), and Kaniovski et al., 1997), etc. And all this appears
even before accounting for environments which are genuinely evolutionary in the sense
that novelties can emerge over time.

But, even more importantly, we can add that, in complex worlds, selection is almost power-
less as an optimization mechanism. If the entities under selection have some internal structure
made up of many interdependent components, such structural properties pose huge constraints
on the evolution, and selection alone cannot break such a constraint (Kauffman, 1993). For
instance, in biological evolution, the question whether an organism is optimal is nonsensical.

494
495

Beyond “bounded rationality”

No doubt that standing on two legs has given us some useful evolutionary advantage, but is
also a cause of many “inefficiencies” (back and knee weakness, difficulties in giving birth,
etc.). Again, Gigerenzer and colleagues suggest an idea of “ecological rationality” rather than
Olympic rationality, i.e., a system of heuristics that have co-evolved (Todd and Gigerenzer,
2012) and may all be suboptimal (whatever this may mean) if taken separately, but together
produce a decently working system.
Another major perspective maintains that cognitive and behavioural assumptions must keep
some empirical foundations and, thus, when needed, account for constraints on memory, on
the maximum levels of complexity of problem-solving algorithms, and on computational time.
It is, in a broad sense, the bounded rationality approach, pioneered by the works of Simon
(1986) and developed in quite different fashions in e.g., organizational studies (starting from
March and Simon, 1958 and Cyert and March, 1963); evolutionary theories (building on
Nelson and Winter, 1982; see also Dosi et al., 1988; Hodgson, 1993; Andersen, 1994); “evo-
lutionary games” (for a rather technical overview, cf. Weibull, 1995; for insightful remarks on
bounded rationality and games in general, Kreps, 1996, and also in otherwise quite orthodox
macroeconomics, see e.g., Sargent, 1993). Again, this is not the place to undertake any review
of this vast literature. However, few comments are required.
Of course, the very idea of “bounds” on rationality implies that, at least in finite time, agents
so represented fall short of full substantively rational behaviours, the latter involving among other
things, (1) a full knowledge of all possible contingencies; (2) an exhaustive exploration of the
entire decision tree; and (3) a correct appreciation of the utility evaluations of all mappings
between actions, events and outcomes (Simon, 1986).
Given that, a first issue concerns the characterization of the origins and nature of the
“boundedness” itself. It is not at all irrelevant whether it relates mainly to limitations on the
memory that agents carry over from the past, or to algorithmic complexity of the decision
problem to be addresses, or to limited ability of defining preferences over (expected) outcomes.
Or, more radically, couldn’t it be due to the fact that agents get it basically wrong (in terms
of representation of the environment, etc.)?
Here the theory faces a subtle but crucial crossroads. One alternative – unfortunately found
too often in economic models, and especially but not only, in game theory – is to select the
bounded-rationality assumptions with extreme casualness, suspiciously well-fitted to the math-
ematics the authors know and to the results one wants to obtain. We have no problem in asso-
ciating ourselves to those who denounce the ad hocry of the procedure. The other alternative
entails the acknowledgement of an empirical discipline upon the restrictions one puts upon the
purported rationality of the agents. No doubt, we want to advocate here the scientific soundness
of this procedure, notwithstanding the inevitable “phenomenological” diversity of cognitive
and behavioural representations one is likely to get. That is, whether and how “rationality is
bound” is likely to depend on the nature of the decision problem at hand, the context in which
the decision-maker is placed, the pre-existing learning skills of the agents, etc. Taxonomical
exercises are inevitable, with their seemingly clumsy reputation. But, in a metaphor inspired by
Keith Pavitt, this is a bit like the comparison of Greek to modern chemistry. The former, based
on the symmetry of just four elements, was very elegant, grounded in underlying philosoph-
ical principles, utterly irrelevant, and, from what we know nowadays, essentially wrong. The
latter is clumsy, taxonomic, and for a long time (until quantum mechanics) lacking underlying
foundations, but is certainly descriptively and operationally more robust.
A second major issue, regards procedural rationality. Granted the bounds on “substan-
tive” rational agency, as defined above, when and to what extent should one maintain any
assumption of coherent purposefulness and logical algorithmic consistency of the agents?4 In

495
496

Giovanni Dosi, Marco Faillo, and Luigi Marengo

a first approximation, Simon’s approach suggests such a theoretical commitment (associated

indeed to major contributions to the identification of constructive procedures for learning and
problem-solving in this vein (Newell and Simon, 1972; Simon, 1976). However, even proced-
ural consistency might not be at all a generic property of empirical agents, and a lot of evidence
from most social disciplines seems to point in this direction.
Third, and relatedly, the very notion of “bounded rationality” (of, we repeat, the vast
majority of contemporary economists, though not of Herbert Simon) commits from the start
to an implicit idea that “full rationality” is the underlying yardstick for comparison. In turn, this
implies the possibility of identifying some metrics upon which “boundedness” and, dynamic-
ally, learning efforts could be measured and assessed. In quite a few circumstances this can be
fruitfully done5 but in others it might not be possible either in practice or even in principle. In
particular, this applies to search and learning in complex functional spaces (as many problems
within and outside the economic arena commonly do).6 And, of course, this is also the case of
most problems involving discovery and/or adaptation to novelty.
Since indeed these features are typical of evolutionary environments, an implication is
that one might need to go well beyond a restricted notion of “bounded rationality”, simply
characterized as an imperfect approximation to a supposedly “full” one – which, in these
circumstances, one is even unable to define what it should precisely be.
But then, again, how does one represent learning agents in these circumstances? Our some-
what radical suggestion is that evolutionary theories ought to make a much greater and system-
atic use of the evidence from other cognitive and social sciences as sort of “building blocks” for
the hypotheses on cognition, learning and behaviours that one adopts. We fully realize that such
a perspective almost inevitably entails the abandonment of any invariant axiomatics of decision
and choice. But, to paraphrase Thaler (1992), this boils down again to the alternative between
being “vaguely right” or “precisely wrong”: we certainly advocate the former (however, com-
pare McGrattan and Marimon (1995) for a sophisticated contrary view).
In this respect, the discussion of routines as foundational behavioural assumptions of evo-
lutionary models in Nelson and Winter (1982) is an excellent example of the methodology
we have in mind, unfortunately not pursued enough in subsequent evolutionary studies (see
Cohen et al., 1996; Becker 2004; Becker et al., 2005).
There are, however, many other fields where a positive theory of learning in economics can
draw, ranging from cognitive and social psychology all the way to anthropology and sociology
of knowledge.

Cognitive categories and problem solving

A crucial aspect of learning regards most often cognition, that is the process by which decision
makers form and modify representations in order to make some sense of a reality which is gen-
erally too complex and uncertain to be fully understood. Hence, the necessity to acknowledge
the existence (and persistence) of a systematic gap between the agents’ cognitive abilities and
“reality” (were there an omniscient observer able to fully grasp it). Such a gap can take at least,
two often interrelated forms,7 namely, first, the knowledge gap, involving incomplete, fuzzy or
simply wrong representations of the environment and, second, a problem-solving gap between the
complexity of the tasks agents face and their capabilities in accomplishing them.
Regarding both, evolutionary theories of learning might significantly benefit from that
branch of cognitive studies concerned with the nature and changes of categories and mental models
(in different perspectives, cf. Johnson-Laird, 1983; Holland et al., 1986; Lakoff, 1987; Margolis,
1987; and the presentation of a few alternative theories in Mayer, 1992). It is crucial to notice

496
497

Beyond “bounded rationality”

that, if one accepts any “mental model” view, learning cannot be reduced to information-
acquisition (possibly cum Bayesian processing of it), but rather is centred around the construc-
tion of new cognitive categories and “models of the world”.
In turn, a robust evidence shows that cognitive categories are not clear-cut constructions
with sharp boundaries and put together in fully consistent interpretative models. Rather, they
seem to display (in all our minds!) blurred contours, shaded by an intrinsic fuzziness, held
around some cognitively guiding “prototypes”, and organized together in ill-structured systems
kept operational also via a lot of default hierarchies (cf. on all those points, see Tversky and
Kahneman, 1982; Holland et al., 1986; Kahneman and Tversky, 1986; Lakoff, 1987; Hahn
and Ramscar, 2001; Gärdenfors, 2004; Fehr, 2005).8 In this domain, note, however, a subtle
but fundamental difference: is “prototypization” a “bias” or an inherent property of cognitive
categorization? That is, is it similar to e.g., anchoring biases, à la Tversky and Kahneman, in
principle still linkable to Olympic rationality with variable doses of “boundedness”? Or, on the
contrary, is it intimately related to the very nature of mental categories? The answer we suggest
is indeed in favour of the latter interpretation.9

Framing and social embeddedness

Cognitive categories, it has repeatedly been shown, go together with various mechanisms of
framing by which information is interpreted and also rendered operationally meaningful to the
decision makers (cf. Kahneman et al., 1982; Borcherding et al., 1990; March, 1994).
Frames appear to be indeed a ubiquitous feature of both decision making and learning. What
one understands is filtered by the cognitive categories that one holds and the repertoires of
elicited problem-solving skills depend on the ways the problem itself is framed. That is, framing
effects occurs along all stages of the decision process – affecting representations, judgements and
the selection of behaviours (cf. Kahneman et al., 1982), and, concerning the patterns of activa-
tion of experts’ skills (Ericsson and Smith, 1991).
As James March put it:

Decisions are framed by beliefs that define the problem to be addressed, the infor-
mation that must be collected, and the dimensions that must be evaluated. Decision
makers adopt paradigms to tell themselves what perspective to take on a problem,
what questions should be asked, and what technologies should be used to ask the
questions. Such frames focus attention and simplify analysis. They direct attention to
different options and different preferences. A decision will be made in one way if it
is framed as a problem of maintaining profits and in a different way if it is framed as a
problem of maintaining market share. A situation will lead to different decisions if it
is seen as being about “the value of innovation” rather than “the importance of not
losing face”.
1994, p. 14

Note that in this view, “frames” include a set of (non-necessarily consistent) beliefs over “what
the problem is” and the goals that should be achieved in that case; cognitive categories deemed
to be appropriate to the problem; and a related menu of behavioural repertoires.
Moreover, framing mechanisms appear at different levels of cognitive and behavioural
observation: they do so in rather elementary acts of judgement and choice, but are also a
general organizing principle of social experience and collective interactions (Bateson, 1972;
Goffman, 1974).

497
498

Giovanni Dosi, Marco Faillo, and Luigi Marengo

One can intuitively appreciate also the links between framing processes and social embedded-
ness of both cognition and action.10
Frames, in the broad definition given above, have long been recognized in the sociological,
psychological and anthropological literature (whatever name is used to refer to them) as being
grounded in the collective experience of the actors and in the history of the institutions in
which agency is nested.11
Indeed, embeddedness seems to go a striking long way and affect even the understanding
and use of cognitively basic categories, such as that of causality and the very processes by
which humans undertake basic operations, such as inferences, generalisations, deductions,
etc. (Luria, 1976; Lakoff, 1987; D’Andrade, 2001; Kitayama, 2002; Oyserman and
Lee, 2008).
Far away from standard rationality, a long and unjustly forgotten broadly defined Austrian
tradition has tried to capture cognition and decision outside the straitjacket of the “max-
something” framework. Hayek’s “Sensory Order” (Hayek, 1952) is probably the most
sophisticated synthesis of that view which

not only emphasizes that the only ways open to people for making sense of their
environment are the ways they already possess (the environment does not dictate
how they see it), and whose probabilistic analysis of how the ‘ways’ at a person’s
disposal get called upon (the probabilities being a function of recent use and cumu-
lative use) to see if there is a match with incoming stimuli provides a ‘plastic’ view
of the mind.12

Related to Hayek’s Sensory Order, are the ideas of Personal Construct Psychology (Kelly,
1955), in which the organizational structure that an individual creates to make sense of the
world limits her permeability to new ways of thinking. This line of thinking was first applied
in economics in Loasby (1983), where it is argued that organizational change is problematic if
“core” constructs are involved, even if what is going on in the firm’s environment may require
the development of a new construct in order to adapt and survive.
A somewhat parallel and almost entirely distinct literature focuses upon the crucial role
of tacit knowledge (Polanyi, 1966). Building on that notion, behavioural and evolutionary
economists have made a fruitful use of habits and, collectively, routines (see below) in order to
characterize behavioural patterns.
By “genuinely behavioural” we mean that interpretative tradition which tries to characterize
behavioural regularities in their own right (archetypical examples are Cyert and March, 1992,
and March and Simon, 1958) as distinct from the somewhat more restrained approach to the
description of actual behaviours in terms of deviations from some normative notion of prefect
rationality as discussed above.
First, even in the simplest setups including stationary environments, satisficing behaviour
may yield a probability of surviving for ever, while maximizing ones are sure to yield death to
inferior options in finite time (Dutta and Radner, 1999). That is the exact opposite to the “as
if ” hypothesis.
Second, heuristics tend to be “fast and frugal”, meaning that they are rules of thumb for
decision making that are ecologically sound, simple enough to operate when time, informa-
tion and computation are limited and grounded in human psychological capabilities, such as
memory and perception (Gigerenzer and Goldstein, 1996).
Third, in largely unknown environments, even if stationary in their fundamentals, higher
“competence gaps” may hinder the agent’s capacity to assess which behaviour is the most

498
499

Beyond “bounded rationality”

appropriate in which environmental conditions. Behavioural inertia is the outcome, other

things being equal, of higher environmental dynamics: “uncertainty is the basic source of pre-
dictable behaviour ... [T]he flexibility of behaviour to react to information is constrained to
the smaller behavioural repertoires that can be reliably administered” (Heiner, 1983, p. 585).
Indeed, this insightful conjecture from a strikingly neglected path-breaking demonstration is
explored in Dosi, Napoletano, Roventini, Stiglitz, and Trebich, 2017), together with its applic-
ability to non-stationary environment.
Fourth, and closer to our concern here, memory does not involve primarily information on
past events (say, the memory of an econometrician going back in her time series), but rather
memory of heuristics – both in their pattern recognition side and behavioural one.

From individuals to organizations

As already mentioned, one side of the story is, in a broad sense, cognitive. The view
of organizations as fragmented and multidimensional interpretation systems is grounded
on the importance of collective information processing mechanisms that yield shared
understandings (Daft and Weick, 1984), or “cognitive theories” (Argyris and Schön, 1978),
of the environment in which they operate, and that assist organizations to bear uncertainty,
besides, as we shall see, manage environmental and problem-solving complexity. If one
subscribes to the notion that organizational learning is a process of refinement of shared
cognitive frames involving action-outcome relationships (Duncan and Weiss, 1979), and
that this knowledge is retained – at least for some time – and can be recalled upon necessity,
this is like saying that organizational learning is in fact the process of building an organ-
izational memory. This cognitive part of the memory is made up of “mental artefacts”
embodying shared beliefs, interpretative frameworks, codes and cultures by which the
organization interprets the state of the environment and its own “internal states” (Levitt
and March, 1988).
Together, there is an operational side to the organizational memory involving the coupling
between stimuli (events and signals, both external and internal ones) with responses (actions),
making up a set of rules that remain available to guide the orientation of the organization and
execute its operations. In this domain, memory largely relates to the ensemble of organizational
routines – patterned actions that are employed as responses to environmental or internal stimuli,
possibly filtered and elaborated via the elements of cognitive memory (much more on routines
in Nelson and Winter, 1982; Cohen et al., 1996; Becker, 2004; Becker et al., 2005, and the
literature reviewed here). As Cohen and Bacdayan (1994) put it, this procedural side is the
“memory of how things are done”, bearing a close resemblance to individual skills and habits,
often with relatively automatic and unarticulated features (p. 554).
Cognitive and operational memories entail an “if…then” structure. Signals from the envir-
onment, as well as from other parts of the organization, elicit particular cognitive responses,
conditional upon the “collective mental models” that the organization holds, which are in turn
conditional upon the structure of its cognitive memory. Cognitive memory maps signals from
an otherwise unknown world into “cognitive states” (e.g., “… this year the conditions of the
market are such that demand for X is high ...”). Conversely, the operational memory elicits
operating routines in response to cognitive states (“… produce X …”), internal states of the
organization (“…prepare the machine M to start producing piece P…”) and also environmental
feedbacks (“… after all X is not selling too well …”). In turn, the organizational memory
embodies the specific features of what an organization “thinks” and does, and what it is “good
at”, that is, its distinct capabilities.13

499
500

Giovanni Dosi, Marco Faillo, and Luigi Marengo

Modelling routines, memory and learning

For a long time all the way to the present, organizational models have run far behind the
qualitative interpretations briefly discussed above. Some catching-up has occurred, however,
especially in the field of modelling learning processes in high dimensional spaces with relatively
limited adaptation mechanisms. A promising candidate to model routines and memory finds its
roots into the formalism of Classifier Systems (CSs) (Holland, 1975; Holland et al., 1986). In a
nutshell, a CS is a system of interlinked condition/action rules that partly evolves according to
the revealed environmental payoffs.
Dosi, Marengo, Paraskevopoulou and Valente (2017) present a model which links Classifiers
Systems and NK fitness landscape models (Kauffman, 1993). The former provides a model of
a memory system that accounts for both cognitive and operational memory, while they use
the latter to represent an environment in which exogenous environmental traits and organiza-
tional actions or policies interact in a complex way to determine the organization fitness or
payoff. While in standard NK models (e.g., Levinthal, 1997), cognition, actions and resulting
payoffs are folded together in a mapping between “traits” and their “fitness”, they unfold such
a map, explicitly defining the cognition/action/environmental feedbacks and modelling their
(evolving) coupling. This is, we believe, a first major advance with respect to the existing lit-
erature. The organization explores a complex and possibly changing landscape in which some
dimensions are outside its control (the environmental traits) and some are within (the action
traits). Since the former contribute to determine the payoff of the latter, the organization
must base its search over the action landscape on an internal representation (its cognition)
of the environmental landscape. When the landscape is complex enough and the organiza-
tion has cognitive and memory bounds, such an internal representation can only be partial,
imperfect and possibly wrong. However, in practice, through the accumulation of experience,
organizations can develop better representations that enable them to act successfully in such a
complex environment. This is a way of saying that organizations painstakingly and imperfectly
learn and develop models of their environment. However, there is an exogenous world “out
there” which is indeed the object of learning, and which of course is not controlled by the
organization. Rather the organization has to learn what to do – the know-how – conditional
on (what it believes to be) the characteristics of the landscape mapping the combinations of
state-of-the-world and actions into payoffs. This is also another major difference vis-à-vis the
NK modelling style wherein the “blackboxing” renders all the landscape notionally under the
control of the agent. Moreover, the CS formalism allows a straightforward study of learning
via non-local search, which, if undertaken at all in NK frameworks, turns out to be quite
arbitrary.
In fact, the characteristics and evolution of organizational memory mirror the characteristics
and evolution of organizational routines. In the case of routines, the memory elicits a “rela-
tively complex pattern of behaviour triggered by a relatively small number of initiating signals
or choices” (Cohen et al., 1996). How small or big is the initiating set of signals in itself is an
important interpretative question, which has to do with the ways the organization categorizes
environmental and intra-organizational information. And, likewise, the behavioural patterns
are likely to display different degrees of conditionality upon particular sets of signals. So, at one
extreme, the action pattern might be totally unconditional and “robust”: “perform a given
sequence of actions irrespectively of the perceived state of the world”. At the opposite extreme

500
501

Beyond “bounded rationality”

actions might be very contingent on the fine structure of the “if ” part, detailing very precisely
the environmental conditions triggering the action part.

Conclusion
A multi-millennial tradition of Western thought has asked “how do people behave?” and “how
do social organization behave?”, from Aristotle to St Augustin, Hume, Adam Smith, Kant,
to name only a few giants. However, modern economics – and more recently social sciences
colonized by modern economics – have taken up the answer by one of the shallowest thinkers,
Bentham: people decide their courses of action by making calculations on the expected
pleasures and pains associated with them. And, indeed, this Weltanschauung has spread all the
way to the economics of marriage, of child bearing, of church going, of torture… (some more
comments in Dosi and Roventini, 2016).
Our argument is that the Benthamian view is misleading or plainly wrong concerning the
motivations, decision processes and nature of the actions.
First, the drivers of human motivation are many more than one. As Adam Smith masterly
argues in its Theory of Moral Sentiments, utility (what he called “prudence”) is just one of them,
and in a lot of social contexts, not the most important one.
Second, the decision processes are very rarely explicit calculations and comparisons of
outcomes.
Third, the ensuing decisions very seldom look like a “rational” (“as…if ”) outcome of the
foregoing decision processes, even when the latter would be possible to calculate. And in the real
word, complex and evolving as it is, they rarely are.
In such circumstances, we suggest, a positive theory of individual and collective behaviours
has to entirely dispose of the max U(…,…, …) apparatus, either as an actual descriptive
device, and as a yardstick, whatever that means. If we are right, then also relaxations of the
paradigms involving varying degrees of “bounded rationality” in the decision process and an
enlargement of the arguments in the utility function (e.g., adding “intrinsic motivations”, or
even “altruism”) are quite misleading. They are a bit like adding epicycles over epicycles in a
Ptolemaic astronomy.
The radical alternative we advocate is an anthropology of a homo heuristicus (Gigerenzer and
Brighton, 2009), socially embedded, imperfectly learning in a complex evolving environment,
and with multiple drivers of his actions.
A tall task, but it is time to break away from a paradigm that trivializes the analysis of human
behaviour, reducing it to sterile exercises of maximization over some arbitrary and ad hoc
functions.

Acknowledgements
We thank Riccardo Viale for his insightful comments and suggestions. GD acknowledges the
earlier support of the ISIGrowth project under grant agreement No. 649186 ISIGrowth, GD
and LM the current support of the GROWINPRO project grant n. 822781, both from the
European Union Horizon 2020 Programme.
This work draws heavily on previous publications, in particular, Dosi, Marengo and Fagiolo
(2004) and Dosi et al. (2017).

501
502

Giovanni Dosi, Marco Faillo, and Luigi Marengo

Notes
1 Note that here and throughout we address the notion of bounded rationality used by most contem-
porary economists as a “full rationality” minus some frictions, memory limitations, biases, noise, etc.
and not the much richer notion conceived by Herbert Simon which stood from the very cognitive and
perceptual boundaries in the interactions between humans and their environments.
2 Admittedly, the point is not uncontroversial as some scholars suggest that perceptual processes and
categorizations are consistent with Bayes rule (e.g. Griffith, Kemp and Tenenbaum, 2008). We are not
cognitive scientists, but frankly we find it hard to believe that people learn in Bayesian manners how
to swim in the Heraclitean river where you can never bathe twice. More generally, for a thorough
discussion of descriptive (as opposed to normative) theories of cognition and action, see Viale (2018).
3 Incidentally, note that the outcomes of pure “Pavlovian” – i.e., reinforcement-driven, consciously
blind – and “Bayesian” – apparently sophisticated rational – dynamics can be shown to be sometimes
asymptotically equivalent (see the review in Suppes 1995a, 1995b, who develops much older intuitions
from behaviourist psychology, e.g. Bush and Mosteller, 1955). However, in order for that equivalence
to hold, reinforcements must operate in the same direction as the Bayesian inferential machinery,
which is, indeed, a hard demand to make. The so-called condition of “weak monotonicity” in the
dynamics of adjustment that one generally finds in evolutionary games is a necessary, albeit not suf-
ficient, condition to this effect. Moreover, a subtle question regards the interpretative value that one
should attribute to asymptotic results: what do they tell us about finite time properties of empirical
observations? (We shall briefly come back to the issue below.)
4 Note that procedural rationality requires all the “linearity assumptions” mentioned above (ruling out,
for example, state-dependent preferences) and also consistent search heuristics (allowing, for example,
assessment rules along any decision tree which at least in probability lead in the “right” direction).
5 Promising results stem from a better understanding of the formal structure of problem-solving
heuristics (cf. e.g. Pearl, 1984; Vassilakis, 1997; and, in a suggestive experimentally-based instance,
Cohen and Bacdayan, 1994, and Egidi, 1996). See also below.
6 For example, in Dosi et al. (1999) consider quantity- and price-setting as cases to the point.
7 Heiner (1983) introduces a similar concept which he calls the “C-D (competence-difficulty) gap”.
In his definition, such a gap reflects the agent’s imperfect capabilities to correctly process the available
information and act reliably. Heiner’s C-D gap does not properly belong to the realm of cognitive gaps,
but it rather captures their behavioural consequences.
8 “Prototypization” is easy to intuitively understand: you would give a sparrow rather than a penguin
as an example of what a bird is. But with that it is also easier to understand the basic ambiguity of
borderliners, fuzziness and categorical attributions by default (how should one treat a duck-billed
platypus?, as a mammal? or should one create a separate category, that of ovoviviparous?). A discussion
of these issues bearing on economic judgements and behaviours is found in Tordjman (1998).
9 For a thorough discussion of algorithmic processes, see Lakoff (1987) and Bonini et al. (1999).
10 On the notion of “social embeddedness” as from contemporary economic sociology, see Granovetter
(1985) and several contributions in Smelser and Swedberg (2006). A discussion quite germane to the
argument developed here is found in Tordjman (1998).
11 Within an enormous literature, just think of a good deal of the sociological tradition influenced by
the works of Talcott Parson or of the classic Bourdieu (1977); in anthropology, among others, cf. the
discussions of “embeddedness” by Karl Polanyi (1944, 1957); and Geertz (1963); see also Edgerton
(1985).
12 See also Frantz and Leeson (2013) for a recent reappraisal of the Hayekian view in relation to behav-
ioural economics.
13 Within a very large literature, cf. for instance Helfat et al. (2006) and the critical survey in Dosi et al. (2008).

References
Andersen, E. S (1994). Evolutionary Economics: Post-Schumpeterian Contributions, London: Pinter.
Argyris, C., and Schön, D. (1978). Organisational learning: A Theory of Action Perspective. Reading,
MA: Addison-Wesley.
Bateson, G. (1972). Steps to an Ecology of Mind, New York: Ballantine Books.
Becker, M. C. (2004). “Organizational routines: A review of the literature”. Industrial and Corporate Change
13(4): 643–678.

502
503

Beyond “bounded rationality”

Becker, M. C., Lazaric, N., Nelson, R. R., and Winter, S. G. (2005). “Applying organizational routines in
understanding organizational change”, Industrial and Corporate Change 14 (5): 775–791.
Binmore, K. (1990). Essays on the Foundations of Game Theory, Oxford: Blackwell.
Bonini, N., Osherson, D., Viale, R., and Williamson, T. (1999). On the psychology of vague predicatives,
Mind and Language 14: 377–393.
Borcherding, K., Larichev, D. L., and Messick, D. M. (Eds.) (1990). Contemporary Issues in Decision Making,
New York: North-Holland.
Bourdieu, P. (1977). Outline of a Theory of Practice, Cambridge: Cambridge University Press.
Bush, R. R., and Mosteller, F. (1955). Stochastic Models for Learning, New York: Wiley
Cohen, M. D., and Bacdayan, P. (1994). “Organizational routines are stored as procedural memory: Evidence
from a laboratory study”, Organizational Science 5: 554–568.
Cohen, M. D., Burkhart, R., Dosi, G., Egidi, M., Marengo, L., ... Coriat, B. (1996). “Routines and
other recurring action patterns of organisations: Contemporary research issues”, Industrial and Corporate
Change 5: 653–698.
Cyert, R. M., and March, J. G. (1992). A Behavioral Theory of the Firm (2nd ed.), Oxford: Basil Blackwell.
D’Andrade, R. (2001). “A cognitivist’s view of the units debate in cultural anthropology”, Cross-Cultural
Research, 35, 242–257.
Daft, R., and Weick, K. (1984). “Toward a model of organizations as interpretation systems”, Academy of
Management Review, 9(2), 284–295.
Dosi, G., and Egidi, M. (1991). “Substantive and procedural uncertainty: An exploration of economic
behaviours in changing environments”, Journal of Evolutionary Economics, 1(2): 145–168.
Dosi, G., Fabiani, S., Aversi, R., and Meacci, M. (1994). “The dynamics of international differentiation: A
multi-country evolutionary model”, Industrial and Corporate Change, 3: 225–241.
Dosi, G., Faillo, M., and Marengo, L. (2008), “Organizational capabilities, patterns of knowledge
accumulation and governance structure in business firms: An introduction”, Organization Studies,
29(August): 1165–1185.
Dosi, G., Freeman, C., Nelson, R. R., Silverberg, G., and Soete, L. (Eds.) (1988). Technical Change and
Economic Theory, London: Pinter.
Dosi, G., Marengo, L., Bassanini, A., and Valente, M. (1999). “Norms as emergent properties of adaptive
learning: The case of economic routines”, Journal of Evolutionary Economics, 9(1): 5–26.
Dosi G., Marengo, L., and Fagiolo, G. (2004). “Learning in evolutionary environments”, in K. Dopfer
(Ed.), The Evolutionary Foundations of Economics (pp. 255–328). Cambridge: Cambridge University Press.
Dosi, G., Marengo, L., Paraskevopoulou, E., and Valente, M. (2017). “A model of cognitive and oper-
ational memory of organizations in changing worlds”, Cambridge Journal of Economics, 41(1): 775–806.
Dosi, G., Napoletano, M., Roventini, A., Stiglitz, J., and Trebich, T. (2017). “Rational heuristics?
Expectations and behaviors in evolving economies with heterogeneous interacting agents”, S. Anna
School of Advanced Studies, LEM Working Paper, 2017/31.
Dosi, G., and Roventini, A. (2016). “The irresistible fetish of utility theory: From ‘pleasure and pain’ to
rationalising torture”, Intereconomics, 51(1): 286–287.
Duncan, R. B., and Weiss, A. (1979). “Organizational learning: Implications for organizational design”,
in B. Staw (Ed.), Research in Organizational Behavior. Oxford: Elsevier.
Dutta, P. K., and Radner, R. (1999). “Profit maximization and the market selection hypothesis”, Review
of Economic Studies, 66(4): 769–798.
Edgerton, R. B. (1985). Rules, Exceptions and Social Order, Berkeley, CA: University of California Press.
Egidi, M. (1996) “Routines, hierarchies of problems, procedural behaviour: Some evidence from
experiments”, in K. J. Arrow, E. Colombatto, M. Perlman, and C. Schmidt (Eds), The Rational
Foundations of Economic Behaviour (pp. 303–333). London: Macmillan.
Ericsson, K. A., and Smith, J. (Eds) (1991). Toward a General Theory of Expertise, Cambridge: Cambridge
University Press.
Fehr, B. (2005). “The role of prototypes in interpersonal cognition”, in M. Baldwin (Ed.), Interpersonal
Cognition (pp. 180–205), New York: Guilford Press.
Frantz, R., and Leeson, R. (Eds) (2013). Hayek and Behavioral Economics, Basingstoke: Palgrave Macmillan.
Friedman, M. (1953). Essays in Positive Economics, Chicago: University of Chicago Press.
Gärdenfors, P. (2004). Conceptual Spaces: The Geometry of Thought, Cambridge, MA: MIT Press.
Geertz, C. (1963). Peddlers and Princes, Chicago: University of Chicago Press.
Gigerenzer, G., and Brighton, H. (2009). Homo heuristicus: Why biased minds make better inference, Topics
in Cognitive Science, 1: 107–143.

503
504

Giovanni Dosi, Marco Faillo, and Luigi Marengo

Gigerenzer, G., and Goldstein, D. G. (1996). “Reasoning the fast and frugal way: Models of bounded
rationality”, Psychological Review, 103, 650–669.
Gigerenzer, G., Todd, P. M., and the ABC Research Group (Eds) (1999). Simple Heuristics That Make Us
Smart. New York: Oxford University Press.
Gigerenzer, G., Todd, P. M., and the ABC Research Group (Eds) (2012). Ecological Rationality: Intelligence
in the World. New York: Oxford University Press.
Goffman, E. (1974). Frame Analysis: An Essay on the Organisation of Experience, Harmondsworth: Penguin.
Granovetter, M. (1985). “Economic action and social structure: The problem of embeddedness.” American
Journal of Sociology, 91(3): 481–510.
Griffiths, T. L., Kemp, C., and Tenenbaum, J. B. (2008). “Bayesian models of cognition”, in R. Sun
(Ed.), The Cambridge Handbook of Computational Psychology (pp. 59–100). New York: Cambridge
University Press,
Hahn, U., and Ramscar, M. (2001). Similarity and Categorization, Oxford: Oxford University Press.
Hayek, F. A. (1952). The Sensory Order, Chicago: University of Chicago Press.
Heiner, R. A. (1983). “The origin of predictable behaviour”, American Economic Review, 73(4): 560–595.
Helfat, C., Peteraf, E., Singh, M., Mitchell, W., Teece, D. ... Winter, S.G. (2006). Dynamic
Capabilities: Understanding Strategic Change in Organizations. Oxford: Blackwell.
Hodgson, G. M. (1988). Economics and Institutions: A Manifesto for a Modern Institutional Economics,
Cambridge: Polity Press.
Hodgson, G. M. (1993). The Economics of Institutions. Aldershot: Edward Elgar.
Holland, J. H. (1975). Adaptation in Natural and Artificial Systems. Ann Arbor, MI: University of Michigan
Press.
Holland, J. H., Holyoak, K. J., Nisbett, R. E., and Thagard, P. R. (Eds) (1986). Induction: Processes of
Inference, Learning and Discovery, Cambridge, MA: MIT Press.
Johnson-Laird, P. N. (1983). Mental Models, Cambridge, MA: Harvard University Press.
Kaniovski, Y. M., Kryazhimskiy, A. V., and Young, H. P. (1997). Learning equilibria in games played by
heterogeneous populations. IIASA Interim Report. IIASA, Laxenburg, Austria.
Kahneman, D., Slovic, P., and Tversky, A. (Eds) (1982). Judgment under Uncertainty: Heuristics and Biases,
Cambridge: Cambridge University Press.
Kahneman, D., and Tversky, A. (1986). “Rational choice and the framing of decision”, Journal of Business
59: 251–278.
Kauffman, S. A. (1993). The Origins of Order, Oxford: Oxford University Press.
Kelly, G. A. (1955). The Psychology of Personal Constructs, vol. 1: Clinical Diagnosis and Psychotherapy,
New York: Norton.
Kitayama, S. (2002). “Culture and basic psychological processes: Toward a system view of culture: Comment
on Oyserman et al. (2002)”. Psychological Bulletin, 128: 89–96.
Kreps, D. M. (1996). “Market, hierarchies and mathematical economic theory”, Industrial and Corporate
Change 5: 561–595.
Lakoff G. (1987). Women, Fire and Dangerous Things: What Categories Reveal About the Mind, Chicago:
University of Chicago Press.
Levinthal, D. (1997). “Adaptation on rugged landscapes.” Management Science, 43(7): 934–950.
Levitt, B., and March J. G. (1988). “Organizational learning”, Annual Review of Sociology, 14(1), 319–338.
Loasby, B. J. (1983). “Knowledge, learning and the enterprise”, in P. Wiseman (Ed.), Beyond Positive
Economics? London: Macmillan.
Luria, A. R. (1976). Cognitive Development: Its Cultural and Social Foundations, Cambridge, MA: Harvard
University Press.
March, J. G. (1994). A Primer on Decision Making: How Decisions Happen, New York: Free Press.
March, J. G., and Simon, H. A. (1958). Organizations, New York: Basil Blackwell.
Marengo, L. (1992). “Coordination and organizational learning in the firm”, Journal of Evolutionary
Economics, 2(4): 313–326.
Marengo, L., and Pasquali, C. (2011). “The construction of choice: A computational voting model”,
Journal of Economic Interaction and Coordination, 6: 139–156.
Margolis, H. (1987). Patterns, Thinking and Cognition, Chicago: University of Chicago Press.
Mayer, R. E. (1992). Thinking, Problem Solving and Cognition, New York: W. H. Freeman.
McGrattan, E. R., and Marimon, R. (1995). “On adaptive learning in strategic games”. In A. Kirman,
and M. Salmon (Eds), Learning and Rationality in Economics, Oxford: Basil Blackwell Publishers.

504
505

Beyond “bounded rationality”

Nelson, R. R., and Winter, S. G. (1982). An Evolutionary Theory of Economic Change, Cambridge,
MA: Harvard University Press.
Newell, A., and Simon, H. A. (1972). Human Problem Solving, Englewood Cliffs, NJ: Prentice Hall.
Oyserman, D., and Lee, S. W. S. (2008). “Does culture influence what and how we think? Effects of
priming individualism and collectivism”, Psychological Bulletin, 134: 311–342.
Pearl, J. (1984). Heuristics: Intelligent Search Strategies for Computer Problem Solving, Reading, MA:
Addison-Wesley.
Posch, M. (1997). “Cycling in a stochastic learning algorithm for normal form games”, Journal of
Evolutionary Economics, 7: 193–207.
Polanyi, K. (1944). The Great Transformation, Boston: Beacon Press.
Polanyi, K. (Ed.) (1957). Trade and Market in the Early Empires, Glencoe, MN: Free Press.
Polanyi, M. (1966). The Tacit Dimension, London: Routledge & Kegan Paul.
Sargent, T. J. (1993). Bounded Rationality in Economics, Oxford: Clarendon Press.
Savage, L. (1954). The Foundations of Statistics, New York: Wiley.
Silverberg, G. (1988) “Modelling economic dynamics and technical change: mathematical approaches
to self-organisation and evolution”, in G. Dosi, C. Freeman, R. Nelson, G. Silverberg, and L. Soete
(Eds.), Technical Change and Economic Theory, London: Pinter.
Silverberg, G., and Verspagen, B. (1994). “Learning, innovation and economic growth: A long-run model
of industrial dynamics”, Industrial and Corporate Change, 3: 199–223.
Simon, H. A. (1976). “From substantive to procedural rationality”, in S. J. Latsis (Ed.), Method and Appraisal
in Economics (pp. 129–148). Cambridge: Cambridge University Press.
Simon, H. A. (1986). “Rationality in psychology and economics”, The Journal of Business, 59(4).
Smelser, N. J., and Swedberg, R. (Eds) (2006). The Handbook of Economic Sociology, 2nd ed., Princeton,
NJ: Princeton University Press.
Suppes, P. (1995a). “A survey of mathematical learning theory 1950–1995”, mimeo, Stanford University.
Suppes, P. (1995b). “Learning by doing, or practice makes perfect”, mimeo, Stanford University.
Thaler, R. H. (1992). The Winner’s Curse: Paradoxes and Anomalies of Economic Life, New York: Free Press.
Tordjman, H. (1998). “The formation of beliefs on financial markets: Representativeness and prototypes”,
in N. Lazaric and E. Lorenz (Eds), Trust and Economic Learning, London: Edward Elgar.
Tversky, A., and Kahneman, D. (1982). “Judgments of and by representativeness”, in D. Kahneman,
P. Slovic, and A. Tversky (Eds), Judgment under Uncertainty: Heuristics and Biases (pp. 84–98).
Cambridge: Cambridge University Press.
Vassilakis, S. (1997). “Accelerating new product development by overcoming complexity constraints”,
Journal of Mathematical Economics, 28(3): 341–373.
Viale. R. (2018). “The normative and descriptive weaknesses of behavioral economics-informed
nudge: Depowered paternalism and unjustified libertarianism”, Mind and Society, 17: 53–69.
Weibull, J. (1995). Evolutionary Game Theory, Cambridge, MA: MIT Press.
Winter, S. G. (1971). “Satisficing, selection and the innovative remnant”, Quarterly Journal of Economics,
85: 237–61.

505
506
507

PART VI

Cognitive organization
508
509

34
BOUNDED RATIONALITY
AND ORGANIZATIONAL
DECISION MAKING
Massimo Egidi and Giacomo Sillari

Introduction
Rational choice and human decision making are two different processes. The distinction
between the two became more and more apparent as the notion of bounded rationality was
proposed and developed. By the 1950s, rational choice was fully formalized, thanks to the
foundational work on Expected Utility Theory carried out by Von Neumann and Morgenstern
(1947), as well as that of Jimmy Savage (1954). Rational choice assumes that all conditions
surrounding a risky choice are given and known by the decision maker, who links means and
ends in a coherent choice among the acts at her disposal. Human decision-making, on the other
hand, investigates whether and how these conditions can be established, and how individuals
make decisions when they are not established, i.e., when data are not fully available, or not
fully processed, or preferences are ill-defined. In these cases, processes widely different from
rational choice are active: Decision framing, decision setting, search, categorization, representa-
tion are all notions developed to try and capture the subtleties and intricacies of human decision
making. Such notions allow the understanding of the processes behind choice, and reveal the
limits and bounds of human decision-making abilities. Thus, the endpoints of both human and
organizational decision making are not rational, but rather boundedly rational choices.

Are managers rational?

Simon developed the idea of bounded rationality in his PhD dissertation,1 later published in
the book, Administrative Behavior (Simon, 1947). The book opened up the “black box” of
the internal mechanisms of organizations. Simon identified the main characteristics of man-
agerial decision making by analyzing the structure of the organizational process. He recognized
that the core of every organization is the pattern of the division of tasks and their coord-
ination. The organization is thought of as a goal-oriented structure based on internal tasks
that must be coordinated in order to achieve the organization’s overall objectives. Behavior
within organizations is thus goal-oriented, and goals are by and large complex and hierarchic-
ally structured, as many intermediate sub-goals need to be realized, often in a specific order, for
the final goal to be achieved. The dynamics of organizational decision-making may therefore
be very complex, presenting two main aspects. First, goals are often defined in very general and

509
510

Massimo Egidi and Giacomo Sillari

ambiguous terms, thus necessitating continuous revising of the sub-goals’ hierarchy. Second,
hidden conflicting objectives can be unearthed during various organizational decisions, and this
may, again, make it necessary to revise both sub-goals and their hierarchy.

Rationality in organizations
The complexity of the organizational environment is a perfect arena for evaluating the com-
plexity of decision making and the feasibility of the standard model of rational decision making.
Indeed, in this context the rich variety of elements characterizing our decision processes
becomes more apparent, illuminating the fact that standard “rationality,” conceived as means-
ends consistency, is only the final element of a more intricate decision process. Decisions, in
a nutshell, are the result of several operations, such as searching and selecting relevant infor-
mation, framing the context of decision, providing an appropriate categorization of its elem-
ents, attempting to reduce uncertainty, and so on. Such operations are prerequisites of choice
requiring the activation of complex mental activities, well beyond mere means-end consistency.
In particular, according to Simon, the search for relevant information, knowledge acqui-
sition, and learning are the most important processes for achieving a good or rational deci-
sion in complex conditions. In the organizational context, the classic processes of division
of labor and subsequent coordination are pervasive, and decision-making conditions present
themselves highly unstructured or ill-defined. This, as we shall see, makes standard Expected
Utility Theory (EUT) inadequate as a model of choice. Indeed, during the 1950s, when EUT
was the standard tool for the market theory of both consumer and production (and hence of
economic equilibrium), we see that no attempts were made to extend this approach to the
internal mechanisms of organizations. By describing the entrepreneur’s production decision as
an optimal, utility-maximizing choice, the model was assuming as a matter of fact all internal
organizational processes as fixed, since the production function was supposed to fully represent
a given technology and organizational form.
Besides Simon, the only approach to the problem of rationality in organizational and evo-
lutionary settings that suggested the need for new analytical tools (other than EUT-based
approaches), was due to Schumpeter. His analysis presupposes limits of individuals' rationality
both in economics and politics. These limits are clarified and explored under the label “con-
scious rationality,” that is, rationality characterized by individual volition leading to purposeful
action. In his view, individuals may have different degrees of conscious rationality, depending
on how familiar they are with the domain of a given decision, and on how deliberate is the
decision they make.2 Schumpeter’s views are strikingly close to Herbert Simon’s models of
bounded rationality. Despite a successful acceptance of many of his analytic positions,3 the
attempts to formalize Schumpeter’s ideas have been relatively limited, due to the difficulties
in applying the standard theory of rational decision to conditions of innovation. Later in this
chapter, we will argue also through an example that Simon’s bounded rationality approach,
sharing several crucial elements with Schumpeter’s approach to rationality, is key to a deeper
understanding of decision making in contexts of organizational innovation.

Rationality, psychology, economics

To fully appreciate the relevance of the idea of bounded rationality in the context of organ-
ization, it is useful to consider it in the light of the intellectual endeavors to produce a “pure”
theory of human reasoning carried over in various contexts (probability theory, logic, decision
theory) during the nineteenth and twentieth centuries. In the words of Reinhardt Selten:

510
511

BR and organizational decision making

Modern mainstream economic theory is largely based on an unrealistic picture of

human decision making. Economic agents are portrayed as fully rational Bayesian
maximizers of subjective utility. This view of economics is not based on empirical evi-
dence, but rather on the simultaneous axiomatization of utility and subjective prob-
ability. In the fundamental book of Savage the axioms are consistency requirements
on actions with actions defined as mappings from states of the world to consequences
(Savage 1954). One can only admire the imposing structure built by Savage. It has
a strong intellectual appeal as a concept of ideal rationality. However, it is wrong to
assume that human beings conform to this ideal.4

Selten’s quote highlights two main issues. The first issue is that ideal rationality, as defined by
the axioms of rational choice in their more widely accepted version––EUT––requires both
logical consistency and coherence with probability theory. Moreover, ideal rationality also
requires unlimited computing capacity, as the procedure to discover the optimal choice may
necessitate an astronomically vast number of calculations while ideal agents would be able to
perform them subject to no limitations.
The second issue raised by Selten pertains to the nature and available evidence of the dis-
crepancy between real and ideal behavior. Such discrepancies had been spotted and discussed
during the creation and development of logic, of probability theory, and of rational choice,
independently from one another. For instance, probability theory was first developed by
and large in order to answer normative questions about gambling, and it was soon apparent
that gamblers’ behavior deviated significantly from the rules of probability calculus. Laplace
maintains5 that probability theory may help individuals to rationally correct the misleading
illusions generated by the “sensorium” (what today we would call the cognitive system).5 In
logic, too, during the nineteenth century and in particular with the work of Gottlob Frege, it
was apparent that there is a great distance between the psychology of human reasoning, on the
one hand, and the laws on which logic was based, on the other.6 In economics, the formal-
ization of rational choice culminating in EUT followed the same anti-psychologistic stance,
and when the theory was found descriptively wanting by the experimental work of Maurice
Allais (1953) showing that human decision-makers violate the axioms of rational choice, a
natural response to his results was to create more sophisticated versions of EUT by weakening
or dropping some of its axioms.
During the same years in which Allais performed his experiments, Herbert Simon began
pioneering his bounded rationality approach, shedding light on the limits of rationality and
exploring decision mechanisms from the vantage point of the cognitive processes involved
in them. Simon’s approach is orthogonal to attempts of mending EUT by tinkering with its
axioms. Simon, along with many other scholars, chose to follow a different route altogether,
one in which human decision making was modeled on the basis of the complexity of the
underlying psychological mechanisms.

Bounded rationality and problem solving

If rational choice theory is removed from the reality of human reasoning, a theory origin-
ating from empirical observations might offer better insights to identify the main mechanisms
undergirding human decision making. Indeed, in 1956, Cyert, Simon, and Trow carried
out an empirical analysis of managerial decision contexts that highlighted how search and
learning were at the core of human rationality.7 The study revealed an evident “dualism” of
behavior: On the one hand, there is behavior guided by a coherent choice among alternatives

511
512

Massimo Egidi and Giacomo Sillari

typical of structured and repetitive conditions; on the other, behavior characterized by highly
uncertain and ill-defined conditions, where the predominant role was played by problem-
solving activities.
The dualism between repetitive and well-known decision contexts and ill-defined decision
contexts proved to be a key distinction for our understanding of decision processes. In situ-
ations of the former kind, it highlights the process of decision-making routinization. In the
latter, the necessary conditions for applying standard rational choice theory are lacking, and the
most important decision process is the ability of the subjects to formulate and solve problems.8
This suggests that the real restrictions on rational decisions happen during the process of con-
struction of the context of the decision, leading to the working representation of the decision
problem. The notion of bounded rationality refers mostly to these conditions, and hence it is
implicitly intertwined with the notion of problem solving.

The dual process account of reasoning

The 1960s were the years of the greatest challenge against the axiomatic foundations of rational
choice. On the one hand, Allais’s critique aroused renewed interest in psychology; on the
other, Simon made clear that if human intelligence was to be thoroughly understood, it had
to be “decomposed” into its many complex processes and elements. Induction, reasoning,
and problem solving were, in Simon’s view, the true protagonists for comprehending human
bounded rationality, and hence producing more realistic models of economic and organiza-
tional phenomena.
One of the components of the discovery of the cognitive limits of rationality was originated
by Luchins (1942), and Luchins and Luchins (1950), who performed experiments on “mech-
anization of thought.” They conducted experiments with subjects exposed to mathematical
problems that had solutions at different levels of efficiency. The authors showed that subjects,
having identified a simple solution to a task in a given (repetitive) context, may automatically
and systematically apply the solution to other contexts, even if it proves to be suboptimal. The
experiments demonstrate that once a mental computation, deliberately performed to solve
a given problem, has been repeatedly applied to solve analogous problems, it may become
mechanized. Mechanization enables individuals to pass from deliberate effortful mental activity
to partially automatic, unconscious, and effortless mental operations.
This phenomenon has been further explored by psychologists, and our understanding
deepened through a great deal of experiments. Among these, the experiments on chess by
Simon and colleagues provided important new findings. Simon maintains that, in the course
of acquiring their skill, chess players store chunks in long-term memory corresponding to
patterns of pieces. Their recalling such patterns from long-term memory, during the match, is
fast and automatic, and forms the basis for the conscious process of symbolic manipulation of
the recalled mental items.

One key to understanding chess mastery ... seems to lie in the immediate perceptual
processing, for it is here that the game is structured, and it is here in the static analysis
that the good moves are generated for subsequent processing. Behind this percep-
tual analysis, as with all skills ... lies an extensive cognitive apparatus amassed through
years of constant practice. What was once accomplished by slow, conscious deductive
reasoning is now arrived at by fast, unconscious perceptual processing. It is no mistake
of language for the chess master to say that he ‘sees’ the right move.9

512
513

BR and organizational decision making

The experiments made by Simon and colleagues then led to the discovery that the architecture
of thinking is characterized by a complex interaction between the automatic and fast recall of
the elements stored in the long-term memory and the conscious process of symbolic manipu-
lation over the mental items. This dualism between unconscious and deliberate aspects of the
process of thinking has been further explored over the past years, also outside of the context of
chess, by many authors, for instance, in Schneider and Shiffrin (1977), Hogarth (2001), Evans
(2003), Evans and Frankish (2009), Stanovich (1999), Stanovich and West (2000) and many
others. While in Simon and in the subsequent literature the use of chunks was to show how to
construct a decision strategy, Kahneman adopts the dual system theory functionally to intro-
duce accessibility as the main theoretical concept at work in the explanation of rational choice
relative to the problem at hand.10

Organizations and routines

Routinized and not routinized behavior
As we have seen, the observation of managerial decisions within large organizations made clear
the dualism between repetitive decisions in stationary conditions and innovative decisions in
ill-defined and evolving environment. The latter (which Simon and March analyze preponder-
antly) is the extreme case of decisions (much more likely to happen) made in complex situations
in which it is too costly or unfeasible to compute a strategy to achieve an optimal solution.
In Organizations, March and Simon expand on this observation, and provide a definition of
“routinized behaviors”:

We will regard a set of activities as routinized, [then,] to the degree that choice has
been simplified by the development of a fixed response to defined stimuli. If search has
been eliminated, but a choice remains in the form of clearly defined and systematic
computing routine, we will say that the activities are routinized.11

A part of this definition should be highlighted: Routinized behaviors take place when “search
has been eliminated,” i.e., when the individual learning and problem-solving process stops.12
The reference to the process of mechanization of thought previously studied in chess players is
evident in the quote above, where it is considered in the context of a multiplicity of individ-
uals. When individual behaviors are routinized, as suggested in the literature, the mental load
required to decide is reduced and choice becomes an automatic and familiar process. March
and Simon suggest that this process happens also in the minds of individuals while they are
cooperating to achieve a shared, common goal. Cooperation, reciprocal adaptability, and the
shared belief in common goals are the most important requirements for individuals partici-
pating in the same program within an organization. Moreover, they assume the differentiation
among members according to their roles and to the knowledge required to perform tasks, as
well as their interdependence.13 By extending to group (or team) behavior the properties of
Simon’s chunking theory, we can tackle the question of how prior knowledge automatically
recalled from the long-term memory of different individuals must be re-composed, as if it were
a puzzle, and how the collective decision process must be simplified and reduced to automatic
and habitual process.14 March and Simon suggest that this kind of process occurs. They do so
by claiming that if all participants in a program become familiar and expert in their specific role
and do not search for new solutions any further, then a routine is established as a collection of

513
514

Massimo Egidi and Giacomo Sillari

repetitive organizational procedures. At the same time, they offer a “micro-individual” definition
of routinized behaviors of the individuals involved into the same organizational procedure:15

Problem-solving activities can generally be identified by the extent to which they

involve search; search aimed at discovering alternatives of action or consequences
of action. “Discovering” alternatives may involve inventing and elaborating whole
performance programs where these are not already available in the problem solver’s
repertory.16

Problem-solving activity is then the core process within organizations, searching for better
results and higher efficiency. This search is performed in every area in which more efficient
goals might be achieved (aspiration levels). A process ends up when the search leads to a satis-
ficing result, i.e., a new “performance program, or simply a program, has been discovered.”
A “performance program” is intended as an organizational procedure, i.e., it consists of a repository
of rules, that, in response to organizational conditions, prescribe the appropriate behavior for
each of the individuals collaborating to achieve a shared goal.17 The main characteristics of these
rules, is that they are prescriptions that originate during problem solving and that come to be
used again when the same conditions apply. Such procedures have basic analogies with com-
puter programs as both consist in prescriptions about what is to be done (action) when a given
condition applies. However, unlike computer programs in which prescriptions must be explicitly
and specifically given in an artificial language, in human procedures, prescriptions can be largely
implicit. In general, not all contents of a procedure involving humans need to be conceptually
and verbally specified. A large part of the verbal prescriptions may be very general and syn-
thetic,18 assuming that individuals will have sufficient competence to activate it correctly. This
aspect has a fundamental importance because humans, while putting into effect a procedure, are
supposed to retain large discretionary power, enabling them to specify and realize prescriptions
even when they are generic or unclear, or even pushing them to discover new alternatives:

A program may specify only general goals, and leave unspecified the exact activities
to be used in reaching them. Moreover, knowledge of the means-ends connections
may be sufficiently incomplete and inexact that these cannot be very well specified in
advance. Then “discretion” refers to the development and modification of the perform-
ance program through problem-solving and learning processes.19

A procedure then can be not fully specified, and in this case the individuals who put it in effect
must have competences to autonomously construct the parts which are not specified. This shows that
individual micro-creativity is a necessary component of the process of realizing a procedure.
The “discretionary” character of large parts of a program presupposes the “creative” ability
of individuals to fill the gaps of imperfectly specified programs.20

Further evolutions of the notion of organizational routine

The notion of routine was revisited 30 years after Organization by Nelson and Winter’s
Evolutionary Theory of Economic Change.21 In their approach to organizations, they consider
routines as the basic elements of an organization’s life, and innovation as the engine of routine
creation. One relevant element on which they focus attention is the role and transmission of
tacit knowledge.22 Routines may consist of actions that are often not verbalized, and need not be
transmitted in the form of messages. Routines, or better organizational routines, are not formally

514
515

BR and organizational decision making

defined and are implicitly treated as behavioral patterns. Expanding March and Simon’s notion,
Nelson and Winter attribute to organizational routines the role of basic units in the evolu-
tionary process of organizational change. This perspective slightly changes the original meaning
of “routinized programs” in March and Simon's approach, as the evolutionary approach ends up
attributing to the “organizational routine” more properties than March and Simon do in their
problem-solving context.23 Many definitions of organizational routines have been proposed
in these years, with slight differences intended to capture different properties. The attempt to
unify and operationalize these requirements is still ongoing after many decades.24
Besides there being a repository of a stored collective knowledge (as we have seen above, an
idea already present in nuce in March and Simon), an issue often debated in the literature pertains
to how to understand routines both as stable entities (especially in relation to internal conflicts)
and simultaneously as a source of change. We suggest that going back to the original Simon
and March approach solves this apparent contradiction. In fact, as we have previously remarked
a (human) routine is never fully specified, presupposing therefore that the individuals effecting
it need to possess the competences necessary to autonomously supply the underspecified parts
of the routines, hence expressing micro-creativity. This implies that micro search processes are
still active also when an organization has produced a certain set of stable routines. Thus, micro-
innovations (mutations) can still emerge.25 This way, the apparent contradictory property of
being stable while also potentially a source of change can be categorized and modeled.
Of course, more radical innovations can happen within organizations through the introduc-
tion of new top-down programs. The distinction between micro-innovations (mutations) and
radical innovations can be treated by using the properties of complex systems, and in particular
the decomposability features of the rules system. In fact, according to the original meaning
of “routinized programs” in March and Simon, routines can be modeled as systems of if-then
rules (Holland et al., 1986) that evolve through learning and adaptation. A system of condition-
action rules embeds a representation of the problem(s) the organization faces. The condition
part of the rules refers to the set of states (both of the world and internal) that are considered
as equivalent or indistinguishable and therefore activate the same course of action. The action
parts instead prescribe what ought to be done for each of these perceived situations.26
Marengo (2015) presents a simple model of organizational problem-solving where the gen-
eration of solutions is constrained by the representation of the problem and, in particular, by
the conjectural decomposition of the problem it embeds. Routines are temporary solutions
and may undergo small local changes which are reinforced as long as the routine keeps being
functional or even improves. By adding up a number of these small local changes, the initial
routine may finally evolve into a considerably different one. Two elements play a key role in
this evolutionary dynamics of routines: The decomposition of the problem and the selection
mechanism. The former sets the constraints to the generation of variations, in the sense that
small local changes can only take place within the given decomposition (or division of labor).
The latter defines the set of routines which are considered at least equally functional to the
current one and therefore the set of variants which are acceptable. When such a set is small,
i.e., small changes in a functional routine tend to damage its functionality, most variations will
not be viable and routines will tend to be rigid and persistent. When, on the contrary, the set
is relatively large, a routine will more easily evolve into a new one.27

Empirical evidence: switching from routinization to exploration

We select, among many others, three examples of experimental evidence related to routines
and bounded rationality. Two are laboratory experiments, and the third is a field experiment.

515
516

Massimo Egidi and Giacomo Sillari

The experiments illustrate some of the crucial issues in organizational bounded rationality
described above: Routinization of behavior, switching from routinized to exploratory behavior,
and whether routinization hinders creativity.
Target the Two (TTT) is a card game introduced by Cohen and Bacdayan (1994). Pairs of
participants to the experiment need to cooperate in order to achieve a common goal. They are
rewarded according to the success of the cooperative strategy they apply, rather than according
to individual performances. Thus, participants must discover a good strategy jointly, and then
follow it cooperatively in order to achieve their goal. Cohen and Bacdayan observe a labora-
tory tournament of pairs of players to study the emergence of routinized behavior involving
coordination and cooperation. They make the first relevant attempt to explore in laboratory the
emergence of rules of coordination, or organizational routines.
Egidi and Narduzzo (1997) find that Target the Two admits of two different strategies, whose
efficiency depends upon the card distribution. In the lab, players who are trained to play one
strategy continue to use it more frequently also in contexts in which the alternative strategy
would be more efficient. This phenomenon shows that the Einstellung effect may happen also
in a cooperative context.28 Most players' performances become automatic, their prior know-
ledge triggering automatic reactions that direct attention toward the cards relevant to the more
familiar strategy. In this way, they do not search for cards relevant to the strategy that would be
more efficient given the current card distribution on the board, and automatically select familiar
cards instead.29 In terms of Simon’s definitions, each of the two groups of players discovers a
different organizational routine and persists in using it even when it becomes not efficient.
Indeed, in this game only a limited subset of players explore more deeply the space of possible
actions and use the routine performing best in the current conditions. They are fully rational,
in the sense that their behavior is not routinized and they rationally select the more efficient
between the two available routines. In TTT the attention of routinized players is strictly driven
by the familiar strategy (analogously to situations of confirmation bias). Here the question
arises, whether routinization of behavior may prevent the discovery of a new strategy. An
answer comes from a field study by Ohly, Sonnentag, and Pluntke (2006). The authors examine
the relationship between routinization and creativity in a randomly selected sample of 278
employees of a German high-tech company. They look at four characteristics (job control, job
complexity, time pressure, and supervisor support) and a range of other creative and proactive
behaviors. Regression analyses reveal that in addition to work characteristics, routinization is
generally positively related to creative and proactive behaviors due to available resources that
can be used to develop new ideas while working.
The problem of the switch from routinized to exploratory behavior has been studied also
at the level of neural activity. Schuck et al. (2015) show through multivariate neuroimaging
analyses that, in the classification task described below, before subjects spontaneously change to
an alternative strategy, the medial prefrontal cortex (MPFC) encodes information irrelevant for
the current strategy but necessary for the new one. Thus, they indicate that behavioral changes,
the realization of the existence of a “better” strategy, and the decision to implement the new
strategy, all lag significantly behind changes in cerebral activity that indicate brain sensitivity to
the new strategy. Preceding the shift to a new strategy there is therefore a largely unconscious
process. For this reason, the shift from exploitation to exploration strategies does not seem to be
the outcome of a conscious, optimal response to a problem-solving procedure, as neural changes
correlated with the (reported) change in the decision method (hence in representation) happen
before the actual behavior change. The medial prefrontal cortex functions in this situation as a
“planning and evaluation” unit for exploration of new strategies. Furthermore, Schuck et al.
(2015) also establish that the categorization strategy yielding the current representation linked

516
517

BR and organizational decision making

to the problem-solving method becomes entrenched at the physiological level, as it is accom-

panied by inhibition of alternative categorizations of the problem at hand. In other words, once
a given strategy has become established, the MPFC tends to inhibit those elements that could
bring about an alternative categorization of the problem, making it more difficult for the agent
to adopt an innovative strategy. Exploration is hindered by our own devices, yet when we adopt
exploratory behavior, we do so before we become conscious of its adoption.

Creativity and innovation

The cognitive limits of human reasoning were fundamental pillars in the Schumpeterian ana-
lysis scheme, both for the theory of innovation and the theory of democracy.30 Schumpeter
considers the innovative activity as a “creative response” to external changes. He claims that this
activity can be understood ex post, but cannot be predicted “ex ante.” It is illuminating to inter-
pret Schumpeter’s description of the capacity for innovative activity (or “seeing things in a
way which afterwards proves to be true, even though it cannot be established at the moment”)
through the theory of bounded rationality.
As we have seen, according to Simon, solving a problem requires a selective search that
presupposes automatic recollection from long-term memory prior to conscious deliberation. In
this context, the interaction between automatic and deliberate thinking explains, as we know,
the chess master’s superiority over a novice. The master has the advantage of automatically
using heuristics accumulated through years of constant practice.31 In the same vein, the more
a novice learns and memorizes new strategies, the less mental effort he needs to play.32 Now
imagine a chess club where members organize a tournament with a grand master as opponent.
Suppose all of the participants have an average level of competence: During the tournament,
players will be mostly unable to predict the grand master’s moves because he draws upon a vastly
wider repertoire of strategies and can explore in much greater depth than his opponent the
possible sequences of moves and countermoves at her disposal. To the eyes of the club players,
the master’s strategy is unpredictable, and exhibits precisely the characteristics that Schumpeter
attributes to an innovator:

From the standpoint of the observer who is in full possession of all relevant facts, it
can always be understood ex post; but it can practically never be understood ex ante;
that is to say, it cannot be predicted by applying the ordinary rules of inference from
the pre-existing facts.33

Then, we could explain innovation purely on the basis of the competence gap34 between the
innovator and other individuals. This shows that there are possible analytical tools that can help
to decodify innovative behaviors, while a direct application of standard expected utility theory
to this context does not seem to be appropriate.

Conclusion
We conclude by highlighting the momentous impulse that bounded rationality has given to
organizational science. The discovery of the dual nature of decisions within organizations
(routine and search) led March and Simon to completely redefine the definition and analysis
of the notion of “planning.” Planning ceases to be a static and mechanical activity based on
rational decisions immersed in a world of complete information, and becomes instead based
on “organizational learning.” Search, therefore, becomes a key activity in organizations, and as

517
518

Massimo Egidi and Giacomo Sillari

such becomes possible to differential improvement, giving rise to differentiation in organiza-

tional performances. Adaptation is now the crucial element generating differentiation and sub-
optimalities. A relevant conceptual improvement is that not only organizations learn, but they
also make errors during the learning process, and – as March’s behavioral description shows –
because adaptation may easily lead to sub-optimal organizational configurations, errors may be
systematic and stable in the long run. Both of these developments have spawned a vast literature.
We have argued that bounded rationality is a component of the theory of problem solving,
assuming that individuals are natural problem solvers and that their central ability is thinking.
This implies that behind the bounds of rationality there lie the same cognitive processes gen-
erating creativity: problem framing, problem solving, searching, categorizing, memorizing,
etc. Simon’s theory incorporates problem solving into the process of rational decision. In the
bounded rationality approach, requisites of consistency were relaxed, but at the same time
new elements (knowledge acquisition and creativity) became of the essence. Moreover, while
empirically based, the theory of problem solving has developed into a highly abstract setting
foundational of Artificial Intelligence. In their seminal Human Problem Solving, Newell and
Simon (1972) establish a formal theory of problem solving: The goal was to uncover the secrets
of human cognition and transfer it into formal representation. They took up one of Turing’s
central statements – if a problem can be clearly described with appropriate language, then it can
be transferred into a form computable by a machine – and began to build artifacts mimicking
human cognitive skills. The artificial reproduction of certain aspects of human problem solving
was a new strategy with which to understand the human mind. Simon worked in parallel on
giving strong impetus to the empirical analysis of cognitive processes. As we have remarked, the
starting point was analysis of the game of chess, from which he moved beyond the notion of cal-
culation by first introducing the idea of “symbolic manipulation” and then directly considering
the determinants of cognitive processes in order to transfer to formal artificial tools.
The nineteenth-century view considered logic and probability as theories representing
the “laws” of human thought.35 To some extent the development of Artificial Intelligence
started with Simon’s theory of problem solving, under the assumption of bounded rationality,
representing a modern version of the same idea that Turing was explicitly supporting. This
opened up a deep unsolved question of the limits of constructing formal models of intelligence
(or, in the words of Turing, building a mechanical intelligence). A question tackled by Turing
himself and later by Gödel, with the discovery of the limits of computability. Whatever the
opinion we can express on the “great perspective” of discovery of the “laws” of human thought,
the approach of bounded rationality has opened the doors for the study of the mental processes
involved in search and reasoning and started with the formal representation of these processes,
both in the individual and organizational context.

Notes
1 “The dissertation contains the foundation and much of the superstructure of the theory of bounded
rationality that has been my lodestar for nearly fifty years” (Simon 1996, p. 86).
2 Schumpeter (1942, p. 363).
3 The contemporary theory of representative democracy is founded on Schumpeter’s theory of dem-
ocracy (see, for example, Held, 2006). Contemporary theories of innovation also draw widely on
Schumpeter's ideas.
4 Cf. Selten (2001 p. 13).
5 Laplace ([1825] 1995).
6 See, for instance, Frege ([1884] 1934).
7 Cyert et al. (1956, p. 238).

518
519

BR and organizational decision making

8 Of course, much exists in between the two extremes (e.g., well-structured analytical problems such as
logical or mathematical problems, or opaque problems tackled intuitively).
9 Chase and Simon (1973, p. 56).
10 See Kahneman’s Nobel Prize Lecture, delivered on December 8, 2002, revised in Kahneman (2003).
Dual process theory has been criticized (notably by Gigerenzer and the ABC Research Group), see,
for instance, Viale (2018, 2019), where it is argued that there is a continuum rather than a dichotomy.
11 March and Simon (1958, p. 142).
12 It is not the case that individual learning always fully resolves and stops problem-solving processes,
rather, the problem at hand is transformed as new problem-solving processes may arise.
13 See e.g., Orasanu and Salas (1993).
14 See e.g., Orasanu (1990), Glickman et al. (1987).
15 A relatively restricted literature has been dedicated to the development of this point (see Orasanu and
Salas 1993).
16 March and Simon (1958, pp. 160–161).
17 Hereinafter. we will use the word procedure instead of program, to distinguish computer programs from
human programs.
18 A possibly relevant part of the search can be entirely unconscious; see, for instance, Macchi et al.
(2016).
19 March and Simon (1958, p. 170).
20 The discretionary processes emerging in routines are well illustrated in a field experiment by Narduzzo,
Rocco, and Warglien (2002).
21 Nelson and Winter (1982).
22 See Polanyi (1958).
23 Among the most relevant, the role of tacit knowledge, memory, distributed authority structure, and
conflict are deepened.
24 As well described in Becker (2002) and further developed in Becker (2008). See also Feldman (2000).
25 See Ohly, Sonnentag, and Pluntke (2006).
26 Actions can operate on the environment or simply generate new “internal” states which in turn acti-
vate other rules. Therefore, a system of condition-action rule can be interpreted as a kind of organiza-
tional mental model (Johnson-Laird, 1983) and a memory of the organization (Dosi et al., 2017).
27 In both cases simple heuristics may serve the decision maker well, in the former case, by rigidly
designing a specific routinized solution to a stable problem, in the latter, or when the environment is
in flux, by suggesting different possible solutions.
28 See Luchins (1942).
29 This suggests that the automatic recall depends on the accessibility that the different strategies have in
the player’s mind. See Egidi (2016, pp. 199–201).
30 See Egidi (2017b).
31 Chase and Simon (1973, p. 56).
32 See Egidi (2017a).
33 Schumpeter (1947, p. 150).
34 See Heiner (1983, p. 562), in which a competence gap arises through the differential abilities of agents
relative to mapping informational signals to true events.
35 The expression “laws of thought” gained prominence after the publication of An Investigation of the
Laws of Thought on Which are Founded the Mathematical Theories of Logic and Probabilities by George Boole
(1854). Today the distinction between psychology (as a study of mental phenomena) and logic (as a
study of valid inference) is widely accepted even though the links between the two areas are debated
by many authors.

References
Allais, M. (1953). Le comportement de l’homme rationnel devant le risque: critique des postulats et
axiomes de l’Ecole Americaine. Econometrica, 21(4), 503–546.
Becker, M. C. (2004). Organizational routines: A review of the literature. Industrial and corporate change,
13(4), 643–678.
Becker, M. C (Ed.) (2008). Handbook of organizational routines, Cheltenham: Edward Elgar
Boole G. (1854). An investigation of the laws of thought: on which are founded the mathematical theories of logic
and probabilities. New York: Dover Publications

519
520

Massimo Egidi and Giacomo Sillari

Chase, W. G., and Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4(1), 55–81.
Cohen, M. D., and Bacdayan, P. (1994). Organizational routines are stored as procedural memory.
Organization Science, 5, 554–568.
Cyert, R. M., Simon, H. A., and Trow, D. B. (1956). Observation of a business decision. The Journal of
Business, 29(4), 237–248.
Dosi, G., Marengo, L., Paraskevopoulou, E., and Valente M. (2017). A model of cognitive and operational
memory of organizations in changing worlds. Cambridge Journal of Economics, 41, 775–806.
Dosi, G., Nelson, R. R., and Winter, S. G. (Eds.) (2002). The nature and dynamics of organizational cap-
abilities. Oxford: Oxford University Press.
Egidi, M. (2016). Organizational decisions in the lab. In R. Frantz and L. March (Eds.), Minds, models and
milieux (pp. 106–206). Basingstoke: Palgrave Macmillan.
Egidi, M. (2017a). Paths in contemporary economics and sciences of artificial intelligence that originate
from Simon’s bounded rationality approach. PSL Quarterly Review, 70(280), 7–33.
Egidi, M. (2017b). Schumpeter’s picture of economic and political institutions in the light of a cognitive
approach to human behavior. Journal of Evolutionary Economics, 27, 139–159.
Egidi, M., and Narduzzo, A. (1997). The emergence of path dependent behaviors in cooperative contexts.
International Journal of Industrial Organization, 15(6), 677–709.
Evans, J. (2003). In two minds: Dual-process accounts of reasoning. Trends in Cognitive Sciences 7(1010),
454–459.
Evans, J. S. B. T., and Frankish, K. (Eds.) (2009). In two minds: Dual processes and beyond. New York: Oxford
University Press.
Feldman, M. S. (2000). Organizational routines as a source of continuous change. Organization Science,
11(6), 611–629.
Frege, G. ([1884] 1934). Grundlagen der Arithmetik. Breslau: Marcus.
Gigerenzer, G., and Selten, R. (2001). Bounded rationality: The adaptive toolbox. Cambridge, MA: MIT Press.
Glickman, A. S., Zimmer, S., Montero, R. C., Guerette, P. J., Campbell, W. J., ... Salas, E. (1987). The
evolution of teamwork skills: An empirical assessment with implications for training (Tech. Rep. No.
87–016). Orlando, FL: Naval Training Systems Center.
Heiner, R. A. (1983). The origin of predictable behavior. The American Economic Review, 7(4), 560–595.
Held, D. (2006). Models of democracy. Cambridge: Polity.
Hogarth, R. M. (2001). Educating intuition. Chicago: University of Chicago Press.
Holland, J. H., Holyoak, K. J., Nisbett, R. E., and Thagard. P. R. (1986). Induction: Processes of inference,
learning and discovery. Cambridge, MA: MIT Press.
Johnson-Laird, P. N. (1983). Mental models. Cambridge, MA: Harvard University Press.
Kahneman, D. (2003). Maps of bounded rationality: Psychology for behavioral economics. American
Economic Review, 93(5), 1449–1475.
Laplace, P. S. de (1995). Philosophical essay on probabilities. Berlin: Springer Verlag.
Luchins, A. S. (1942). Mechanization in problem solving: The effect of Einstellung. Psychological
Monographs, 54(6).
Luchins, A. S., and Luchins, E. H. (1950). New experimental attempts at preventing mechanization in
problem solving. The Journal of General Psychology, 42(2), 279–297.
Macchi, L., Bagassi, M., and Viale, R. (Eds.) (2016). Cognitive unconscious and human rationality. Cambridge,
MA: MIT Press.
March, J. G. (1991). Exploration and exploitation in organizational learning. Organization Science,
2(1), 71–87.
March, J. G., and Simon, H. A. (1958). Organizations, New York: Wiley.
Marengo, L. (2015). Representation, search and the evolution of routines in problem solving, Industrial
and Corporate Change, 24, 951–980.
Narduzzo, A., Rocco, E., and Warglien, M. (2002) Talking about routines in the field. In G. Dosi, R. R,
Nelson, and S. G. Winter (Eds.), The nature and dynamics of organizational capabilities (pp. 27–50).
Oxford: Oxford University Press.
Nelson, R. R., and Winter, S. G. (1982). An evolutionary theory of economic change. Cambridge, MA: Harvard
University Press.
Newell, A., and Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall.
Ohly, S., Sonnentag, S., and Pluntke, F. (2006). Routinization, work characteristics and their relationships
with creative and proactive behaviors. Journal of Organizational Behavior: The International Journal of
Industrial, Occupational and Organizational Psychology and Behavior, 27(3), 257–279.

520
521

BR and organizational decision making

Orasanu, J., and Salas, E. (1993). Team decision making in complex environments. In G. Klein, J.,
Orasanu, R. Calderwood, and C. Zsambok (Eds.), Decision-making in action: Models and methods.
Norwood, NJ: Ablex.
Orasanu, J. (1990). Shared mental models and crew decision making (Tech. Rep. No. 46). Princeton,
NJ: Princeton University, Cognitive Sciences Laboratory.
Polanyi, M. (1958). Personal knowledge. London: Routledge.
Savage, L. J. (1954). The foundations of statistics. New York: Wiley.
Schneider, W., and Shiffrin, R. (1977). Controlled and automatic human information processing: I.
Detection, search, and attention. Psychological Review 84(2), 127–190.
Schuck, N. W., Gaschler, R., Wenke, D., Heinzle, J., Frensch, P. A., ... Reverberi, C. (2015). Medial pre-
frontal cortex predicts internally driven strategy shifts. Neuron, 86(1), 331–340.
Schumpeter, J. A. (1942). Capitalism, socialism and democracy. New York: Harper and Brothers.
Schumpeter, J. A. (1947). The creative response in economic history. The Journal of Economic History, 7(2),
149–159.
Selten, R. (2001). What is bounded rationality? In G. Gigerenzer and R. Selten (Eds.), Bounded ration-
ality: The adaptive toolbox (pp. 13–37). Cambridge, MA: MIT Press.
Simon, H. A. (1947). Administrative behavior: A study of decision-making processes in administrative organization.
New York: Free Press.
Simon, H. A. (1996). The sciences of the artificial. Cambridge, MA: MIT Press.
Shiffrin, R. M., and Schneider, W. E. (1977). Controlled and automatic human information processing: II.
Perceptual learning, automatic attending, and a general theory. Psychological Review, 84, 128–190.
Stanovich, K. E. (1999). Who is rational? Studies of individual differences in reasoning. Hillsdale, NJ: Erlbaum.
Stanovich, K. E., and West, R. F. (2000). Individual differences in reasoning: Implications for the ration-
ality debate. Behavioral and Brain Sciences 23, 645–726.
Viale, R. (2018). The normative and descriptive weaknesses of behavioral economics-informed
nudge: Depowered paternalism and unjustified libertarianism. Mind & Society, 17(1–2), 53–69.
Viale, R. (2019). Architecture of the mind and libertarian paternalism: Is the reversibility of Type 1 nudges
likely to happen? Minds & Machines (submitted).
von Neumann, J. and Morgenstern, O. (1947). The theory of games and economic behavior. Princeton,
NJ: Princeton University Press.

521
522

35
ATTENTION AND
ORGANIZATIONS
Inga Jonaityte and Massimo Warglien

Bounded rationality, attention, and organizations:

the Carnegie perspective
[I]n an information-rich world, the wealth of information ... creates a poverty of attention
and a need to allocate that attention efficiently among the overabundance of information
sources that might consume it.
Simon, 1971, p. 40

The allocation of attention is commonly seen as a fundamental problem for organizations

(Bouquet and Birkinshaw, 2008; Ocasio, 2011; Joseph and Wilson, 2018). Developing Barnard’s
(1938) intuition that “narrowing choice” is a central function of executive decisions, Simon
(1947) early on identified the process of directing and channeling managers’ attention as a key
function of organizations, and also provided the almost iconic description of attention manage-
ment featured in the epigraph. The foundations for such perspectives rely on the central role
that limited attention provides in defining bounded rationality (Simon, 1983). The “limits”
of decision-making rationality are to a large extent the results of the attentional bottleneck
(Simon, 1947).
The notion that attention is a scarce resource and that its allocation is central to intelligent
behavior has, of course, a longer history and deep roots in psychology. Administrative Behavior
resorts to classical sources, such as James (1890) and Tolman (1932), to provide the conceptual
background and basic definitions of attention.

Everyone knows what attention is. It is the taking possession by the mind, in clear
and vivid form, of one out of what seem several simultaneously possible objects or
trains of thought. Focalization, concentration, of consciousness are of its essence. It
implies withdrawal from some things in order to deal effectively with others, and is a
condition which has a real opposite in the confused, dazed, scatterbrained state which
in French is called distraction, and Zerstreutheit in German.
James, 1890, pp. 403–404

522
523

Attention and organizations

However, what characterizes organizational views of attention is “a dual emphasis on cogni-

tion (limited attention capacity) and structure (how organization shapes individual’s attention)”
(Festré and Garouste, 2015). Just as the cognitive limitations of individuals help explain how
organizations are structured, organizations in turn help understand how individual attention is
distributed and coordinated in collective systems.
Early work from the Carnegie School (Simon, 1947; March and Simon, 1958) emphasized
some fundamental ways in which organizations provide “attention-directors” as a response to
human cognitive limits. To that end, two general, hierarchical systems were proposed to design
the architecture of attention in organizations.

1 Decision hierarchies. “Prior controlling decisions” create a narrow frame of action for indi-
vidual decision makers. Organizations provide a stratification of decisions levels that allow
decisions on the spot to be guided by broader rationality considerations. Thus, organiza-
tional hierarchies can be seen as nested systems of decision premises, ensuring that lower-
level decisions are constrained and coordinated by higher-level ones.
2 Division of labor. Organizations design and assign tasks to single members, and thus direct
their attention to those tasks. By reducing interdependences between different tasks,
organizations reduce the number of features that each agent has to pay attention to. Tasks
have to be designed in ways that are compatible with individual attention capacity (the
“span of attention,” March and Simon, 1958). Division of labor determines selective per-
ception by agents specialized in a task. For example, Dearborne, DeWitt, and Simon (1958)
have shown how industrial executives see the aspects of a situation that relate to the tasks
and goals of their department.

In Simon’s original framework, decision premises and division of labor create a close connection
between attention and (sub)goals. Both processes define a hierarchy of subgoals attached to
tasks that selectively orient individual actions. The conceptual frame here is one of collective
problem solving. Cyert and March (1963) introduce a more genuine political perspective on
goals and attention. It is a natural implication of the notion of limited attention that individuals
cannot simultaneously attend to all dimensions of a decision, and thus attention to different
features can only be allocated sequentially (Simon, 1947). If attention to goals is limited, shifts
in attention will affect preference orderings. Cyert and March (1963) focus on how the logic
of organizational coalition making can affect sequential attentional dynamics. They suggest
that stakeholders in organizations generally have conflicting goals that make it impossible to
find stable and consistent set of organizational objectives. Instead, organizations can resolve
conflicting demands on their decisions by paying attention to different goals in a sequential
rather than simultaneous way. For example, when there are conflicting pressures to “smooth
production” and “satisfy consumers” the problem could be resolved by first satisfying the former
and then the latter. “Quasi-resolution of conflict” becomes an alternative to the standard view
of organizational coalition making, opposing shifting attention to side-payments, and dynamic
inconsistency of goals to coherent solutions. Phenomena like the present bias (O’Donoghue
and Rabin, 1999) and related forms of temporal myopia further facilitate quasi-resolution of
conflict and may be complementary to attentional shifts.
Two important further contributions stemmed from the original Carnegie approach, pro-
viding new insights and a different perspective on attention in organizations: the garbage can
model and the attention-based theory of the firm.
While Cyert and March emphasize the political implications of the sequential allocation of
attention, the garbage can model (Cohen, March, and Olsen, 1972) explores its coordination

523
524

Inga Jonaityte and Massimo Warglien

implications in non-routine decision making. In a radical departure from traditional models

of organizational decision making, the garbage can model explores “organized anarchies” in
which preferences are problematic, technologies are unclear, and participation is fluid. Despite
its quite radical language, the garbage can model combines important elements of the Carnegie
tradition: the “ambiguity of goals” further develops Cyert-March’s sequential attention to goals.
Attention is a scarce resource, participants to decision processes can pay attention to only one
choice at a time, and they have a preference for focusing on choices being close to be made.
The main novelty of the model comes from the way these ingredients are combined into a
flow of independent streams of decision makers, choice opportunities, and problems. As has
been noted (Ocasio, 2012), this introduces an ecological view of how attention is allocated in
organizations. Patterns of coordinated organizational decision making arise from the competing
demands on the participants’ attention and the way they are streaming over time. The result
is a subversion of the classical problem-solving structure. Most choices are made by flight (as
problems move to other choice opportunities) or by oversight (as most problems still have to be
attached to a choice arena), rather than by resolution. Again, some similarity with Cyert and
March’s quasi-resolution of conflict can be noticed. In organized anarchies, choices are made
because the attention of decision makers is focused on issues from which conflicting demands
have been temporarily removed. Distraction has an adaptive value.
Building on the dual emphasis on structure and cognition by Simon (1947) and other
earlier works (March and Simon, 1958; Cyert and March, 1963; Cohen et al., 1972; Weick,
1979), Ocasio (1997) developed the attention-based view of the firm (ABV). Ocasio describes
organizations as systems of structurally distributed attention and defines attention as a cognitive
process that encompasses the “noticing, encoding, interpreting, and focusing of time and effort
by organizational decision-makers” on issues and answers (Ocasio, 1997, p. 189). Ocasio, whose
definition echoes the one provided by James (1890), emphasizes the relevance of attention in
decision making, as already underscored by Simon (1947, p. 110). This theory highlights the
relationship between individual and organizational information processing: (1) actions of deci-
sion makers and subsequent organizational moves depend on the issues and answers the former
focus their attention on; (2) in turn, decision makers' focus of attention depends on specific
situation and the organizational and environmental context they are in. Later, Ocasio (2011)
stresses that attention can be classified into three interacting varieties: (1) attentional perspective,
which refers to the focus of attention across space and time; (2) attentional engagement, that is,
the extent by which a firm attends to organizational agendas and sustains vigilance in problem
solving over time; and (3) attentional selection, and thus the decision as to which stimuli to
address at any given time. Subsequent ABV studies focused mostly on topics such as the effect
of attention structures on decision-making (e.g., Maula, Keil, and Zahra, 2013; Wilson and
Joseph, 2015) and top-down and bottom-up attentional processes (e.g., Shepherd, McMullen,
and Jennings, 2007; Zbaracki and Bergen, 2015). Joseph and Wilson (2018) amalgamate these
two topics: Whereas previous ABV studies mostly focused on the consequences of the distribu-
tion of attention, these authors also stress the role played by organizational tensions and archi-
tectural complexity in influencing the distribution of attention.

Attention at work: organizational mechanisms

While the Carnegie classics deal with attention in broad terms, they also provide a way to
look at familiar organizational phenomena in a new perspective. Different organizational
mechanisms have been analyzed in terms of attentional processes, allowing a reinterpretation of
their functions and ways of operating. Here we will focus on some examples.

524
525

Attention and organizations

Control systems (and more generally accounting systems) provide an interesting case, and they
were among the first to be analyzed in an attention framework. Control systems are usually
described as information feedback systems. Goals and targets are set, current results are regu-
larly compared with the targets, and feedback is used to correct deviations and keep the organ-
ization focused on implementing its original strategies (Anthony and Govindarajan, 2007).
However, a focus on attention may suggest a different interpretation. In their study of the
controller’s department in seven organizations, Simon et al. (1954) had already made a distinc-
tion between score-card and attention-directing use of information. The score-card question
is “Am I doing well or badly?” The attention-directing one is “what problems should I look
into?” (Simon et al., 1954, p. 3). It was observed that the attention-directing use of control
systems is always paired with the score-card one. Attentional uses of control systems are tightly
related to the exertion of the “principle of exception”: accounting data direct the attention of
managers toward “out of line” situations and trigger direct hierarchical intervention to address
them. In a subsequent study encompassing 19 major companies, Simons (1991) found that
top managers selectively use control systems to focus attention on strategic uncertainties. As a
result, control systems favor the generation of new strategic initiatives in response to emergent
uncertainties, rather than just supporting the implementation of current strategies. In a similar
vein, Vanderbosch (1999) finds that executive support systems have the fundamental function
of concentrating management attention and helping managers to have a focusing influence on
the organization.
March and Simon (1958) suggested that the structure of communication channels in organizations
will affect attention-allocation phenomena by determining which type of stimuli will reach
which type of organizational members, and the awareness agents have of the consequences of
their actions. In turn, ease of access, saliency, and channel affordances may feed back on organ-
izational flows of communication and the use of different channels (Treem and Leonardi, 2012).
Recent research, mostly in the light of Ocasio’s ABV theory, has stressed the role of communi-
cation channels in the process of attentional engagement. Ocasio and Joseph (2005) extended
the ABV theory by explaining the structure and role of communication channels in the distri-
bution of organizational attention, and formulation and implementation of the strategic plan.
Distribution and integration of attention occur through the organization’s communication
channels (Joseph and Ocasio, 2012). Recent studies demonstrate how distribution of attention
is affected by specific communication channels. For example, Bouquet and Birkinshaw (2008)
explain how subsidiary firms may actively gain or lose the positive attention and efforts of
their parent corporate headquarters managers, and how they can have more control over the
attention their parent companies dedicate to them. Dutton, Ashford, O’Neill, and Lawrence
(2001) explore issue selling and describe how managers’ initiatives may shape the attention of
top management, and thereby, influence the organizational moves and performance.
Most attention research primarily focuses on structures of communication channels
to measure the response to the environmental stimuli (Joseph and Ocasio, 2012). Ocasio,
Laamanen, and Vaara (2018) propose communication as a process by which actors can attend
to and engage with organizational and environmental issues and initiatives. Comparing the
examples of Apple and Motorola companies, Ocasio and Joseph (2018) demonstrate that inte-
grating attention through communication channels with the communications within those
channels can be useful in reaching coherence in strategy formulation and implementation.
Apple’s focus on sustaining organizational attention over the implementation of its strategy, as
opposed to the lack of it by Motorola, contribute to the explanation of why Apple succeeded
in making smartphones the digital hub, integrating different technologies, while Motorola
experienced strategic failure despite moving with similar intents.

525
526

Inga Jonaityte and Massimo Warglien

Corporate governance mechanisms also affect patterns of attention at the top of

organizations, sometimes in subtle ways. In particular, the board composition matters, as
it affects the diversity of cognitive lenses through which board members look at issues.
Tuggle, Simmon, Reutzel, and Bierman (2010) and Tuggle, Schnatterly, and Johnson
(2010) argue that heterogeneity on a board of directors influences their patterns of
attention by affecting how boards allocate their attention between monitoring functions
and discussing new entrepreneurial issues. Tuggle et al. (2010) analyzed a sample of 210
firms by collecting data on board compositions and the text of their board minutes. They
suggest that tenure variance, firm/industry background heterogeneity, and the propor-
tion of directors with output-oriented backgrounds are positively associated with the
discussion of entrepreneurial issues. Some recent studies examine how governance may
affect the attention given to issues and the responses that are formulated. Galbreath’s
findings (2018) highlight that certain attention structures link boards, inasmuch boards
engage in environmental scanning and stakeholder debate. Galbreath also finds interaction
between environmental scanning and women on boards, and between stakeholder debate
and women on boards. Gender diversity is likely to shift attention through a change in
the nature of the attention-directing structures (e.g., environmental scanning, stakeholder
debate), rather than through a direct effect.
Surprisingly, little heed has been paid to the role of incentives in directing attention in
organizations, despite Simon’s (1947) early remark that motivation (together with emotion) is
the mechanism responsible for the allocation of attention to competing tasks. It is somehow
ironic that inquiry into the role of incentive mechanisms as attention allocation systems has
been mostly left to “rational” theories of organizations, in particular to agency models. In their
seminal paper on multitasking, Holmstrom and Milgrom (1991) have pointed out that “when
there are multiple tasks, incentive pay not only serves to allocate risks and motivate hard work,
but also direct the allocation of the agents’ attention among their various duties” (p. 25). In their
view, attention is a scarce resource that agents have to allocate among tasks that exert competing
pressure. Incentives will affect which dimensions of his/her job an agent will pay attention to.
As a result, “an increase in an agent’s compensation in any specific task will cause some reallo-
cation of attention away from other tasks.”
Despite its original rationalistic flavor, the problem of multitasking offers an important
window on mechanisms directing attention in organizations and invites closer empirical ana-
lysis. There is ample psychological evidence supporting the claim that incentives have an
important role in directing human attention among competing stimuli. Stimuli associated with
rewards gain prominence in individual attention and are recognized more promptly (O’Brien
and Raymond, 2012). If this can generally help to maintain attention focused on relevant
items, it can also generate peculiar distortions. For example, stimuli that are associated with
rewards but are irrelevant to the current task can capture attention and distract from task-
relevant attention (Anderson, Laurent, and Yantis, 2011).
Similar phenomena seem to emerge at the collective level of organized activity. In her
excellent review on the provision of incentives on firms, Prendergast (1999) provides ample
documentation on the potentially distortive effects of incentives on the allocation of attention
(and consequently, selective effort), especially in multitasking situations. For example, when
agents are subject to piece-rate incentives, they are less likely to help other workers (Drago and
Garvey, 1998). This applies also to temporal allocation: when incentives are provided in rather
distant moments of time, a distorted temporal allocation of effort is often observed, with much
effort concentrated before the evaluation date, and then falling again after evaluation (Asch,
1990). While these effects can be attributed to rational calculation (e.g., distorted temporal

526
527

Attention and organizations

allocation of effort may result from intertemporal discounting; Prendergast 1999), an interpret-
ation in terms of limited attention seems equally plausible. More research is needed to disen-
tangle these two explanations.
An especially interesting example of how attention relates to incentives is provided by
the so-called crowding-out problem (Frey and Oberholzer-Gee, 1997; Fehr and List, 2004;
Gneezy, Meier, and Rey-Biel, 2011). It has repeatedly been observed that explicit incentives
addressed at inducing behaviors, such as cooperation, trust, or education may actually produce
a contrary effect. For example, pro-social behavior may actually be discouraged rather than
favored by the introduction of monetary incentives meant to support it. A well-known case is
that of blood donation: the introduction of a monetary reward for blood donors actually causes
a drop in blood donation, unless an option is left to leave money to a charity (Mellström and
Johannesson, 2008). In the employment relationship, the introduction of incentives that rely on
monitoring and fining agents who underperform may significantly decrease the agents’ effort
(Fehr and Gächter, 2002). A basic explanation of the crowding-out effect relies on the fact that
incentives direct attention toward certain features of social interaction (Heyman and Ariely,
2004) and thus induce different frames of the decision situation, e.g., from social to monetary
(Gneezy et al., 2011). On the contrary, positive affect factors may activate prosocial frames
and direct attention toward non-job-specific behaviors and tasks, triggering Organizational
Citizenship behavior (Organ, 1988).

The adaptive value of (in)attention

Organizations can be conceived as adaptive responses to the cognitive limits of individuals.
“Organizations will have structure … insofar as there are boundaries of rationality … If there
were not boundaries to rationality … there could be no stable organizational structure” (March
and Simon, 1958, p. 192). A fundamental function of organizations is to manage the gap
between the overabundance of information in the world and the limited attention of individual
minds. Many features of organizational structures and mechanisms are responses to such limi-
tation (Simon, 1971).
However, attention is not just a problem that organizations have to solve. Attentional
processes significantly contribute to how organizations adapt to their environments, and, as
such, they have their own, relevant adaptive value. First of all, attentional processes are a fun-
damental trigger of organizational adaptive responses (Simon, 1947; Nigam and Ocasio, 2010).
Sequential shifts in attention are especially relevant to understanding adaptive change, also
because they connect cognitive components to the reshaping of organizational coalitions. In
his analysis of corporate environmentalism from 1960 to 1993, Hoffman (1999) shows how
changes in the chemical industry were triggered by attention shifts related to specific events.
These changes were accompanied by transformations in the patterns of interactions among
stakeholders.
Attentional processes have adaptive relevance also because they affect which features of the
environment organizations pay attention to as they change. Individuals and organizations do
not consider all stimuli from their environment but pay attention selectively to only a few of
them (Ocasio, 1997). This leads to selective responses that end up determining to which envir-
onmental subsystems organizations will adapt to. Weick (1979) has aptly defined such a process
as enactment: through focused attention and response, an organization selects (and to some
extent “creates”) the environment to which it is adapting. This means that the fitness criteria
that determine successful adaptation are to some extent endogenous, and that attention actively
contributes to generating them.

527
528

Inga Jonaityte and Massimo Warglien

The adaptive value of attention is enhanced by the shifting nature of organization goals. The
very notion of adaptive success depends upon the criteria with which organizational perform-
ance emerge, often as a result of quasi-resolution of conflict. As Cyert and March (1963) have
suggested, a fundamental way in which organizations adapt is by learning which performance
criteria to attend for. This also means that organizational learning defines its own criteria of
success. Attentional shifts triggering change and the subsequent organizational adaptation may
often result from the dynamics of aspiration levels connected to different goals. By analyzing
growth patterns in the insurance industry, Greve (2008) has shown how firm’s growth may be
the result of a shift of attention to size when profitability goals are satisfied.
Whereas attentional constraints may be seen as a cognitive limitation, they also bring some
advantages as far as adaptive processes are concerned. The “less is more” argument (Gigerenzer
and Todd, 1999) seems to apply also in the context of attention allocation. In this context, it
is worth noticing that scholars of language acquisition have suggested that processing limits of
children increase their ability to learn a language (Newport, 1990; Entman, 1993) because pro-
cessing limits favor focusing on the crucial components of language. In a related essay, Kareev
(1995) has shown that processing limitations can produce positive bias in detecting correlations
in the environment, which in turn may enhance early detection of potentially informative
relationships among variables. A first important implication of attentional limit is that only a few
features of experience are indeed considered feed learning. In a dimensionally rich environment,
limiting attention to a few features can certainly limit in the long term the accuracy of learning
processes. However, it can considerably improve the speed of learning and adaptive perform-
ance in the short run. “Speed of learning favors selective attention” (Kruschke and Hullinger,
2010). In turn, selective attention affects the speed of learning (Knudsen and Warglien, 2018). If
organizations have limited time to learn, in some environments this advantage may turn out to
be important. Knudsen and Warglien (2018) show that in skewed environments where different
cues have different informative value, agents with limited attention will learn from a subset of
features only and adapt faster than “broad-minded” agents who consider all features presented by
experience. Moreover, if short-term learning-speed advantage leads organizations to have access
to more information (e.g., because they acquire more customers), the adaptive advantage might
become a permanent one, even in the long run. Finally, attention limits may reduce the negative
effects of overfitting that may be caused by considering a too-large asset of variables in learning
(Goldstein and Gigerenzer, 2009; see also Artinger, Petersen, Gigerenzer, and Weibler, 2015); in
fact, they act as a sort of “model selection” constraint that improves the robustness of inferences.
Whether these advantages of limited processing apply to organizational learning, especially in
non-stationary environments and facing changes, is still an open and under-investigated issue
(Bingham and Eisenhardt, 2011) that certainly deserves more attention.

Inattention: from the economist’s point of view

Limited attention (often: inattention) has become a hot topic in economics. Many economists
have no problem in assuming that attention is a scarce resource, and as such an excellent subject
for economic inquiry. Gabaix (2017) offers an excellent introduction to this rapidly growing
field of inquiry and suggests that the concept of limited attention has the potential to explain
and unify many behavioral anomalies.
The basic paradigm of inattention is that when agents receive a signal with n dimensions,
agents “pay attention” only to m < n dimensions, substituting the signal value with a default
one in the remaining n-m dimensions. In more graded version, they can pay “partial” attention
to each dimension, adjusting the default value in direction of the signal value. Regardless,

528
529

Attention and organizations

inattention generates an “anchoring and adjustment” process in which the set of defaults is
partially adjusted toward the signal values. Gabaix (2017) shows how a large set of familiar
decision-making biases can be modeled as special cases of the inattention paradigm. For
example, quasi-hyperbolic discounting (Laibson, 1997) can be modeled as inattention to the
future, providing a simple interpretation of intertemporal-choice anomalies. This interpretation
is also normatively manageable, as exponential discounting can be preserved, in this case, the
normative reference.
Organizational economics has provided some interesting applications of the inattention
paradigm to classical organizational issues. Consider, for example, the tradeoffs between the two
most general forms of organizational coordination (March and Simon, 1958): decentralized
communication and standardized behavior. Decentralized communication relies on continuous,
multilateral feedback among agents. Standardized behavior resorts to predefined rules that
established ex ante which behavior is appropriate in which context, e.g., through standard oper-
ating procedures. It is possible to frame such tradeoffs in attentional terms. Decentralized com-
munication taxes agents’ attention, while standard operating procedures save attention resources
by making use of a default. Dessein and Santos (2006) build a simple model in which the quality
of communication between two organizational agents is a function of the cost of attention.
Agents have aligned interests (a team model), must perform multiple tasks, and their payoff
depends on (1) how accurately their action matches the current state of the environment and
(2) on how well they coordinate their actions. The “standard operating procedure” captures the
optimal response to the average state of the environment. There are n tasks to coordinate, and
each agent is allocated a task – and only he/she can observe the current state of the environment
in such dimension and transmit it to the other agents – but successful transmission is a function
of attention. Dessein and Santos show that attentional factors determine the adaptability of
an organization: under low-attention conditions, standard operating procedures minimize the
cost of miscoordination but make the organization unresponsive, as the agent who observes
the true state of the environment will prefer to act according to standard operating procedures
rather than respond to the current state, thereby incurring the risk of miscoordination with
other agents. Moreover, they show that attention limits the division of labor: as tasks are more
fractioned, organizations become more rigid as the cost of miscoordination by feedback will
increase with the number of tasks to coordinate.
In a subsequent development of the model, Dessein, Galeotti, and Santos (2016) show
that organizations can structure in asymmetric ways the distribution of attention, by focusing
all coordination by feedback on a few focal tasks while leaving other tasks to standardized
behavior. This example shows the remarkable potential for a dialogue between the economics
of inattention and the models of bounded rationality. The role of standard-operating procedures
as economizers of attention, the interactions of attention and division of labor, and hybrid
models of coordination in which standardization of some tasks frees attention for non-routine
coordination, are classical themes of the Carnegie tradition, and these recent models offer
new insights and qualifications that enrich our understanding of these issues. An interesting
aspect is that such models are typically equilibrium ones, and require that attentional limits
are common knowledge, which is a problem leading to the well-known problems of “super-
rational” awareness of bounded rationality.

Open questions
The numbers of studies on “attention to attention” have been steadily increasing in recent years,
but there are important theoretical areas and empirical issues that still need to be addressed and

529
530

Inga Jonaityte and Massimo Warglien

offer relevant opportunities for new research (e.g., van Knippenberg, Dahlander, Haas, and
George, 2015). Here we suggest a few of them that we find of special priority.

1. Better integrating organizational attention and organizational learning models. Models of organiza-
tional learning do not usually consider the effect of attentional factors on adaptive perform-
ance. Particularly important issues concern the potential benefits of attention limits: Does
bounded attention improve learning performance? In which type of environments? What
attention mechanisms favor learning? A second major terra incognita is how attentional
mechanisms help cope with changes in the structure of the environment. Most organiza-
tional learning models consider only stationary environments. On the other hand, empir-
ical studies suggest that attention plays a major role in how organizations perceive disruptive
events and trigger response to changing environments. The study of adaptation to non-
stationary environments opens a set of new questions for students of organizational learning.
Some of them are exquisitely attentional. How is non-stationarity recognized? And, if it
is recognized, what forces may determine under-reaction (or maybe over-reaction)? How
does responding to novelty imply redistributing attentional across organizational members?
2. Organizational politics and attention. Organizational cognition and the study of political
processes in organization have remained substantially separate domains of inquiry. As
originally suggested by Cyert and March (1963), the allocation of attention is a crucial
connecting link between these two domains. One area where this is especially evident is
organizational communication. Communication affects selective attention (Entman 1993)
and helps determine the frames through which organizational actors perceive and inter-
pret decision-making issues and their own interest. This offers opportunities to model
“cognitive coalitions” in organizations in which the patterns of alignment of organiza-
tional actors in coalitions depend on which dimensions of the coalitional problem are
activated by selective attention. Interestingly enough, this seems also an issue that lends
itself naturally to experimental treatment. A model of quasi-resolution of conflict might
be mature.
3. Attention failures. The negative side of the attention limits is the failure of individuals and
organizations to detect relevant events and understand changes in their environment (Bansal,
Kim, and Wood, 2018), reinforcing the structural (Hannan and Freeman, 1984) and cogni-
tive (Tripsas and Gavetti, 2000) forces leading to organizational inertia. Bounded awareness
is the other face of focusing, as it prevents people from considering relevant, available,
and perceivable information for informed decision-making (Chugh and Bazerman, 2007).
This focusing failure is closely related to well-documented psychology research, such as
“inattentional blindness” (Neisser, 1979), “focalism” (Wilson, Wheatley, Meyers, Gilbert,
and Axsom, 2000), and “information avoidance” (Golman, Hagmann, and Loewenstein,
2017). This a general problem: for example, Zegart (2006) pointed out that despite the
many signals collected about the possibility of a severe terroristic threat, the counter-
terrorism agencies did not adjust rapidly enough to prevent the 11 September 2001 attacks.
This adaptation failure, which originated in politics, had the consequence that the agencies
involved failed to achieve the changes they believed were needed very urgently. Identifying
the organizational mechanisms responsible for collective attention failures may require
going beyond analogies with individual processes and, instead, articulating how the distri-
bution of attention and the communication channels integrating it work or do not work in
detecting threats (e.g., Rerup, 2009; Vuori and Huy, 2016) and opportunities (Shepherd,
Mcmullen, and Ocasio, 2017) outside the attentional focus of an organization. This topic
may require looking more carefully into the processes through which organizations update

530
531

Attention and organizations

or fail to update their attention rules (Cyert and March, 1963) by responding to weak
signals from the environment.
4. Developing experimental paradigms. Ocasio (2011) has lamented a lack of cumulativity in
research on organizational attention. One natural research strategy to increase cumulativity
would be to put both theoretical propositions and inspirations from field observations
into hypotheses that can be tested through careful, reproducible laboratory experimenta-
tion. Despite the burgeoning development of experimental paradigms to study attention
in psychology (e.g., Fawcett, Risko, and Kingstone, 2015), very little has been done in
the field of organization research. Experiments on organizational attention should address
aspects that are usually absent in psychological research but would be key parameters in the
design of an organizational lab setting. Three of them are especially relevant. (a) The dis-
tribution of attention. A fundamental feature of organizations is that attention is structurally
distributed. Experiments should enable to specify the distribution of attention of agents.
(b) Mechanisms of attention integration. Organizations provide a multiplicity of mechanisms
integrating individual attention processes (hierarchies, decision premises, communication,
etc.). An organizational experiment should enable evaluation of how these mechanisms
interact with attention distribution in affecting organizational performance. (c) Strategic
aspects of attention. Experiments should enable control of the effect of how different interests
affect attention allocation. Examples include the effect of incentives, attention capture, and
strategic manipulation of agendas.

References
Anderson, Brian A., Laurent, Patryk A., and Yantis, Steven. 2011. Value-driven attentional capture.
Proceedings of the National Academy of Sciences of the U.S.A., 108(25): 10367–10371.
Anthony, Robert N., and Govindarajan, Vijay, 2007. Management Control Systems. 12th ed. Boston:
McGraw-Hill/Irwin.
Artinger, Florian, Petersen, Malte, Gigerenzer, Gerd, and Weibler, Jürgen. 2015. Heuristics as adaptive
decision strategies in management. Journal of Organizational Behavior, 36(S1): S33–S52.
Asch, Beth J. 1990. Do incentives matter? The case of navy recruiters. Industrial and Labor Relations Review,
43(3): 89S–106S.
Bansal, Pratima, Kim, Anna, and Wood, Michael O. 2018. Hidden in plain sight: The importance of scale
in organizations’ attention to issues. The Academy of Management Review, 43(2): 217–241.
Barnard, Chester I. 1938. The Functions of the Executive. Cambridge, MA: Harvard University Press.
Bingham, Christopher B., and Eisenhardt, Kathleen M. 2011. Rational heuristics: The ‘simple rules’ that
strategists learn from process experience. Strategic Management Journal, 32(13): 1437–1464.
Bouquet, Cyril and Birkinshaw, Julian. 2008. Weight versus voice: How foreign subsidiaries gain
attention from corporate headquarters. Academy of Management Journal, 51(3): 577–601, doi:10.5465/
amj.2008.32626039.
Chugh, Dolly, and Bazerman, Max H. 2007. Bounded awareness: What you fail to see can hurt you. Mind
& Society, 6(1): 1–18.
Cohen, Michael D., March, James G., and Olsen, Johan P. 1972. A garbage can model of organizational
choice. Administrative Science Quarterly, 17(1): 1–25.
Cyert, Richard M., and March, James G. 1963. A Behavioral Theory of the Firm. Englewood Cliffs,
NJ: Prentice-Hall.
Dearborn, DeWitt C., and Simon, Herbert A. 1958. Selective perception: A note on the departmental
identifications of executives. Sociometry, 21(2): 140–144.
Dessein, Wouter, Galeotti, Andrea, and Santos, Tano. 2016. Rational inattention and organizational focus.
American Economic Review, 106(6): 1522–1536, doi:10.1257/aer.20140741.
Dessein, Wouter, and Santos, Tano. 2006. Adaptive organizations. Journal of Political Economy, 114(5): 956–
995, doi:10.1086/508031.
Drago, Robert, and Garvey, Gerald T. 1998. Incentives for helping on the job: Theory and evidence.
Journal of Labor Economics, 16(1): 1–25.

531
532

Inga Jonaityte and Massimo Warglien

Dutton, J. E., Ashford, S. J., O’Neill, R. M., and Lawrence, K. A. 2001. Moves that matter: Issue selling
and organizational change. Academy of Management Journal, 44(4): 716–736, doi:10.2307/3069412.
Entman, Robert M. 1993. Framing: Toward clarification of a fractured paradigm. Journal of Communication,
43(4): 51–58.
Fawcett, Jonathan, Risko, Evan, and Kingstone, Alan. 2015. The Handbook of Attention. Cambridge,
MA: MIT Press.
Fehr, Ernst, and Gächter, Simon. 2002. Do incentive contracts crowd out voluntary cooperation? IEW
Working Papers 034, Institute for Empirical Research in Economics, University of Zurich.
Fehr, Ernst, and List, John A. 2004. The hidden costs and returns of incentives: Trust and trustworthiness
among CEOs. Journal of the European Economic Association, 2(5): 743–771.
Festre´, Agnes, and Garrouste, Pierre. 2015. The ‘Economics of Attention’: A history of economic
thought perspective. Oeconomia: History, Methodology, Philosophy, 5(1): 3–36.
Frey, Bruno S., and Oberholzer-Gee, Felix. 1997. The cost of price incentives: An empirical analysis of
motivation crowding-out. The American Economic Review, 87(4): 746–755.
Gabaix, Xavier. 2017. Behavioral inattention. NBER Working Paper No. w24096, National Bureau of
Economic Research, Cambridge, MA.
Galbreath, Jeremy. 2018. Do boards of directors influence corporate sustainable development? An
attention-based analysis: Board influence on corporate sustainable development. Business Strategy and
the Environment, doi:10.1002/bse.2028.
Gigerenzer, Gerd, and Todd, Peter M. 1999. Simple heuristics that make us smart. New York: Oxford
University Press.
Gneezy, Uri, Meier, Stephan, and Rey-Biel, Pedro. 2011. When and why incentives (don’t) work to
modify behavior. Journal of Economic Perspectives, 25(4): 191–210, doi:10.1257/ jep.25.4.191.
Goldstein, Daniel G., and Gigerenzer, Gerd. 2009. Fast and frugal forecasting. International Journal of
Forecasting, 25(4): 760–772.
Golman, Russell, Hagmann, David, and Loewenstein, George. 2017. Information avoidance. Journal of
Economic Literature, 55(1): 96–135, doi:10.1257/jel.20151245.
Greve, Henrich R. 2008. A behavioral theory of firm growth: Sequential attention to size and perform-
ance goals. Academy of Management Journal, 51(3): 476–494, doi:10.5465/amj.2008.32625975.
Hannan, M. T., and Freeman, J. 1984. Structural inertia and organizational change. American Sociological
Review, 49: 149–164.
Heyman, James, and Ariely, Dan. 2004. Effort for payment: A tale of two markets. Psychological Science,
15(11): 787–793.
Hoffman, Andrew J. 1999. Institutional evolution and change: Environmentalism and the us chemical
industry. Academy of Management Journal, 42(4): 351–371.
Holmstrom, Bengt, and Milgrom, Paul. 1991. Multitask principal-agent analyses: Incentive contracts, asset
ownership, and job design. Journal of Law, Economics, & Organization, 7: 24–25.
James, William. 1890. The Principles of Psychology. New York: Henry Holt and Company.
Joseph, John, and Ocasio, William. 2012. Architecture, attention, and adaptation in the multibusiness
firm: General Electric from 1951 to 2001. Strategic Management Journal, 33(6): 633–660, doi:10.1002/
smj.1971.
Joseph, John, and Wilson, Alex J. 2018. The growth of the firm: An attention-based view. Strategic
Management Journal, 39(6): 1779–1800, doi:10.1002/smj.2715.
Kaplan, Sarah. 2008. Framing contests: Strategy making under uncertainty. Organization Science, 19(5):
729–752, doi:10.1287/orsc.1070.0340.
Kareev, Yaakov. 1995. Through a narrow window: Working memory capacity and the detection of covari-
ation. Cognition, 56(3): 263–269.
Knudsen, Thorbjørn, and Warglien, Massimo. 2018. Why less is more: The power of simple cogni-
tive representations in organizational search and learning. SOD-SDU Working Paper, Strategic
Organization Design, University of Southern Denmark.
Kruschke, John K., and Hullinger, Richard A. 2010. Evolution of attention in learning. In Nestor
A. Schmajuk (Ed.), Computational Models of Conditioning (pp. 10–52). Cambridge: Cambridge
University Press.
Laibson, David. 1997. Golden eggs and hyperbolic discounting. The Quarterly Journal of Economics, 112(2):
443–478, doi:10.1162/003355397555253.
March, J. G., and Simon, H. 1958. Organizations. New York: John Wiley & Sons.

532
533

Attention and organizations

Maula, Markku V. J., Keil, Thomas, and Zahra, Shaker A. 2013. Top management’s attention to discon-
tinuous technological change: Corporate venture capital as an alert mechanism. Organization Science,
24(3): 926–947.
Mellström, Carl, and Johannesson, Magnus. 2008. Crowding out in blood donation: Was Titmuss right?
Journal of the European Economic Association, 6(4): 845–863.
Neisser, Ulric. 1979. The concept of intelligence. Intelligence, 3(3): 217–227.
Newport, Elissa L. 1990.Maturational constraints on language learning. Cognitive Science, 14(1): 11–28.
Nigam, Amit, and Ocasio, William. 2010. Event attention, environmental sensemaking, and change in
institutional logics: An inductive analysis of the effects of public attention to Clinton’s health care
reform initiative. Organization Science, 21(4): 823–841.
O’Brien, Jennifer L., and Raymond, Jane E. 2012. Learned predictiveness speeds visual processing.
Psychological Science, 23(4): 359–363.
Ocasio, William. 1997. Towards an attention-based view of the firm. Strategic Management Journal,
18(S1): 187–206.
Ocasio, William. 2011. Attention to attention. Organization Science, 22(5): 1286–1296.
Ocasio, William. 2012. Situated attention, loose and tight coupling, and the garbage can model.
In Alessandro Lomi, and J. Richard Harrison (Eds.), The Garbage Can Model of Organizational
Choice: Looking Forward at Forty, volume 36 of Research in the Sociology of Organizations (pp. 293–317).
Bingley, UK: Emerald Group Publishing Limited.
Ocasio, William, and Joseph, John. 2008. Rise and fall – or transformation? the evolution of strategic
planning at the General Electric Company, 1940–2006. Long Range Planning, 41(3): 248–272.
Ocasio, William, and Joseph, John. 2018. The attention-based view of great strategies. Strategy Science,
3(1): 289–294.
Ocasio, William, Laamanen, Tomi, and Vaara, Eero. 2018. Communication and attention dynamics: An
attention-based view of strategic change. Strategic Management Journal, 39(1): 155–167.
O’Donoghue, T., and Rabin, M. 1999. Doing it now or later. American Economic Review, 89(1): 103–124.
Organ, Dennis W. 1988. Organizational Citizenship Behavior: The Good Soldier Syndrome. Washington,
DC: Lexington Books/DC Heath and Com.
Prendergast, Canice. 1999. The provision of incentives in firms. Journal of Economic Literature, 37(1): 7–63.
Rerup, Claus. 2009. Attentional triangulation: Learning from unexpected rare crises. Organizational
Science, 5(20): 876893.
Shepherd, D. A.. McMullen, J. S., and Jennings, P. D. 2007. The formation of opportunity
beliefs: Overcoming ignorance and reducing doubt. Strategic Entrepreneurship Journal, 1(1–2): 75–95.
Shepherd, Dean A., McMullen, Jeffery S., and Ocasio, William. 2017. Is that an opportunity? An
attention model of top managers’ opportunity beliefs for strategic action. Strategic Management Journal,
38(3): 626–644.
Simon, Herbert A. 1947. Administrative Behavior: A Study of the Decision-Making Process in Administrative
Organization. New York: Macmillan.
Simon, Herbert A. 1954. Centralization vs. decentralization in organizing the controller’s department: A
research study and report. Number 4 in A1. Controllership Foundation.
Simon, Herbert A. 1971. Designing organizations for an information-rich world. Computers,
Communications, and the Public Interest, 72: 37.
Simon, Herbert A. 1983. Reason in Human Affairs, vol. 1982. Stanford, CA: Stanford University Press.
Simons, Robert. 1991. Strategic orientation and top management attention to control systems. Strategic
Management Journal, 12(1): 49–62.
Tolman, Edward C. 1932. Purposive Behavior in Animals and Men. New York: The Century Co.
Treem, J. W., and Leonardi, P. 2012. Social media use in organizations: Exploring the affordances of visi-
bility, editability, persistence and association. Communication Yearbook, 36: 143–189.
Tripsas, Mary, and Giovanni Gavetti. 2000. Capabilities, cognition, and inertia: Evidence from digital
imaging. Strategic Management Journal, 21(10–11): 1147–1161.
Tuggle, Chris S., Simmon, David G., Reutzel, Chris R., and Bierman, Leonard. 2010. Commanding
board of director attention: Investigating how organizational performance and CEO duality affect
board members attention to monitoring. Strategic Management Journal, 31(9): 946–968.
Tuggle, Christopher S., Schnatterly, Karen, and Johnson, Richard A. 2010. Attention patterns in the
boardroom: How board composition and processes affect discussion of entrepreneurial issues. The
Academy of Management Journal, 53(3): 550–571.

533
534

Inga Jonaityte and Massimo Warglien

Vandenbosch, Betty. 1999. An empirical analysis of the association between the use of executive
support systems and perceived organizational competitiveness. Accounting, Organizations and Society,
24(1): 77–92.
van Knippenberg, Daan, Dahlander, Linus, Haas, Martine R., and George, Gerard. 2015. Information,
attention, and decision making. Academy of Management Journal, 58(3): 649–657, doi:10.5465/
amj.2015.4003.
Vuori, Timo O., and Huy, Quy N. 2016. Distributed attention and shared emotions in the innovation pro-
cess: How Nokia lost the smartphone battle. Administrative Science Quarterly, 61(1): 9–51. doi:10.1177/
0001839215606951.
Weick, Karl E. 1979. The social psychology of organizing (2nd ed.). New York: Random House.
Wilson, Alex James, and Joseph, John. 2015. Organizational attention and technological search in the
multibusiness firm: Motorola from 1974 to 1997. In Giovanni Gavetti, and William Ocasio (Eds.),
Cognition and Strategy, vol. 32 of Advances in Strategic Management: A Research Annual (pp. 407–435).
Bingley: Emerald Group Publishing Limited.
Wilson, Timothy D., Wheatley, Thalia, Meyers, Jonathan M., Gilbert, Daniel T., and Axsom, Danny.
2000. Focalism: A source of durability bias in affective forecasting. Journal of Personality and Social
Psychology, 78(5): 821–836.
Zbaracki, Mark J., and Bergen, Mark. 2015. Managing market attention. In Giovanni Gavetti and William
Ocasio (Eds.), Cognition and Strategy, volume 32 of Advances in Strategic Management: A Research Annual
(pp. 371–405). Bingley: Emerald Group Publishing Limited.
Zegart, Amy B. 2006. An empirical analysis of failed intelligence reforms before September 11. Political
Science Quarterly, 121(1): 33–60.

534
535

36
THE BOUNDED RATIONALITY
OF GROUPS AND TEAMS
Torsten Reimer, Hayden Barber, and Kirstin Dolick

Like many teenage boys in Liverpool in 1960, John Lennon and Paul McCartney wanted
to start a rock band. However, as their fledgling band went from record company to record
company with their manager Brian Epstein, it became clear: Britain was getting tired of rock
and roll. According to Dick Rowe, who famously turned them down from Decca Records,
“Not to mince words, Mr. Epstein, we don’t like your boys’ sound. Groups of guitarists are
on their way out” (Lewisohn, 2013, p. 557). On the other hand, America at that time had
never seen anything like The Beatles. The Beatles received a very different reaction in the
US than in Britain. Seventy-three million Americans—the largest audience to view a show
at that time—tuned in to the Ed Sullivan Show on February 9, 1964, to watch the opti-
mism and joviality of four young Brits. The Beatles had enjoyed only moderate success in
Britain, but their success skyrocketed in the US. It was not just The Beatles, but the pairing
of The Beatles with a particular time and place that made their music so successful. The
story of The Beatles points to an important aspect of the bounded rationality of groups and
teams: The success or failure of a group cannot be understood without looking at the envir-
onment in which the group is embedded. The same strategy and behavior may be successful
in one environment but result in a failure in another environment. Thus, it can be adaptive
to change strategies such that they match the characteristics of environments; and, at times,
it can also be adaptive to change or switch environments. Different from classic standards of
rationality such as logical consistency, bounded rationality focuses on the match between a
decision strategy or problem procedure and characteristics of the environment (Gigerenzer
& Selten, 2001). Common characteristics of environments that have been studied in indi-
vidual decision-making research include characteristics of the structure of the information
environment, such as the similarity of choice alternatives and the similarity of the attributes
that describe the choice alternatives (Gigerenzer & Gaissmaier, 2011). In the context of
groups and teams, the environment also includes social characteristics of the group members
such as the distribution of knowledge about the choice alternatives and the distribution of
preferences among the involved members for specific attributes (see Reimer, Roland, &
Russell, 2017).
The concept of bounded rationality offers a new perspective on the study of groups and
teams. Bounded rationality can be seen as a guiding principle to conduct research that focuses

535
536

Torsten Reimer, Hayden Barber, and Kirstin Dolick

on three inter-related questions: (1) The strategy: Which strategies should groups use to form
decisions and solve problems (prescriptive question) and which strategies do groups use to form
decisions (descriptive question)?; (2) The environment: What are the characteristics of the social
and non-social environments in which groups form decisions?; and (3) Group adaptivity: Are
groups able to form ecologically rational decisions by selecting strategies that match the struc-
ture of their social and non-social environments?
The descriptive strategy question is the only question out of the three questions that has
repeatedly received attention in the literature on groups and teams (e.g., see Davis, 1982;
Kerr, MacCoun, & Kramer, 1996). Hardly any studies have looked at the second or third
questions regarding the information environments that are encountered by groups and the
match between group decision strategies and those environments. Likewise, only few studies
have addressed the question of how groups should form their decisions (e.g., see Hastie &
Kameda, 2005). The history of group research, though, is filled with studies that have evaluated
group performance, including the quality of group decisions and problem solutions. This his-
tory is—by and large—a history of failure, documenting the irrationality and poor performance
of groups. Accordingly, there is an abundance of studies on process losses in groups, whereas
only few studies on groups and teams identified process gains (for an example of a process gain,
see research on the Koehler effect, Kerr & Seok, 2011).
The notion of bounded rationality offers new conceptual and methodological perspectives
on the study of groups that hold the promise of providing alternative interpretations of at least
some of the process losses that have been described in the literature. In the remainder of our
chapter, we will first introduce a prominent task in group research, the hidden-profile task, that
is often cited as an example demonstrating that groups fail to form good decisions. Extensive
research on this task has suggested that groups are not able to connect the dots and integrate
relevant knowledge and, as a consequence, form poor decisions (for overviews, see Lu, Yuan,
& McLeod, 2012; Sohrab, Waller, & Kaplan, 2015). We demonstrate that approaching this
task from the perspective of bounded rationality can alter the interpretation of those findings.
Guided by the bounded rationality perspective, we illustrate in which situations groups can
solve hidden-profile tasks. The study of the bounded rationality of groups and teams is in
its infancy but has much to offer to the field of group communication and decision making.
We conclude by highlighting key insights from group research that has explored the bounded
rationality of groups and offer questions for future research.

Bounded rationality as an eye-opener: the case of hidden profiles

Research on groups and teams has often taken an information-processing perspective by
describing how task-related knowledge is distributed in groups and which pieces of informa-
tion groups discuss when forming decisions (e.g., Hinsz, Tindale, & Vollrath, 1997; Reimer
et al., 2017). A prominent task in this research tradition is the hidden-profile task (Stasser &
Titus, 1985; Stasser & Abele, 2020). Hidden-profile tasks are choice tasks in which groups are
asked to choose among a number of choice alternatives. The tasks are typically constructed
such that one choice alternative has more positive and fewer negative attributes than the other
alternatives in the choice set and is therefore assumed to be the best alternative (Stasser & Titus,
1985, Note 1). However, it is unknown to the individual group members at the outset that this
“best choice” has the most positive attributes as each member receives only a partial and biased
set of the information on the choice alternatives prior to group deliberation: Information items
on the choice alternatives are distributed such that each member has more positive and fewer
negative information items on a choice alternative that is different from the best choice.

536
537

The BR of groups and teams

As an example, consider the following simplified scenario (see Committee 1 in Table 36.1).
A three-member personnel committee has to decide which of two candidates, Steven or Peter,
is better suited for an open position for a sales manager. Each of the two candidates is described
in terms of several attributes including specific skills and personality traits. The columns in
Table 36.1 list the attributes for each candidate, and the first three rows depict the particular set
of information items that each group member is given prior to the group deliberation. The last
row (Group knowledge) describes the information that is known to the group as a whole. For the
sake of simplicity, only positive and fewer attributes as in experiments with hidden profiles are
used in the example (see Reimer, Kuendig, Hoffrage, Park, & Hinsz, 2007, for the description
of a more complex problem).
As illustrated in Table 36.1, each group member of Committee 1 knows that Steven has a
business degree and that he is a reliable and kind person. Regarding Peter, the group knows
that Peter has an engineering degree and that he can be described by a number of posi-
tive attributes—he is organized, punctual, well respected, open-minded, and hard working.
Overall, there are six positive attributes in favor of Peter compared to three attributes that speak
in favor of Steven. Based on the mere number of positive attributes, one would expect that the
group would pick Peter (see the Group knowledge row in Table 36.1). However, note that the three
committee members all have the same information about Steven (this information is shared),
whereas their knowledge about Peter is unshared. As a consequence, each individual group member
is likely to prefer Steven at the outset as each group member knows more positive attributes about
Steven than they do about Peter. Thus, the question arises: Are groups able to connect the
dots in their group discussions and discover that, overall, Peter is described by more positive
attributes than Steven?
The literature on hidden profiles suggests that groups are unable to connect the dots in this
situation: A vast majority of groups fail to detect hidden profiles by choosing the alternative that
is preferred by most members at the outset (Steven in our example; for overviews, see Sohrab
et al., 2015; Luan et al., 2012). Empirical studies included various types of groups (e.g., pro-
fessional teams in organizations; teams of doctors and nurses) as well as a variety of tasks (e.g.,
medical choice tasks, murder mysteries, and personnel selection tasks).
Why are groups not able to detect hidden profiles? The most prominent explanations for
the hidden-profile effect focus on why those pieces of information that are shared by all group
members have more impact on group decisions than unshared pieces of information. One
explanation is based on the sampling advantage of shared information, that is, the observa-
tion that groups tend to exchange more of their shared than unshared information items and
also pay more attention to shared information when it is brought up during group discussions
(Stasser & Titus, 1985; Reimer, Reimer, & Hinsz, 2010; Wittenbaum & Park, 2001). If the
group members of Committee 1 primarily exchange their shared information on Steven, they
will fail to become aware that Peter’s profile has more positive attributes than Steven’s profile.
A second prominent explanation is based on the observation that shared compared to
unshared information also has a greater impact on the preferences of the individual group
members at the outset because shared information is known to more group members (the
common-knowledge effect; see Gigone & Hastie, 1993). Hidden-profile tasks are set up such
that shared information favors the preferences that group members form before they enter
group discussions. If the group integrates members’ initial preferences on the basis of a social
combination rule like the majority rule, groups will fail to choose the hidden-profile alterna-
tive. Groups only rarely choose an alternative that is not proposed by at least one member at the
outset (e.g., Brodbeck, Kerschreiter, Mojzisch, Frey, & Schulz-Hardt, 2002; Reimer, 1999). In
short, according to these two explanations, groups fail to connect the dots in hidden-profile

537
538

Torsten Reimer, Hayden Barber, and Kirstin Dolick

Table 36.1 Two hidden profiles in information environments with two alternatives (Candidate A and
B), three group members (Members 1 to 3), and unique (Committee 1) or common attributes
(Committee 2)

Committee 1: unique attributes

Member/group Knowledge about Steven Knowledge about Peter Decision based on

a count of positive
attributes

Member 1 Reliable, kind, business Organized, open-minded Steven

degree
Member 2 Reliable, kind, business Hard working, respected Steven
degree
Member 3 Reliable, kind, business Punctual, engineering Steven
degree degree
Group knowledge Reliable, kind, business Organized, open-minded, Peter
degree punctual, respected,
hardworking, engineering
degree

Committee 2: common attributes

Member/group Knowledge about Steven Knowledge about Peter Decision based on

a count of positive
attributes
Member 1 Reliable, kind, business Reliable, organized Steven
degree
Member 2 Reliable, kind, business Kind, open-minded Steven
degree
Member 3 Reliable, kind, business Business degree, hardworking Steven
degree
Group knowledge Reliable, kind, business Reliable, kind, business degree, Peter
degree organized, open-minded,
hard working

Notes: In the examples, attributes are assumed to be positive throughout. Thus, each attribute speaks,
if present, in favor of the particular candidate. Attributes that do not discriminate between alternatives
are written in italics. The numbers of information items that are known to the group and to the
individual group members as well as the two candidates’ sum scores are held constant across the two
committees. The profile on Peter is hidden to the individual group members. Group knowledge displays
the information items that are known to the group as a whole.

tasks because their shared information has more impact on the group members’ individual
preferences (common-knowledge effect) and is also more likely to be mentioned during discus-
sion than unshared information (sampling advantage of shared information).
The starting point of our journey was the conviction that the notion of bounded rationality
can help us better understand the failure of groups in experiments using hidden-profile tasks.
Approaching hidden profiles from a bounded-rationality perspective suggests we apply the three

538
539

The BR of groups and teams

questions introduced at the beginning of the chapter: (1) Which strategies should groups use
when facing hidden-profile tasks (prescriptive question), and which strategies do they use when
solving these tasks (descriptive question)? (2) What are the characteristics of the information
environments that are used in hidden-profile research? and (3) How would ecologically rational
groups behave when facing hidden-profile tasks? To answer these questions, we conducted a
number of simulation studies, a meta-analysis, and experiments with interacting groups. This
journey enabled us to identify a condition in which all groups were able to connect the dots and
choose the hidden-profile alternative. The following sections summarize the lessons learned by
addressing the three questions and provide open questions for future research.

Which strategy should groups use to solve hidden-profile tasks?

Whereas most research on hidden profiles has focused on factors that affect the amount of
pooled information, the question of how the available information should be processed and
integrated into a decision has been addressed by only a few studies (e.g., see Stasser, 1988;
Chernyshenko, Miner, Baumann, & Sniezek, 2003).
One strategy would be to discuss all the available information, particularly the unshared infor-
mation, and to choose the alternative that has the highest sum score. Such a communication-
based unit weight linear model (Reimer & Hoffrage, 2005) would always detect hidden profiles.
It is intuitive that group decisions would not be affected by the distribution of information if
members exchanged and integrated all of their knowledge. However, because this would be a
very inefficient and laborious endeavor, we conducted several simulation studies, in which we
aimed to identify plausible decision strategies that detect hidden profiles and limit information
processing at the same time (Reimer & Hoffrage, 2005; Reimer et al., 2007).
In one set of simulations, we first generated random information distributions by distrib-
uting information on a choice task randomly among the simulated members of a group. In these
simulations, four-member committees selected one of three candidates in an information envir-
onment that had an outside criterion. Introducing an outside criterion allowed us to determine
the strategies’ accuracies under various conditions. In addition, we constructed four environ-
ments in which we systematically varied the distributions of cue validities. The validity of a
cue refers to the percentage of cases in which a cue makes a correct inference (Katsikopoulos,
2011). In two of the four environments, the distribution of cue validities was linear (L), and in
the other two they followed a J-shaped distribution. The two linear distributions differed with
respect to their overall means (L-high versus L-low), and the two J-shaped distributions differed
in their skewness, which mainly affected the validity of the most valid cues (J-flat versus J-steep;
see Reimer & Hoffrage, 2006, for details).
In the next step, we selected from all the generated information environments those envir-
onments that contained a hidden profile—environments in which one choice alternative
dominated the remaining alternatives by having the most positive and least negative attributes
in the choice set but where information was distributed such that each individual group
member’s information favored another alternative. The simulations taught us several lessons.
The simulated groups failed in more than 50 percent of the cases to solve the hidden-profile
task when integrating individual preferences on the basis of a majority rule, irrespective of the
strategy the individual members used. This result accords well with the common-knowledge
effect (Gigone & Hastie, 1993): Hidden-profile tasks typically cannot be solved on the basis
of a social combination rule that integrates individual preferences but require the exchange
and integration of unshared information items that are known by different individual group
members.

539
540

Torsten Reimer, Hayden Barber, and Kirstin Dolick

Unlike social combination rules, social communication rules integrate individual informa-
tion items rather than individual choices or preferences. Similar to decision strategies for indi-
vidual deciders, we implemented two types of communication-based strategies—strategies that
process information alternative-wise and strategies that process information attribute-wise. The
communication-based unit weight model described above is an example of a decision strategy
that processes information alternative-wise: A group using this strategy discusses each choice
alternative by exchanging all information on that alternative and forming an overall sum score.
Once all the alternatives are discussed, the group will choose the alternative with the highest
overall sum score. Conversely, the following strategy processes information attribute-wise: The
strategy assumes that one group member draws the group’s attention to one attribute on which
this member has at least some knowledge. The group then compares the alternatives on this
attribute by pooling all available values on this attribute. If the attribute discriminates among
the alternatives such that one alternative has a positive characteristic that sets this alternative
apart, group discussions and information processing stop, and the group will decide in favor
of that alternative (see Reimer & Hoffrage, 2005, in which we also discussed variants where
group members were not drawn randomly but according to their expertise). Both strategies are
communication-based as they require group members to pool and share information. However,
different from the alternative-based strategy, the attribute-based strategy has a stopping rule and
is, thus, more frugal than the communication-based unit weight linear model. The simulations
revealed that the described attribute-based strategy chose the hidden-profile alternative in a
high percentage of cases, despite its frugality. As a general result, communication-based strat-
egies led groups to choose the hidden-profile alternative much more often than combination
rules. The described attribute-based strategy was particularly successful in identifying hidden
profiles.
The simulation studies, thus, did not accord with the results of empirical research described
in the literature. Whereas the vast majority of groups in empirical studies failed to connect the
dots and choose hidden profiles in experiments, the groups in our simulations performed much
better. Thus, the question arose: Why are interacting groups not able to detect hidden profiles if
there are effective and efficient strategies? When we went through the literature, we discovered
that all studies we could find used a peculiar information environment that might have hindered
groups from using an attribute-based strategy: Groups were presented with unique attributes
throughout that described only one of the choice alternatives.

Peculiar information environments trigger the use of

sub-optimal strategies
The bounded rationality perspective suggests that not only characteristics of the task (e.g.,
Davis, 1982) but also characteristics of the information environment influence which decision
strategies groups use. Research on hidden profiles provides a good example for this claim as
studies on hidden profiles have used very peculiar information environments. We offer as an
explanation for groups’ failure to solve hidden-profile tasks that the specific information envir-
onments that have been used in the literature trigger decision strategies that are not adaptive
nor goal-serving, as they do not enable groups to discover hidden profiles. Even though the
sampling advantage for shared information and the hidden-profile effect have been evident in
a variety of groups performing a variety of decision tasks, all of the studies held an important
dimension of the information environment constant: Group members were provided with
shared and unshared information for unique attributes that described only one of the alternatives
(see Fraidin, 2004; Reimer et al., 2007). However, there is evidence from research relating

540
541

The BR of groups and teams

to individuals that attribute commonality can affect which information a receiver samples and
integrates into a decision.
The two dimensions of information sharedness and attribute commonality are orthog-
onal: The values of attributes can be shared or unshared, and attributes can be common or
unique. Sharedness of information refers to whether an information item (i.e., an attribute value)
is one that multiple group members know (shared information) or one that only one group
member knows (unshared information). Conversely, attribute commonality refers to whether
multiple (common attribute) or only one of the decision alternatives (unique attribute) has
a known value for this particular attribute, irrespective of how many group members possess
this knowledge (see Reimer et al., 2007). If multiple decision alternatives are described by this
attribute, then this attribute is common; if only one alternative is described by this attribute,
then the attribute is unique.
As in the information distribution among the members of Committee 1 (see Table 36.1),
research on hidden profiles exclusively implemented unique attributes. Some of the informa-
tion items on Peter and Steven are shared among group members and some information items
are unshared. However, all information items refer to unique attributes. For each and every
attribute, Committee 1 only has knowledge about one alternative. For example, Committee 1
knows that Peter is organized and that Steven is reliable and kind; however, the committee does
not have any knowledge about whether or not Steven is organized and Peter is reliable and kind.
How does the exclusive use of unique attributes and lack of any common attributes affect
the selection of strategies? Research on individuals suggests that the presentation of unique
attributes triggers the use of alternative-based decision strategies (Burke, 1990, 1995; also see
Payne, Bettman, & Johnson, 1988). Because the potential alternatives cannot be compared
along the same attributes, deciders form a global impression of each alternative. Then, he or she
chooses the alternative that creates the best overall impression, for example, the one with the
most positive and least negative features. Conversely, the presentation of common attributes,
for which all alternatives have a value, facilitates the implementation of attribute-based deci-
sion strategies—deciders compare alternatives on common attributes or features (Burke, 1990,
1995; Rieskamp & Hoffrage, 1999). Research on individuals suggests that attribute common-
ality can systematically influence which decision strategies are adopted at the individual level.
We expected and observed that attribute commonality also has an effect on how groups form
decisions and solve hidden-profile tasks.
To illustrate, a hidden profile within an information environment that contains common
attributes is shown in the lower part of Table 36.1 (see Committee 2). As is indicated by the last
row, which displays the information that is known to the group as a whole (Group knowledge),
there are again three attributes in favor of Steven and six attributes in favor of Peter. However,
in Committee 2, both candidates––Steven as well as Peter––can be described as being reliable
and kind; and the group knows that both have a business degree. If group members mention
Steven with respect to those three attributes during their discussion, they might also bring up
what they know about Peter in regard to these attributes. This process should lead to several
outcomes. First of all, the group members should realize that the attributes that support Steven
do not discriminate between Steven and Peter. In other words, Steven does not possess any
positive attribute that Peter would not have. Second, this process should prompt the discussion
of unshared information. Imagine that Committee Member 1 tells the group that a plus for
Steven is that he is kind. The other members of the group will agree because this is a shared
piece of information that they all know. However, after Committee Member 1 mentions the
attribute kind, Committee Member 2 may, in turn, mention that Peter has also been described
as being kind, something the group does not already know.

541
542

Torsten Reimer, Hayden Barber, and Kirstin Dolick

As expected, we observed that providing common attributes facilitated the exchange of

unshared pieces of information in group discussions (Reimer et al., 2007; Reimer et al., 2010).
Note that the information environments in terms of the number of information items that are
available to the individual group members as well as the candidates’ sum scores are identical
in the two examples displayed in Table 36.1. However, the information structure provided in
the second example is decision-maker friendly insofar as a group will always detect the hidden
profile if they process the available information in an attribute-wise fashion by looking for
attributes that discriminate between the alternatives.
The concept of bounded rationality served as an eye-opener which enabled us to identify
a major shortcoming in research on hidden profiles and to set up an experimental condition
in which all groups were able to solve the hidden-profile task: In experiments systematically
manipulating the attribute commonality (Reimer et al., 2007; Reimer et al., 2010), none of
the studied groups selected the hidden-profile alternative when group members were given
their information prior to group discussions, when they did not have access to their infor-
mation during discussion, and when the information structure was exclusively composed of
unique attributes (as in Committee 1 in Table 36.1). Conversely, all groups selected the hidden-
profile alternative when group members were given their information about the alternatives
at the onset of the group session, had access to their information sheets during group deliber-
ation, and when information environments entailed common attributes (as in Committee 2 in
Table 36.1). Distributing information at the beginning of group discussions prevented group
members from forming strong individual preferences at the outset and, thus, prevented the
common-knowledge effect (see Reimer et al., 2007; Reimer et al., 2010).

Bounded rationality as a research program and paradigm

for group research
The study of the bounded rationality of groups and teams is in its infancy. Even though
information-processing approaches in group research have a long tradition, and group research
has acknowledged that group performance depends on the characteristics of the tasks that are
studied (Davis, 1982; Kerr et al., 1996), only few studies have taken a bounded rationality per-
spective by looking at the match of decision strategies and characteristics of the information
environment.
As the hidden-profile example illustrates, the concept of bounded rationality offers a new
perspective to the study of groups and teams. In the remainder, we highlight two important
quests for the study of group decision making from a bounded rationality perspective: (1) the
focus on group adaptivity; and (2) the need to move beyond as-if models by providing process
models of group decision making.

Group adaptivity
At the heart of the concept of bounded rationality is the idea of adaptive decision making.
Group adaptivity refers to the match of decision strategies and the structure of information in
the environment. Adaptivity can be achieved through the selection of decision strategies by
groups as well as a change or even switch of the environment (e.g., as in the success story of the
Beatles). As with individual decision making, there is no golden group decision strategy that
fits all purposes and works well in all possible environments. Studying the bounded ration-
ality of groups includes the methodological quest to describe not only possible decision strat-
egies but also the information environments in which they are used, as specific strategies may

542
543

The BR of groups and teams

yield better decisions in some environments than in others. Moreover, in group research, it
is important also to consider characteristics of the social environments. For example, Luan,
Katsikopoulos, and Reimer (2012) identified information environments, in which diversity
in a group trumped individual accuracy and other environments in which individual accuracy
trumped diversity. Specifically, individual accuracy was more important for the performance
of a simulated group in task environments in which cues differed greatly in the quality of their
information (as in environments with J-shaped validities as discussed above), and diversity
mattered more when such differences were relatively small (as in environments with linear
distributions of validities).
To the extent that groups encounter environments that demand different decision strategies
and to the extent that groups encounter dynamic environments in which task demands change,
it is important that groups are adaptive. It has been acknowledged in the team literature that
adaptiveness is a vital competency and a component of effective teamwork (Kozlowski, 1998;
Klein & Pierce, 2001). Team adaptation and the ability to select and change strategies based
on information in the environment have been identified as a relevant skill in group decision
making because group members must modify or replace routine performance strategies when
they detect that the characteristics of the environment and task change (Salas, Sims, & Burke,
2005; Reimer, Opwis, & Bornstein, 2005; Burke, Stagl, Salas, Pierce, & Kandall, 2006). To be
adaptive, teams have to be able to focus their attention on relevant information to accurately
understand the situation and achieve a shared team situation model (Sonesh, Rico, & Salas,
2014). Some research suggests that groups are able to show some adaptivity in their strategy
selection. Kaemmer, Gaissmaier, Reimer, and Schermuly (2014), for example, observed that
groups could be best modeled by strategies that matched the predictive validity of a strategy
in their groups (also see Reimer & Katsikopoulos, 2004). At the same time, groups often have
difficulties in changing and adapting their behaviors in situations in which task dependencies
and coordination requirements are high. This holds in particular for homogeneous groups in
which group members have similar knowledge and apply the same decision or problem-solving
strategies (Reimer et al., 2005). Organizations as well as teams within organizations often use
routines to approach problems and form decisions (Reimer et al., 2017). Routines can have
many advantages: As long as environments are stable, routines will allow teams to form decisions
efficiently and reliably. Typically, members of teams who are in a routine mode have clear roles
and do not have to negotiate and coordinate who is doing what. However, routine decision
making is a burden in environments in which groups and teams have to make decisions under
high uncertainty. In particular, if environments are dynamic and changing quickly, routines
can hinder groups and teams from being adaptive. An important quest for boundedly rational
groups consists in learning when to develop and apply routines and when to abandon routines
or not to develop routines, to begin with. Likewise, more research is needed to better under-
stand in which situations groups should actively alter or change their environment instead of
changing the strategies and procedures that are used.
Despite the insight that the study of a group’s information and social environments is
necessary to understand and evaluate group decision making, only few studies have taken
this quest seriously. Research on the Condorcet paradox provides an example of such an
approach that can serve as a role model. The Condorcet paradox describes a situation in
which plurality voting rules yield intransitive preferences. Specifically, plurality voting can
yield a situation in which majorities prefer Candidate A over B, B over C, and yet C over
A. However, empirical research suggests that the Condorcet paradox, even though it provides
a theoretically fascinating challenge, practically hardly ever occurs (Regenwetter, 2009).
Regenwetter (2009) developed a methodological framework for Behavioral Social Choice

543
544

Torsten Reimer, Hayden Barber, and Kirstin Dolick

in which the mathematical properties of social choice procedures can be evaluated against
empirical behavioral data. The application of this framework to large data sets revealed that the
logically possible intransitivity is practically irrelevant as it does not occur in political elections
(Regenwetter, 2009). The notion of bounded rationality comes with an invitation for group
researchers to start systematically studying characteristics of information environments that
groups and teams encounter.
By the same token, it would be worthwhile to understand how often groups face hidden-
profile tasks. It is obvious that groups practically never encounter an environment in which
they have only access to unique cues. We also know that the chance of the occurrence of a
hidden profile depends on the distribution of cue validities in an environment (Reimer &
Hoffrage, 2012). However, we do not have systematic data and insights about the prevalence
and characteristics of hidden profiles that groups face outside of laboratories. These data would
help understand whether hidden profiles are encountered by natural groups and how they
could be solved.

As-if models vs. process models

The notion of bounded rationality focuses on the strategies that are used by deciders, on the
adaptivity of their decision strategies, and the match between strategies and the structure of the
information environment in which the strategies are used. This approach hinges on the appro-
priateness of the described decision strategies. Even though group research has a long trad-
ition of formulating group decision strategies that integrate individual preferences or choices
(such as the majority rule), the studied decision strategies have the character of as-if models.
Different from actual process models, as-if models of decision making do not claim to provide
a psychologically viable description of the actual decision process or mechanism but only pro-
vide a model that predicts the decision outcome (see Gigerenzer, Todd, & the ABC Research
Group, 1999). Group research suggests that groups often behave as if they applied a majority
rule; however, that does not mean that groups necessarily engage in voting, which would be
an underlying process model.
The description of decision strategies has a long tradition in group research. Studies on
social decision schemes, for example, have distinguished between majority/plurality rules, the
truth-win principle, proportionality, and equiprobability as possible strategies to form a group
decision. All of these strategies integrate individual opinions or preferences into a group deci-
sion (Davis, 1982). Empirical research across a variety of different groups, teams, tasks, and
contexts revealed that the majority/plurality rule describes many group decisions well. If we
predict that a group will go with the choice alternative that is preferred by most members at the
outset, the prediction of the group choice will hold in many environments. Except for hidden-
profile tasks, the majority rule is not only descriptive but also often prescriptive in that it yields
good decisions in many situations (Hastie & Kameda, 2005).
The majority rule and other social decision schemes have typically been studied as as-if
models, that is, groups behaved as if they integrated individual preferences in a certain way.
However, there are many potential processes that would allow groups to implement a majority
rule and alternative rules that yield the same decisions. One possible process model would be
voting. However, groups only rarely engage in formal voting in empirical experiments unless
they are explicitly asked to do so. Alternative process models to implement the majority rule
include a process by which individual group members express their preference for a specific
choice early in the group deliberation. Non-verbal signals such as nodding and the lack of

544
545

The BR of groups and teams

disagreement can serve the function of communicating agreement. Another plausible mech-
anism consists in the imitation of group members. Future research may describe and test spe-
cific process models that are used by groups to form a joint group decision.

Conclusion
The concept of bounded rationality offers a new perspective on the study of groups and teams
that moves beyond traditional group research. As a signature characteristic, group research that
subscribes to the approach of bounded rationality focuses on the match between group strat-
egies and characteristics of both the information and social environment. The history of group
research using outcome and performance measures is, by and large, a history of demonstrating
group failure. We illustrated how the concept of bounded rationality can broaden our view on
process losses. It may well turn out that several process losses that are described in the literature
reflect group behaviors that are functional in many environments.
It is the match in strategy and environment that leads to successful group outcomes. The
presented research on hidden profiles provides evidence for that claim, and there is also anec-
dotal evidence that illustrates this principle. To return to the opening anecdote on The Beatles,
“Those later world-changing Beatles … [were the same] Beatles, just lesser-known, local not
global” (Lewisohn, 2013, p. 3). The Beatles adapted to their initial British environment, as
evidenced by a number of successful hits on the UK singles’ charts prior to their American
debut. However, it was the perfect match of group and the social environment of their audi-
ence that allowed them to truly shine. As Epstein put it: “One has thought about America in
connection with the Beatles … I always thought—I was always quite sure, really—that the
Beatles would make it big over there” (Lewisohn, 2013, p. 506). More research is needed that
enables us to understand to what extent groups and teams adapt to the information and social
environments in which they make decisions.

References
Brodbeck, F. C., Kerschreiter, R., Mojzisch, A., Frey, D., & Schulz-Hardt, S. (2002). The dissemin-
ation of critical, unshared information in decision-making groups: The effects of prediscussion dissent.
European Journal of Social Psychology, 32, 35–56.
Burke, C. S., Stagl, K. C., Salas, E., Pierce, L., & Kandall, D. L. (2006). Understanding team adaptation: A
conceptual analysis and model. Journal of Applied Psychology, 91, 1189–1207.
Burke, S. J. (1990). The effects of missing information on decision strategy selection. Advances in Consumer
Research, 17, 250–256.
Burke, S. J. (1995). The dimensional effects of missing information on choice processing. Journal of
Behavioral Decision Making, 8, 223–244.
Chernyshenko, O. S., Miner, A. G., Baumann, M. R., & Sniezek J. A. (2003). The impact of information
distribution, ownership, and discussion on group member judgment: The differential cue weighting
model. Organizational Behavior and Human Decision Processes, 91, 12–25.
Davis, J. H. (1982). Social interaction as a combinatorial process in group decisions. In H. Brandstaetter, J.
H. Davis, & G. Stocker-Kreichgauer (Eds.), Group decision making (pp. 27–58). London: Academic Press.
Fraidin, S. N. (2004). When is one head better than two? Interdependent information in group decision
making. Organizational Behavior and Human Decision Processes, 93, 102–113.
Gigerenzer, G., & Gaissmaier, W. (2011). Heuristic decision making. Annual Review of Psychology, 62,
451–482.
Gigerenzer, G., & Selten, R. (Eds.) (2001). Bounded rationality: The adaptive toolbox. Cambridge.
MA: MIT Press.
Gigerenzer, G., Todd, P. M., & the ABC Research Group (Eds.) (1999). Simple heuristics that make us smart.
New York: Oxford University Press.

545
546

Torsten Reimer, Hayden Barber, and Kirstin Dolick

Gigone, D., & Hastie, R. (1993). The common knowledge effect: Information sampling and group
judgment. Journal of Personality and Social Psychology, 65, 959–974.
Hastie, R., & Kameda, T. (2005). The robust beauty of majority rules in group decisions. Psychological
Review, 112, 494–508.
Hinsz, V. B., Tindale, R. S., & Vollrath, D. A. (1997). The emerging conceptualization of groups as infor-
mation processors. Psychological Bulletin, 121, 43–64.
Johnson, E. J., Payne, J. W., & Bettman, J. R. (1988). Information displays and preference reversals.
Organizational Behavior and Human Decision Processes, 42, 1–21.
Kaemmer, J., Gaissmaier, W., Reimer, T., & Schermuly, C. C. (2014). The adaptive use of recognition in
group decision making. Cognitive Science, 38, 911–942.
Katsikopoulos, K. V. (2011). Psychological heuristics for making inferences: Definition, performance, and
the emerging theory and practice. Decision Analysis, 8, 10–29.
Kerr, N. L., MacCoun, R. J., & Kramer, G. P. (1996). Bias in judgment: Comparing individuals
and groups. Psychological Review, 103, 687–719.
Kerr, N. L., & Seok, D.-H. (2011). “… with a little help from my friends”: Friendship, effort norms, and
group motivation gain. Journal of Managerial Psychology, 26, 205–218.
Klein, G. A., & Pierce, L. (2001). Adaptive teams. Paper presented at 6th International Command and
Control Research and Technology Symposium, Annapolis, MD, June 19–21.
Kozlowski, S. W. (1998). Training and developing adaptive teams: Theory, principles, and research. In J.
A. Cannon-Bowers & E. Salas (Eds.), Decision making under stress: Implications for training and simulation
(pp. 115–153). Washington, DC: APA Books.
Luan, S., Katsikopoulos, K., & Reimer, T. (2012). When does diversity trump ability (and vice versa) in
group decision making? A simulation study. PLoS ONE, 7, 1–9.
Lewisohn, M. (2013). Tune in: The Beatles: all these years. New York: Crown Archetype.
Payne, J. W., Bettman, J. R., & Johnson, E. J. (1988). Adaptive strategy selection in decision making.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 14, 534–552.
Regenwetter, M. (2009). Perspectives on preference aggregation. Perspectives on Psychological Science, 4,
403–407.
Reimer, T. (1999). Argumentieren und Problemlösen [Argumentation and problem solving]. Lengerich,
Germany: Pabst Science Publishers.
Reimer, T., Bornstein, A.-L., & Opwis, K. (2005). Positive and negative transfer effects in groups. In T.
Betsch & S. Haberstroh (Eds.), The routine of decision making (pp. 175–192). Mahwah, NJ: Lawrence
Erlbaum Associates.
Reimer, T., & Hoffrage, U. (2005). Can simple group heuristics detect hidden profiles in randomly
generated environments? Swiss Journal of Psychology, 64, 21–37.
Reimer, T., & Hoffrage, U. (2006). The ecological rationality of simple group heuristics: Effects of group
member strategies on decision accuracy. Theory and Decision, 60, 403–438.
Reimer, T., & Hoffrage, U. (2012). Simple heuristics and information sharing in groups. In R. Hertwig,
U. Hoffrage, & the ABC Research Group (Eds.), Simple heuristics in a social world (pp. 266–286).
New York: Oxford University Press.
Reimer, T., & Katsikopoulos, K. (2004). The use of recognition in group decision-making. Cognitive
Science, 28, 1009–1029.
Reimer, T., Kuendig, S., Hoffrage, U., Park, E., & Hinsz, V. (2007). Effects of the information environ-
ment on group discussions and decisions in the hidden-profile paradigm. Communication Monographs,
74, 1–28.
Reimer, T., Reimer, A., & Hinsz, V. (2010). Naïve groups can solve the hidden-profile problem. Human
Communication Research, 36, 443–467.
Reimer, T., Roland, C., & Russell, T. (2017). Groups and teams. In L. Lewis & C. Scott (Eds.), International
encyclopedia of organizational communication (pp. 1–21). Hoboken, NJ: Wiley-Blackwell.
Rieskamp, J., & Hoffrage, U. (1999). When do people use simple heuristics, and how can we tell? In
G. Gigerenzer, P. M. Todd, & the ABC Research Group (Eds.), Simple heuristics that make us smart
(pp. 141–167). New York: Oxford University Press.
Salas, E., Sims, D. E., & Burke, C. S. (2005). Is there a “big five” in teamwork? Small Group Research, 36,
555–599.
Sohrab, S. G., Waller, M. J., & Kaplan, S. (2015). Exploring the hidden-profile paradigm: A literature
review and analysis. Small Group Research, 46, 489–535.

546
547

The BR of groups and teams

Sonesh, S., Rico, R., & Salas, E. (2014). Team decision making in naturalistic environments: A frame-
work for and introduction to illusory shared cognition. In S. Highhouse, R. S. Dalal, & E. Salas (Eds.),
Judgment and decision making at work (pp. 199–227). New York: Routledge.
Stasser, G. (1988). Computer simulation as a research tool: The DISCUSS model of group decision
making. Journal of Experimental Social Psychology, 24, 393–422.
Stasser, G., & Abele, S. (2020). Collective choice, collaboration, and communication. Annual Review of
Psychology, 71, 589–612.
Stasser, G., & Titus, W. (1985). Pooling of unshared information in group decision making: Biased infor-
mation sampling during discussion. Journal of Personality and Social Psychology, 48, 1467–1478.
Wittenbaum, G. M., & Park, E. S. (2001). The collective preference for shared information. Current
Directions in Psychological Science, 10, 70–73.

547
548

37
COGNITIVE BIASES AND
DEBIASING IN INTELLIGENCE
ANALYSIS
Ian K. Belton and Mandeep K. Dhami

Introduction
Simon (1947, p. 79) boldly asserted that “It is impossible for the behavior of a single, isolated
individual to reach any high degree of rationality.” He recognized that rationality is bounded
by limitations of the unaided human mind and by the complexity and uncertainty of the task
environment (Simon, 1957). Simon (1956, 1990) believed that organisms are adapted to the
structure of their environments. “Human rational behavior … is shaped by a scissors whose two
blades are the structure of task environments and the computational capabilities of the actor”
(Simon, 1990, p. 7). He believed that under such conditions people may opt to satisfice, i.e.,
settle on a “good enough” solution (Simon, 1956). Indeed, research has now identified a wide
range of systematic deviations from normatively rational behavior, often referred to collectively
as “cognitive biases” (for reviews, see Gilovich, Griffin, & Kahneman, 2002; Kahneman, 2011;
Kahneman, Slovic, & Tversky, 1982).
However, as Katsikopoulos and Lan (2011) note, Simon also explored ways in which per-
formance could be improved so that satisficing and boundedly rational behavior did not lead
to poor outcomes. Similarly, in response to the findings regarding cognitive biases, researchers
have begun to identify and test possible debiasing strategies (for reviews, see Arkes, 1991;
Fischhoff, 1982; Larrick, 2004; Lilienfeld, Ammirati, & Landfield, 2009). Several classifications
of cognitive bias have been proposed, and debiasing strategies may be informed by theories of
how and why a bias may occur.
Fischhoff (1982) distinguishes between faulty decision makers, faulty tasks, and deci-
sion maker–task mismatches. Arkes (1991) focuses on the type of error made, distinguishing
association-based errors (i.e., side-effects of the fact that triggering one concept in memory
activates other, related concepts) from psychophysical errors based on risk aversion and our
biased response to losses and gains, and strategy-based errors resulting from a person using a
poor analytic strategy. Stanovich, Toplak, and West (2008) base their approach to bias on dual-
process theories of cognition which distinguish System 1 thinking (i.e., fast, unconscious, auto-
matic, associative, parallel processing, effortless, and intuitive) from System 2 thinking (i.e., slow,
conscious, controlled, rule-based, sequential processing, effortful, and deliberative). Cognitive
bias is considered to be the product of System 1, which can be mitigated when an individual
uses System 2 to override a System 1 response. A successful override may fail to occur if, for
example, there is a lack of effort/self-control (“override failures/cognitive miserliness”), a lack

548
549

Cognitive biases in intelligence analysis

of knowledge of the correct strategy to use (“mindware gaps”), or use of a flawed strategy
(“contaminated mindware”; see also Stanovich & West, 2000). Similarly, Kahneman, Slovic,
and Tversky (1982) distinguish between comprehension errors (i.e., mindware gaps) and appli-
cation errors (i.e., override failures or cognitive miserliness).
Dual-process theories of mind have been criticized and alternative accounts exist. Viale
(2018; 2019) argues that the dual-process account is undermined by recent psychological and
neuroscientific evidence which suggests that the cognitive processes involved in creative incu-
bation and mind wandering are both unconscious and analytical. Alternatively, Hammond’s
(1996, 2000) cognitive continuum theory proposes that there are many “quasi-rational” modes
of cognition that lie in between the poles of intuitive and analytic thinking, and which com-
prise different combinations of intuition and analysis (see also Sloman’s 1996 view that intuition
and analysis are interactive). According to Hammond, the mode of cognition applied to a task
is determined by task properties and/or the experience the individual has with the task. Success
on a task inhibits movement along the cognitive continuum (or change in cognitive mode)
while failure stimulates it.
When working in an organizational context, Simon (1947, p. 79) believed that organizations
can place “members in a psychological environment that will … provide them with the infor-
mation needed to make decisions correctly.” For instance, organizations can establish standard
working practices, train individuals, and structure the work environment so that it encourages
rational thinking and consistency or regularization of practice. In this chapter, we critically
evaluate the solutions that intelligence analysis organizations have offered to combat cogni-
tive bias in their intelligence analysts. We identify cognitive biases that may affect the practice
of intelligence analysis and review debiasing strategies developed and tested by psychological
research.

Intelligence analysis
The intelligence analyst must produce a coherent report that precisely communicates his/her
conclusions about a current or future situation to a variety of consumers who may then make
strategic and tactical decisions that affect national and international security. For example, what
forces does North Korea have along the border with South Korea? What will be Russia’s offen-
sive cyber capability in 2025? Where are the headquarters of terror group X?
At its core, intelligence analysis is a cognitive task. Analysts must plan, search for, select,
process, and interpret data in order to gain situational awareness and/or forecast an outcome of
interest to customers. This is articulated in the generic analytic workflow (Dhami & Careless,
2015) which applies to different sorts of intelligence analysis (e.g., HUMINT, SIGNIT, as well
as single and multi-source), conducted individually or in teams, for different purposes (e.g.,
strategic, tactical). The workflow is separated into six meaningfully different stages of activity
that follow from one another, namely, capture requirements, plan analytic response, obtain data,
process data, interpret outputs, and communicate conclusions.
The task of analysis is made difficult partly because the human mind is limited in terms of
attention, perception, memory, and processing capacity, and partly because the task itself can
be extremely constraining and demanding. Indeed, there may be not enough relevant data
or there may be large volumes of data, the credibility of data sources may vary, the data may
be formatted in different ways (e.g., structured/unstructured, textual/visual/audio), it may be
ambiguous, unreliable, and sometimes intentionally misleading, and there may be time pressure
and high stakes involved. This is further compounded by the lack of feedback which limits
learning on how to perform analytic tasks.

549
550

Ian K. Belton and Mandeep K. Dhami

It is no surprise, therefore, that critics have accused analysts of resorting to using simple
heuristics which can lead to cognitive bias and error (Heuer, 1999). Some would even suggest
that the only way to decide rationally is to use heuristics because no rules following normative
canons of economic rationality may be applied (e.g., Gigerenzer, Todd, & the ABC Research
Group, 1999). Nonetheless, heuristics can lead to systematic cognitive bias, and cognitive bias
has been implicated in well-known intelligence failures such as the failure to find weapons
of mass destruction in Iraq (Jervis, 2006). Although intelligence analysts are experts in their
domain, this is unlikely to protect them from cognitive bias. Indeed, studies demonstrate that
intelligence analysts are as susceptible to cognitive bias as anyone else (e.g., Cook & Smallman,
2008), and may be even more susceptible than non-experts (Reyna, Chick, Corbin, & Hsia,
2014). As we discuss in the next section, cognitive biases can manifest at each stage of the ana-
lytic workflow.

Cognitive biases in intelligence analysis

After a review of the extant literature on intelligence analysis and on cognitive biases, we iden-
tified at least 21 key biases that could affect individual analysis (rather than analysis conducted
in groups/collaboratively). The biases can be assigned to stages of the analytic workflow based
on an assessment of the cognitive tasks involved at each stage and how those might induce bias.
Placing the biases along the workflow allows analysts, managers, and trainers to identify where
issues may arise and can aid in tailoring debiasing support efforts.
For present purposes, we focus on eight biases that are particularly meaningful to one or
more stages of the analytic workflow. These are described briefly below:

1 Belief bias (Evans, Barston, & Pollard, 1983) is the tendency to evaluate the logical strength
of an argument based on the plausibility of its conclusion.
2 Confirmation bias can manifest in a variety of ways (see Klayman, 1995), i.e., remaining
overconfident in an initial position, searching for evidence in a way that supports a favored
viewpoint, interpreting evidence in a way that favors a preferred viewpoint, and resisting
change or insufficiently adjusting confidence in a viewpoint in response to new conflicting
evidence or when existing evidence is discredited.
3 Explanation bias (Ross, Lepper, Strack, & Steinmetz, 1977) refers to the idea that if you
think about/imagine how or why an event may happen, you will then consider it more
likely to happen than if you had not thought about it.
4 Fluency effects (Schwartz et al., 1991) refer to the idea that information which can be
retrieved and/or processed fluently (e.g., because it is familiar) tends to be preferred and
judged more likely and credible than less easily processed information (Reber & Schwarz,
1999; Tversky & Kahneman, 1973; Zajonc, 1968). When evaluating sources, this can lead
to a preference for evidence received from an expert even if that expertise is irrelevant (also
called expertise bias); overestimating the association between independent characteristics
of a favored person or object (halo effect; Nisbett & Wilson, 1977); and evaluating people
who share your characteristics more favorably than others who do not (similarity bias;
Rand & Wexley, 1975). Some have suggested that fluency may sometimes be the most
appropriate way to make decisions, at least where no other knowledge is available (Hertwig,
Herzog, Schooler, & Reimer, 2008).
5 The framing effect has many facets (Tversky & Kahneman, 1981; see also Levin, Schneider, &
Gaeth, 1998), i.e., the tendency for risk-aversion when a choice is framed as a gain (relative
to the status quo), but risk-seeking when a choice is framed as a loss; making an evaluation

550
551

Cognitive biases in intelligence analysis

based on whether something is described as positive or negative; and choosing to engage in

a behavior based on whether participation is described as advantageous or disadvantageous.
6 Order effects (see Hogarth & Einhorn, 1992) refer to the fact that the order in which infor-
mation is presented affects the relative importance attached to it. Information presented first
and last is particularly biasing.
7 The planning fallacy (Kahneman & Tversky, 1979) is the tendency to underestimate the time
(and cost) required to complete a task by overlooking potential difficulties.
8 Overconfidence (Fischhoff, Slovic, & Lichtenstein, 1977) refers to when an individual’s
subjective confidence in the accuracy of his/her judgments is greater than the objective
accuracy of those judgments.

These biases may manifest at various stages of the analytic workflow. At the capture requirements
stage, analysts must understand the context for the intelligence question, including the customer’s
goal and how it will be achieved, and analysts should challenge this if necessary. Analytic per-
formance at this stage can be affected by confirmation bias and framing. For instance, analysts
might set out to look for evidence that confirms a customer’s pre-existing assumptions or
hypotheses rather than challenging these. In addition, the way an intelligence request is framed
may bias the analysts’ understanding of it (e.g., looking to determine how many casualties are
likely to occur as a result of civil unrest rather than considering how many lives could be saved
by one course of action or another).
In the plan analytic response stage, analysts must identify alternative methods that could be
employed to answer the question, evaluate their potential effectiveness and efficiency, and
prioritize how to proceed. Analytic performance may be affected by fluency effects and the
planning fallacy. For instance, analysts may favor specific analytic lines to follow, consider par-
ticular information to be necessary to test hypotheses, and select methods for obtaining data
that have been used recently or frequently in the past and therefore come to mind most easily.
Analysts may also underestimate the time involved in completing the analytic task.
At the obtain data stage, analysts must select relevant data from the most appropriate sources
in an efficient manner, as well as establish new sources of data if necessary. Performance at this
stage may be affected by fluency effects, confirmation bias, and order effects. For instance,
analysts’ choice of data sources and search terms may be limited to those that come to mind
most easily, partly because they have been used regularly or recently. In addition, analysts may
be biased toward obtaining data from trusted experts (even if their expertise is inappropriate for
the context in question), favored or familiar sources as well as sources that are similar to them
(i.e., other intelligence organizations), believing them to be more credible. Analysts may search
for new evidence in a way that favors their existing hypothesis (e.g., by avoiding sources likely
to contradict the hypothesis). Data obtained first and last (e.g., in a list of search results) may be
more likely to be filtered/selected even though these are not more reliable or valid.
At the process data stage, analysts must understand the ‘raw’ data, and this may involve
collating, reformatting, and manipulating it using relevant tradecraft, tools, and technology.
Here, analysts may suffer from fluency effects and confirmation bias. For instance, analysts may
over-use analytic techniques that they are familiar with or use regularly. Data processing tasks
such as identifying unexpected or anomalous results and generating visualizations may be biased
toward favored analytic lines/hypotheses.
At the interpret outputs stage, analysts must evaluate alternative explanations for the (often
incomplete) facts, and construct logical arguments to support conclusions as well as dismiss
alternative ones, determine the degree of uncertainty in these conclusions, and identify any
ambiguities. Here, analysts may be affected by belief bias, confirmation bias, and explanation

551
552

Ian K. Belton and Mandeep K. Dhami

bias. For instance, analysts may over-weight arguments that produce plausible conclusions rather
than properly assess their logical strength. Analysts may be overconfident in an initial belief and
so, even if they take an unbiased approach to new information, they will remain overconfi-
dent in their initial position. Analysts may reach conclusions about a hypothesis based on the
presence of supporting rather than conflicting evidence. Analysts may interpret new evidence
in a way that favors their existing hypothesis (e.g., by regarding supporting evidence as reliable
and conflicting evidence as unreliable, or by interpreting ambiguous information in a way that
supports a favored hypothesis). Analysts may be resistant to change so they insufficiently adjust
their confidence in a hypothesis in response to new conflicting evidence or when existing evi-
dence is discredited. By considering possible explanations for the data, analysts may conclude
that those explanations are more likely than is really the case.
Finally, at the communicate conclusions stage, analysts must present the outcome of analysis in
a clear and meaningful way; distinguishing fact from inference, highlighting the alternatives
that were considered, and expressing uncertainty and confidence. Analytic performance at
this stage may be affected by belief bias and overconfidence. For instance, where analysts have
reached a plausible conclusion, they may overstate the strength of their supporting arguments
or evidence. Analysts may not allow sufficiently for uncertainties and ambiguities in their
reports.

Debiasing strategies
Intelligence organizations have made an attempt to help analysts overcome cognitive bias by
investing in training analysts to use so-called structured analytic techniques (SATs) and in the
development of specialized computer technologies to support and aid analysis. The intelligence
community has largely eschewed psychologically informed and empirically tested cognitive
debiasing interventions (Dhami, Mandel, Mellers, & Tetlock, 2015). Next, after a brief review
of the intelligence community’s preferred debiasing methods, we consider psychologically
informed debiasing strategies relevant to the eight biases described above.
SATs (see Dhami, Belton, & Careless, 2016; Heuer & Pherson, 2014) are a collection of
techniques designed to reduce cognitive bias. The primary rationale behind SATs is that exter-
nalizing and decomposing the cognitive process will result in bias mitigation. Some SATs,
such as the Analysis of Competing Hypotheses (ACH), were developed specifically for use by
intelligence analysts, while other SATs (e.g., the Delphi method) were originally used in other
contexts but have since been applied in the intelligence analysis domain. However, to date, very
few studies have tested whether SATs actually reduce bias. The research that does exist suggests
that they may actually lead to errors such as base-rate neglect (e.g., Dhami, Belton, & Mandel,
2019; Mandel, Karvetski, & Dhami, 2018).
As Dhami et al. (2016) point out, SATs cannot guarantee accuracy. This is partly because
they rely on the judgment skills of the analyst and his/her subjective input of the information
and interpretation of the outputs. In addition, SATs may overcorrect one bias, potentially
triggering an opposing bias, and the decomposition process can make judgments less reliable
rather than more so (Chang, Berdini, Mandel, & Tetlock, 2017). Despite the limitations of
SATs and the lack of evidence attesting to their efficacy, SATs often form part of the core skill
set taught in analytic training programs (e.g., Marrin, 2008) and analysts are expected to apply
them (e.g., UK MoD, 2013; US Government, 2009).
As mentioned, the intelligence community has also adopted different types of computer-
based tools to support analysis (see Dhami, 2017). Indeed, there are currently a vast array of
analytic tools available, and these can be used at different stages of the analytic workflow as

552
553

Cognitive biases in intelligence analysis

they serve a variety of purposes. For instance, at the processing data stage, tools can be used
to visualize data (e.g., Stasko, Gorg, Liu, & Singhal, 2007), perform network analysis and
geospatial analysis, support argumentation (e.g., Kang & Stasko, 2011), and decision making
(e.g., Svenson et al., 2010). In addition, some SATs, such as ACH, have been automated.
Finally, “serious games” are video games that use technology and techniques from the enter-
tainment sector to teach individuals to recognize and reduce cognitive biases. However, as
with SATs, the ability of computer-based tools to reduce cognitive bias has generally not
been empirically tested. A notable exception is serious games, which have been found to
successfully reduce several biases, namely, confirmation bias, the fundamental attribution
error, the bias blind spot, anchoring, the representativeness heuristic, and projection bias
(e.g., Dunbar, Miller, et al., 2014; Morewedge et al., 2015; Mullinix et al., 2013; Shaw et al.,
2016). These games involve interactive learning about biases and activities aimed at reducing
them. The development of serious games has been informed by psychology.
The practical utility of computer-based tools may be limited because, as Dhami (2017)
found in a recent survey of intelligence analysts, the lack of usability can be an important barrier
to analysts’ uptake of analytic tools. When there was a tradeoff between the usability and use-
fulness of a tool, analysts preferred usable tools over useful ones. In addition, like SATs, many
computer-based tools rely on potentially biased subjective human inputs and interpretations
of output (Büyükkurt & Büyükkurt, 1991; Montibeller & von Winterfeldt, 2015), which
limits their ability to help analysts avoid bias and error. Despite this, the intelligence com-
munity considers analytic technology as potentially helpful, as illustrated by projects funded
by the US Intelligence Advanced Research Projects Activity: (www.iarpa.gov/index.php/
research-programs).

Psychologically informed interventions

Psychologists have investigated many different interventions for reducing cognitive bias. These
are usually based on theories about the sources of bias, and empirically tested to examine their
effectiveness in countering the bias. Table 37.1 provides a summary of the empirically tested
debiasing strategies for the eight cognitive biases identified described earlier.
Psychologically informed and empirically tested debiasing interventions typically involve two
elements. First, participants are given training or instructions that aim to increase understanding
and awareness of cognitive biases. The goal here is to improve an individual’s ability to identify
tasks or situations where an intuitive response is likely to be biased, so that they can override
their intuition with an appropriate deliberative strategy. Second, interventions aim to fill
mindware gaps by teaching relevant rules (e.g., probability or logic), or specific strategies to
use in a given task (e.g., consider the opposite of your original answer or unpack a task into
component parts).
An alternative approach to debiasing involves restructuring the task environment to reduce
biased behavior, either by encouraging more deliberative thinking or by inducing unbiased
intuition. This kind of so-called “choice architecture” that “nudges” individuals towards better
decisions (Thaler & Sunstein, 2008) has proved itself to be effective for promoting pro-social
behaviors such as paying tax or becoming an organ donor and has become an increasingly
popular policy-making tool (e.g., Hallsworth, Egan, Rutter & McCrae, 2018). In the public
policy context, questions remain about whether some nudges may be unacceptably coercive
(e.g., Viale, 2018, 2019). However, it is likely that there are many contexts within the intel-
ligence community where choice architecture could usefully be employed to reduce biased
thinking.

553
554

Ian K. Belton and Mandeep K. Dhami

Table 37.1 Psychologically informed and empirically tested debiasing interventions

Cognitive bias Psychological intervention

Belief bias Evans, Newstead, Allen, and Pollard (1994) found that brief “instructional training in
logical principles” reduced belief bias in basic logical reasoning tasks.
Confirmation Several variations of the “consider-the-opposite” strategy have been found to reduce
bias confirmation bias. For example, instructing individuals to imagine their response
if given evidence pointed in the opposite direction reduced the tendency to
discount conflicting evidence, and presenting conflicting evidence in advance of a
search for information reduced the bias of that search towards supporting evidence
(Lord, Lepper, & Preston, 1984). Similarly, instructions to consider multiple
alternative hypotheses reduced subsequent belief perseverance (Lewandowsky
et al., 2012). If too many alternatives are generated, this may undermine the
mitigating effect of “consider-the-opposite” interventions because the difficulty
of generating alternatives can undermine their perceived plausibility (Sanna,
Schwartz, & Stocker, 2002). Fortunately, this “backfire” effect can be reduced by
raising awareness of the difficulty of generating alternatives (Sanna & Schwartz,
2003). Actively searching for disconfirming evidence to falsify an initial hypothesis
may also reduce confirmation bias (Lam, 2007).
Explanation Considering multiple alternatives has been shown to reduce the explanation bias, as
bias long as not too many alternatives are generated (Hirt & Markman, 1995; Sanna
et al., 2002).
Fluency effects “Consider-the-opposite” strategies of the kind successfully tested on confirmation
bias and order effects may be effective here, since they aim to induce consideration
of alternatives beyond those that spring to mind immediately. A simple, practical
solution for biases relating to source evaluation is to get a second opinion from
a colleague who is not familiar with the source and/or aggregate evidence from
multiple sources wherever possible.
Framing effect A range of strategies have been found to reduce the framing effect. These
include: considering the opposite (Cheng, Wu & Lin, 2014); reframing the
problem in an opposite way (Korobkin & Guthrie, 1998); listing advantages and
disadvantages of two options before choosing between them (Almashat, Ayotte,
Edelstein, & Margrett, 2008); giving reasons for the chosen option (Leboeuf &
Shafir, 2003); presenting information in a frame-neutral way, for example, through
graphical representations (Garcia-Retamero & Dhami, 2013); “causal cognitive
mapping” or a node-link diagram of the variables used to make a decision and
their cause-effect relationships (Hodgkinson, Bown, Maule, Glaister, & Pearman,
1999); scenario planning (Meissner & Wulf, 2013); and a strong warning message
about the risk of the framing effect (Cheng & Wu, 2010)
Order effects Mumma and Wilson (1995) found that considering the opposite of the first
information item given and also sorting information based on diagnosticity removed
the primacy effect, as did simply writing a list of the information items after viewing
them on-screen. Ashton and Kennedy (2002) found that requiring individuals
to rank information items by importance and then review them before making a
judgment successfully removed the recency effect. Wickens, Ketels, Healy, Buck-
Gengler, & Bourne (2010) found that instructing participants that newer data is
likely to be more reliable increased the recency effect (when appropriate) but did
not reduce the primacy effect. Logan and Fischman (2011) found that performing
a simple manual task (i.e., rearranging a pair of water glasses) between viewing and
recalling a list of words removed the recency effect, but not the primacy effect.

554
555

Cognitive biases in intelligence analysis

Table 37.1 Cont.

Cognitive bias Psychological intervention

Overconfidence Koriat, Lichtenstein, and Fischhoff (1980) found that overconfidence could be
reduced using a form of the “consider-the-opposite” strategy, which involved
listing counter-arguments to a preferred hypothesis or prediction. Multi-step
strategies for eliciting forecasts in the form of confidence intervals have been
found to reduce overconfidence (e.g., Speirs-Bridge et al., 2010). Simply
instructing people on the risk of overconfidence when making judgments has also
been found to have a debiasing effect (McGraw, Mellers, & Ritov, 2004).
Planning fallacy Breaking down or “unpacking” the task into component parts and estimating the
time needed for each can be an effective debiasing strategy, especially when the
unpacking task is made relatively difficult, e.g., breaking the task into five rather
than two steps (Kruger & Evans, 2004) and if individuals focus on obstacles to
task completion (Peetz, Buehler, & Wilson, 2010). Thinking of multiple scenarios
in which the task in question or similar tasks could take much more or less time
has been found to reduce the planning fallacy in subsequent task time estimates
(Newby-Clark, Ross, Buehler, Koehler, & Griffin, 2000). Taking an “outside
view” of the task may also reduce the planning fallacy (Kahneman & Lovallo,
1993). Estimating the time another person would take to complete a task, based
on that person’s own estimate, improves estimation accuracy (Buehler, Griffin, &
Ross, 1994). Estimates are somewhat more accurate when based on memories of
the time taken to complete comparable tasks in the past (Buehler et al., 1994) and
are likely to be even more accurate when benchmarked against the actual time
taken to complete comparable tasks previously (Roy, Christenfeld, & McKenzie,
2005). Visualizing a third party completing the task can also mitigate bias
(Buehler, Griffin, Lam, & Deslauriers, 2012). Backward planning, which involves
working step-by-step back from the desired end goal, can also reduce the planning
fallacy (Wiese, Buehler, & Griffin, 2016).

Conclusion
Intelligence analysts have been criticized for suffering from cognitive biases (Heuer, 1999),
although there have been few empirical demonstrations of this. In this chapter, we identified
eight biases applicable to the analytic workflow and we reviewed psychologically informed
debiasing strategies that aim to counter those biases. The intelligence community’s efforts to
help their analysts become more rational thinkers may be inadequate because organizations
rely on ad hoc SATs and a proliferation of computer-based tools that largely lack empirical
evidence attesting to their efficacy. In addition, there seems to be little acknowledgment of the
fact that SATs and computer-based tools may simply replace existing biases with new ones. In
this sense, one could argue that intelligence organizations have failed to fulfill their function, as
identified by Simon (1947), to provide analysts with an environment that encourages rational
thinking.
The intelligence community has, to date, largely eschewed psychologically informed and
empirically tested interventions. In essence, many of these interventions require the individual
to stop and think before pursuing a course of action. As Simon (1947, p. 89) suggested, “If
rationality is to be achieved, a period of hesitation must precede choice.” Although further
research is needed to confirm how well the effects of such interventions can be transferred to
the intelligence analysis domain, this should not preclude their use in the meantime. Below, we

555
556

Ian K. Belton and Mandeep K. Dhami

conclude with a consideration of the potential challenges associated with debiasing people and
the limitations of debiasing strategies, before highlighting the main barrier to debiasing in the
intelligence analysis domain.
There are many challenges involved in successfully implementing any debiasing strategy. The
first is the bias blind spot which is a meta-cognitive bias that refers to the tendency to believe
that we are free of cognitive bias even though we recognize bias exists in others (Pronin, Lin, &
Ross, 2002; West, Meserve, & Stanovich, 2012). This can make analysts resistant to training or
instruction designed to mitigate cognitive biases since they will tend to think it does not apply
to them. Second, individuals need to be convinced that debiasing is relevant to their work, and
in order to reduce threats to self-esteem, they may need to be reassured that debiasing is about
further enhancing their thinking rather than correcting cognitive faults. Indeed, it is often
observed that analysts can be resistant to using debiasing techniques such as SATs or computer
support tools (e.g., Trent, Voshell, & Patterson, 2007; Treverton & Gabbard, 2008). Third,
the durability of the effects of most debiasing strategies is currently unknown (for exceptions,
see the recent work on serious games which used follow-up periods between 8 and 12 weeks,
e.g., Dunbar, Miller, et al., 2014; Morewedge et al., 2015; Mullinix et al., 2013; Shaw et al.,
2016). Fourth, it is likely that individual differences such as personality, cognitive ability/style,
and culture may affect how well people respond to debiasing strategies (Oreg & Bayazit, 2009;
Stanovich & West, 2000), particularly as research suggests that individuals differ widely in their
susceptibility to cognitive bias (e.g., Peters, 2012; Toplak, West, & Stanovich, 2011). Finally,
some debiasing interventions may backfire, increasing rather than reducing the target bias (e.g.,
Sanna et al., 2002) or may even stimulate some other bias.
In addition, features of the intelligence analysis domain may make it difficult for intelligence
organizations to build and maintain a psychological environment that is conducive to rational
thinking. In particular, the intelligence community has traditionally held to the belief that intel-
ligence analysis is a “tradecraft”––a skill that must be learnt “on-the-job” over a long period and
which generates implicit knowledge handed down from one analyst to another. This hinders
efforts to improve analysis through top-down approaches.
Despite the many challenges and barriers, there is scope for optimism. Simon (1947) observed
that effective organizations enable individuals to think rationally. We suggest that by adopting
an appropriately skeptical attitude to untested approaches such as SATs and some computer
technologies, and by embracing a psychologically informed and empirically tested approach to
reducing cognitive bias, intelligence organizations can assist their analysts to improve analytic
performance.

Acknowledgments
The work presented in this chapter was supported by funding provided to Mandeep K. Dhami
by HM Government. We would like to thank Kathryn Careless for providing us with feedback
on the work.

References
Almashat, S., Ayotte, B., Edelstein, B., & Margrett, J. (2008). Framing effect debiasing in medical decision
making. Patient Education and Counselling, 71(1), 102–107. doi:10.1016/j.pec.2007.11.004.
Arkes, H. (1991). Costs and benefits of judgment errors: Implications for debiasing. Psychological Bulletin,
110(3), 486–498. doi:10.1037/0033-2909.110.3.486.
Ashton, R. H., & Kennedy, J. (2002). Eliminating recency with self-review: The case of auditors’ ’going
concern’ judgments. Journal of Behavioral Decision Making, 15(3), 221–231. doi:10.1002/bdm.412.

556
557

Cognitive biases in intelligence analysis

Buehler, R., Griffin, D., Lam, K. C. H., & Deslauriers, J. (2012). Perspectives on prediction: Does
third-person imagery improve task completion estimates? Organizational Behavior and Human Decision
Processes, 117(1), 138–149. doi:10.1016/j.obhdp.2011.09.001.
Buehler, R., Griffin, D., & Ross, M. (1994). Exploring the ‘planning fallacy’: Why people underestimate
their task completion times. Journal of Personality and Social Psychology, 67(3), 366–381. doi:10.1037/
0022-3514.67.3.366.
Büyükkurt, B. K., & Büyükkurt, M. D. (1991). An experimental study of the effectiveness of three
debiasing techniques. Decision Sciences, 22(1), 60–73. doi:10.1111/j.1540-5915.1991.tb01262.x.
Chang, W., Berdini, E., Mandel, D. R., & Tetlock, P. E. (2017). Restructuring structured analytic techniques
in intelligence. Intelligence and National Security, 33(3), 337–356. doi:10.1080/02684527.2017.1400230.
Cheng, F., & Wu, C. (2010). Debiasing the framing effect: The effect of warning and involvement.
Decision Support Systems, 49(3), 328–334. doi:10.1016/j.dss.2010.04.002.
Cheng, F., Wu, C., & Lin, H. (2014). Reducing the influence of framing on internet consumers’ decisions: The
role of elaboration. Computers in Human Behavior, 37, 56–63. doi:10.1016/j.chb.2014.04.015.
Cook, M. B., & Smallman, H. S. (2008). Human factors of the confirmation bias in intelligence ana-
lysis: Decision support from graphical evidence landscapes. Human Factors: The Journal of the Human
Factors and Ergonomics Society, 50(5), 745–754. doi:10.1518/001872008X354183.
Dhami, M. K. (2017). A survey of intelligence analysts’ perceptions of analytic tools. Paper presented at
2017 European Intelligence and Security Informatics Conference. doi:10.1109/EISIC.2017.26.
Dhami, M. K., Belton, I. K., & Careless, K. E. (2016). Critical review of analytic techniques. Paper presented
at 2016 European Intelligence and Security Informatics Conference. doi:10.1109/EISIC.2016.039.
Dhami, M. K., Belton, I. K., & Mandel D. R. (2019). The “analysis of competing hypotheses” in intelli-
gence analysis. Journal of Applied Cognitive Psychology, 33(6), 1080–1090. doi: 10.1002/acp.3550.
Dhami, M. K., & Careless, K. E. (2015). Ordinal structure of the generic analytic workflow: A survey
of intelligence analysts. Paper presented at 2015 European Intelligence and Security Informatics
Conference. doi:10.1109/EISIC.2015.37.
Dhami, M. K., Mandel, D. R., Mellers, B. A., & Tetlock, P. E. (2015). Improving intelligence analysis with
decision science. Perspectives on Psychological Science, 10(6), 753–757. doi:10.1177/1745691615598511.
Dunbar, N. E., Miller, C. H., Adame, B. J., Elizondo, J., Wilson, S. N., Lane, B. L., … Zhang, J. (2014).
Implicit and explicit training in the mitigation of cognitive bias through the use of a serious game.
Computers in Human Behavior, 37, 307–318. doi:10.1016/j.chb.2014.04.053.
Evans, J. St. B. T., Barston, J. L., & Pollard, P. (1983). On the conflict between logic and belief in syllo-
gistic reasoning. Memory & Cognition, 11(3), 295–306. doi:10.3758/BF03196976.
Evans, J. St. B. T., Newstead, S. E., Allen, J. L., & Pollard, P. (1994). Debiasing by instruction: The case
of belief bias. Cognition, 6(3), 263–285. doi:10.1080/09541449408520148.
Fischhoff, B. (1982). Debiasing. In D. Kahneman, P. Slovic, & A. Tversky (Eds.), Judgement under uncer-
tainty: Heuristics and biases (pp. 422–444). New York: Cambridge University Press.
Fischhoff, B., Slovic, P., & Lichtenstein, S. (1977). Knowing with certainty: The appropriateness of
extreme confidence. Journal of Experimental Psychology: Human Perception and Performance, 3(4), 552–
564. doi:10.1037/0096-1523.3.4.552.
Garcia-Retamero, R., & Dhami, M. K. (2013). On avoiding framing effects in experienced decision makers.
Quarterly Journal of Experimental Psychology, 66(4), 829–842. doi:10.1080/17470218.2012.727836.
Gigerenzer, G., Todd, P. M., & the ABC Research Group (Eds.) (1999). Simple heuristics that make us smart.
New York: Oxford University Press.
Gilovich, T., Griffin, D., & Kahneman, D. (Eds.) (2002). Heuristics and biases: The psychology of intuitive
judgment. Cambridge: Cambridge University Press.
Hallsworth, M., Egan, M., Rutter, J., & McCrae, J. (2018). Behavioural government: Using behavioural
science to improve how governments make decisions. Retrieved from the Behavioural Insights Team
at: www.bi.team/
Hammond, K. R. (1996). Human judgment and social policy: Irreducible uncertainty, inevitable error, unavoidable
injustice. New York, NY: Oxford University Press.
Hammond, K. R. (2000). Judgments under stress. New York, NY: Oxford University Press.
Hertwig, R., Herzog, S. M., Schooler, L. J., & Reimer, T. (2008). Fluency heuristic: How the mind
exploits a by-product of information retrieval. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 24(5), 1191–1206. doi: 10.1037/a0013025.
Heuer, R. J., Jr. (1999). The psychology of intelligence analysis. Washington, DC: CQ Press.

557
558

Ian K. Belton and Mandeep K. Dhami

Heuer, R. J., Jr., & Pherson, R. H. (2014). Structured analytic techniques for intelligence analysis. Washington,
DC: CQ Press.
Hirt, E. R., & Markman, K. D. (1995). Multiple explanation: A consider-an-alternative strategy for
debiasing judgments. Journal of Personality and Social Psychology, 69(6), 1069–1086. doi:10.1037/
0022-3514.69.6.1069.
Hodgkinson, G. P., Bown, N. J., Maule, A. J., Glaister, K. W., & Pearman, A. D. (1999). Breaking the
frame: An analysis of strategic cognition and decision making under uncertainty. Strategic Management
Journal, 20(10), 977–985.
Hogarth, R. M., & Einhorn, H. J. (1992). Order effect in belief updating: The belief-adjustment model.
Cognitive Psychology, 24(1), 1–55. doi:10.1016/0010-0285(92)90002-J.
Jervis, R. J. (2006). Reports, politics, and intelligence failures: The case of Iraq. Journal of Strategic Studies,
29(1), 3–52. doi:10.1080/01402390600566282.
Kahneman, D. (2011). Thinking, fast and slow. London: Penguin Books.
Kahneman, D., & Lovallo, D. (1993). Timid choices and bold forecasts: A cognitive perspective on risk
taking. Management Science, 39(1), 17–31. doi:10.1287/mnsc.39.1.17.
Kahneman, D., Slovic, P., & Tversky, A. (Eds.) (1982). Judgement under uncertainty: Heuristics and biases.
Cambridge: Cambridge University Press.
Kahneman, D., & Tversky, A. (1979). Intuitive prediction: biases and corrective procedures. In
S. Makridakis & S. C. Wheelright (Eds.), Forecasting: TIMS studies in the management sciences, vol. 12
(pp. 313–327). Amsterdam: North-Holland.
Kang, Y., & Stasko, J. (2011). Characterizing the intelligence analysis process: Informing visual
analytics design through a longitudinal field study. In G. Santucci & M. Ward (Eds.), Proceedings of
IEEE Conference on Visual Analytics Science and Technology (VAST) 2011 (pp. 21–30). doi:10.1109/
VAST.2011.6102438.
Katsikopolous, K. V., & Lan, C. (2011). Herbert Simon’s spell on judgment and decision making. Judgment
and Decision Making, 6(8), 722–732.
Klayman, J. (1995). Varieties of confirmation bias. In J. Busemeyer, R. Hastie, & D. L. Medin (Eds.),
Decision making from a cognitive perspective (pp. 385–418). New York: Academic Press.
Koriat, A., Lichtenstein, S., & Fischhoff, B. (1980). Reasons for confidence. Journal of Experimental
Psychology: Human Learning and Memory, 6(2), 107–118. doi:10.1037/0278-7393.6.2.107.
Korobkin, R., & Guthrie, G. (1998). Psychology, economics and settlement: A new look at the role of the
lawyer. Texas Law Review, 76, 77–141.
Kruger, J., & Evans, M. (2004). If you don´t want to be late, enumerate: Unpacking reduces the planning
fallacy. Journal of Experimental Social Psychology, 40(5), 586–598. doi:10.1016/j.jesp.2003.11.001.
Lam, C. (2007). Is Popper’s falsificationist heuristic a helpful resource for developing critical thinking?
Educational Philosophy and Theory, 39(4), 432–448. doi:10.1111/j.1469-5812.2007.00349.x.
Larrick, R. P. (2004). Debiasing. In D. J. Koehler & N. Harvey (Eds.), Blackwell handbook of judgment and
decision making (pp. 316–337). Oxford: Blackwell.
Leboeuf, R. A., & Shafir, E. (2003). Deep thoughts and shallow frames: On the susceptibility to framing
effects. Journal of Behavioral Decision Making, 16(2), 77–92. doi:10.1002/bdm.433.
Levin, I. P., Schneider, S. L., & Gaeth, G. J. (1998). All frames are not created equal: A typology and crit-
ical analysis of framing effects. Organizational Behavior and Human Decision Processes, 76(2), 149–188.
doi:10.1006/obhd.1998.2804.
Lewandowsky, S., Ecker, U. K. H., Seifert, C. M., Schwarz, N., & Cook, J. (2012). Misinformation and
its correction: Continued influence and successful debiasing. Psychological Science in the Public Interest,
13(3), 106–131. doi:10.1177/1529100612451018.
Lilienfeld, S. O., Ammirati, R., & Landfield, K. (2009). Giving debiasing away: Can psychological research
on correcting cognitive errors promote human welfare? Perspectives in Psychological Science, 4(4), 390–
398. doi:10.1111/j.1745-6924.2009.01144.x.
Logan, S. W., & Fischman, M. G. (2011). The relationship between end-state comfort effects and
memory performance in serial and free recall. Acta Psychologica, 137(3), 292–299. doi:10.1016/
j.actpsy.2011.03.009.
Lord, C. G., Lepper, M. R., & Preston, E. (1984). Considering the opposite: A corrective strategy
for social judgment. Journal of Personality and Social Psychology, 47(6), 1231–1243. doi:10.1037/
0022-3514.47.6.1231.
Mandel, D. R., Karvetski, C. W., & Dhami, M. K. (2018). Boosting intelligence analysts’ judgment
accuracy: What works, what fails? Judgment and Decision Making, 13(6), 607–621.

558
559

Cognitive biases in intelligence analysis

Marrin, S. (2008). Training and educating U.S. intelligence analysts. International Journal of Intelligence and
Counterintelligence, 22(1), 131–146. doi:10.1080/08850600802486986.
McGraw, A. P., Mellers, B. A., & Ritov, I. (2004). The affective costs of overconfidence. Journal of
Behavioral Decision Making, 17, 281–295. doi:10.1002/bdm.472.
Meissner, P., & Wulf, T. (2013). Cognitive benefits of scenario planning: Its impact on biases and deci-
sion quality. Technological Forecasting & Social Change, 80, 801–814. doi:10.1016/j.techfore.2012.09.011.
Montibeller, G., & von Winterfeldt, D. (2015). Cognitive and motivational biases in decision and risk
analysis. Risk Analysis, 35(7), 1230–1251. doi:10.1111/risa.12360.
Morewedge, C. K., Yoon, H., Scopelliti, I., Symborski, C. W., Korris, J. H., & Kassam, K. S. (2015).
Debiasing decisions: Improved decision making with a single training intervention. Policy Insights from
the Behavioral and Brain Sciences, 2(1), 129–140. doi:10.1177/2372732215600886.
Mullinix, G. Gray, O., Colado, J., Veinott, E., Leonard, J., Papautsky, E. L, … Flach, J. (2013).
Heuristica: Designing a serious game for improving decision making. Paper presented at IEEE Games
Innovation Conference, 23–25 September, Vancouver, BC. doi:10.1109/IGIC.2013.6659159.
Mumma, G. H., & Wilson, S. B. (1995). Procedural debiasing of primacy/anchoring effects in clinical-like
judgments. Journal of Clinical Psychology, 51(6), 841–853.
Newby-Clark, I. R., Ross, M., Buehler, R., Koehler, D. J., & Griffin, D. (2000). People focus on opti-
mistic scenarios and disregard pessimistic scenarios while predicting task completion times. Journal of
Experimental Psychology: Applied, 6(3), 171–182. doi:10.1037//1076-898X.6.3.171.
Nisbett, R. E., & Wilson, T. D. (1977). The halo effect: Evidence for the unconscious alteration of
judgments. Journal of Personality and Social Psychology, 35, 250–256. doi:10.1037/0022-3514.35.4.250.
Oreg, S., & Bayazit, M. (2009). Prone to bias: Development of a bias taxonomy from an individual
differences perspective. Review of General Psychology, 13(3), 175–193. doi:10.1037/a0015656.
Peetz, J., Buehler, R., & Wilson, A. E. (2010). Planning for the near and distant future: How does tem-
poral distance affect task completion predictions? Journal of Experimental Social Psychology, 46(5), 709–
720. doi:10.1016/j.jesp.2010.03.008.
Peters, E. (2012). Beyond comprehension: The role of numeracy in judgments and decisions. Current
Directions in Psychological Science, 21(1), 31–35. doi:10.1177/0963721411429960.
Pronin, E., Lin, D. Y., & Ross, L. (2002). The bias blind spot: Perceptions of bias in self versus others.
Personality and Social Psychology Bulletin, 28(3), 369–381. doi:10.1177/0146167202286008.
Rand, T. M., & Wexley, K. N. (1975). Demonstration of the effect, ‘similar to me’, in simulated employ-
ment interviews. Psychological Reports, 36(2), 535–544. doi:10.2466/pr0.1975.36.2.535.
Reber, R., & Schwarz, N. (1999). Effects of perceptual fluency on judgments of truth. Consciousness and
Cognition, 8(3), 338–342. doi:10.1006/ccog.1999.0386.
Reyna, V. F., Chick, C. F., Corbin, J. C., & Hsia, A. N. (2014). Developmental reversals in risky deci-
sion making: Intelligence agents show larger decision biases than college students. Psychological Science,
25(1), 76–84. doi:10.1177/0956797613497022.
Ross, L. D., Lepper, M. R., Strack, F., & Steinmetz, J. (1977). Social explanation and social expect-
ation: Effects of real and hypothetical explanations on subjective likelihood. Journal of Personality and
Social Psychology, 35(11), 817–829. doi:10.1037/0022-3514.35.11.817.
Roy, M. M., Christenfeld, N. J. S., & McKenzie, C. R. M. (2005). Underestimating the duration of future
events: Memory incorrectly used or memory bias? Psychological Bulletin, 131(5), 738–756. doi:10.1037/
0033-2909.131.5.738.
Sanna, L. J., & Schwarz, N. (2003). Debiasing the hindsight bias: The role of accessibility experiences
and (mis)attributions. Journal of Experimental Social Psychology, 39, 287–295. doi:10.1016/
S0022-1031(02)00528-0.
Sanna, L. J., Schwarz, N., & Stocker, S. L. (2002). When debiasing backfires: Accessible content and
accessibility experiences in debiasing hindsight. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 28(3), 497–502. doi:10.1037/0278-7393.28.3.497.
Schwartz, N., Bless, H., Strack, F., Klumpp, G., Rittenauer-Schatka, H., & Simons, A. (1991). Ease
of retrieval as information: Another look at the availability heuristic. Journal of Personality and Social
Psychology, 61(2), 195–202. doi:10.1037/0022-3514.61.2.195.
Shaw, A., Kenski, K., Stromer-Galley, J., Martey, R. M., Clegg, B. A., ... Strzalkowski, T. (2016). Serious
efforts at bias reduction: The effects of digital games and avatar customization on three cognitive biases.
Journal of Media Psychology, 30(1), 16–28. doi:10.1027/1864-1105/a000174.
Simon, H. A. (1947). Administrative behavior. New York: Macmillan.

559
560

Ian K. Belton and Mandeep K. Dhami

Simon, H. A. (1956). Rational choice and the structure of the environment. Psychological Review, 63(2),
129–138. doi:10.1037/h0042769.
Simon, H. A. (1957). Models of man: Social and rational. Oxford: Wiley.
Simon, H. A. (1990). Invariants of human behavior. Annual Review of Psychology, 41, 1–20. doi:10.1146/
annurev.ps.41.020190.000245.
Sloman, S. (1996). “The empirical case for two systems of reasoning”. Psychological Bulletin, 119, 3–22.
doi: 10.1037/0033-2909.119.1.3.
Speirs-Bridge, A., Fidler, F., McBride, M., Flander, L., Cumming, G., & Burgman, M. (2010). Reducing
overconfidence in the interval judgments of experts. Risk Analysis, 30(3), 512–523. doi:10.1111/
j.1539-6924.2009.01337.x.
Stanovich, K. E., Toplak, M. E., & West, R. F. (2008). The development of rational thought: A tax-
onomy of heuristics and biases. Advances in Child Development and Behavior, 36, 251–285. doi:10.1016/
S0065-2407(08)00006-2.
Stanovich, K. E., & West, R. F. (2000). Individual differences in reasoning: Implications for the rationality
debate? Behavioral and Brain Sciences, 23(5), 645–726. doi:10.1017/S0140525X00003435.
Stasko, J., Görg, C., Liu, Z., & Singhal, K. (2007). Jigsaw: Supporting investigative analysis through inter-
active visualization. In W. Ribarsky & J. Dill (Eds.), Proceedings of IEEE Symposium on Visual Analytics
Science and Technology (VAST) 2007 (pp. 131–138). doi:10.1109/VAST.2007.4389006.
Svenson, P., Forsgren, R., Kylesten, B., Berggren, P., Fah, W. R. ... Hann, J. K. Y. (2010). Swedish-
Singapore studies of Bayesian modelling techniques for tactical intelligence analysis. Paper presented
at 13th Conference on Information Fusion (FUSION) 2010. Available at: http://ieeexplore.ieee.org/
xpl/mostRecentIssue.jsp?punumber=5706806
Thaler, R., & Sunstein, C. R. (2008). Nudge: Improving decisions about health, wealth, and happiness. New
Haven, CT: Yale University Press.
Toplak, M. E., West, R. F., & Stanovich, K. E. (2011). The cognitive reflection test as a predictor of
performance on heuristics-and-biases tasks. Memory and Cognition, 39(7), 1275–1289. doi:10.3758/
s13421-011-0104-1.
Trent, S., Voshell, M., & Patterson, E. (2007). Team cognition in intelligence analysis. Proceedings of the
Human Factors and Ergonomics Society Annual Meeting, 51(4), 308–312.
Treverton, G. F., & Gabbard, C. B. (2008). Assessing the tradecraft of intelligence analysis. Technical
report. Retrieved from the RAND Corporation at: www.rand.org/content/dam/rand/pubs/tech-
nical_reports/2008/RAND_TR293.pdf
Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability.
Cognitive Psychology, 5(1), 207–232. doi:10.1016/0010-0285(73)90033-9.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science,
185(4157), 1124–1131. doi:10.1126/science.185.4157.1124.
Tversky, A., & Kahneman, D. (1981). The framing of decisions and the rationality of choice, Science,
211(4481), 453–458. doi:10.1126/science.7455683.
UK MoD (2013). Quick wins for busy analysts. London: Defence Intelligence, UK.
US Government (2009). A tradecraft primer: Structured analytic techniques for improving intelligence
analysis. Retrieved from the CIA at: www.cia.gov/library/center-for-the-study-of-intelligence/csi-
publications/books-and-monographs/Tradecraft%20Primer-apr09.pdf
Viale, R. (2018). The normative and descriptive weakness of behavioral economics-informed
nudge: Depowered paternalism and unjustified libertarianism. Mind & Society, 17(1–2), 53–69.
Viale, R. (2019). Architecture of the mind and libertarian paternalism: Is the reversibility of System 1
nudges likely to happen? Thinking & Reasoning, (submitted).
West, R. F., Meserve, R. J., & Stanovich, K. E. (2012). Cognitive sophistication does not attenuate the
bias blind spot. Journal of Personality and Social Psychology, 103(3), 506–519. doi:10.1037/a0028857.
Wickens, C., Ketels, S. L. Healy, A. F., Buck-Gengler, C. J., & Bourne, L. E., Jr. (2010). The anchoring
heuristic in intelligence integration: A bias in need of de-biasing. Proceedings of the Human Factors and
Ergonomics Society Annual Meeting, 54, 2324–2328. doi:10.1177/154193121005402722.
Wiese, J., Buehler, R., & Griffin, D. (2016). Backward planning: Effects of planning direction on
predictions of task completion time. Judgment and Decision Making, 11(2), 147–167.
Zajonc, R. B. (1968). Attitudinal effects of mere exposure. Journal of Personality and Social Psychology, 9(2),
1–27. doi:10.1037/h0025848.

560
561

PART VII

Behavioral public policies

Nudging and boosting
562
563

38
“BETTER OFF, AS JUDGED
BY THEMSELVES”
Bounded rationality and nudging

Cass R. Sunstein

Some nudges are designed to reduce externalities; consider fuel economy labels that draw
attention to environmental consequences, or default rules that automatically enroll people in
green energy (Reisch and Sunstein 2014). But many nudges are designed to increase the likeli-
hood that people’s choices will improve their own welfare. Richard Thaler and I argue that the
central goal of such nudges is to “make choosers better off, as judged by themselves” (Thaler and
Sunstein 2008, p. 5; italics in original; Thaler 2015). Social planners – or in our terminology,
choice architects – might well have their own ideas about what would make choosers better off,
but in our view, the lodestar is people’s own judgments. To be a bit more specific: The lode-
star is welfare, and people’s own judgments are a good (if imperfect) way to test the question
whether nudges are increasing their welfare.
The last sentence raises many questions, and it is certainly reasonable to wonder about
potential ambiguities in the “as judged by themselves” (hereinafter AJBT) criterion. Should
the focus be on choosers’ judgments before the nudge, or instead after? What if the nudge
alters people’s preferences, so that they like the outcome produced by the nudge, when they
would not have sought that outcome in advance? What if preferences are constructed by
the relevant choice architecture (Lichtenstein and Slovic 2006)? Or what if people’s ex ante
judgments are wrong, in the sense that a nudge would improve their welfare, even though
they do not think that it will (Dolan 2014)? Do we want to ask about choosers’ actual,
potentially uninformed or behaviorally biased judgments, or are we entitled to ask what
choosers would think if they had all relevant information and were unaffected by relevant
biases (Goldin 2015)?
My goal here is to explore the meaning of the AJBT criterion and to sort out some of
the ambiguities. As we shall see, three categories of cases should be distinguished: (1) those
in which choosers have clear antecedent preferences, and nudges help them to satisfy those
preferences; (2) those in which choosers face a self-control problem, and nudges help them to
overcome that problem; and (3) those in which choosers would be content with the outcomes
produced by two or more nudges, or in which ex post preferences are endogenous to or
constructed by nudges, so that the AJBT criterion leaves choice architects with several options,
without specifying which one to choose. Cases that fall in category (1) plainly satisfy the AJBT
criterion, and there are many such cases. From the standpoint of the AJBT criterion, cases that
fall in category (2) are also unobjectionable, indeed they can be seen as a subset of category (1);

563
564

Cass R. Sunstein

and they are plentiful. Cases that fall in category (3) create special challenges, which may lead
us to make direct inquiries into people’s welfare or to explore what informed, active choosers
typically select.
An important question, pressed by Robert Sugden (Sugden 2016), is simple: “Do people
really want to be nudged towards healthy lifestyles?” That is an empirical question, and we
have a great deal of evidence about it. The answer is “yes,” at least in the sense that in numerous
nations – including the United States, the United Kingdom, Germany, France, Italy, and
Australia – strong majorities endorse nudges toward healthy lifestyles (Jung and Mellers 2016;
Reisch and Sunstein 2016; Sunstein 2016a, 2016b; Sunstein, Reisch, and Rauber 2017).
One might object that general attitudes toward nudges do not specifically answer Sudgen’s
question, and that it is best not to ask people generally (1) whether they approve of nudges,
but instead to ask more specifically (2) whether they themselves would like to be nudged (on
potential differences between answers to the two questions, see Cornwell and Krantz 2014).
But general approval of health-related nudges strongly suggests that the answer to (2) is prob-
ably “yes” as well, and in any case, the existing evidence suggests that the answer to (2), asked
specifically, is also “yes” (Jung and Mellers 2016). To be sure, more remains to be learned on
these issues.
These findings cannot speak to conceptual and normative questions. What does it even
mean to say that people want to make the choices toward which they are being nudged? What
is the relationship between people’s preferences and imaginable nudges? In light of behavioral
findings, how confidently can we speak of people’s “preferences”? If people want to be nudged,
why are they not already doing what they would be nudged to do? Those are important
questions (and they do have empirical features).
Thaler and I are interested in “libertarian paternalism,” in the form of approaches that pre-
serve freedom of choice, but that steer people in a direction that will promote their welfare.
Coercive paternalism is very different (Conly 2014). Reminders, warnings, information dis-
closure, invocations of social norms, and default rules are examples of what we have in mind
(Sunstein 2016a, 2016b). Contrary to a common misconception, the AJBT criterion is not
at all designed to counter the charge of paternalism. On the contrary, Thaler and I explicitly
embrace (a mild form of) paternalism, and the purpose of the criterion is to discipline the content
of paternalistic interventions. Consider, for example, a GPS device. It will identify the best route
for you, given the direction that you identify; it makes you better off by your own lights. It
increases navigability. At the same time, it is paternalistic in the sense that it purports to know,
better than you do, how to get where you want to go. Its paternalism is one of means, not ends
(Sunstein 2014). Means paternalism is unquestionably a form of paternalism, but it will typic-
ally satisfy the AJBT criterion, because it is respectful of people’s ends. To be sure, we might
find cases in which the distinction between means and ends is not straightforward (Sunstein
2014), but most cases are easy.
It is not possible to understand the operation of the AJBT criterion without reference to
examples. In countless cases, we can fairly say that given people’s antecedent preferences, a nudge will
make choosers better off AJBT. For example:

1 Luke has heart disease, and he needs to take various medications. He wants to do so, but
he is sometimes forgetful. His doctor sends him periodic text messages. As a result, he takes
the medications. He is very glad to receive those messages.
2 Meredith has a mild weight problem. She is aware of that fact, and while she does not suffer
serious issues of self-control, and does not want to stop eating the foods that she enjoys,
she does seek to lose weight. Because of a new law, many restaurants in her city have clear

564
565

Bounded rationality and nudging

calorie labels, informing her of the caloric content of various options. As a result, she some-
times chooses low-calorie offerings – which she would not do if she were not informed.
She is losing weight. She is very glad to see those calorie labels.
3 Edna is a professor at a large university, which has long offered its employees the option to
sign up for a retirement plan. Edna believes that signing up would be a terrific idea, but she
has not gotten around to it. She is somewhat embarrassed about that. Last year, the univer-
sity switched to an automatic enrollment plan, by which employees are defaulted into the
university’s plan. They are allowed to opt out, but Edna does not. She is very glad that she
has been automatically enrolled in the plan.

There is nothing unfamiliar about these cases. On the contrary, they capture a great deal of the
real-world terrain of nudging, both by governments and by the private sector (Halpern 2015;
Sunstein 2016a; 2016b). Choosers have a goal, or an assortment of goals, and the relevant
choice architecture can make it easier or harder for them to achieve it or them. Insofar as we
understand the AJBT criterion by reference to people’s antecedent preferences, that criterion is
met. Note that it would be easy to design variations on these cases in which nudges failed that
criterion, because they would make people worse off by their own lights.
We could complicate the cases of Luke, Meredith, and Edna by assuming that they have clear
antecedent preferences, that the nudge is inconsistent with those preferences, but that as a result
of the nudge, their preferences are changed. For example:

Jonathan likes talking on his cell phone while driving. He talks to friends on his
commute to work, and he does business as well. As a result of a set of vivid warnings,
he has stopped. He is glad. He cannot imagine why anyone would talk on a cell phone
while driving. In his view, that is too dangerous.

After the nudge, Luke, Meredith, Edna, and Jonathan believe themselves to be better off. Cases
of this kind raise the question whether the AJBT criterion requires reference to ex ante or ex
post preferences. That is a good question, which might be answered by making direct inquiries
into people’s welfare; I will turn to that question below. My main point is that as originally
given, the cases of Luke, Elizabeth, and Edna are straightforward. Such cases are common
(Halpern 2015; Thaler 2015).
Some cases can be seen as different, because they raise questions about self-control (or
akrasia):

1 Ted smokes cigarettes. He wishes that he had not started, but he has been unable to quit.
His government has recently imposed a new requirement, which is that cigarette packages
must be accompanied with graphic images, showing people with serious health problems,
including lung cancer. Ted is deeply affected by those images; he cannot bear to see them.
He quits, and he is glad.
2 Joan is a student at a large university. She drinks a lot of alcohol. She enjoys it, but not that
much, and she is worried that her drinking is impairing her performance and her health.
She says that she would like to scale back, but for reasons that she does not entirely under-
stand, she has found it difficult to do so. Her university recently embarked on an educa-
tional campaign to reduce drinking on campus, in which it (accurately) notes that four out
of five students drink only twice a month or less. Informed of the social norm, Joan finally
resolves to cut back her drinking. She does, and she is glad.

565
566

Cass R. Sunstein

In these cases, the chooser suffers from a self-control problem and is fully aware of that fact.
Ted and Joan can be seen as both planners, with second-order preferences, and doers, with
first-order preferences (Thaler and Sunstein 2008; Thaler 2015). A nudge helps to strengthen
the hand of the planner. It is possible to raise interesting philosophical and economic questions
about akrasia and planner-doer models (Stroud and Tappolet 2003), but insofar as Ted and Joan
welcome the relevant nudges, and do so ex ante as well as ex post, the AJBT criterion is met.
In a sense, self-control problems require their own GPS devices and so can be seen to involve
navigability; but for choosers who face such problems, the underlying challenge is qualitatively
distinctive, and they recognize that fact.
In such cases, the AJBT criterion is met. But do people acknowledge that they face a self-
control problem? That is an empirical question, of course, and my own preliminary research
suggests that the answer is “yes.” On Amazon’s Mechanical Turk, I asked about 200 people this
question:

Many people believe that they have an issue, whether large or small, of self-control.
They may eat too much, they may smoke, they may drink too much, they may not
save enough money. Do you believe that you have any issue of self-control?

A whopping 70 percent said that they did (55 percent said “somewhat agree,” while 15 percent
said “strongly agree”). Only 22 percent disagreed (8 percent were neutral).
This is a preliminary test, of course. Whatever majorities say, the cases of Ted and Joan cap-
ture a lot of the territory of human life, as reflected in the immense popularity of programs
designed to help people to combat addiction to tobacco (Halpern et al. 2015) and alcohol. We
should agree that nudges that do the work of such programs, or that are used in such programs
(Halpern et al. 2015), are likely to satisfy the AJBT criterion.
There are harder cases. In some of them, it is not clear if people have antecedent preferences
at all. In others – as in the case of Jonathan – their ex post preferences are an artifact of, or
constructed by, the nudge. Sometimes these two factors are combined (as marketers are well
aware). As Amos Tversky and Richard Thaler put it long ago, “values or preferences are com-
monly constructed in the process of elicitation” (Tversky and Thaler 1990). If so, how ought
the AJBT criterion to be understood and applied?
For example:

1 George cares about the environment, but he also cares about money. He currently receives
his electricity from coal; he knows that coal is not exactly good for the environment, but it
is cheap, and he does not bother to switch to wind, which would be slightly more expen-
sive. He is quite content with the current situation. Last month, his government imposed
an automatic enrollment rule on electricity providers: People will receive energy from
wind, and pay a slight premium, unless they choose to switch. George does not bother to
switch. He says that he likes the current situation of automatic enrollment. He approves of
the policy and he approves of his own enrollment.
2 Mary is automatically enrolled in a Bronze Health Care Plan – it is less expensive than Silver
and Gold, but it is also less comprehensive in its coverage, and it has a higher deductible.
Mary prefers Bronze and has no interest in switching. In a parallel world (a lot like ours,
but not quite identical, Wolf 1990), Mary is automatically enrolled in a Gold Health Care
Plan – it is more expensive than Silver and Bronze, but it is also more comprehensive in its
coverage, and it has a lower deductible. Mary prefers Gold and has no interest in switching.

566
567

Bounded rationality and nudging

3 Thomas has a serious illness. The question is whether he should have an operation, which
is accompanied with potential benefits and potential risks. Reading about the operation
online, Thomas is not sure whether he should go ahead with it. Thomas’s doctor advises
him to have the operation, emphasizing how much he has to lose if he does not. He decides
to follow the advice. In a parallel world (a lot like ours, but not quite identical), Thomas’s
doctor advises him not to have the operation, emphasizing how much he has to lose if he
does. He decides to follow the advice.

In the latter two cases, Mary and Thomas appear to lack an antecedent preference; what they
prefer is an artifact of the default rule (in the case of Mary) or the framing (in the case of
Thomas). George’s case is less clear, because he might be taken to have an antecedent preference
in favor of green energy, but we could easily understand the narrative to mean that his prefer-
ence, like that of Mary and Thomas, is endogenous to the default rule.
These are the situations on which I am now focusing: People lack an antecedent preference,
and what they like is a product of the nudge. Their preference is constructed by it. After being
nudged, they will be happy and possibly grateful. We have also seen that even if people have
an antecedent preference, the nudge might change it, so that they will be happy and possibly
grateful even if they did not want to be nudged in advance.
In all of these cases, application of the AJBT criterion is less simple. Choice architects cannot
contend that they are merely vindicating choosers’ ex ante preferences. If we look ex post,
people do think that they are better off, and in that sense the criterion is met. For use of the
AJBT criterion, the challenge is that however Mary and Thomas are nudged, they will agree that they
are better off. In my view, there is no escaping at least some kind of welfarist analysis in choosing
between the two worlds in the cases of Mary and Thomas. There is a large question about
which nudge to choose in such cases (for relevant discussion, see Dolan 2014). Nonetheless,
the AJBT criterion remains relevant in the sense that it constrains what choice architects can
do, even if it does not specify a unique outcome (as it does in cases in which people have clear
ex ante preferences and in which the nudge does not alter them).
Again: Thaler and I embrace paternalism, and so the AJBT criterion is emphatically not
designed to defeat a charge of paternalism. It is psychologically fine (often) to think that
choosers have antecedent preferences (whether or not “latent”), but that because of a lack of
information or a behavioral bias, their choices will not satisfy them. (Recall the cases of Luke,
Meredith, and Edna.) To be sure, it is imaginable that some forms of choice architecture will
affect people who have information or lack such biases; an error-free cafeteria visitor might
grab the first item she sees, because she is busy, and because it is not worth it to her to decide
which item to choose (Ullmann-Margalit 2017). Consider this case:

Gretchen enjoys her employer’s cafeteria. She tends to eat high-calorie meals, but she
knows that, and she likes them a lot. Her employer recently redesigned the cafeteria
so that salads and fruits are the most visible and accessible. She now chooses salad and
fruit, and she likes them a lot.

By stipulation, Gretchen suffers from no behavioral bias, but she is affected by the nudge. But in
many (standard) cases, behaviorally biased or uninformed choosers will be affected by a nudge,
and less biased and informed choosers will not; a developing literature explores how to pro-
ceed in such cases, with careful reference to what seems to me a version of the AJBT criterion
(Goldin 2015; Goldin and Lawson 2016).

567
568

Cass R. Sunstein

In Gretchen’s case, and all those like it, the criterion does not leave choice architects at sea: If
she did not like the salad, the criterion would be violated. From the normative standpoint, it
may not be entirely comforting to say that nudges satisfy the AJBT criterion if choice architects
succeed in altering the preferences of those whom they are targeting. (Is that a road to serfdom?
Recall the chilling last lines of Orwell’s 1984: “He had won the victory over himself. He loved
Big Brother” (cf. Elster 1983).) But insofar as we are concerned with subjective welfare, it is
a highly relevant question whether choosers believe, ex post, that the nudge has produced an
outcome of which they approve.
Countless nudges increase navigability, writ large, in the sense that they enable people to get
where they want to go, and therefore enable them to satisfy their antecedent preferences. Many
other nudges, helping to overcome self-control problems, are warmly welcomed by choosers,
and so are consistent with the AJBT criterion. Numerous people acknowledge that they suffer
from such problems. When people lack antecedent preferences or when those preferences are
not firm, and when a nudge constructs or alters their preferences, the AJBT criterion is more
difficult to operationalize, and it may not lead to a unique solution. But it restricts the universe
of candidate solutions, and in that sense helps to orient choice architects.

Acknowledgments

The author is grateful to the Program on Behavioral Economics and Public Policy for support
and to Lucia Reisch for valuable comments on a previous draft.

References
Conly, S. (2014). Against Autonomy. Oxford: Oxford University Press.
Cornwell, J. F., and Krantz, D. H. (2014). Public policy for thee, but not for me: Varying the grammat-
ical person of public policy justifications influences their support. Judgment and Decision Making, 5,
433–444.
Dolan, P. (2014). Happiness by Design. London: Penguin.
Elster, J. (1983). Sour Grapes. Cambridge: Cambridge University Press.
Goldin, J. (2015). Which way to nudge? Uncovering preferences in the behavioral age. Yale Law Journal,
125, 226–271.
Goldin, J., and Lawson, N. (2016). Defaults, mandates, and taxes: Policy design with active and passive
decision-makers. American Journal of Law and Economics, 18, 438–462.
Halpern, D. (2015). Inside the Nudge Unit: How Small Changes Can Make a Big Difference, London: W.H. Allen.
Halpern, S. D. et al. (2015). Randomized trial of four financial-incentive programs for smoking cessation.
The New England Journal of Medicine, 372, 2108–2111.
Infante, G., Lecouteux, G., and Sugden, R. (2016). Preference purification and the inner rational agent: a
critique of the conventional wisdom of behavioural welfare economics. Journal of Economic Methodology,
23, 1–25.
Jung, J. Y., and Mellers, B. A. (2016). American attitudes toward nudges. Judgment and Decision Making,
11(1), 62–74.
Lichtenstein, S., and Slovic, P. (2006). The Construction of Preference. Cambridge: Cambridge University
Press.
Reisch, L., and Sunstein, C. R. (2016). Do Europeans like nudges? Judgment and Decision Making, 11,
310–325.
Sudgen, R. (2016). Do people really want to be nudged towards healthy lifestyles? International Review of
Economics, 64, 113–123.
Stroud, S., and Tappolet, C. (Eds.) (2003). Weakness of Will and Practical Irrationality, Oxford: Clarendon Press.
Sunstein, C. R. (2014). Why Nudge? New Haven, CT: Yale University Press.
Sunstein, C. R. (2016a). The Council of Psychological Advisers. Annual Review of Psychology, 67, 713–737.
Sunstein, C. R. (2016b). The Ethics of Influence. Cambridge: Cambridge University Press.

568
569

Bounded rationality and nudging

Sunstein, C. R., Reich, L., and Rauber, J. (2017). Behavioral insights all over the world? Public
attitudes toward nudging in a multi-country study, available at https://papers.ssrn.com/sol3/papers.
cfm?abstract_id=2921217
Sunstein, C. R., and Reisch, L. A. (2014). Automatically green: Behavioral economics and environmental
protection. Harvard Environmental Law Review, 38, 127–158.
Thaler, R. H. (2015). Misbehaving. New York: Norton.
Thaler, R. H., and Sunstein, C. R. (2008). Nudge: Improving Decisions about Health, Wealth, and Happiness.
New Haven, CT: Yale University Press.
Tversky, A., and Thaler, R. H. (1990). Anomalies: Preference reversals. Journal of Economic Perspectives, 4,
201–211.
Ullmann-Margalit, E. (2017). Normal Rationality. Oxford: Oxford University Press.
Wolf, F. (1990). Parallel Universes: The Search for Other Worlds. New York: Simon & Schuster.

569
570

39
AN ALTERNATIVE
BEHAVIOURAL PUBLIC POLICY
Adam Oliver

Introduction
The application of behavioural insights to public policy, defined here as behavioural public
policy, is in any substantive sense, a relatively recent endeavour, although decades of social
science research underpin the approach (see, for example, Oliver, 2017). Several conceptual
frameworks have been developed within the field of behavioural public policy, but the dom-
inant frameworks have thus far been forms of paternalism, with paternalistic approaches defined
here as those that focus upon improving the position of those targeted by behaviour change
initiatives, often (although not always) as judged deliberatively by those persons themselves. In
other words, these approaches aim to improve internalities, rather than reduce negative exter-
nalities – i.e., harms – to third parties.
Moreover, the dominant behavioural public policy frameworks tend to retain the normative
assumption of standard economic theory and rational choice theory that people ought to maxi-
mise utility (considered here to be synonymous with welfare and happiness). Yet they differ
from standard theory by contending that, descriptively, people often fail to maximise utility
due to being influenced by innate behavioural influences, such as present bias and loss aversion.
As such, these frameworks can be defined as behavioural welfare economics and propose that
policy makers ought to influence people’s decisions such that they move to increase their utility.
In this chapter, I will contend that the focus on both internalities and utility is less than
ideal for the future development of behavioural public policy. In its stead, I will propose a
political economy of behavioural public policy that sits alongside the liberal economic trad-
ition of John Stuart Mill, albeit with my approach being somewhat more interventionist than
Mill would have allowed (and far more interventionist than followers of the Austrian School
of Economics (e.g., von Mises, [1927] 2005)). But first let us unpick a little more some of the
details introduced thus far.

Addressing internalities
Of the forms of paternalism, libertarian paternalism – policy applications of which are the
fabled nudges (Thaler and Sunstein, 2003, 2008) – has thus far been the most prominent in
the development of behavioural public policy. Like asymmetric paternalism (Camerer et al.,

570
571

An alternative behavioural public policy

2003), libertarian paternalism is a soft form of paternalism. In short, the approach seeks to guide
people in particular directions so that they are more likely to make better (i.e. more welfare-
enhancing) decisions for themselves but does not propose the use of force or mandates. People
are free to continue with their existing behaviours if they wish, with Thaler and Sunstein
(2008) contending that retaining the freedom to choose is the best safeguard against any mis-
guided policy intervention.
Underpinning libertarian paternalism is the assumption that of the many decisions that each
of us make quickly and automatically each day – decisions that are guided by simple rules of
thumb (i.e., the heuristics famously associated with Herbert Simon (1956)) and influenced by
various innate behavioural influences (e.g., present bias, loss aversion) – a few will lead us to
act in ways that, if we deliberated a little more, we would prefer not to do (e.g., present bias
might lead us to eat more doughnuts than we would ideally consume if we thought about our
decisions a little more). It is noteworthy that these heuristics may have evolved because most of
the time they guide us efficiently through our daily lives (e.g., Gigerenzer, 2015), but the basic
idea in libertarian paternalism is that with knowledge of the behavioural influences, the ‘choice
architecture’ (i.e., the context or environment) that people face can be redesigned, such that
their automatic choices are even more likely to better align with their deliberative preferences.
Thus, to summarise, for an intervention to be a nudge, it has to target internalities, preserve
liberty and be informed by behavioural science. Each of these three requirements is represented
by an axis in Figure 39.1.
In Figure 39.1, moving towards the origin on the horizontal axis, the vertical axis and the
diagonal axis respectively indicates that a policy is increasingly addressing internalities rather
than externalities, is increasingly liberty-preserving rather than regulatory, and is increasingly
informed by behavioural science rather than standard economic theory or rational choice
theory. Consequently, a nudge in its purest form must lie at the point where the axes intersect.
An oft-mentioned pure nudge, for instance, is the placing of apples at the front and cheesecake

Regulatory

Pure nudge

Liberty

Behavioural

Internalities Externalities

Rational

Figure 39.1 The requirements of libertarian paternalism

571
572

Adam Oliver

at the back of canteen shelves, to motivate those who would ideally like to eat more healthily
to choose the healthier option (but people can still choose the cheesecake if they wish).
Some argue that by retaining the notion of liberty, soft forms of paternalism do not go far
enough in that they will ultimately prove to be insufficiently effective at affecting behaviour
change for those who threaten their own health and financial security (Conly, 2013). Those
who adhere to this line of reasoning call instead for hard paternalism or coercive paternalism.
They retain the notion that activities that they argue are not good for people – such as smoking
and saving insufficiently for one’s retirement – may be driven by behavioural influences such
as present bias and the like. In short, whereas soft paternalists allow people to continue their
pre-existing behaviours, coercive paternalists call for some activities, such as smoking, to be
banned entirely.
By advocating behaviour change for some or all of those targeted for their own good, all
forms of paternalism are, however, open to the criticism that they infantilise those whom
they target, and that politicians and policy makers ought to have no role in influencing the
behaviours of adults of sound mind if their activities impose no harms on others, a view
consistent with Mill’s harm principle (Mill, 1859/1972). The behavioural welfare economics
counterargument is predicated on retaining the normative postulate of utility maximisation,
i.e., libertarian paternalists (and, indeed, other types of paternalist) believe that due to the
behavioural influences that contravene rational choice theory, people will sometimes fail to act
and choose in their own best interests (i.e. to maximise utility) – and thus they contend that it
is legitimate for policy makers to steer them (or, for coercive paternalists, to force them) in that
normative direction. But is this assumption of legitimacy acceptable?

The view from nowhere

As aforementioned, behavioural welfare economics, like standard economic theory and rational
choice theory, postulates that people ought to maximise utility (welfare or happiness). All of
these approaches are thus underpinned by utilitarianism, the founding father of which was
Jeremy Bentham. Bentham (1988 [1781] ) believed that humans are governed by the two “sov-
ereign masters” of pleasure and pain, and he viewed these as feelings that are experienced on a
continuum and can thus be compared. He contended that they guide us on what we ought to
do, and what we shall do.
By the beginning of the twentieth century, measures of Benthamite utility – i.e., numerical
interpersonally comparable indicators of how much pain and pleasure individuals experience in
the moment – were thought to be impossible to uncover, but modern neoclassical economic
theory, which is based upon decisions over future experiences rather than experiences in the
moment, retained the notion that people ought to, and will want to, maximise their utility.
Therefore, other than making random mistakes, we can infer from this that the decisions that
people make over future episodes, and their retrospective assessments of their previous courses
of action, will be consistent with them wanting to maximise the amount of utility that they
experience. There is now evidence to suggest that this inference is not always accurate.
Specifically, it has been demonstrated in experimental settings that when people are asked
to offer retrospective preferences over different events, there is a tendency for their relative
preferences to differ systematically from that which would be suggested by aggregating moment-
to-moment instant utilities that they experienced as they lived through each event (similar
observations have sometimes been observed with respect to prospective evaluations). The cause
of these systematic discrepancies are salient features of events, known as the gestalt characteristics
(Ariely and Carmon, 2000). The gestalt characteristics include the tendency for people, at

572
573

An alternative behavioural public policy

least over relatively short events, to prefer worse outcomes to precede better outcomes rather
than vice versa (Loewenstein and Prelic, 1993), an aversion to sudden, steep rates of change in
outcomes (Hsee and Abelson, 1991), and the tendency for people to place a heavy emphasis on
the peak and end moments of an episode (Fredrickson and Kahneman, 1993). This last gestalt
characteristic, called peak-end evaluation, is the most studied, and can cause people to neglect
the duration of the event (for a general review of the gestalts, see Kahneman et al., 1997).
Peak-end evaluation is nicely illustrated by a study by Redelmeier et al. (2003), who divided
682 sigmoidoscopy patients into two groups. For the patients in one of the groups, unbe-
knownst to them, the sigmoidoscopy tube was left inserted in their bodies for an additional
short period at the end of the procedure for no clinical reason, which would have been uncom-
fortable but not as painful as when the tube was being moved around. The patients’ discom-
fort was recorded every 60 seconds to measure their moments of instant disutility, and by
aggregating these moments at the end of the procedures a measure of total experienced dis-
utility of the procedure for each patient could be calculated. Following their procedures, the
patients’ retrospective evaluations of the total discomfort that they experienced was recorded on
a ten-point scale. By comparing the retrospective evaluations to the moments of instant utility,
Redelmeier et al. observed peak-end evaluation and a neglect of duration. They also reported
that the group for whom the tube was left inserted tended to remember the procedure as less
painful than those for whom it was removed as soon as the clinical procedure was over. That
the former group must on average have experienced greater total experienced disutility as a
consequence of having the procedure duration extended unnecessarily means that the authors
recorded a violation of dominance. That is, in general, those who remembered the experience
as less bad were exposed to an experience that in general caused greater aggregate pain.
Since the gestalt characteristics can cause people to choose options that conflict with the
assumption of experienced utility maximisation, we must ask the question of whether they are
causing people to make errors, or whether the gestalts are legitimate influences on preferences.
Behavioural welfare economics would view preferences that are influenced as such as mistakes,
which may sometimes be a fair conclusion to draw (after all, who, other than a masochist,
would want to experience more pain?). But in many circumstances and for many people, it
is legitimate to question whether a simple aggregation of moments of instant utility gives an
accurate assessment of the underlying value of an experience. If the experience in question is a
whole life, Daniel Hausman, for instance, has stated that

a good life is not a sum of the net goodness of its moments … The same sum of
momentary experiences can add up to a wonderful life or an incoherent and medi-
ocre one, depending on how the experiences are ordered and what overall narrative
they sustain.
Hausman, 2015, p. 114

When we look forward to an episode, or reflect back on it, the gestalts may often matter
because they give meaning – a narrative – to the story. They may in part determine how ful-
filling the event was to a person, and whether the person feels they had the opportunity, or will
have the opportunity, to flourish.
Some economists might contend that the gestalts, and the concern with how the moments of
an experience fit together, can be encapsulated by additional arguments in the utility function.
Following this line of reasoning, although the gestalts may contrast with the assumptions of
Benthamite utility maximisation, they might be consistent with a broader notion of utility
maximisation. Placing anything in the utility function, however, presents the risk that utility

573
574

Adam Oliver

maximisation can justify everything, but can really explain and specifically predict nothing.
That is to say, if everything that a person does is attributed to her simply maximising her utility
and we accept that this is what she ought to do, the theory is so general that it becomes empty
and the person is allowed to do anything.
Rather than accept that the drive behind human behaviour and decision making is and
ought to be the drive to increase utility/welfare/happiness, we might more modestly admit
that we cannot know why people do the things they do. It may well be that people have
various and varied legitimate reasons for their actions, both interpersonally and, across contexts,
intrapersonally. To assume that utility (or welfare or happiness) maximisation is the universal
normative condition is, as suggested by the economist, Robert Sugden (2018), the view from
nowhere. On this basis, it is my contention that the role of policy makers should not be to
seek to maximise utility, but rather to seek to improve the opportunities for people to flourish
as they themselves see fit (which may for some be to seek to maximise utility, but that is their
prerogative). Encouraging people to reciprocate with others, as opposed to acting upon their
selfish egoistic instincts, is likely to facilitate the individuals that comprise a community in
the pursuit of their own goals (for discussions of the benefits of reciprocity, see Bowles, 2016;
Oliver, 2019; Sugden, 2018).

To reciprocate, to flourish
If the aim of policy makers is to help create the conditions for people to flourish as they
themselves see fit, then, to reiterate, facilitating reciprocity, as an aspect of behavioural public
policy, is potentially an important arm of this effort. Sugden suggests that since people are
fundamentally reciprocators, then the tendency and scope for reciprocal actions can be aided
simply by increasing the number of choices and opportunities that are available to them.
However, although reciprocity is indeed a fundamental motivator of human actions, it can be
crowded out by our baser motivations (namely, selfish egoism), and thus conditions need to be
created in order to nurture reciprocity. Some suggestions for how policy makers might do this
include emphasising the importance of this motivator of human behaviour in their rhetoric,
decentralising public policy decision making to sub-national levels, and ensuring that income
and wealth are not concentrated excessively in the hands of a small proportion of the popula-
tion. Elaborations on these policy directions is beyond my scope here; more detailed discussions
and references to support these propositions can be found in Oliver (2019).
The approach proposed here does not therefore aim to instil a view from nowhere onto
targeted populations. Instead, it postulates that people in local communities organise themselves
in any way they see fit to achieve the objectives that they wish to pursue in their private choices
and actions, on the condition that they do not impose harms on others, although – consistent
with Mill ([1859] 1972) – this does not preclude attempts at educating people openly about
the possible self-harm inflicting consequences of some of their activities (and allows scope
for “boosting”, or, in other words, educating people openly in relation to how the heuristics
and behavioural phenomena might influence their choices with a view to them making more
informed decisions (Hertwig, 2017)). The members of the community need not share the same
goals. They may reciprocate and collaborate in their pursuit of different personal conceptions
of a flourishing life. If the conditions for mutually supportive efforts towards meeting personal
goals are strengthened, then the suggestion here is that it is reasonable to expect that their goals
will more likely be realised.
Allowing people great freedom to live their lives in the ways that they see fit is, however, not
without risks, in that without interference, there will be much opportunity for selfish egoists

574
575

An alternative behavioural public policy

to act upon their instincts. Mill’s ([1859] 1972) harm principle, which allows the regulation
of private actions that are imposing harms – or negative externalities – on others, of course
recognises this danger. But Mill did not explicitly foresee that an actor or organisation could
use the behavioural influences to impose harms on others. For more than a century the private
marketing industry has used what are essentially behavioural insights to sometimes nefarious
ends, activities that remain rife (Akerlof and Shiller, 2015). If providers of products and services
implement measures that undermine the concept of a fair, reciprocal exchange, then policy
action to regulate against these behaviourally-informed harms is justified.

Budging
An alternative (or, at the very least, a complement) to paternalistic interventions, budges are
regulations against activities that rely on their effectiveness by being informed by behavioural
insights, and where their effectiveness imposes harms on others (Oliver, 2013, 2015, 2017).
For example, the marketing arms of confectionary companies know implicitly that salience and
immediacy (i.e., factors associated with present bias) can have a large effect on consumer buying
patterns, and this is why they have traditionally paid supermarkets substantial amounts of money
so that their products are displayed close to checkout counters. If we were to conclude that
consumers, as a consequence, are purchasing more confectionary than they otherwise would
(i.e., more than in the absence of manipulation, which is particularly nefarious if the products
or services in question impose harms on characteristics, such as health or finances, that might
be considered important factors underlying a flourishing life, irrespective of what it means to
flourish), then policy makers would have an intellectual justification to regulate against – to
budge – this activity.
Budges can also be placed within a three-dimensional space, depicted as Figure 39.2.
Compared to Figure 39.1, the horizontal and vertical axes have been inverted, such that a

Liberty

Pure budge

Regulatory

Behavioural

Externalities Internalities

Rational

Figure 39.2 The requirements of behavioural regulation

575
576

Adam Oliver

policy placed at the origin would now be a regulation against a behaviourally-informed nega-
tive externality: a pure budge.

Conclusion
In this chapter, I have argued that the principal overarching political economy of behavioural
public policy should not be to focus upon moving internalities towards some conception of
utility (or welfare or happiness) maximisation, but ought to rather be focused on fostering
reciprocity and cooperation so that people are better equipped to flourish and find fulfilment
in their lives. This proposed approach is in the spirit of the liberal economic tradition (albeit
recognising that the individual’s conception of flourishing does not necessarily equate to con-
sequentialist notions of utility, welfare or happiness), but goes somewhat further than that
tradition specifies by recognising that giving people great freedom affords more opportunities
for the egoistically inclined to use behavioural insights in order to manipulate people, and
potentially impose harms on them. Thus, it is argued that when behavioural influences are
being used by individuals or organisations towards self-serving and harmful ends, then there is a
legitimate, intellectual justification for regulating against those activities. So, to sum up, the pol-
itical economy of behavioural public policy proposed in this chapter has two arms: to nurture
reciprocity so as improve the stock of human flourishing, and to regulate against harm-inducing
egoism to protect the capacity of people to flourish as they themselves see fit.

References
Akerlof, G. A., and Shiller, R. J. 2015. Phishing for Phools: The Economics of Manipulation and Deception.
Princeton, NJ: Princeton University Press.
Ariely, D., and Carmon, Z. 2000. Gestalt characteristics of experiences: The defining features of
summarized events. Journal of Behavioral Decision Making 13: 191–201.
Bentham, J. [1781] 1988. The Principles of Morals and Legislation. New York: Prometheus Books.
Bowles, S. 2016. The Moral Economy: Why Good Incentives Are No Substitute for Good Citizens. New Haven,
CT: Yale University Press.
Camerer, C., Issacharoff, S., Loewenstein, G., O’Donoghue, T., and Rabin, M. 2003. Regulation for
conservatives: Behavioral economics and the case for ‘asymmetric paternalism’. University of Pennsylvania
Law Review 1151: 1211–1254.
Conly, S. 2013. Against Autonomy: Justifying Coercive Paternalism. Cambridge: Cambridge University Press.
Fredrickson, B. L., and Kahneman, D. 1993. Duration neglect in retrospective evaluations of affective
episodes. Journal of Personality and Social Psychology 65: 44–55.
Gigerenzer, G. 2015. On the supposed evidence for libertarian paternalism. Review of Philosophy and
Psychology 6: 361–383.
Hausman, D. M. 2015. Valuing Health: Well-Being, Freedom and Suffering. Oxford: Oxford University Press.
Hertwig, R. 2017. When to consider boosting: Some rules for policy-makers. Behavioural Public Policy
1: 143–161.
Hsee, C. K., and Abelson, R. P. 1991. Velocity relation: Satisfaction as a function of the first derivative of
outcome over time. Journal of Personality and Social Psychology 60: 341–347.
Kahneman, D., Wakker, P. P., and Sarin, R. 1997. Back to Bentham? Explorations of expected utility. The
Quarterly Journal of Economics 112: 375–405.
Loewenstein, G., and Prelec, D. 1993. Preferences for sequences of outcomes. Psychological Review
100: 91–98.
Mill, J. S. [1859] 1972. On Liberty. London: Dent.
Oliver, A. 2013. From nudging to budging: Using behavioural economics to inform public sector policy.
Journal of Social Policy 42: 685–700.
Oliver, A. 2015. Nudging, shoving and budging: Behavioural economic-informed policy. Public
Administration 93: 700–714.
Oliver, A. 2017. The Origins of Behavioural Public Policy. Cambridge: Cambridge University Press.

576
577

An alternative behavioural public policy

Oliver, A. 2019. Reciprocity and the Art of Behavioural Public Policy. Cambridge: Cambridge University Press.
Redelmeier, D., Katz, J., and Kahneman, D. 2003. Memories of colonoscopy: A randomized trial. Pain
104: 187–194.
Simon, H. A. 1956. Rational choice and the structure of the environment. Psychological Review 63: 129–138.
Sugden, R. 2018. The Community of Advantage: A Behavioural Defence of the Liberal Tradition of Economics.
Oxford: Oxford University Press.
Thaler, R. H., and Sunstein, C. R. 2003. Libertarian paternalism. The American Economic Review
93: 175–179.
Thaler, R. H., and Sunstein, C. R. 2008. Nudge: Improving Decisions about Health, Wealth and Happiness.
New Haven, CT: Yale University Press.
von Mises, L. [1927] 2005. Liberalism. Indianapolis: Liberty Fund.

577
578

40
AGAINST NUDGING
Simon-inspired behavioral law and economics
founded on ecological rationality

Nathan Berg

Introduction
Paternalistic policy making in general—and Thaler and Sunstein's (2008) nudge program, in
particular—assume that we have a reliable and relatively stable expert consensus regarding what
we should eat, how much we should save for retirement, which types of securities should
be held in our retirement portfolios, and how we ought to make sundry other decisions for
which many lack conviction to take decisive action without a nudge (e.g., whether to become
an organ donor). Critiques of the nudging program are many. Some of these critiques follow
from: (1) debates over competing definitions of axiomatic rationality versus Simon-inspired
bounded or ecological rationality (Gigerenzer and Selten, 2001; Berg, 2014a; Gigerenzer,
2016); (2) philosophical and methodological concerns over autonomy and the option to change
one’s mind raised by Sugden (2016, 2018) and Infante, Lecouteux, and Sugden (2016); (3) evi-
dence regarding the stability of expert advice and social epistemologies on which appeals to
authority rest discussed in Viale (2001, 2012, 2019, forthcoming); (4) and well-known incen-
tive problems in the signaling literature that may impede the unbiased transmission of expert
advice (Berg, 2018; Berg and Kim, 2019). A less frequently recognized implication of eco-
logical rationality (in contrast to that of axiomatic rationality) is that heterogeneity of beliefs and
decision processes is a beneficial social good responsible for numerous positive externalities (Berg
and Watanabe, 2020).1 How nudging risks reducing these important yet difficult-to-observe
indirect benefits of heterogeneous beliefs and behavior is the focus of this chapter.
Heterogeneous beliefs and decisions can, of course, be costly. Not all behavioral deviations
from what experts say we should do are good. Putting aside the uniformly distributed trem-
bling hand (because that is not the kind of unsystematic deviation one often finds), many of the
compelling behavioral findings in the last three decades reflect highly non-uniform deliberate
deviations from normative decision theory—or purposively systematic deviations when they are
unconscious or autonomomic (whether helpful or otherwise in the particular environment in
which they are used, e.g., evolutionary mismatch). Although not yet mainstream (because it
contradicts the Kahneman-inspired view among many behavioral economists that deviations
from axiomatic rationality must somehow be costly and therefore pathological), there are,
by now, abundant empirical evidence and numerous theoretical models demonstrating that
some types of deviations, when well-matched to the environments in which they are used,

578
579

Against nudging

confer individual-level advantages or improved payoffs to deviators relative to non-deviators

(Gigerenzer, Todd, and the ABC Research Group, 1999; Selten and Gigerenzer, 2001; Berg and
Hoffrage, 2008; Berg, Biele, and Gigerenzer, 2016). My aim in this chapter is to focus instead
on the less frequently considered social benefits of heterogeneity in beliefs and behavior. I argue
that advocates of nudging policies—which risk jeopardizing both population- and individual-
level benefits of behavioral diversity, in part, by inducing less resilient monocultures—would do
well to incorporate these risks into their analysis.
Peter Kropotkin and Glen Shafer express the idea that heterogeneity among the deci-
sion processes that individuals employ—and heterogeneity among the metrics that individ-
uals use to assess how well they perform—is broadly normative (Berg, 2003, 2018; Berg and
Gigerenzer, 2010). In many real-world (and theoretical) decision-making environments, there
is no one-size-fits-all “optimal” or “best” process that people would agree should be used
to make decisions, form beliefs, or evaluate outcomes (Simon, 1957, 1979; Gigerenzer and
Selten, 2001).2 Social benefits from behavioral diversity include diversification of population-
level risks, efficiencies from competition among differing views, ideas, and beliefs, and—less
obviously—beneficial social coordination services achieved through voluntary choice in het-
erogeneous populations, which may be (inadvertently) blocked when paternalistic policies
induce behavioral monocultures. For example, the social networks generated from voluntarily
choosing to eat vegetarian or give money to charity may lose social significance once these
behaviors become policy makers’ desired outcomes. Just as financial incentives may crowd out
cooperative or prosocial behavior (e.g., Eckel, Grossman, and Johnston, 2005; Mellström and
Johannesson, 2008; Brown and Knowles, 2019), so, too, policy prescriptions may crowd out
voluntary choice by individuals of the policy maker’s preferred behavior. Policy prescriptions
can be expected to alter coordination services, social meanings, and what can be inferred
from voluntarily choosing behaviors that policy makers wish to induce with nudging. Another
important aspect of social benefits of heterogeneity follow from the so-called triple-helix
model in which nurturing distinct views and objectives that lead to complementary but distinct
processes of innovation within a complex adaptive system raises special challenges for policy
makers seeking to encourage innovation (Viale and Pozzali, 2010).
In addition to benefits, there are also costs of new complexities to consider when policy
makers introduce framings and nudges. Nudges introduce strategic communication between
experts and non-experts, which forces non-experts to filter information in new ways. Non-
experts notice when experts shift from disseminating information without paternalism toward
paternalism (even in the absence of coercion where the choice set is simply re-described but not
materially altered, e.g., nudging).3 Non-experts therefore cannot be sure of experts’ objectives
when disseminating information, which may reduce information flows, trust and social welfare
(Berg, 2018; Berg and Kim, 2010).
There are additional informational benefits that heterogeneity may confer that are not
mentioned above, which include discovery of hidden benefits from minority behaviors (e.g.,
therapeutic uses of tobacco in irritable bowel syndrome and Parkinson’s and Alzheimer’s
Disease), innovation and welfare gains from competition and trade based on opposing beliefs,
and subjective benefits from the option value and exercise of individual freedom (e.g., Sugden,
2018). In this context in which public health is used to justify restrictions on consumer choice
(i.e., paternalism that goes beyond nudging), Popper’s critiques of seemingly benign sympathy
with totalitarian means to virtuous ends (among political philosophers’ interpretations of Plato)
as developed in The Open Society and Its Enemies are relevant. Similarly relevant is Hayek’s dis-
cussion of world views about whether order,4 as a foundational public good, should be under-
stood as cosmos (endogenous or spontaneous order that arises through purposeful action in

579
580

Nathan Berg

the face of “irremediable ignorance on everyone’s part of most of the particular facts which
determine the actions of all the several members of human society” (Hayek, 1973, p. 12)) or as
if it were designed by policy makers, or choice architects, or some other unitary entity, as taxis
(exogenously given or “made order” (Hayek, 1973, p. 36)).
Benefits of decentralization—once a foundational insight of welfare economics based in
part on the First Fundamental Welfare Theorem—risk being mislaid (Berg, 2018). The possi-
bility of social costs attributable to market failures are also underscored by the stringency of the
Welfare Theorem’s hypothesis, namely, the joint absence of market power, externalities, and
information asymmetry. When failures of this hypothesis arise, as they often do, of course it is
not automatic that intervention will satisfy the benefit-cost criterion and improve social wel-
fare. And it would be naïve to fail to recognize numerous rent-seeking motives when political
support for intervention forms.
As appealing and reasonable as it may sound to introduce nudges aimed at influencing indi-
viduals to make (what some experts claim would be) better choices as judged by themselves
(about which many well-published experts would themselves disagree, e.g., fewer saturated
fats; more cancer screenings; greater contributions to employer-sponsored retirement savings
programs,5 etc.), there are real risks of inadvertently hurting people who respond to nudges pre-
cisely in the way that those policies are intended to work.6 If populations respond to nudges as
they are designed to and bring the population’s profile of beliefs and behaviors into closer con-
formity, then multiple beneficial streams of belief and behavioral heterogeneity also risk being
reduced. This latter risk involves more subtle and indirect social benefits of heterogeneity for-
gone, which motivates its special consideration in evaluating the benefits and costs of nudging
and paternalistic policies more broadly.

Expert advice and political uses of scientific claims

There is substantial disagreement in the nutrition and medical research literatures regarding
macronutrient percentages (e.g., fat, protein, carbohydrate) in the diet and, specifically, whether
polyunsaturated or saturated fats reduce or increase risk of chronic disease.7 Policies concerning
the screening of asymptomatic women and men for breast and prostate cancer, respectively, are
another important area of contention among experts where proponents of nudging have never-
theless appeared and influenced governments.8 Encouraging savers to accept more exposure to
financial risk (which often means equities and high-yield bond funds) to earn higher rates of
return is yet another case where behavioral economists have argued that influencing people
to shift their behavior is obvious (e.g., Benartzi and Thaler, 1995) even though we know that
large equity draw-downs, money-market-fund defaults, and severe liquidity crises are far from
unlikely events that savers will face over their lifetimes.
Despite the standard advice from financial advisers to max out employer-matches in
401ks and the compounding tax benefits of individual retirement accounts, it should at least
be acknowledged that this advice, when acted on, raises new risks of having one’s wealth
concentrated in brokerage accounts, Exchange Traded Funds, and asset classes with coun-
terparty risk adverse liquidity events (e.g., New York money-market specialist, The Reserve,
whose Primary Fund with exposure to Lehman-Brothers-issued commercial paper “broke
the buck” on September 18, 2008). If a nudging program is “successful” by the narrow cri-
terion of increasing savings rates and balances held in retirement accounts, the social-welfare
consequences of homogenizing a population’s otherwise more heterogeneous retirement
savings decisions remain far from obvious.

580
581

Against nudging

Viale (2001) argues that expert knowledge could be reconfigured to strengthen the neces-
sary appeals to authority in evaluating scientific claims. See Viale (2001, 2012) for more discus-
sion of social epistemologies and the methodological cognitivism advocated in this context, which
bear special relevance for evaluating prescriptive claims upon which nudging policies rest. It
must be acknowledged, of course, how challenging it is to consistently apply any political prin-
ciple (e.g., ranging from laissez-faire to totalitarian health dictatorship by public health experts),
or a fixed set of normative principles, to guide policy makers’ responses to serious problems
such as insufficient retirement savings (Harrison and Ross, 2018), even when there is agreement
that it is indeed a problem. Observers can agree on the population-level outcome they would
like to see without finding common ground on the role of centralized power in bringing that
outcome about. The question of “Should there be more or less government?” applied to such
problems depends crucially on the counterfactuals considered. Part of the challenge involves
how to appropriately circumscribe the answers we regard as admissible evidence when we ask
“Without intervention, which harms would have occurred if not for the intervention?” versus
the much subtler question of “Which socially beneficial behavioral innovations and new private
organizations to address this problem would have appeared without the intervention?”
Given the available evidence regarding regulatory capture, falsehoods propagated by
governments based on lobbying rather than scientific evidence (e.g., the “food pyramid”),
regulatory failures to protect the public from dangerous drugs, and regulatory failures (of a dis-
tinct kind) that block research and availability of un-patentable approaches to treating disease
and improving health—one would hope that advocates for more intervention would acknow-
ledge that our answers to the question of whether more government regulation based on sci-
entific expertise would improve social welfare are often ambiguous. Well-informed empiricists
can, and do, frequently disagree. And transparent access to that disagreement is beneficial for
those who want to make up their own minds. Disagreement is, once again, socially benefi-
cial when it leads to collection and dissemination of new evidence. False harmonization of
discordant expert views, as well as disparaging those who question evidence regarding what a
political “consensus” claims to be “settled science,” can do real harm.

Ecological rationality
Inspired by Herbert Simon,9 the research program on ecological rationality (Gigerenzer and
Selten, 2001; Smith, 2003; Berg and Gigerenzer, 2010; Berg, 2014a, 2015), focuses on a novel
normative criterion that diverges from definitions of rationality used elsewhere in economics.
Rather than rationality being an attribute solely of the decision maker—or the decision-making
process a decision maker employs (as in axiomatic definitions of a preference ordering’s ration-
ality or the choice data it generates)—the definition of ecological rationality also depends on
the external reward-generating environment in which the decision process is used.
Ecological rationality is a matching criterion. It requires that a decision process perform
sufficiently well when measured by one or more performance metrics relevant to the envir-
onment in which it is used. There are virtually no decision processes that are universally eco-
logically rational. For any decision process (including constrained optimization), there exists
some choice domains where it performs poorly according to one or more relevant performance
metrics. Ecological rationality is therefore not an inherent attribute of a decision maker or a
decision process. Decision processes can be ecologically rational in one set of environments (or
decision domains) and ecologically irrational in others.
One theme in the ecological rationality research program is to clearly define domains of
reward-generating environments and performance metrics with respect to which a given

581
582

Nathan Berg

decision process performs well. Another theme relevant to Simon-inspired behavioral law and
economics is how law and regulation—interaction with the population profile of decision
processes actually in use—jointly construct the reward-generating environment. A heteroge-
neous ecology of decision rules employed by many different individuals and constrained by
the institutional particulars of whichever laws and regulations are in effect at a particular time
and place can perform well, by an appropriately specified aggregation rule or social welfare
function. Epstein (1995), for example, argues that legal codes based on six simple rules could
resolve many seemingly intractable complexities in contemporary societies arising from both
technological innovation and timeless or nearly universal human moral principles.
As a normative criterion for evaluating the rationality of a decision process, ecological
rationality is not domain-general, whereas axiomatic rationality is. Ecological rationality situates
normative evaluations of rationality in a circumscribed set of environments or domains. The
important role of the reward-generating environment and environment-specific metrics of per-
formance make ecological rationality amenable to formal modeling. They also make ecological
rationality falsifiable and, thus, not tautological. Such models describe rules of belief formation
and behavior that do not necessarily arise as the solution to a constrained optimization problem.
Like axiomatic definitions of rationality, the possibility of failing to achieve ecological ration-
ality is critical to its value as a normative concept. The possibility of failure gives meaning
to statements or boundary conditions that circumscribe the set of environments in which a
decision process performs well enough to achieve ecological rationality (given a threshold of
performance in a particular class of environments). Clear partitioning of this set into disjoint
sets of environments—one in which a decision process performs badly, and another in which it
performs well enough—provides valuable information to the study of institutions, in general,
and law and regulation in particular.
Having established that ecological rationality is distinct from axiomatic rationality,10 it is
also worthwhile to observe that both concepts of rationality support normative evaluations
of a decision process that can be conceptually identified as characterizing individual perform-
ance. The criterion of ecological rationality requires good-enough performance in a particular
environment. In contrast, axiomatic rationality requires that observed choice data conform to
domain-general axioms (e.g., transitivity), which first came to prominence in economics as
technical requirements for well-known utility representation theorems. These distinct concepts
of rationality also map into distinct normative evaluations of public policies.

Public policy
Another relevant methodological observation when comparing the performance of public pol-
icies based on ecological versus axiomatic rationality is that they can lead to different rankings
of policies based on policy makers’ own objectives—apart from the social welfare effects they
produce. For example, Berg and Gigerenzer (2007) show that a society of satisificers requires
less paternalistic intervention from risk regulators than a society of risk-averse expected utility
maximizers to achieve the policy makers’ goal of limiting behavior they regard as dangerous.
In contrast, the nudging program assumes that violations of standard axiomatic definitions
of rationality based on internal logical consistency (e.g., preference reversals, framing effects,
transitivity violations, time inconsistency) are themselves pathological. The assumption that
violations of logical invariance and axiomatic rationality must incur economically meaningful
harms lacks much empirical evidence and is contradicted in many studies outside of Kahneman
and Thaler’s research program (only a small subset of which is cited in this chapter). Nevertheless,
the de-biasing program proposes that violations of axiomatic rationality be “corrected” so that

582
583

Against nudging

allegedly “biased” decision processes undergo paternalistic interventions to “de-bias” (e.g.,

Jolls, Sunstein, and Thaler, 1998). Paternalistic policy advice based on behavioral economics
(e.g., Sunstein, 2014, 2016a, 2016b; Sunstein and Reisch, 2014; Sunstein, Reich, and Rauber,
2017)—and Sunstein and Thaler’s nudge program, in particular—furthermore fail to explain
why alleged biases that afflict people in general (not to mention rent-seeking motives) do not
afflict paternalistic policy makers and choice architects.
Following Simon, March, Gigerenzer and others, one can point to compelling evidence
that high-performing organisms, humans, and organizations (e.g., Alphabet, Amazon, or the
New Zealand rugby team, the All Blacks) are not optimizers (e.g., Cyert and March, 1963;
Gage, 2012). Instead, they perform to a high level (i.e., succeed in their respective endeavors)
by using heuristics that are well-matched to the class of decisions and inference tasks in which
they are used.11 Note (once again) that ecological rationality measures “success”—of procedures
for making decisions or inferences and the social institutions that influence those decisions and
inferences—by domain-specific performance metrics (e.g., wealth, health, happiness, objective
accuracy, etc.).

Heterogeneity of beliefs and behavior as a public good

Although one should be cautious about attributing specific policy views to Simon, my reading
of him suggests a Simon-inspired approach to the design of institutions in general—and legal
and regulatory frameworks in particular—which rests on the principle that behavioral heterogen-
eity is itself a public good that confers many important social benefits which are often difficult
to observe directly. Based on this insight, one may consider heterogeneity itself to be a public
good and that efforts to defend against policies that risk encroaching upon heterogeneity are
worthwhile to consider in social-welfare analyses of nudging and other paternalistic policies.
In contrast, the nudging program and related work in behavioral law and economics prescribe
public policies aiming to induce greater conformity with “expert” views of what constitutes
optimal behavior (based on the biases and heuristics program). In so doing, it tends to view
behavioral heterogeneity as a problem requiring heterogeneity-reducing policy solutions.
By aiming to induce conformity with expert recommendations (e.g., Sunstein and Vermeule,
2008; Thaler and Sunstein, 2008), which are relatively static compared to the speed at which
the adaptive dynamics of decentralized systems generate value from individual experimenta-
tion with non-orthodox behavior and beliefs, the nudging program risks reducing individual
payoffs, as well as population-level robustness. The nudging program also risks blocking reve-
lation of valuable information generated by heterogeneous individual approaches to decision
making. In blocking the revelation of new information generated by heterogeneity, beneficial
social equilibria based on voluntary choice instead of paternalistic intervention—or worse,
authoritarian prescription and proscription—are disrupted. These social costs are separate
from any direct consideration of individual autonomy as a weighted term in the social welfare
function. Including autonomy or liberty as weighted terms in the social welfare function would
only compound losses from nudging.
Viale (forthcoming) argues that the hypothesized “System 1,” which uses fast, non-deliberative
heuristic shortcuts, is one mechanism through which “choice architects” who advocate nudging
policies such as defaults for organ donation believe that nudges affect decisions. Therefore, the
claim that these nudges are “libertarian” is weakened because the vitally important preservation
of a nudged decision maker’s choice set and his or her capacity to choose to not take up the
nudge (i.e., Rebonato’s (2012) so-called “reversibility” criterion, required for the libertarian
claim) is never engaged. (This is according to the choice architects’ own theory of why the

583
584

Nathan Berg

nudge will work.) The libertarian claim for nudging is therefore undermined and its coercive
intent revealed. Nudges that are effective because of how System 1 reacts to choice architects’
strategic framing stand in contrast to a second set of nudges (such as “cooling off” periods) that
are designed to cue System 2 (i.e., more deliberation and analytic reflection of benefits and
costs). Viale (2019) further analyses the implausibility of the dual theory of mind (i.e., Systems
1 and 2) upon which advocates of nudging rely.
Distinct normative interpretations of behavioral heterogeneity in Simon-inspired (as
opposed to Kahneman-Thaler-inspired) schools of behavioral economics rest on opposing
views about how stable the reward-generating environment is relative to the much slower
speed at which institutions can change policies that seek to paternalistically influence people
toward expert views of optimal behavior and belief. Is what we know about financial markets,
human physiology and risk taking really stable enough to justify nudging individuals toward
consensus views of optimal behavior in the current environment and given any currently avail-
able body of evidence, which nearly always include conflicting views and interpretations? Do
institutions that provide expert advice and influence public policy (e.g., recommending that
people invest more in the stock market, give more to charity, avoid sugar, and eat “healthy
fats”—whichever those turn out to be) wind up processing new information fast enough,
updating expert recommendations flexibly enough, and avoiding regulatory capture and the
influence of lobbyists12 reliably enough for us to have confidence that reducing heterogen-
eity by inducing conformity with expert recommendations is a good idea? These unsettled
questions reveal how different assumptions regarding the stability of experts’ partial knowledge
about the reward-generating play an important and largely unrecognized role in driving these
competing research programs within behavioral law and economics.

Heterogeneity forgone: costs and risks of nudging

In a world governed by a stable reward-generating environment that is simple enough to be
described with a scalar-valued objective function, and exhaustively known (possibly vector-
valued) constraint set which gives rise to a well-defined optimal choice (i.e., a “small world”
(Savage, 1954; Shafer, 1986)), it is indisputable (within the confines of such a model) that
deviations (i.e., suboptimal choices) are costly. Even in this case, however, where it can be
calculated how much is lost by deviating from optimal choice (or, aggregating over individuals,
lost social welfare), the costs of nudging programs and paternalistic intervention are not guaranteed
to be fully offset by aggregate social welfare improvements achieved by nudging a non-
optimizing population to optimize. Although gains are possible (as long as the payoff functions
and contributions of heterogeneity to payoffs are correctly specified), new largely unforeseen
risks emerge as a result of shifting the population profile of beliefs and actions toward greater
homogeneity.
Advocates of nudging do not seem to take into account how changes in the population dis-
tribution of beliefs and behavior could adversely affect the social welfare function. Instead, they
appear to rely on a tacit assumption of a Benthamite social welfare function under which payoff
improvements at the individual level (predicated on their explicitly maintained assumption that
individuals who respond to nudges will judge their own payoff to have improved as a result)
cleanly aggregate into social welfare improvements.
Yet another problem when assessing possible payoff improvements achieved by paternal-
istic intervention—from the perspective of process-dependent (rather than consequentialist)
preferences—concerns the loss of liberty and the subjective devaluation of payoffs associated
with an outcome when that outcome is mandatory rather than voluntary (cf. Conly, 2014).

584
585

Against nudging

If the social welfare gains hypothesized by choice architects are the result of non-deliberative
responses to nudges and fail the reversibility criterion emphasized by Rebonato (2012) and
Viale (2019, forthcoming), then we should consider whether those subpopulations with
process-dependent preferences that put positive weight on self-determination, autonomy, and
liberty might evaluate the resulting population profiles of nudged beliefs and actions less favor-
ably than if those same profiles were arrived at without nudging. Insofar as some members of
society do put weight on self-determination, autonomy and liberty, even a utilitarian social
welfare analysis should take these non-consequentialist aspects of preferences into account.
It is not difficult to imagine substantial violations of consequentialists’ invariance-over-
outcomes principle in subjective valuations attached to an outcome such as weight loss. The
decision maker can lose weight by voluntarily paying to enter a “fat farm” where food choice is
restricted. Or the decision maker could achieve an identical slimming outcome as the result of
a legislative mandate, government control over the distribution of food, or active and successful
nudging campaigns that change social norms and meanings associated with voluntarily
preventing obesity. It does not seem far-fetched that many people would have strict preferences
over an identical slimming outcome, ranking self-determined slimming (“I did it myself!”) over
market-assisted slimming (“I chose to go to the fat farm and achieved the desired result”) over
nudges (“I didn't notice that the desserts in the lunch room had been placed out of sight in the
corner, and it seems they have caused me to lose weight”) over coercive paternalism (“I stopped
eating as much after sin taxes raised the prices of foods I like to eat”). A successful slimming
nudge could reduce payoffs from the self-determined slimming outcome while achieving the
same outcome by nudging. The social value associated with the slimming outcome for some
likely depends on: (1) the population distribution of slimming outcomes, and (2) the process
by which the outcome was brought about (e.g., slimming has greater subjective value when
chosen autonomously rather than as the result of deception, coercion or command).
One could object to these points by recalling two justifications in Thaler and Sunstein
(2008). First, proponents of “libertarian paternalism” argue that if all elements of the choice set
remain available and a nudging policy only changes the default, or re-describes the choice set
with a new framing, then “choice” is preserved. Although influenced by a centralized power,
authority or government, Thaler and Sunstein claim that preservation of the availability of all
elements in the choice set should make us regard nudging policies as libertarian. If the preserved
items in the choice set are never considered or deliberated about (Rebonato, 2012; Viale, 2019,
forthcoming), however, or if the social and individual values associated with those elements in
the decision maker’s choice set are materially altered (as I argue in this chapter), then the liber-
tarian claim is unjustified.
A second justification put forward by Thaler and Sunstein, with which I agree, is that, in
many cases, there is no neutral description of the choice set. For example, in arranging food
items on a food buffet line, some item must be placed at the front. Similarly (their argument
goes), regarding organ-donor status, the law must take a non-neutral stand on opt-in (as in
Germany) versus opt-out (as in Austria). Granted that a non-neutral choice-influencing default
must be selected, there is still a collision of conflicting principles. The libertarian principles of
self-ownership and voluntarism clearly suggest that defaults should be set so that posthumous
use of one’s body reverts to the individual’s estate in the absence of any directives “opting in” to
organ-donor status. On the other hand, advocates of nudging claim that changing the default
helps more individuals realize an otherwise unexpressed preference to be an organ donor. Social
welfare analysis should at least acknowledge that there are conflicting normative principles and
reasonable people can disagree about the appropriate weights to place on each principle and
possible unintended consequences.

585
586

Nathan Berg

Another risk to consider more carefully before accepting the social welfare claims of those
advocating for more nudging is that expert consensus is mistaken, or that no one-size-fits-all
recommendation applies to everyone in the population. Expert recommendations are notorious
for flip-flopping over time. Tragic episodes where regulators have recommended medicines that
caused great harm (e.g., more than 10,000 disabled by thalidomide given to pregnant women
for several years) are hardly unknown. When policy makers set out to influence individuals to
make decisions with their money and health, risks of harm should be considered.
Honest errors and revision are to be expected as new medical, financial or behavioral research
contravenes previous recommendations. Then there are more nefarious strategic distortions,
such as the influence of lobbyists, rent seeking or deep capture. In my view, it makes little sense
to exclude consideration of policy errors and unintended consequences from social welfare ana-
lysis of proposals to introduce nudges as public policy.

Beyond “as-if ” to policy in a profoundly uncertain world

One of Gigerenzer’s key arguments against modeling choice and inference as if it were the
solution to a well-defined constrained optimization problem (Berg and Gigerenzer, 2010) is
the instability of the reward-generating environment. What basis of evidence about behav-
ioral “errors” could be relied upon for policy recommendations about savings, dietary choice,
and charitable giving in such environments? Even with a well-defined and stable reward dis-
tribution, if that distribution is sufficiently fat-tailed so that the theoretical mean and condi-
tional means do not exist (e.g., the Cauchy distribution), forming beliefs and basing actions on
averages and correlations from the past would not provide a stable and informative set of mental
processes to learn about the reward environment and make accurate predictions—let alone
make public policies that prescribe optimal behavior from on high (i.e., nudging people to
conform to expert recommendation).
This is not to say that public policy is impossible in such environments. Rather, Epstein
(1995) argues, in effect, that the Simon-inspired behavioral and law economics founded on
ecological rationality in un-learnably unstable and unknowably complex environments must rest on
principles of liberty and voluntary choice, while mitigating the most harmful negative exter-
nalities (e.g., murder and theft) based on long-established legal principles that otherwise allow
for wide-ranging preference and belief heterogeneity in decision making about money, health,
and information (e.g., Epstein, 2003, 2018).
If behavioral heterogeneity consisted of random deviations from expert recommendations,
then perhaps the lost payoffs from suboptimal decision making could be used as justification for
nudging. Because real-world deviations are, by and large, non-random and constituted by pur-
poseful (although oftentimes inconsistent) action, nudging programs that seek to limit or restrict
those deviations lack both normative and moral authority (Epstein, 1995, 2011, 2018). Epstein’s
arguments in favor of long-established legal traditions and restraint against complexifying the
legal environment with new policy objectives and their unintended consequences are largely
compatible with Simon.13
Gigerenzer (2016, p. 364) argues for education, drawing motivation from Simon’s (1985)
assertion that “people are generally quite rational; that is, they usually have reasons for what
they do:”

the dismal picture of human nature painted by behavioral economists and libertarian
paternalists is not justified by psychological research. Rather, it is largely the product
of narrow logical norms of rationality and selective reporting of the psychological

586
587

Against nudging

literature. Most important for public policy, by comparing cognitive illusions with
visual illusions, libertarian paternalists misleadingly suggest that attempts to liberate
people from their biases through education are largely doomed to fail. However,
as I will show, there is experimental evidence that even children can learn to deal
with risk and uncertainty—if they are taught how. I will conclude that democratic
governments should invest less in nudging and more in educating people to become
risk savvy.

This basic normative question is raised too infrequently: Which normative principle is the
basis for designing policies to “correct” alleged violations of rationality that are frequently often
cited in the behavioral economics literature? Many arguments in favor of nudging begin by
reciting extensive empirical evidence—which is indeed compelling and descriptively convin-
cing that such empirical patterns exist. This evidence, by and large, reveals that axioms, which
were originally used as technical criteria in representation theorems, are routinely violated.
Observed violations of consistency axioms are not in doubt. Rather it is the interpretation of
those important empirical findings that requires deeper investigation.
It may be the case that healthy and successful people do not typically have well-defined
domain-general preferences (i.e., no single objective function explains the decisions they make).
They may rather use a toolkit of simple action rules or heuristics. The use of a particular deci-
sion heuristic is thought to be applied when cued by context or situation. Without disputing
the damning evidence against the descriptive realisms of stable and consistent preferences, the
normative implications are, in contrast, far from straightforward.
Promoters of nudging argue that this evidence implies a need to roll out new policies to help
people avoid “errors.” This conclusion, however, reflects an odd normative interpretation given
to the observation that most people’s choice data do not conform to the strictures of perfect
internal logical consistency. Their argument is that nudging policies that help restore behavioral
consistency according to the neoclassical model of rationality will make people better off—or
better off “as judged by themselves” as Sunstein writes.

The authoritarian turn

Economists undertaking normative analysis could, at this point, instead dispense with the
assumption of stable internally consistent preferences which are required in both nudgers’ and
neoclassical revealed preference theorists’ welfare claims based on the common premise of stable
preferences represented by smooth utility functions that can be inferred or “recovered” based
on observed choice data. They could circumscribe the universality of axiomatic definitions of
rationality which, by their definitions, must hold across all choice domains. They could follow
Simon and allow for contextually-cued piecewise-defined objective functions, which would
mean applying distinct welfare criteria or multiple performance metrics depending on the
choice domain.
For example, many of us behave as if we are free riders in some domains and, at the same
time, inconsistently take the lead contributing to public goods in other domains (Kameda et al,
2011). Many of us are dogged own-payoff maximizers in some domains and generously pro-
social in others. Sometimes we expend great effort to think analytically, and sometimes we
use shortcuts, such as imitation, or follow intuition instead of cognitive analysis. Sometimes
we exhibit such inconsistency with self-awareness and reflection. Yet sometimes our incon-
sistent behavior is autonomic or unconscious. Setting aside neoclassical definitions of ration-
ality, an empiricist observing the regularity of inconsistent beliefs and behavior could more

587
588

Nathan Berg

easily interpret them as descriptive of normal, healthy and productive decision making than
pathological irrationality.
Behavioral economists could become more committed empiricists by observing the richly
heterogeneous human population and seeking to describe the multiple decision processes and
normative evaluations that people use to achieve success, thrive, and live “a life well-lived.”
Instead, the nudging program channels Orwellian visions of centralized control under guidance
by experts. This unfortunate exclusion of the multiplicity of decision processes in use, and the
multiplicity of relevant normative criteria, reveal an authoritarian turn in the intellectual his-
tory of behavioral economics. This authoritarian turn is based on inflexible and narrow norms.
Perhaps incidentally, the authoritarian turn also generates new demand for academics and
“expert advice.” Cynics could be forgiven for not writing it off as coincidence when they
observe increased reports of “irrationality” in behavioral economists’ academic findings and
concurrent increases in public money and power over others allocated to them (as policy
influencers, providers of research aimed at influencing policy, policy designers, choice architects,
and program administrators).
Behavioral economists discovered that observed choice data fail to conform to axiomatic
rationality. These axiomatic definitions of rationality require internal logical consistency above
all else (even though the reward-generating process is dynamic and unstable). They were first
used in utility representation theorems by Samuelson and von Neumann and Morgenstern, and
later re-interpreted by Kahneman-inspired behavioral economists as a normative standard of
rational or optimal behavior. There is little evidence that violations of axiomatic definitions of
rationality in choice data lead to measurable economic harm, and some evidence that deviations
from axiomatic rationality are positively associated with performance. Successful people (by
many metrics of success) violate the axiomatic definitions of rationality as well as those less
successful. Given that people’s choice data generally fail to conform to that axiomatic standard,
what should policy makers do about it? It would be surprising if many non-economists regarded
this brief synopsis of the stylized facts surrounding the Kahneman-Thaler program as strong
motivation for paternalistic intervention rather than revising the shortcomings of the narrow
norms used and expanding the way rational behavior is defined.

A Simon-inspired alternative
Following Simon’s notion of satisficing, ecological rationality requires that a behavioral rule or
heuristic—a procedure for making a decision or inference—satisfies a threshold condition (i.e.,
is “good enough”) when evaluated by a metric of performance that makes sense in the par-
ticular context in which it is used. This threshold requirement in ecological rationality provides
an objective standard that usually (but not always) allows for more behavioral heterogeneity
than in Kahneman-inspired models of rationality with prescriptive logical invariance. Although
optimization problems may also have a large number of solutions and allow for an arbitrarily
large degree of behavioral heterogeneity among optimizers, economic models used by those
who advocate nudging programs typically have narrower views on what constitutes successful
behavior—in decisions about retirement savings, diet, organ donation, charitable giving, etc.
This narrowness gives substance to the critique of nudging as reducing heterogeneity of beliefs
and behavior.
The standard of ecological rationality allows for multiple ways of being objectively successful.
For example, if there is a threshold level of wealth that is required to be financially viable, then it
follows that there is a plurality of purposeful decision processes that can achieve that criterion.
The problem of designing an effective reward environment (in both private organizations and

588
589

Against nudging

as public policy, e.g., legal institutions, restrictions on action in the legal code, fines, subsidies,
and—perhaps—strategic framings used to induce law-abiding behavior) defines a naturally
Simon-inspired approach to behavioral law and economics founded on the matching concept
(between heuristics and reward environments) that underlies ecological rationality. Given the
vast literature documenting legal institutions that fail to induce behavior consistent with policy
makers’ objectives (e.g., the War On Drugs aiming to reduce illicit drug use; Sarbanes-Oxley
aiming to fix systematic risk caused by “too big to fail” financial institutions, which arguably
led to increased market concentration in banking and finance since the Great Financial Crisis
of 2008–2009; and immigration policies across most G30 countries that often bring results
opposite to stated intentions), it would seem that more veridical study of decision processes—
and the challenge of designing robust rules of the game that enable value generated by inter-
action among the heterogeneous decision processes that individuals actually use—should be
a research priority. Positive externalities from behavioral heterogeneity are missing in most
social-welfare analyses of nudging. When policy is based on academic experts’ views of optimal
retirement savings behavior, optimal dietary choice, or optimal medical decision making, a
number of social costs are incurred that rarely enter into benefit-cost analyses of such programs.
Insofar as nudging succeeds in concentrating more of a population’s decisions on an
allegedly optimal decision, population-level behavioral heterogeneity is reduced. The popula-
tion loses portfolio diversification that would otherwise have been afforded by more heteroge-
neous retirement savings decisions. Nudging workers of a certain age into similar retirement
portfolios increases payoff variance in the event of a financial crisis. A generation of Japanese
retirement savers would have been far worse off had they been nudged out of low-return
“postal savings accounts” into professionally managed funds with greater exposure to equi-
ties in the NIKKEI Index during the 1980s—which peaked in December 1989 and traded
80 percent lower more than 20 years later and currently, after a long bull market, trades at
more than 40 percent below its peak. Japan’s aggregate retirement savings would have endured
greater volatility and sharply lower levels had Save More Tomorrow (as advocated by Thaler
and Sunstein’s (2008) nudging program) been used as a template for the design of Japan’s
retirement savings schemes.
One may argue that the importance of the Save More Tomorrow nudge was to increase
the level of savings rather than shift the composition of retirement savings accounts (i.e., how
retirement funds were invested). If the level of savings in retirement accounts is the sole nor-
mative criterion used to evaluate the nudge, then we can regard it as successful. If we expand
the normative criteria to include savers’ subjective judgment of their own consumption-
savings path, which includes the reduction in disposable income before retirement and related
hardships that Zywicki (2018) reports—not to mention stress-testing its effects on wealth
over time horizons that include a depression or 50 percent drawdown—then the question
of whether the nudge is objectively beneficial to savers remains ambiguous. For more on
species-, population- and individual-level benefits from heuristics that generate behavioral
heterogeneity rather than uniformity (at an optimum at one particular time in one particular
reward environment), see Bookstaber and Langsam (1985); Todd, Gigerenzer, and the ABC
Research Group (2012); Hertwig, Hoffrage, and the ABC Research Group (2013); Mousavi
and Kheirandish (2014).
As mentioned earlier, Zywicki (2018) reports evidence of individual-level harm inflicted
by Save More Tomorrow nudging programs. Controlling for other differences, savers enrolled
in Save More Tomorrow programs, on average, made larger contributions to their retire-
ment savings accounts by direct debit each month but carried larger credit card balances and
used other high-interest debt products such as payday lending at significantly higher rates

589
590

Nathan Berg

than un-nudged workers. Thus, the seemingly worthy goal of encouraging workers to save
more for retirement had unintended costs in the form of greater expenditures on interest and
financing fees—and possibly greater stress and reduced financial flexibility with less disposable
income.
Expert nutritional advice is famously contradictory, and perhaps justifiably so, given that
contradictory information is frequently revealed by new studies, which includes randomized
control trials, longitudinal population studies, and small-sample findings that bring new infor-
mation to light despite lacking statistical precision or external validity (also potentially valuable
in revealing previous mistakes and new breakthroughs). Optimal macronutrient composition
(fats, carbs, protein), for example, remains unknown and likely varies across people and within
person over time. More difficult questions would include: Which fats?, Which carbs?, Which
proteins?
Well-informed scientists frequently disagree, sometimes reaching opposite conclusions
based on the same available evidence. This disagreement, in turn, provides socially valuable
informational benefits for those who have access to open debate rather than facing top-down,
centralized one-size-fits-all food choice environments and incentive systems with sin taxes,
agricultural subsidies, regulation of food imports, etc., based on a static conceptualization of
“expert consensus.” One thing that most nutrition researchers seem to agree on, however, is
the inappropriateness and ineffectiveness of food pyramids promulgated by the U.S. and for-
eign governments. (e.g., the U.S. Department of Agriculture (USDA) introduced but then
halted publication of its “Eating Right Pyramid” in 1991, which was followed by subsequent
food pyramids, including “My Plate” (Rowland, 2016),14 each eliciting controversial responses
from nutrition experts and industry groups.) Nudging programs designed to influence what
food people choose risks inflicting harm by influencing choices toward food choices that will
subsequently be modified or contra-indicated. They also risk wasting information about dis-
covery of the best practices in food choice, which may not follow any simple algorithm at the
population level and, instead, require substantially different approaches to eating for different
people.

Against nudging
Sunstein (Chapter 38 in this volume) acknowledges that the welfare implications of nudging
are, in general, ambiguous. He asserts that there are three cases to consider. In the first case,
there are stable preferences and nudging is unambiguously helpful as judged by the individual’s
own standard. The second case involves a self-control problem where it is argued there are mul-
tiple selves with different preferences. In this case, Sunstein claims that nudging is once again
helpful by the individual’s own standard, so long as the individual has a clear meta-preference
(e.g., ranking the cool and sober self, who patiently deliberates about future-versus-future
intertemporal trade-offs, over the impulsive and emotionally “hot” self, who prefers an imme-
diate small payoff over a later larger payoff). In Sunstein’s third case, preferences are endogenous
to nudging, which therefore has ambiguous welfare implications. Sunstein claims that the first
two cases comprise a “large” set of choice domains that gives ample justification (and enthu-
siasm) for nudging programs. In the third case with ambiguous welfare implications, he calls for
case-by-case investigation into the welfare consequences, suggesting that they can be gleaned
by “explor[ing] what informed, active choosers typically select.”
Sunstein’s list of three cases is incomplete, however. There are omitted cases that come
up frequently: undefined preferences, undefined or dynamic discovery of new objectives,
and multiple context-specific objectives across which there may be no one-size-fits-all

590
591

Against nudging

nudging program whose social-welfare improvements exceed its expected costs from new
risks imposed. The claim that the welfare implications in cases 1 and 2 are unambiguously
positive can also be challenged, for example, with the example of Save More Tomorrow
and the observation that its application in other places and times would not have led to
social welfare improvements using an expanded set of reasonable performance metrics, such
as net wealth at time of retirement. If applied decades prior to a 50 percent draw-down
in stock market indexes and major dislocations in bond markets, as has occurred in mul-
tiple countries including the U.S. and Japan over the last 100 years, that nudge could have
materially damaged savers’ aggregate wealth and well-being, although it succeeded in its
narrow objective of increasing retirement savings rates during most working years. Given
the instability introduced by unconventional monetary policy since the GFC, who is to say
whether increased savings into 401k retirement plans will prove to be an enabler or detriment
to savers’ well-being?15 Serious risks of harming individual and aggregate well-being by intro-
ducing nudging programs as government policies should be considered with more attention
to new costs and risks they introduce.
Controlling impulsiveness is widely acknowledged as a genuine behavioral challenge by
people with a broad range of views on the advisability of nudging. It is instructive to recall
here that, in addition to impulsive under-saving, Strotz (1955) also considered the problem of
excessive saving among multiple time-inconsistent behavioral profiles in intertemporal choice
domains. For many of us, introspection reveals important cases—perhaps just as many (as one
wonders how these should be counted, weighted, or averaged)—in which excessive moder-
ation, or too much emotional cool, can hurt us, too.16 We sometimes regret staying cool and
deliberative and wish we had been more strongly moved by emotion. Proponents of nudging
and “System 1 versus System 2” dualism contend that System 2 is generally helpful and the
true seat of rationality. But there are times when we should have made noise and we erred by
not acting on emotion—not asking for a raise, not negotiating a better deal, or not giving into
impulsively and asking someone who would have been a great life partner for a phone number
for a first date.
Generalized prescriptions on the basis of behavioral economics at this early date, while
we still lack conclusive evidence of economic harms attributable to violations of consist-
ency axioms, would seem to reflect hubris on the part of behavioral economists. No doubt,
many advocates of nudging are genuine, and their concern for helping others avoid mistakes
is sincere. Given the risks of paternalistic policies uncovered already and lack of investiga-
tion into the unintended consequences of nudging and new paternalism, however, it would
seem that recalling some of the fundamental observations by classical economists would be
wise: social-welfare benefits from decentralization; heterogeneous endowments, preferences
and beliefs as fundamental underpinnings of socially beneficial exchange; and the corrupting
influence of power. These fundamentals generally encourage caution about introducing new
policies. To these should be added the link between population heterogeneity, on one
hand, and creativity and innovation on the other. Simon’s corpus of work alerts us to
many mechanisms by which successful adaptation can occur in the face of unstable reward
environments. Its implication is a behavioral law and economics that depends beneficially
on high-dimension heterogeneity of beliefs and behavior (e.g., Bennis et al., 2012; Berg,
2003, 2006, 2010, 2014b, 2017; Berg, Abramczuk, and Hoffrage, 2013; Berg, El-Komi
and Kim, 2016; Berg and Gabel, 2015, 2017; Berg and Hoffrage, 2008; Berg, Hoffrage and
Abramczuk, 2010; Berg and Kim, 2014, 2015, 2016, 2018a, 2018b, 2019; Berg and Maital,
2007; Berg and Murdoch, 2008; Dold and Schubert, 2018; Kameda et al., 2011; Rizzo and
Whitman, 2018).

591
592

Nathan Berg

Notes
1 This chapter draws substantially on Berg and Watanabe’s (2020) “Conservation of behavioral diver-
sity: Nudging, paternalism-induced monoculture, and the social value of heterogeneous beliefs and
behaviors.”
2 According to Simon:
The first consequence of the principle of bounded rationality is that the intended rationality of
an actor requires him to construct a simplified model of the real situation in order to deal with
it. He behaves rationally with respect to this model, and such behavior is not even approximately
optimal with respect to the real world.
1957, p. 198
Simon also states:
The first is to retain optimization, but to simplify sufficiently so that the optimum (in the sim-
plified world!) is computable. The second is to construct satisficing models that provide good
enough decisions with reasonable costs of computation. By giving up optimization, a richer set
of properties of the real world can be retained in the models.
1979, p. 498
3 For example, we can expect some proportion of the population to formulate different beliefs and take
different actions depending on whether expert advice is disseminated as “Consider this information and
then you decide what is best'” as opposed to “This decision has been structured to influence you to
choose what experts believe is best for you.”
4 In Law, Legislation and Liberty (1973, p. 35), Hayek wrote: “Order is an indispensable concept for the
discussion of all complex phenomena, in which it must largely play the role the concept of law plays in
the analysis of simpler phenomena.”
5 Zywicki (2018) reports that participants in the private nudging policy in the Save More Tomorrow
Program succeeded at contributing more to tax-advantaged 401k retirement accounts. On average,
those successfully nudged savers also wound up cash-strapped with significantly higher revolving
balances on credit cards and other high-interest-rate credit products such as payday lending. In his
evaluation, the evidence is far from clear that this nudge—when it worked as intended by inducing
greater contributions to 401s—improved the targeted population’s well-being. Of course, one can
attribute the increased use of high-interest-rate financing as another behavioral mistake requiring
further paternalistic regulation (and perhaps nudging) to discourage uptake. The observed pattern
of one nudge leading to unintended secondary problems is unlikely to resolve disagreements about
whether paternalism (or lack of more paternalism) is the root problem. The pattern does under-
score, however, the conspicuous absence of serious consideration of unintended consequences from
paternalistic interventions in the nudging literature that classical and neoclassical economists have
identified.
6 These three allegedly “reasonable” behaviors are examples of decision domains where advocates of
nudges have influenced real-world policy making despite remarkable disagreement among experts
regarding what constitutes a reasonable inference based on available evidence and decisions.
7 Forouhi et al. (2018) report in the British Medical Journal that there is controversy rather than evidenced-
based consensus regarding dietary recommendations on fat:
The medical literature is still full of articles arguing opposing positions. For example, in 2017,
after a review of the evidence, the American Heart Association Presidential Advisory strongly
endorsed that “lowering intake of saturated fat and replacing it with unsaturated fats, especially
polyunsaturated fats, will lower the incidence of CVD”. Three months later, the 18-country
observational Prospective Rural Urban Epidemiology (PURE) Study concluded much the
opposite: “Total fat and types of fat were not associated with cardiovascular disease, myocardial
infarction, or cardiovascular disease mortality”. ...
In the absence of long term randomised controlled trials, the best available evidence on which
to establish public health guidelines on diet often comes from the combination of relatively short
term randomised trials with intermediate risk factors (such as blood lipids, blood pressure, or body
weight) as outcomes and large observational cohort studies using reported intake or biomarkers of
intake to establish associations between diet and disease. Although a controversial practice, many,

592
593

Against nudging

if not most, public health interventions and dietary guidelines have relied on a synthesis of such
evidence ...
Although authorities still disagree, most consider that public health decisions should be made on
the weight of the available evidence, acknowledging its limitations, and seeking to obtain further,
better evidence when indicated. Equally important is to acknowledge when evidence is insufficient
to formulate any guidance, in which case all the relevant options should be clearly outlined to enable
informed choice.
8 In JAMA, Pace et al. (2014) report a newly updated (i.e., reversed) recommendation against default
screening with mammography of asymptomatic women under 50, given the large numbers of false
positives, biopsies with complications, anxiety, and overtreatment of cancers that would not have been
lethal. Similarly, Tikkinen et al. (2018) report a revised recommendation against PSA screening of
asymptomatic males for prostate cancer. In both cases, the revised guidelines suggest that the pros and
cons should be presented to patients who should then make their own decisions based on personal
weights applied to pros and cons of the test. Berg, Biele and Gigerenzer (2016) report on male
economists’ subjective beliefs about the PSA test.
9 According to Simon (1969, p. 53): “Human beings, viewed as behaving systems, are quite simple.
The apparent complexity of our behavior over time is largely a reflection of the complexity of the
environment in which we find ourselves.” And Simon states (1990, p. 1): “ Human rational behaviour
is shaped by a scissors whose blades are the structure of task environments and the computational cap-
abilities of the actor.”
10 Behavioral economics often identifies rationality with the axioms of internal logical consistency used in
neoclassical theory’s utility representation theorems (e.g., Samuelson’s use of transitivity; von Neumann
and Morgenstern’s use of the continuity and independence axioms; Kolmogorov axioms and Bayes
rule as used in subjective probability theory). Identification of rationality with internal logical consist-
ency underpins the biases and heuristics research program inspired largely by Kahneman, which focuses
on deviations from neoclassical rationality axioms and its prescriptive program of inducing greater
behavioral conformity with those consistency requirements (de-biasing as in Jolls, Sunstein, and Thaler
(1998) or Thaler and Sunstein’s (2008) Nudge; cf. Sheffrin, 2017).
11 See, for example, Gigerenzer and Selten (2001), Smith (2003), Berg and Gigerenzer (2010), Berg
(2014a) and Mousavi and Kheirandish (2014) for definitions of ecological rationality and research
programs based on Herbert Simon’s seminal work on bounded rationality.
12 Hawkes (2018) reports allegations of influence from lobbyists representing pharmaceutical firms and
complaints about removal of board members from Cochrane (a widely respected UK charity and
Limited Liability Company). Many observers regard Cochrane as one of most influential and well-
executed institutions designed to objectively evaluate drugs and medical procedures with evidence-
based reasoning and sophisticated meta-analyses of the medical research literature. These allegations
and recent controversy among former Board members of the Cochrane are just one example of how
vulnerable attempts at evidenced-based medicine (and policy making) are to rent-seeking and non-
transparent influence.
13 Of course, we cannot be sure that people’s deviations are purposeful, given Hayek’s “irremediable
ignorance” about other people and the causal forces that structure various choice environments we
encounter. Neither can advocates of nudging be sure that deviations are harmful or lack purpose
and value.
14 The Chairman of Nutrition at the Harvard School of Public Health is quoted as saying: “Unfortunately,
like the earlier U.S. Department of Agriculture pyramids, My Plate mixes science with the influence
of powerful agricultural interests, which is not the recipe for healthy eating.”
15 In financial crises, the correlation of most paper assets (e.g., bonds and stocks) approaches 1. In some
instances, “safe” treasury bonds and “risky” stocks both decline, and the restricted universe of financial
assets in many pension accounts may not provide an effective hedge against such risks. For those who
assign greater probabilistic or subjective weight to wealth preservation in those states of the world, the
nudge into greater exposure to the restricted asset classes offered in 401k accounts may hurt rather than
help investors.
16 For example, Presidential Candidate Michael Dukakis’ dispassionate response to journalist Bernard
Shaw’s question “about whether he would support the death penalty should his wife, Kitty, be raped
and murdered” (Shepley, 2019) hurt Dukakis despite its logical consistency and valid evidence from his
home State of Massachusetts about the death penalty’s weak deterrent effect.

593
594

Nathan Berg

References
Benartzi, S. and Thaler, R. H. (1995). Myopic loss aversion and the equity premium puzzle. Quarterly
Journal of Economics 110(1), 73–92.
Bennis, W., Katsikopolous, K., Goldstein, D., Dieckmann, A., and Berg, N. (2012). Designed to fit
minds: Institutions and ecological rationality. In P. M. Todd, G. Gigerenzer, and the ABC Research
Group (Eds.), Ecological Rationality: Intelligence in the World (pp. 409–427). New York: Oxford
University Press.
Berg, N. (2003). Normative behavioral economics, Journal of Socio-Economics 32, 411–427.
Berg, N. (2006). A simple Bayesian procedure for sample size selection in an audit of property value
appraisals, Real Estate Economics 34(1), 133–155.
Berg, N. (2010). Behavioral economics, In R. C. Free (Ed.), 21st Century Economics: A Reference Handbook,
(vol. 2., pp. 861–872). Los Angeles, CA: Sage.
Berg, N. (2014a). The consistency and ecological rationality schools of normative economics: Singular
versus plural metrics for assessing bounded rationality, Journal of Economic Methodology 21(4),
375–395.
Berg, N. (2014b). Success from satisficing and imitation: Entrepreneurs’ location choice and implications
of heuristics for local economic development, Journal of Business Research 67(8), 1700–1709.
Berg, N. (2015). Gerd Gigerenzer. In M. Altman (Ed.), Real World Decision Making: An Encyclopedia of
Behavioral Economics. Santa Barbara, CA: Praeger, ABC-CLIO.
Berg, N. (2017). Smart people’s rational mistakes. In M. Altman (Ed.), Handbook of Behavioral Economics and
Smart Decision-Making: Rational Decision-Making Within the Bounds of Reason (pp 43–67). Cheltenham:
Edward Elgar Publishing.
Berg, N. (2018). Decentralization mislaid: On new paternalism and skepticism toward experts, Review of
Behavioral Economics 5(3–4), 361–387. http://dx.doi.org/10.1561/105.00000099
Berg, N., Abramczuk, K., and Hoffrage, U. (2013). Schelling’s neighborhood segregation model with
FACE-Recognition. In R. Hertwig, U. Hoffrage, and The ABC Research Group (Eds.), Simple
Heuristics in a Social World (pp. 225–257). New York: Oxford University Press.
Berg, N., Biele, G., and Gigerenzer, G. (2016). Consistent Bayesians are no more accurate than non-
Bayesians: Economists surveyed about PSA. Review of Behavioral Economics (ROBE) 3(2), 189–219.
http://dx.doi.org/10.1561/105.00000034
Berg, N., El-Komi, M., and Kim, J. Y. (2016). The puzzle of uniform standards and market segmentation
among Islamic banks. Journal of Economic Behavior and Organization 132, 39–49.
Berg, N. and Gabel, T. (2015). Did Canadian welfare reform work? The effects of new reform strategies
on Social Assistance participation, Canadian Journal of Economics 48(2), 494–528.
Berg, N. and Gabel, T. (2017). Did tax cuts on earned income reduce Social Assistance participation in
Canada? Australian Tax Forum 32(1).
Berg, N., and Gigerenzer, G. (2007). Psychology implies paternalism?: Bounded rationality may reduce
the rationale to regulate risk-taking, Social Choice and Welfare 28(2), 337–359.
Berg, N., and Gigerenzer, G. (2010). As-if behavioral economics: Neoclassical economics in disguise?,
History of Economic Ideas 18(1), 133–166.
Berg, N., and Hoffrage, U. (2008). Rational ignoring with unbounded cognitive capacity, Journal of
Economic Psychology 29, 792–809.
Berg, N., Hoffrage, U., and Abramczuk, K. (2010). Fast Acceptance by Common Experience: FACE-
recognition in Schelling’s model of neighborhood segregation, Judgment and Decision Making 5(5),
391–410.
Berg, N, and Kim, J. Y. (2014). Prohibition of Riba and Gharar: A signaling and screening explanation?,
Journal of Economic Behavior and Organization 103, 146–159.
Berg, N, and Kim, J. Y. (2015). Quantity restrictions with imperfect enforcement in an over-used
commons: Permissive regulation to reduce over-use?, Journal of Institutional and Theoretical Economics
171(2), 308–329.
Berg, N, and Kim, J. Y. (2016). Equilibrium national border and its stability, Prague Economic Papers 25(6),
637–654.
Berg, N., and Kim, J. Y. (2018a). Plea bargaining with multiple defendants and its deterrence effect.
International Review of Law and Economics 55, 58–70.
Berg, N., and Kim, J. Y. (2018b). Free expression and defamation. Law, Probability and Risk 17(3), 201–223.
Berg, N., and Kim, J. Y. (2019). A good advisor. Bulletin of Economic Research 71(3), 558–572.

594
595

Against nudging

Berg, N., and Maital, S. (2007). Tailoring globalization to national needs and wellbeing: One size never
fits all, Global Business and Economics Review 9(2/3), 319–334.
Berg, N., and Murdoch, J. (2008). Access to grocery stores in Dallas, International Journal of Behavioural and
Healthcare Research 1(1), 22–37.
Berg, N. and Watanabe, Y. (2020). Conservation of behavioral diversity: Nudging, paternalism-induced
monoculture, and the social value of heterogeneous beliefs and behaviors. Mind and Society 19, 103–120.
Bookstaber, R., and Langsam, J. (1995). On the optimality of coarse behavior rules. Journal of Theoretical
Biology 116(2), 161–193.
Brown, P., and Knowles, S. (2019). Cash is not king in incentivising online surveys. Working Paper,
University of Otago. Available at: www.otago.ac.nz/economics/otago707883.pdf
Conly, S. (2014). Against Autonomy. Oxford: Oxford University Press
Cyert, R., and March, J.G. (1992 [1963]). A Behavioral Theory of the Firm (2nd ed.). Oxford: Blackwell.
Dold, M. F., and Schubert, C. (2018). Toward a behavioral foundation of normative economics, Review of
Behavioral Economics 5(3–4), 221–241. http://dx.doi.org/10.1561/105.00000097
Eckel, C., Grossman, P. J., and Johnston, R. M. (2005). An experimental test of the crowding out hypoth-
esis. Journal of Public Economics 89(8) 1543–1560.
Epstein, R. A. (1995). Simple Rules for a Complex World. Cambridge, MA: Harvard University Press.
Epstein, R. A. (2003). Skepticism and Freedom: A Modern Case for Classical Liberalism. Chicago: University
of Chicago Press.
Epstein, R. A. (2011). Design for Liberty: Private Property, Public Administration, and the Rule of Law.
Cambridge, MA: Harvard University Press.
Epstein, R. A. (2018). The dangerous allure of libertarian paternalism, Review of Behavioral Economics
5(3–4), 389–416.
Forouhi, N. G., Krauss, R. M., Taubes, G., and Willet, W. (2018). Dietary fat and cardiometabolic
health: Evidence, controversies, and consensus for guidance, British Medical Journal 361: k2139.
Gage, D. (2012). The venture capital secret: 3 out of 4 start-ups fail. Wall Street Journal February 18.
Available at: www.wsj.com/articles/SB10000872396390443720204578004980476429190
Gigerenzer, G. (2016). On the supposed evidence for libertarian paternalism. Review of Philosophy and
Psychology 6, 361–383.
Gigerenzer, G., and Selten, R. (2001). Rethinking rationality. In G. Gigerenzer and R. Selten (Eds.),
Bounded rationality: The adaptive toolbox (pp. 1–12). Cambridge, MA: The MIT Press.
Gigerenzer, G., Todd, P. M., and the ABC Research Group (Eds.) (1999). Simple Heuristics that Make Us
Smart. New York: Oxford University Press.
Harrison, G. W., and Ross, D. (2018). Varieties of paternalism and the heterogeneity of utility structures.
Journal of Economic Methodology 25(1), 42–67.
Hawkes, N. (2018). Cochrane director’s expulsion results in four board members resigning, British Medical
Journal 362. https://doi.org/10.1136/bmj.k3945.
Hayek, F. A. (1973). Law, Legislation and Liberty. New York: Routledge. Available at: https://libsa.files.
wordpress.com/2015/01/hayek-law-legislation-and-liberty.pdf
Hertwig, R., Hoffrage, U., and the ABC Research Group (Eds.) (2013). Simple Heuristics in a Social World.
New York: Oxford University Press.
Infante, G., Lecouteux, G., and Sugden, R. (2016). Preference purification and the inner rational agent: a
critique of the conventional wisdom of behavioural welfare economics. Journal of Economic Methodology
23(1), 1–25.
Jolls, C., Sunstein, C. R., and Thaler, R. (1998). A behavioral approach to law and economics. Stanford
Law Review 50, 1471–1480.
Kameda, T., Tsukasaki, T., Hastie, R., and Berg, N. (2011). Democracy under uncertainty: The wisdom
of crowds and the free-rider problem in group decision making, Psychological Review 118, 76–96.
Kropotkin, P. (1892). The Conquest of Bread. Chapter IV. Available at: http://dwardmac.pitzer.edu/
anarchist_archives/kropotkin/conquest/ch9.html
Mellström, C., and Johannesson, M. (2008). Crowding out in blood donation: Was Titmuss right? Journal
of the European Economic Association 6(4), 845–863, https://doi.org/10.1162/JEEA.2008.6.4.845
Mousavi, S., and Kheirandish, R. (2014). Behind and beyond a shared definition of ecological ration-
ality: A functional view of heuristics. Journal of Business Research 67, 1780–1785.
Pace, L. E. et al. (2014). A systematic assessment of benefits and risks to guide breast cancer screening
decisions. JAMA 311(13), 1327–1335.
Rebonato, R. (2012). Taking Liberties. New York: Palgrave Macmillan.

595
596

Nathan Berg

Rizzo, M. J., and Whitman, G. (2018). Rationality as a process, Review of Behavioral Economics 5(3–4),
201–219. http://dx.doi.org/10.1561/105.00000098
Rowland, M. P. (2016). The food pyramid of the future, Forbes, November 2. Available at: www.forbes.
com/sites/michaelpellmanrowland/2016/11/02/the-food-pyramid-of-the-future/#50d3d2324da3
Savage, L. J. (1954). The Foundations of Statistics. New York: Wiley.
Shafer, G. (1986). Savage revisited, Statistical Science 1(4), 463–501.
Sheffrin, S. M. (2017). Behavioral law and economics is not just a refinement of law and economics,
Oeconomia 7(3), 331–352.
Shepley, M. J. (2019). Dukakis’ deadly response. Time. Available at: http://content.time.com/time/
specials/packages/article/0,28804,1844704_1844706_1844712,00.html
Simon, H. A. (1957). Models of Man. Chichester: John Wiley.
Simon, H. A. (1969). The Sciences of the Artificial. Cambridge, MA: MIT Press.
Simon, H. A. (1979). Rational decision making in business organizations, Nobel Memorial Lecture 1978.
American Economic Review 69(4), 493–513.
Simon, H. A. (1985). Human nature in politics: The dialogue of psychology with political science.
American Political Science Review 79, 293–304.
Simon, H. A. (1990). Invariants of human behavior. Annual Review of Psychology 41, 1–19.
Smith, V. L. (2003). Constructivist and ecological rationality in economics. American Economic Review
93(3), 465–508.
Strotz, R. H. (1955). Myopia and inconsistency in dynamic utility maximization, The Review of Economic
Studies 23(3), 165–180.
Sudgen, R. (2016). Do people really want to be nudged towards healthy lifestyles? International Review of
Economics.
Sugden, R. (2018). Paternalism and entrepreneurship, Review of Behavioral Economics 5(3–4), 243–259.
http://dx.doi.org/10.1561/105.00000089
Sunstein, C. R. (2014). Why Nudge? New Haven, CT: Yale University Press.
Sunstein, C. R. (2016a). The Council of Psychological Advisers. Annual Review of Psychology 67,
713–737.
Sunstein, C. R. (2016b). The Ethics of Influence. Cambridge: Cambridge University Press.
Sunstein, C. R., Reich, L., and Rauber, J. (2017). Behavioral insights all over the world? Public
attitudes toward nudging in a multi-country study, available at https://papers.ssrn.com/sol3/papers.
cfm?abstract_id=2921217
Sunstein, C. R., and Reisch, L.A . (2014). Automatically green: Behavioral economics and environmental
protection. Harvard Environmental Law Review 38, 127–158.
Sunstein, C., and Vermeule, A. (2008). Conspiracy theories. Harvard Public Law Working Paper No.
08–03; U of Chicago, Public Law Working Paper No. 199; University of Chicago Law & Economics,
Olin Working Paper No. 387.
Thaler, R. H., and Benartzi, S. (2004). Save More Tomorrow™: Using behavioral economics to increase
employee saving, Journal of Political Economy 112, S164–S187.
Thaler, R. H., and Sunstein, C. R. (2008). Nudge: Improving Decisions about Health, Wealth, and Happiness.
New Haven, CT: Yale University Press.
Tikkinen, K. A. O., et al. (2018). Prostate cancer screening with prostate-specific antigen (PSA) test: a
clinical practice guideline. British Medical Journal 362: k3581.
Todd, P. M., Gigerenzer, G., and the ABC Research Group (Eds.) (2012). Ecological Rationality: Intelligence
in the World (pp. 409–427). New York: Oxford University Press,
Viale, R. (Ed.) (2001). Knowledge and Politics. Berlin: Springer.
Viale, R. (Ed.) (2012). Methodological Cognitivism, vol. 1. Berlin: Springer.
Viale, R. (2019). Architecture of the mind and libertarian paternalism: Is the reversibility of System 1
nudges likely to happen? Herbert Simon Society Working Paper.
Viale, R. (forthcoming). The normative and descriptive weaknesses of behavioral economics-informed
nudge: Depowered paternalism and unjustified libertarianism. Mind and Society.
Viale, R., and Pozzali, A. (2010). Complex adaptive systems and the evolutionary triple helix. Critical
Sociology 36(4), 575–594. https://doi.org/10.1177/0896920510365923
Zywicki, T. J. (2018). The behavioral economics of behavioral law & economics, Review of Behavioral
Economics 5(3–4), 439–471. http://dx.doi.org/10.1561/105.00000094

596
597

41
BOUNDED RATIONALITY
IN POLITICAL SCIENCE
Zachary A. McGee, Brooke N. Shannon, and Bryan D. Jones

Bounded rationality’s origins and principles

The father of bounded rationality, Nobel Prize-winner Herbert A. Simon was principally
a political scientist with a special focus on public administration. Simon saw his scientific
career as being motivated by the development of two big ideas, both originally developed in
Administrative Behavior, bounded rationality and organizational identification1 (Simon 1996).
We will discuss each idea in turn and then show how it applies to different literatures in pol-
itical and policy sciences. Administrative Behavior, a version of Simon’s doctoral dissertation at
the University of Chicago, argued that organizations need to be understood in terms of their
decision-making processes (Simon 1945). Put another way, Simon thought the limits of cog-
nitive processing must apply to individuals within organizations and those effects would carry
over into how organizations operate. This book essentially combined bounded rationality and
the study of organizations to explore the processes that determine outputs similar to the study
of rational choices and decision-making processes at the individual level (Jones 1997; Ostrom
1998; Selten 2001).
Bounded rationality is, in part, based on the limitations that all actors have in cognitive
processing. What does Simon mean when he refers to limitations of cognitive processing?
A full exposition is beyond the scope of this chapter,2 but central to his explanation are the
failure to have perfect knowledge, imprecision in anticipation of consequences, and emotions
related to imagined consequences, all of which contribute to our shortcomings as individual
utility maximizers (Simon 1945). Limitations in cognitive processing explain the bounds of the
individual’s ability to process the overabundance of information, left to them to distill. Bounded
rationality is the relationship between the individual’s information processing ability, complicated
by ever-present limitations and the complexity of the problems faced (Bendor 2003).
Complexity alone does not limit individual information processing in bounded rationality.
The uncertainty of the decision-making environment plays an important role as well. Such
uncertainty depends, in part, on whether decisions are being made in so-called “small world”
or “large world” environments (Savage 1954). Consider a game of chess, the decision for which
piece to move next and to where may be complex but have a finite and knowable number of
variables. This describes a small world where choices may be risky (i.e., reliably assigned a prob-
ability) but not particularly uncertain (Knight 1921; Binmore 2006). The political world, and

597
598

Zachary A. McGee et al.

the policy process, are large worlds. These so-called large worlds have significant uncertainty
because many alternatives are unknowns and, in fact, unknowable. Therefore, decision making
is not only a problem of computational difficulty and risk, as in a small-world chess game, but
also has the added caveat that some outcomes will never be assessed. We will return to these
principles shortly.
Bounded rationality challenges the central set of assumptions outlined by comprehensive
rationality, also known as rational choice theory, which is the standard for formal models of pol-
itics. Rational choice hinges on behaviors of the individual, primarily focusing on vote choice
in elections and legislatures. The behavioral theory of comprehensive rationality has a durable
tradition in political science, conceptualizing individuals as rational, calculating actors who avail
themselves of the entirety of available information regarding alternatives. Bounded rationality
challenges these assumptions, fundamentally the infinite abilities of humans to recognize and
process all information available in their environment.
Realistically, no human has the boundless ability to process information as a computer does.
Human rationality is bounded both by limited cognitive abilities but also by the brain’s limited
size and speed, “but it can often function well by making coherence judgments rather than
overextending itself by making too many deductions and calculations” (Thagard 2018, p. 8).
Due to biological constraints as well as the complex environment humans make decisions in,
humans can exclusively focus attention on only a small number of issues at once. For these
reasons, decisions influenced by a complicated, or uncertain, environment are unable to meet
the criteria for rationality under rational choice. The implication for political decisions is that
an individual’s ability to make rational responses to their environment can fail, particularly at
critical moments and with complicated problems. This mismatch characterizes bounded ration-
ality (Simon 1996; Jones 1999).

Bounded rationality’s influence on political science

In political science, there are two main branches of research utilizing bounded rationality
(Bendor 2003). We focus here on the first branch, which retains Simon’s name and his core
argument that cognitive limitations only show through in a person’s decision making when the
problem they are handling is sufficiently difficult.3 That is, we can expect people to be able
to solve simple problems in limited choice environments rationally, but beyond simple choice
environments, our decision-making is impaired. The Simonian branch of bounded rationality
is also heavily influenced by Charles Lindblom’s “The Science of Muddling Through” (1959),
imagining decision making as a set of successive limited approximations to problems rather than
a comprehensive analysis of alternatives (Bendor 2015).
As time went on, bounded rationality gained acceptance in political science, especially as
the discipline trended increasingly toward empirical work (Jones 1999). Lindblom (1959) builds
on Simon’s theory by characterizing decision making in policy formulation as a metaphorical
tree’s branches and roots. Comprehensive rationality takes a root approach; solving complex
problems, formulating policy, or pulling a tree up from the root requires comprehensive action.
The means to achieve any end must include all possible alternatives, isolating the means is
necessary to make a rational decision toward an end. In this way, “analysis for the root method
is comprehensive; every important relevant factor is taken into account” (Lindblom 1959). As
one may imagine, creation or destruction by this method is drastic and not necessarily an incre-
mental process.
Bounded rationality, on the other hand, assumes the branch method, for what Lindblom calls
successive limited comparisons. Like a tree’s branches building out from the center structure,

598
599

Bounded rationality in political science

branch analysis of policy formation improves the central problem in small steps, adding to an
initial pre-existing foundation. As it relates to the policy process, rationality is bounded by
a number of limitations, including cognitive processes and time constraints. For this reason,
policy processes most often subscribe to the branch method, as policy making builds on itself
in an endless cycle of formulating new policies and improving existing ones. As Simon fam-
ously argued, “it is impossible for the behavior of a single, isolated individual to reach any high
degrees of rationality” (1997 [1945]], p. 92). Contrasted with comprehensive rationality as root
analysis which relies heavily on its own theoretical framework and builds from the past only
as embodied in the theory (Lindblom 1959), bounded rationality establishes new foundations
for rationality by including institutional constraints, limited processing capability, and values.
As Wildavsky (1964) showed, boundedly rational decision making led to incremental policy
changes.

An institutional bridge between the individual and organizations

An institutional focus provides the link between Simon’s emphasis on human nature and
organization theory. The term “institution” refers to the rules that govern different kinds of
organizations (and the norms that solidify them), as well as the legitimacy given to them.
This legitimacy is invisible and tangible first through shared understanding and meaning,
and more importantly through policy (Ostrom 1998; 2009). Political scientist Elinor Ostrom
argued for a bounded rationality frame on institutional political science because it better
reflects realistic politics. For example, in situations where an individual’s preferences are
not maximized but must nevertheless reach an outcome, rational choice predicts no action
toward cooperation. Although action is not predicted, it regularly occurs; rational choice
predictions fall flat here.
Institutions are able to bridge the divide between schools of thought in bounded rationality
because they (exist) in the political environment and hold the norms and formal rules of action
(Ostrom 2009). Bounded rationality adapts its rational choice predecessor to include the envir-
onment in which individuals make decisions, comprised of its institutions. Simon (1945) and
Lindblom (1959) point to the relative success of the majority to handle simple tasks, and note
that cognitive limitations show through for difficult situations. For simple problems, most make
decisions rationally. Institutions support individuals’ capacities to do better in terms of rational
decision making, and minimize mistakes made with even simpler problems, and in this way
bridge the two schools of thought in bounded rationality.
While much of the understanding of bounded rationality is rooted in individual decision-
making processes, Simon’s original target was organizational behavior. We argue for an
understanding of institutions as a connection between the two schools of thought in bounded
rationality because institutions moderate and inform individuals’ behavior, to help them solve
problems.
Understanding the institutional connection between the individual and organization helps
to provide a more robust understanding of choice. Institutions exist to facilitate successes that
would be impossible by individuals acting alone. In the more pessimistic view of bounded
rationality, Tversky and Kahneman develop a view of actors that shows a dependence on
mental shortcuts, or heuristics, that allow lower-information actors to behave like actors who
have more knowledge of the process and alternatives. The role of heuristics in political decision
making is to help citizens make up for a lack of information with mental shortcuts.
For the general public, heuristics help individuals make rational decisions in vote choice in
elections. For policy makers, heuristics seek to benefit leaders in developing policy solutions

599
600

Zachary A. McGee et al.

to complex problems. Tversky and Kahneman (1974) point to the mistakes some make even
in simple tasks and the prevalence of heuristics and bias in decision making for even highly
knowledgeable individuals, and the universal dependence on mental shortcuts or heuristics, to
help even low-information individuals make rational decisions (Gigerenzer et al. 1999; Boyd
and Richerson 2001; Tversky and Kahneman 1974; Sniderman, Brody, and Tetlock 1991;
Cherniak, Chapter 24 in this volume). We argue for a refocus on the role institutions play in
political decision-making in part due to Simon’s emphasis on one’s environment during the
decision-making process. This refocus further underlines the decision-making processes of both
institutions and organizations in human nature.
Tversky and Kahneman show the disadvantages of bounded rationality. Heuristics are used
by nearly everyone in their experiments, and their use is not restricted to low-information
laymen (Tversky and Kahneman 1974, p. 1130). Intuitive thinking, a distinctive feature of
bounded rationality, leads nearly all individuals to fail in rational decision making. These
experiments conducted by Tversky and Kahneman are examples of risky environments and
these same biases may become adaptive tools in uncertain, but more realistic, environments.
Furthermore, when people can learn and have environmental feedbacks from their actions,
they may also show unexpected convergent conformity. Even in uncertain environments, how-
ever, people will overestimate how much they know and rely again on heuristics and simple
decision rules when making decisions.
Norms, “the result of shared notions of appropriate behavior and the willingness of indi-
viduals to reward appropriate behavior and punish inappropriate behavior” (Boyd & Richerson
2001; McAdams 1997), can also dictate behavior for individuals. Much of the extant literature
on norms focuses on public goods and coordination dilemmas (March & Olsen 1984; Ostrom
1998, 2000). The institutional link with political heuristics is found in norms; the norm of util-
izing heuristics is due to cost and risk. Simply put, it is easier to use shortcuts or copy others
to determine the best behaviors, especially in complex environments and for individuals facing
difficult decisions (Boyd & Richerson 2001). Instead of inventing a new behavior of each
instance to test its utility, humans observe and imitate others (Boyd & Richerson 2001), akin to
Lindblom’s branch method of decision-making.
Institutions represent a link between the two schools of thought in bounded rationality.
As Simon believed, in spite of cognitive limitations and attention constraints, some do very
well and most do reasonably well in making rational decisions––even for very complex
issues. Simon viewed organizations as adaptive, structuring choices made by individuals.
Most importantly, hierarchies and specialization of labor allow for tasks to be accomplished
that would be impossible for boundedly rational individuals. The institutions in government,
with rules and processes for creating and passing policy, are based on rules and norms in the
formal branches of government, interest groups, and bureaucracy (May 1991; Wagner 2010).
Institutions allow individuals to participate, learn policy processes, and specialize attention
to prioritize, which can serve to support and augment individuals’ abilities to make rational
political decisions with more information than would be available for individuals. Because
cognitive limits show through in decision-making processes most when those processes are
exceedingly complex, institutions are capable of easing the difficulty in specializing. When
organizations specialize, full attention can be given to a specific problem within the noisy
environment. Lindblom’s branch method is assisted by institutions as well, as previous
decisions constitute the base to which new decisions are added in incrementalism. The base is
sustained by institutions, a key example being budgetary decisions, which will be considered
in depth below.

600
601

Bounded rationality in political science

Coming to prominence: bounded rationality and theories of the

policy process
In the policy process literature, scholars took Simon’s words to heart and began assuming that
policy makers were boundedly rational actors, and like all individuals, the choices they made
were rooted in a mixture of self-interest, budgetary constraints, and the demands of their con-
stituency. The three considerations do not always mirror each other, and policy consideration
typically reflects compromise. Because policy makers are rational but constrained by capacity,
the institutions of government, and re-election, comprehensive rationality, as a model for policy
makers, is too basic a theory to explain policy processes. Policy makers respond to both internal
and external pressures when faced with policy considerations. The external environment puts
pressure on actors and how one acts is a reflection on the environmental incentives, including
institutional pressures from government actors and constituency, and internally, policy makers
act based on preconceptions that make up preferences that may cause deviations from the
external environment (Simon 1996; Jones 1999).
Bounded rationality is the microfoundation for agenda setting and policy processes litera-
ture because it expands the shared characteristics of individuals and institutions. Assuming at
once that “actors are goal-oriented and takes into account the cognitive limitations of decision
makers in attempting to achieve these goals” (Jones 1999), the theory of bounded rationality
can easily be expanded to include institutions within the theoretical framework. Much like
organizations described in Administrative Behavior, institutions are made up of individuals. This
seems at once obvious but is fundamental to understanding actions in policy processes, because
organizational studies rely on organizational behavior mimicking the behavior of individuals
within the institutions (March 1994).
People in the policy process are boundedly rational goal-seeking actors who have limited
cognitive abilities and capacities for processing information (Jones 2001). Critically important
in policy process theories is attention allocation. Of course, if the environment is uncertain,
it will not be an inability to allocate attention that limits decision making but instead the
reality that some outcomes will never even be considered. The limited capacity of decision-
makers to allocate attention in risky environments guarantees actors will miss issues they should
pay attention to while building agendas to prioritize issues (Baumgartner and Jones 1993).
Institutions like interest groups, political parties, and congressional committees expand the
application of bounded rationality to the macro level of analysis. Bounded rationality is the
microfoundation of policy processes because it maps individual theories of decision making,
information processing, and agenda setting onto institutions and systems (Jones 2017; Jones and
McGee 2018). The Lindblom-Wildavsky model of choice led to incremental policy change,
but adding attention allocation, which must be disjoint and episodic, leads to punctuated equi-
librium (Jones and Baumgartner 2005).
A definitive characteristic of American governmental decision making is its slow pace of
change. Members of Congress are boundedly rational individuals, and they comprise the
boundedly rational institution of Congress. If the status quo of the policy process is no change
in policy, then taking action to incite is risky. Ostrom frames the policy process in government
as a classic rational choice problem (Ostrom 1998). Collective actions arise when decisions are
difficult; any option chosen will lead to short-term benefits for self-interests but ultimately
leaves everyone worse off than before (Ostrom 1998). Due to the boundedly rationality and
omnipresent self-interest of members of Congress, these social dilemmas occur in politics regu-
larly, but theories of comprehensive rationality are unable to explain how decisions are made
in them.

601
602

Zachary A. McGee et al.

Bounded rationality is a more useful theory for explaining legislators’ actions in social
dilemmas; action is taken although the status quo is for no action to be taken and cooperation
occurs that may not maximize the immediate self-interest of legislators in the short term.
Legislators, as boundedly rational individuals, must make difficult decisions regularly and have
a limited amount of time, cognitive processing ability, and attention to choose the best alter-
native, and often “satisfice,” or choose an alternative that is both satisfactory and sufficient, the
best option at the time (Simon 1945). With an institution that expects action and holds norms
of cooperation (at the very least within the party), boundedly rational individuals will work
within the expectations of the structure to make decisions and decide between alternatives.
Members of Congress are influenced not only by their own self-interest, but their constituents,
and the institutional structure of Congress itself. Assuming the boundedly rational character
of individual decision makers, who inhabit a boundedly rational institution like government,
cooperation and action are easier to understand. The decisions made in the policy process are
likewise more easily understood, as they are created within a boundedly rational institutional
framework.
One of the first sets of scholars to implement bounded rationality as a microfoundation were
scholars studying public budgets. Budgetary considerations are suitable subjects for exploring
institutional decision making, since there is a clear budgetary process, mimicking the policy
process itself. The budgetary process begins with a proposal and ends with an output in measur-
able units. Objectives are also clearly stated in budgets; which objectives take priority reveal the
values held internally by policy makers (Lindblom 1959). Budgets are also incremental actions
that happen on a consistent basis from an existing base that methodically modify and build on
the decisions made in previous years (Lindblom 1959; Davis, Dempster, and Wildavsky 1966).
The incremental nature of budgets avoids mistakes by policy makers, because the decisions of
individuals (including interest groups) “are powerful, persistent and strongly grounded in the
expectations of others as well as in the internal requirements of the positions” (Davis, Dempster,
and Wildavsky 1966).
Budgets prove to be branch processes as Lindblom posits, because models of the budgetary
process borrow heavily from the existing base and are strategic in character. “For budgets,
‘base’ is previous appropriations for agency, similar to a bank of past decisions for individual
decision makers” (Davis, Dempster, and Wildavsky 1966). Previous years’ appropriations in the
budgetary base are comprised of “sunk costs,” commitments made by lawmakers in previous
years. These commitments operate in a path-dependent relationship, as previous commitments
can dictate what this year’s lawmakers choose what to invest in and which organizational pri-
orities to support. Sunk costs reinforce path dependence and the canalization of organiza-
tional practices (Pierson 1993; Jones and Baumgartner 2005, p. 57). In budgetary policy, sunk
cost commitments provide the mechanism for understanding bounded rationality of institu-
tional decision-making by demonstrating the direct link between current policy and previous
decisions.
Sunk costs reflect the strength of the base in incremental budgets and how bounded ration-
ality exists as a backdrop of the policy process. As a path-dependent continuation of previous
years’ budgets, sunk costs strengthen lawmakers’ identification with the means. As a result of
limited attention capacity and information processing ability, previous policy decisions become
entrenched as foregone conclusions, and individuals come to identify emotionally and cogni-
tively with these operating procedures. Identification with the means is a nonrational process
motivated by bounded rationality, that exacerbates the additive and incremental branch-like
nature of budgets, leading to inevitable trade-offs in decision-making, because the means
become prioritized over the policy goal (Simon 1945; Jones 2003; Jones et al. 2014, p. 153).

602
603

Bounded rationality in political science

Given the omnipresence of trade-offs in budgetary decisions, budgets are a consistent example
of the trade-offs and outputs characteristic of bounded rationality and information processing
abilities of individuals and institutions. They inherently hold prioritization of issues and values
and reveal the capacity for attention in policy makers. The institutional constraint on policy
makers in Washington for budgets is strong, as interest groups, rival parties, and a scarce amount
of resources frame debate and compromise prior to policy output, two parameters for the
budgetary process.
Using the budget process as an example, we can again consider that the political world
is a large world. There are going to be budget alternatives that are never considered, but
actors shaping budget policy fall back onto their decision rules and favor items that have been
prioritized in the past despite this uncertainty (Jones 2001; Jones and Baumgartner 2005).
While uncertainty, as opposed to risk, is a feature of the decision-making process for budgeting,
the critical distinction is that attribute uncertainty (i.e., information about a policy problem
or solution) and statistical uncertainty (i.e., being able to assign a probability to all possible
outcomes or events) are distinct. Put another way, the policy problems being dealt with in
major countries around the world are sufficiently multidimensional and complex that decision
makers rely on attention to inform their preferences, despite the constant threat of not knowing
all possible alternatives (Jones 1994). Relying on the policy problems plaguing the country or
an identification with the means (whether that is to their district, state, party, congressional
committee, or something else) legislators make decisions based on uncertainty surrounding
attributes of policy problems or solutions when making budget decisions, even when operating
in large world contexts (Simon 1996; Jones and Baumgartner 2005).
Following the budget scholars, Kingdon (1984) elaborated on the theory of bounded ration-
ality. He expanded the policy process application by introducing policy streams. In the pre-
decision-making phase, even preceding agenda setting, elites identify problems and then are
able to offer solutions. When problems are identified, they are placed on the agenda, and then
solutions can begin to be sought and offered.
Behavioral rationality still acknowledges the individuals and institutions have similar
characteristics. While both individuals and institutions work towards fulfilling their agendas
and are goal-oriented, their goals are impeded by a limited capacity for processing information.
Since attention-space is limited, agendas must be set to change policy on the most pressing issues
first in case time expires before reaching the end of the agenda. These shared characteristics
make policy change not incrementally, but instead, in bursts.
Often, policy change does not occur in slow, incremental, steps but instead in great
punctuations where there are many changes in a short period. Punctuated Equilibrium Theory
(PET) (Baumgartner and Jones 1993), perceives policy change in this way. Like tectonic
plates, policy change faces impassable friction most of the time. Policy processes are typically
characterized as incremental, with slow, methodical, moving change through debate and grid-
lock in Congress. However, for some policies like health care, criminal justice, and drug policy,
change may not occur at a constant and slow pace, but in dramatic bursts following long periods
of no change at all. Policy problems build up without policy solutions to alleviate them, and
with them there is a creation of friction. Most of the time, problems in society continue to exist
in this state, as problems without clear policy solutions. Once policies begin to be introduced to
try to fix the problem, salience will increase and the problem is debated publicly by legislators.
During punctuated periods, the system may overcorrect itself, and a dearth of solutions may
be proposed, even passed into policy, for the same problem. In the time between punctuations,
friction prevents policy change from getting through, then punctuations occur like earthquakes
in stick-slip fashion.

603
604

Zachary A. McGee et al.

Bounded rationality, the foundation for PET, hinges upon an actor’s ability to process infor-
mation in a complex environment. Both individuals and organizations process information, par-
ticularly applied to the policy process. The criteria for information processing are “collecting,
assembling, interpreting, and prioritizing signals in the environment” (Jones & Baumgartner
2005, p. 7), which alludes to the inescapability of limited attention in bounded rationality due
to the overabundance of information in one’s environment. Boundedly rational individuals and
organizations prioritize what they pay attention to, and for higher-order processing that require
more conscious thinking or attract high levels of attention, processing is serial in nature (Bendor
2003; Jones & Baumgartner 2005; Workman et al. 2009). Information is processed serially
when one issue at a time is processed and prioritized at the expense of others. Information is
processed disproportionately due to attention limits, complex problems, and an overabundance
of information in the environment.
For information to be processed serially, issues must be prioritized in comparison to others.
Reasoning in prioritizing of information takes on a dual system framework. Characterizing
mental processes that help to prioritize attention, System 1 is fast, automatic, and often uncon-
scious reasoning. On the contrary, System 2 is deliberate, slow, and conscious (Kahneman 2011;
Thagard, Chapter 25 in this volume). Applied to political decision-making, information pro-
cessing of System 2 requires more time and attention than System 1 reasoning and will result
in developing expertise through deliberate and reflective reasoning. Because individuals are
boundedly rational, constrained by time and information processing capacity, and faced with
trade-offs in attention allocation, System 2 reasoning occurs much less frequently than System
1 due to higher demands on finite attention.
For political organizations and systems, information processing is remedied by delegating
processing of information to its subsystems. At the political system level, organizations special-
izing in a specific issue or policy area develop expertise and consequently are able to devote
attention to this topic, informing policy makers and other organizations. The oversupply
of information necessitates this delegation, called parallel processing. In the policy process,
bounded rationality provides the theory of behavior connecting attention and information pro-
cessing for individual policy makers, organizations involved, and the institutions that dictate the
process via rules of the game.
Government systems are evaluated on their ability to solve social problems rather than
from top-down democratic accountability of elections, which is an institutional metric for
assessing government (Jones and Baumgartner 2005). Due to the uncertainty and complexity
of the environment and the multidimensionality of issues in the complex problem environ-
ment, it is impossible to assume clear and static problems and preferences. Instead, government
is boundedly rational as a system, its bureaucracy and organizations make decisions in the
same complex environment and have limited attention to pay, therefore are also boundedly
rational and must make calculated decisions for which issues to focus on at a time. In this way,
government is a complex system that is constantly adapting to react to issues and problems
becoming prominent and demanding attention. Policy makers, organizations, and government
systems interact with the environment by processing information. Policy change is the result
of information processing, which is often disproportionate due to environmental and cognitive
constraints; moreover, attention and policy solutions are not often proportional to changes in
the environment itself (Jones and Baumgartner 2005).
In an attempt to solve issues in the problem environment, the political system reacts by
offering policy solutions. Low or high attention to an issue dictates whether policy reaction
will be an under- or over-reaction. Over-reaction leads to policy punctuations, which are large
changes in policy output representing drastic surges in attention to an issue. Disproportionate

604
605

Bounded rationality in political science

information processing provides the framework for policy reactions and punctuations, and a
resulting effect is the inability for the policy response to be proportionate to the issue’s severity
(Jones et al. 2014). Responses to the overabundance of information and myriad issues in the
environment are disproportionate, following familiar patterns of bounded rationality found in
individuals. The inability of organizations to focus attention on numerous issues at once and
differences in information processing mechanisms such as serial processing, determine imper-
fect reactions to issues in the environment.
The status quo for policy change is no change, or an under-reaction to problems in the envir-
onment. Policy over-reaction is more unusual because policy punctuations are rare due to the
myriad of conditions necessary for a drastic change in output. First, attention to the issue must
be exceedingly high. Second, institutional friction slows down policy change, often debilitating
rapid action, assuring policy underreaction (Maor 2014). Institutional friction, often described
as “gridlock,” illustrates the preference for the status quo, found in policy making rules and
procedures, governmental bodies, interest groups, and bureaucracy. Friction guarantees incre-
mental policy change a majority of the time but is overcome from time to time to create a
punctuation in policy change, or a burst in policy change. “Friction, in other words, is not an
absolute barrier to action, but rather a major hindrance. As it operates, pressures can mount,
making change, when it does occur, more profound” (Jones & Baumgartner 2005, p. 88).
Policy over-reaction attempts to address an issue, imposing “objective and/or social costs”
without producing equal benefits (Maor 2012). A disproportionate reaction is a mismatch
between issues and solutions, missing the mark. Likewise, a policy bubble becomes “wildly
dissociated from its instrumental value in achieving a policy goal” (Jones et al. 2014, 147). Like
asset bubbles in economics, which occur when an asset’s price is substantially higher than its
intrinsic value, policy bubbles are easily spotted in hindsight, but much more difficult to assess
as they are happening.
Bubbles are often caused by imitation heuristics, such as when the actions of a single actor
on a financial market trading floor, make an independent action that is seen and imitated by
many others. This cascading effect, caused by many actors imitating the action of one, can
lead to dramatic fluctuations or crashes, an effect that also characterizes asset bubbles in eco-
nomics (Jones and Baumgartner 2005). Policy bubbles occur following a period of sustained
overinvestment, demonstrating a mismatch between problems and policy action, and the accu-
mulation of this mismatch leads to more severe shifts in policy action if and when the issue is
addressed by policy. Policy overreaction that continues for a long period, even after attention
has moved on to other problems, can lead to policy bubbles when policy “takes on a mind of its
own,” exacerbating the mismatch between policy and problem (Baumgartner and Jones 1993;
Jones and Baumgartner 2005; Jones et al. 2014; Hallsworth et al. 2018).
Policy bubbles reflect an overinvestment in an issue beyond its initial value, which is extended
for a long period of time (Jones et al. 2014; Maor 2014). Crime policy represents the clearest
example of a policy bubble in twentieth-century American politics. In the early 1990s, as the
national crime rate declined, policy output continued growing; as the problem indicator, crime
rates decreased throughout the 1990s, and the policy instrument of incarceration increased
(Jones et al. 2014). Ultimately, throughout the 1990s, the problem continued to decrease as
prison populations remained static, which displays the mismatch between problem and policy
response.
Bounded rationality serves as the foundational theory for other theories of social science,
which offer theories inspired by bounded rationality that contribute to studies of public policy
processes. Behavioral economic theories incorporate concepts familiar to bounded rationality,
including attention allocation, heuristics, and framing, to explore how behavioral science affects

605
606

Zachary A. McGee et al.

governmental decision-making in creating policy (Amir et al. 2005; Shafir 2013; Hallsworth
et al. 2018). Representative of behavioral economics is the concept of nudges, paternalistic
intervention by government into decisions by individuals to make them better off, as judged by
themselves (Thaler and Sunstein 2008; Sunstein, Chapter 38 in this volume).
Attention is a finite resource for everyone, and policy makers seek to nudge people in the
direction they would have chosen had they not been subject to constraints and limitations of
rationality. Key to the concept is the liberty in libertarian paternalism, that the government does
not restrict information but frames it using heuristics to encourage people to make decisions in
their best interest. Consider government as choice architects in healthy food and anti-smoking
campaigns or encouraging sign-ups for health care plans through the Affordable Care Act.
A nudge is a strategy to guide people to make decisions in their best interest, maintaining indi-
vidual freedom of choice (including the choice to opt out), while “trying to influence people’s
behavior in order to make their lives longer, healthier, and better” (Thaler and Sunstein 2008,
p. 7). These concepts show that the way policy is created and framed can enhance its effect-
iveness and is rooted in selective attention and heuristics, reflecting Simon’s original work on
bounded rationality (Jones 2017).
More directly, bounded rationality is the foundational theory for other theories of the policy
process, including the Advocacy Coalition Framework (Sabatier 1986; Sabatier and Jenkins-
Smith 1993; Jenkins-Smith et al. 2014), the Social Construction Framework (Schneider and
Ingram 1993), the Multiple Streams Framework (Kingdon 1984), and even in some applications
of diffusion (Boushey 2012) and the Institutional Analysis and Design Framework (Ostrom
2011). In the next section, we will discuss how bounded rationality has shaped contemporary
political science research and speculate on future uses of the model.

The future of bounded rationality in political science: bridging

organizational and individual choice
Bounded rationality allows for the connection of two disparate decision-making systems: the
individual level and the systems level (Jones 2017). By connecting these two systems of decision
making and applying an information-processing framework, bounded rationality will continue
to be the micro-foundation of policy studies for decades to come. Since the publication of
Agendas and Instability in American Politics, Baumgartner and Jones have generalized their theory
of the public policy process to stress the importance of information processing and attention
allocation for how lawmakers decide which problems to prioritize and act upon (Jones and
Baumgartner 2005; Baumgartner and Jones 2015). Much like Simon did in Administrative
Behavior, Baumgartner and Jones stress the decision-making process of elites as being a crit-
ical causal mechanism. This development provides a critical pathway for future work utilizing
bounded rationality.
Attention allocation by organizations is inherently linked to attention allocation at the indi-
vidual level (Jones 2017; Jones and McGee 2018). As we have already noted, attention allo-
cation at the individual level is constrained; individuals can only focus on one thing at a time.
While organizations expand individuals’ abilities to process information (March and Simon
1958), organizations also tend to canalize the choices of individuals and lead to institutional
attention being focused on the same recurring problems and/or solutions time and time again
(Jones 2001). These canals can be difficult to escape and within organizations they can turn
into bureaucratic decision-rules. Decision rules often lead to incrementalism (e.g., the early
work on budgets) and in some cases can lead to policy bubbles (Jones et al. 2014). These policy
bubbles are the result of overinvestment of attention in a given policy area and have been

606
607

Bounded rationality in political science

demonstrated to exist in many policy areas. The existence of policy bubbles demonstrates that
canals in individuals have some sort of link to organizational attention allocation. Congressional
committees provide another example. The issues committees pay attention to previously tend
to get attention again; and, the solutions prescribed previously tend to be proposed again,
sometimes in slightly different forms (Baumgartner and Jones 1993).
Information processing and a reintegration of institutions provide easy links between indi-
vidual and organizational choice. Information processing, in the broadest sense, examines the
supply and prioritization of information. Usually congressional committees will receive infor-
mation and incrementally adjust the policies they deal with. Sometimes, however, new infor-
mation or shifts in issue definitions might cause rapid changes in the problem and solution
definitions. These changes can result in rapid changes in the proposed policy solutions. Taken
together, these conditions are known as the general punctuation thesis (Jones and Baumgartner
2005). Information is not scarce within government; in fact, information is in oversupply. This
oversupply of information leads actors to be overwhelmed by their choice environment. To deal
with the oversupply, they must winnow out the information that is not useful; this winnowing
process is boundedly rational. How actors search for and weight the information they receive
(e.g., members in a committee hearing) is crucial to whether or not a policy problem is resolved,
or a specific solution is chosen (Baumgartner and Jones 2015). Ultimately, the decisions made
about what information is important are agenda-setting decisions. Therefore, to think about
agenda setting, attention allocation, or information processing is to confront bounded ration-
ality and its influence on the policy process literature (Jones and McGee 2018).
Scholars have already started utilizing the information-processing framework in exam-
ining political institutions, including Congress (Baumgartner and Jones 1993, 2015; Jones and
Baumgartner 2005; Adler and Wilkerson 2012; Lewallen, Theriault, and Jones 2016), the bur-
eaucracy (Workman 2015), and the media (Wolfe 2012; Boydstun 2013). It has been applied
comparatively as well (Breunig 2006; Walgrave 2008; Green-Pedersen and Mortensen 2010;
Bevan and Jennings 2014).
We are at a turning point in political science for modeling decision making. As Jones (2017)
has stated, “few social scientists have any faith that the rational model can take us any further
than it has. But we have not thought deeply enough about what elements are necessary and
which are expendable.” The information-processing perspective, rooted in Simon’s model of
bounded rationality, opens a new path for understanding choice in the study of politics. We
are confident that this framework will continue to shed light on new phenomena in political
science, and it would not have been possible with the foundation built by Herb Simon.

Notes
1 Sometimes known as “identification with the means.”
2 See Jones (2001) for a comprehensive discussion and thoughtful extrapolation.
3 The second branch is the namesake of Tversky and Kahneman and is rooted in the idea that cognitive
processes are limited in even the simplest of tasks. We will reference this school of thought, but see
Bendor (2003) for a full explication of the differences between the two.

References
Adler, E. S., and Wilkerson, J. D. (2012). Congress and the Politics of Problem Solving, New York: Cambridge
University Press.
Amir, O., Ariely, D., Cooke, A., Dunning, D., Epley, N. ... Silva, J. (2005). Behavioral economics, psych-
ology, and public policy, Marketing Letters 16(3/4), 443–454.

607
608

Zachary A. McGee et al.

Baumgartner, F. R., and Jones, B. D. (1993). Agendas and Instability in American Politics, Chicago: University
of Chicago Press.
Baumgartner, F. R., and Jones, B. D. (2015). The Politics of Information: Problem Definition and the Course of
Public Policy in America, Chicago: University of Chicago Press.
Bendor, J. (2003). HERBERT A. SIMON: Political scientist, Annual Review of Political Science 6(1),
433–471.
Bendor, J. (2015). Incrementalism: dead yet flourishing, Public Administration Review 75, 194–205.
Bevan, S., and Jennings, W. (2014). Representation, agendas and institutions, European Journal of Political
Research 53(1), 37–56.
Binmore, K. (2006). Making decisions in large worlds. Paper presented at the ADRES Conference,
Marseilles, France.
Boushey, G. (2012). Punctuated Equilibrium Theory and the diffusion of innovations, The Policy Studies
Journal 40(1), 127–146.
Boyd, R., and Richerson, P. (2001). Norms and bounded rationality, in G. Gigerenzer and R. Selten
(Eds.), Bounded Rationality: The Adaptive Toolbox, Cambridge, MA: MIT Press.
Boydstun, A. E. (2013). Making the News: Politics, the Media, and Agenda Setting, Chicago: University of
Chicago Press.
Breunig, C. (2006). The more things change, the more things stay the same: A comparative analysis of
budget punctuations, Journal of European Public Policy 13(7), 1069–1085.
Davis, O. A., Dempster, M. A. H., and Wildavsky, A. (1966). A theory of the budgetary process, The
American Political Science Review 60(3), 529.
Gigerenzer, G., Todd, P., and the ABC Research Group (Eds.) (1999). Simple Heuristics That Make Us
Smart, New York: Oxford University Press.
Green-Pedersen, C., and Moretensen, P. B. (2010). Who sets the agenda and who responds to it in the
Danish parliament? A new model of issue competition and agenda-setting, European Journal of Political
Research 49(2), 257–281.
Hallsworth, M., Egan, M., Rutter, J., and McCrae, J. (2018). Behavioral Government: Using Behavioral
Science to Improve how Governments Make Decisions, London: The Behavioral Insights Team.
Jenkins-Smith, H. C., Nohrstedt, D., Weible, C. M., and Sabatier, P. A. (2014). The Advocacy Coalition
Framework: Foundations, evolution, and ongoing research, in P. A. Sabatier and C. M. Weible (Eds.),
Theories of the Policy Process (3rd edn, pp. 183–224). Boulder, CO: Westview Press.
Jones, B. D. 1994. A change of mind or a change of focus? A theory of choice reversals in politics. Journal
of Public Administration Research and Theory 4(2), 141–177.
Jones, B. D. (1997). The rational decision-making model in politics, Technical report, University of
Washington, Seattle.
Jones, B. D. (1999). Bounded rationality, Annual Review of Political Science 2, 297–321.
Jones, B. D. (2001). Politics and the Architecture of Choice, Chicago: University of Chicago Press.
Jones, B. D. (2003). Bounded rationality and political science: Lessons from public administration and
public policy, Journal of Public Administration Research and Theory 13(4), 395–412.
Jones, B. D. (2017). Behavioral rationality as a foundation for public policy studies, Journal of Cognitive
Systems Research 43, 63–75.
Jones, B. D., and Baumgartner, F. R. (2005). The Politics of Attention, Chicago: University of Chicago
Press.
Jones, B. D., and McGee, Z. A. (2018). Agenda setting and bounded rationality, in A. Mintz and L. Terris
(Eds.), The Oxford Handbook of Behavioral Political Science, Oxford: Oxford University Press.
Jones, B. D., Thomas, H. F., and Wolfe, M. (2014). Policy bubbles, Policy Studies Journal 42(1), 146–171.
Kahneman, D. (2011). Thinking, Fast and Slow, New York: Farrar, Straus and Giroux.
Kingdon, J. W. (1984). Agendas, Alternatives, and Public Policies, Boston: Little, Brown.
Knight, F. (1921). Risk, Uncertainty, and Profit, Boston: Houghton-Mifflin.
Lewallen, J., Theriault, S. M., and Jones, B. D. (2016). Congressional dysfunction: An information pro-
cessing perspective, Regulation and Governance 10(2), 179–190.
Lindblom, C. (1959). The science of muddling through, Public Administration Review 19(2), 79–88.
Maor, M. (2012). Policy overreaction, Journal of Public Policy 32(3), 231–259.
Maor, M. (2014). Policy bubbles: Policy overreaction and positive feedback, Governance 27(3), 469–487.
March, J. G. (1994). Primer on Decision Making: How Decisions Happen, New York: Simon & Schuster.
March, J. G., and Olsen, J. P. (1984). The new institutionalism: Organizational factors in political life,
American Political Science Review 77(2), 281–296.

608
609

Bounded rationality in political science

March, J. G., and Simon, H. A. (1958). Organizations (2nd ed.), Oxford: Blackwell.
May, P. J. (1991). Reconsidering policy design: Policies and publics, Journal of Public Policy 11(2), 187–206.
McAdams, R. H. (1997). The origin, development, and regulation of norms, Michigan Law Review 96(2),
338–433.
Ostrom, E. (1998). A behavioral approach to the rational choice theory of collective action: Presidential
Address, American Political Science Association, 1997, The American Political Science Review 92(1), 1.
Ostrom, E. (2000). Collective action and the evolution of social norms, Journal of Economic Perspectives
14(3), 137–158.
Ostrom, E. (2009). Institutional rational choice: An assessment of the institutional analysis and develop-
ment framework, in Theories of the Policy Process (2nd ed.), Boulder, CO: Westview Press.
Ostrom, E. (2011). Background on the institutional analysis and development framework, Policy Studies
Journal 39(1), 7–27.
Pierson, P. (1993). Review: When effect becomes cause: Policy feedback and political change, World
Politics 45(04), 595–628.
Sabatier, P. A. (1986). Top-down and bottom-up approaches to implementation research: A critical ana-
lysis and suggested synthesis, Journal of Public Policy 6(01), 21.
Sabatier, P. A., and Jenkins-Smith, H. (1993). Policy Change and Learning: An Advocacy Coalition Framework,
Boulder, CO: Westview Press.
Savage, L. J. (1954). The Foundations of Statistics, New York: John Wiley & Sons.
Schneider, A., and Ingram, H. (1993). Social construction of target populations: Implications for politics
and policy, American Political Science Review 87(02), 334–347.
Selten, R. (2001). What is bounded rationality?, in G. Gigerenzer, and R. Selten (Eds.), Bounded
Rationality: The Adaptive Toolbox, Cambridge, MA: MIT Press.
Shafir, E. (Ed.) (2013). The Behavioral Foundations of Public Policy, Princeton, NJ: Princeton University Press.
Simon, H. A. (1997 [1945]). Administrative Behavior (4th ed.), New York: Free Press.
Simon, H. A. (1996). The Sciences of the Artificial, Cambridge, MA: MIT Press.
Sniderman, P. M., Brody, R. A., and Tetlock, P. E. (1991). Reasoning and Choice: Explorations in Political
Psychology, New York: Cambridge University Press,
Thaler, R. H., and Sunstein, C. R. (2008). Nudge: Improving Decisions about Health, Wealth, and Happiness,
New Haven, CT: Yale University Press.
Tversky, A., and Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases, Science
185(4157), 1124–1131.
Wagner, W. E. (2010). Administrative law, filter failure, and information capture, Duke Law Journal 59,
1321–1432.
Walgrave, S. (2008). Again, the almighty mass media? The media’s political agenda-setting power according
to politicians and journalists in Belgium, Political Communication 25(4), 445–459.
Wildavsky, A. B. (1964). Politics of the Budgetary Process, Boston: Little, Brown.
Wolfe, M. (2012). Putting on the brakes or pressing on the gas? Media attention and the speed of policy-
making, Policy Studies Journal 40(1), 109–126.
Workman, S. (2015). The Dynamics of Bureaucracy in the US Government: How Congress and Federal Agencies
Process Information and Solve Problems, New York: Cambridge University Press,.
Workman, S., Jones, B. D., and Jochim, A. E. (2009). Information processing and policy dynamics, Policy
Studies Journal 37(1), 75–92.

609
610

42
LAYERING, EXPANDING,
AND VISUALIZING
Lessons learned from three “process boosts”
in action

Valentina Ferretti

Introduction
Strategic decision making is a challenging process as the transformation of an undesirable state
of things into a desirable one usually requires the identification of a set of often conflicting deci-
sion objectives, the careful consideration of multiple perspectives, the proper elicitation of value
judgements and trade-offs, a creative effort in designing suitable solutions, as well as the con-
sideration of high levels of uncertainty, to name just some of the relevant complexities involved
in making important decisions (e.g., Keeney, 1982). Indeed, real-life problems and opportun-
ities typically develop within a complex and uncertain environment, which leads to unclear
consequences and difficulty in measuring probabilities and utilities (Viale, 2019). Within this
context, human rationality is bounded because there are limits to our thinking capacity, to the
available information regarding alternatives and their possible consequences, and to the time we
can take to make the decision (Simon, 1955, 1982).
Since Tversky and Kahneman’s seminal paper (1974), behavioural scientists have identified
a large number of distortions in human judgement and decision-making which have now been
studied in many different fields, ranging from economics (Camerer and Lowenstein, 2003),
to finance (Bruce, 2010; Thaler, 1999), accounting (Birnberg et al., 2007), strategic manage-
ment (Powell et al., 2011), operations management (Bendoly et al., 2015), Decision and Game
Theory (Camerer 2003; von Winterfeldt and Edwards, 1986), multi-criteria decision ana-
lysis (Morton and Fasolo, 2009), risk analysis (Slovic, 1987), operational research (Franco and
Hämäläinen, 2016), environmental economics (Shogren and Taylor, 2008) and environmental
modelling (Hämäläinen, 2015). This multi-directional line of research is based on a concept of
bounded rationality focused mainly on the study of the correspondence between judgement
and decision-making performance in tests and laboratory simulations and canonical models
of economic rationality, i.e., the maximization of subjective expected utility (Viale, 2019).
Another perspective on bounded rationality is the one offered by Baumol and Quandt (1964),
who provided an insight into how much information gathering, the amount of its retention,
and reasoning in decision making is likely to be economically optimal, thus outlining a theory
of optimally imperfect decisions.

610
611

Layering, expanding, and visualizing

The concept of bounded rationality underpinning the research presented in this chapter
focuses instead on how the interactions between the mind and the uncertain and often
constrained external environment bound human rationality as a form of ecological and adaptive
rationality (Gigerenzer and Selten, 2002; Viale, 2012). These bounds on people’s time, know-
ledge and computational powers do not necessarily prevent them from making good decisions,
to the extent that they succeed in employing simple decisions strategies or techniques in the
appropriate contexts, i.e., where there is a fit between cognition and environment (Gigerenzer
et al., 2011). In this context, an emerging and promising approach is the one of boosts, i.e.,
interventions that target long-lasting competences (domain-specific or generalizable) rather than
immediate behaviour and that foster these competences through changes in skills, knowledge,
decision tools, or the external environment (Hertwig and Grune-Yanoff, 2017). As opposed to
nudges (Thaler and Sunstein, 2008), boosts are necessarily transparent to the boosted individual
and require cooperation. Indeed, people can then harness the new or boosted competence to
make choices for themselves, thus benefitting from a capital stock that can be engaged at will
and across situations (Hertwig and Grune-Yanoff, 2017). Boosts range from interventions that
require little time and cognitive effort on the individuals’ part to ones that require substantial
amounts of training, effort and motivation. The boosting techniques that I propose in this
chapter to improve strategic decision-making processes belong to the second category, as the
intention is to effectively change the decision makers’ repertoire of decision capabilities. In
particular, what I propose is a multi-methodology (Myllyviita et al., 2014; Henao and Franco,
2016; Ferretti, 2016a) “process boost” aimed at improving both the decision-making process
and its outcome.
Multi-methodology interventions are attracting increasing interest in both management and
public contexts as they integrate qualitative and quantitative approaches to better support all
stages of a decision-making process, from the initial divergent one to the final convergent one
(Ferretti and Gandino, 2018). The multi-methodology process boosts presented in this chapter
target the following two specific stages of the decision-making process: the framing stage,
which requires a clear understanding of the problem or opportunity to be addressed as well as a
clear understanding of what needs to be achieved (i.e., the decision objectives), and the prefer-
ence elicitation stage, which requires sound reasoning about heterogeneous performances and
value trade-offs.
Although there is a wide scholarly discussion on multi-methodology interventions, the
assumed benefits of using mixed methods have not yet been systematically tested (Myllyviita
et al., 2014; Henao and Franco, 2016). There is thus an evident need to pursue, and to better
communicate, the benefits of mixing and the research presented in this chapter is an attempt
to fill this gap. In particular, the aim of this chapter is twofold. First, to illustrate promising
solutions to support human bounded rationality in the key steps of a decision-making process
for both private organizations and public policy makers. Second, to discuss the lessons learned
by implementing the proposed process boosts in three different interventions.
Following from these aims, the contributions of this research can be summarized as
follows: (1) a new perspective that will feed the debate about multi-methodology interventions
and their positive impacts from a bounded rationality point of view; and (2) real examples and
guidelines on how to use the proposed solutions in other similar as well as different decision-
making contexts.
The following sections report on the design and implementation of three interventions,
carried out by the author, to provide support for complex decision-making problems
in different settings, ranging from rural regeneration, to talent selection and territorial
planning.

611
612

Valentina Ferretti

The three interventions are presented and discussed using the same lens, i.e., contextual-
ization of the cognitive challenge with examples from real projects, proposed solutions, and
observed results and impacts.

Framing the decision context: gaining focus for a rural regeneration

project in a new World Heritage Site
The cognitive challenge and the context of the intervention
Decision framing is the critical first step in any decision-making process, as the decision frame
helps us define the decision by answering the question “what problem or opportunity are
we addressing?” (Tani and Parnell, 2012). Indeed, an appropriate frame, including a clear
understanding of the problem and what needs to be achieved, is the first of the six requirements
for Decision Quality (Spetzler et al., 2016). However, framing a decision in the appropriate
way is a challenging task, as evidenced by the fact that inadequate or poor framing is an all-
too-common cause of failure to achieve good decision making within an organisation (Tani
and Parnell, 2012).
To give a few examples of common cognitive traps in the framing step, we can start by
discussing the dangers of a too narrow frame. Indeed, a decision with a narrow frame has
usually a very limited focus and may thus miss relevant objectives or solutions. On the other
hand, also a too broad frame has drawbacks, as usually decisions with broad frames have long
time horizons which result in higher uncertainty. They lead to more general objectives, which
are difficult to measure, and they encompass many issues, which make it more difficult to dis-
entangle the key decision that has to be made now. Another common tendency is to distort
situations to fit our preconceptions. This is also known as comfort zone bias and consists in
dragging a problem into our comfort zone and solving the problem we know how to solve
rather than the problem than needs solving (Spetzler et al., 2016).
Finding the frame that is most appropriate for the situation is extremely important. If we get
the frame wrong, we will be solving the wrong problem or addressing the opportunity in the
wrong way (Spetzler et al., 2016).
Although narrow or broad representations may occur at the individual level for basic cog-
nitive reasons (e.g., selective attention, anchoring, availability, confirmation, accessibility, satis-
ficing tendencies), they are also often compounded by the social external environment in which
we live and work (Larrick, 2009). Indeed, common training, common experiences, frequent
interactions between people working within the same organization and group environments
focusing on harmony or conformity to the boss’s opinion, all have the effect of leading to
shared and often incomplete views of the problem (Cronin and Weingart, 2007).
The following paragraphs provide an account of the framing challenge and a proposed
solution tested within a rural regeneration project in the vineyard landscape of Langhe, Roero
and Monferrato (Piedmont Region, Italy).1 The region under analysis recently became a new
UNESCO World Heritage Site as an exceptional living testimony of the historical tradition of
grape growing and winemaking processes, of a social context, and a rural economy based on
the culture of wine.
Gaining this important recognition means that conflicting needs coexist in the same area,
i.e., conservation and protection needs as well as new development needs. The context under
analysis represents thus a complex decision-making environment. As often happens in these
environments, situations which are not acceptable to stakeholders may not yield at first glance a
statement of a problem to be solved, or may yield multiple problems whose statements may be

612
613

Layering, expanding, and visualizing

contradictory or messy (Johnson, 2012). Indeed, this project did not start with a clear statement
of a problem to be solved but rather with a need expressed by the Tourism organization of Alba,
Bra, Langhe and Roero to better know the territorial strengths and weaknesses of the region
in order to properly support the planning of the future of an area considered of universal value
(Ferretti and Gandino, 2018). We can consider this as a quite broad initial frame for the decision
context, as “strengths” and “weaknesses” of a region can mean many different things.

Proposed solutions
To gain the knowledge needed to properly frame the decision context under analysis, we
proposed the combined use of three complementary tools. More precisely, we integrated
Strengths, Weaknesses, Opportunities and Threats (SWOT) analysis with Geographic
Information Systems (GIS) and Multicriteria Decision Analysis (MCDA), all within a partici-
pative framework consisting of subsequent focus groups with the client organization (i.e., the
Tourism organization of Alba, Bra, Langhe and Roero) and experts from the SiTI Research
Institute (www.siti.polito.it/index.php?l=ENG), which was the leading actor involved in the
UNESCO site candidacy preparation (SiTI, 2013). The rationale for the use and integration
of these three tools can be summarized as follows: first, SWOT analysis is a powerful and com-
monly used tool to provide an overview of strengths and weaknesses of a concept/project/
system as a basis for creating strategies. Second, given that the context under analysis is a ter-
ritorial one, the spatial distribution of its characteristics (e.g., extension of the vineyards, prox-
imity to negative aesthetic interferences to the landscape such as industrial buildings, etc.) plays
a crucial role. To visualize the spatial distribution of each key element, we combined GIS and
SWOT analysis by developing a geographical map for each indicator worth consideration in the
analysis (e.g., Comino and Ferretti, 2016). Third, the indicators belonging to each of the four
categories of the SWOT should be overlaid in order to obtain for each category an overview
of the situation where hotspots requiring monitoring or mitigation measures can be identified.
To do this, MCDA (Belton and Stewart, 2002) plays a crucial role as it compares heterogeneous
information and properly assesses trade-offs among different performances.
We thus conducted a focus groups with experts who discussed, identified and agreed on the
key indicators that better represented the Strengths, Weaknesses, Opportunities and Threats for
the region under analysis (Table 42.1). We then studied the spatial distribution of each indicator
with GIS and developed a map for each of them. Finally, we used MCDA to elicit the relative
weight of each spatial indicator, taking into account its spread of values for each and used those
weights to overlay the maps in each of the four categories. The reader interested in learning the
details of the process can refer to Ferretti and Gandino (2018).

Results and impacts of the multi-methodology boost

The result of the framing stage in this project was the discovery that the network of abandoned
historical rural buildings (one of the indicators spatially analysed in the SWOT analysis)
represented both a weakness and an opportunity for the whole World Heritage Site. Indeed,
this analysis initiated the creation of a geographical database with 292 abandoned rural historical
buildings scattered across the World Heritage Site, which includes 101 municipalities. Fourteen
of these buildings are concentrated in the Municipality of La Morra, making it the municipality
with the highest concentration of abandoned rural buildings. These buildings represent a true
landmark of the rural architectural and historical legacy and are usually located in beautiful

613
614

Valentina Ferretti

Table 42.1 SWOT indicators to be transformed into maps

Strengths Weaknesses Opportunities Threats

Presence of the full wine Lack of public Network of cultural Land use
production chain transportation associations consumption
Wine biodiversity Abandoned train stations Future planned Ageing population
requalification
projects
Historical continuity Natural risks Touristic pressure
Traditional festivals Presence of natural Loss or increase of
distribution interferences with the autochthonous
landscape vineyards
Landscape valuable Abandoned architectural
elements heritage
Cultural and
architectural heritage
Presence of ongoing
requalification
projects

surroundings. Their regeneration would thus create significant added value for the area, both
from the residents’ and from the tourists’ points of view.
The key impacts observed after this transparent and transferable multi-methodology boost
can be summarized as follows:

1 Actionable knowledge about the territory (online open spatial database).

2 New awareness of problem (i.e., density of abandoned buildings) and sense of urgency,
which provided purpose for the organization.
3 Nudge to start a second phase of the project aimed at identifying the most strategic
abandoned building to be regenerated first in the municipality of La Morra.
4 Transition from a very broad frame of the problem under analysis (i.e., need to improve
the knowledge about the territory under analysis) to a more focused one (i.e., there is
a problem with the density of abandoned buildings within the World Heritage site that
can be turned into an opportunity for rural regeneration, but with which building is it
more strategic to start the requalification process?). In particular, this multi-methodology
intervention improved the frame by acting on all three components considered key to
reach an appropriate frame, i.e., purpose, perspective and scope (Tani and Parnell,
2012). The integrated SWOT-GIS-MCDA approach developed through focus groups
with experts allowed the team indeed: (a) to clarify why and when resources need to be
allocated for the regeneration of abandoned buildings (i.e., purpose); (b) to highlight
the advantage of including different perspectives in a focus group setting with both
the client organization and external technical experts to avoid using the usual organ-
izational frame even if the challenge is new (i.e., perspective), and (c) progressively to
zoom in on the nature of alternative solutions, starting from the whole region under
analysis, then zooming in to the municipal scale and finally focusing on the building
scale (i.e., scope).

614
615

Layering, expanding, and visualizing

Expanding the set of relevant objectives: the case of an Educational

Foundation for underprivileged children in Hungary
The cognitive challenge and the context of the intervention
The first of the six requirements for Decision Quality (Spetzler et al., 2016) mentioned in
the contextualisation of the first intervention is an appropriate frame, which includes a clear
understanding of the problem but also of what needs to be achieved, i.e., the fundamental
objectives for the decision under analysis. Indeed, objectives should be the driving force in
decision making as they represent what we care about, while alternatives are just means to
achieve more fundamental values (Keeney, 1996).
Unfortunately, scientific research has demonstrated that the identification of the fundamental
objectives associated with a decision is not an easy task from the cognitive point of view and
that, without support, people are often aware of only half of the objectives that later turn out
to be relevant to them (Bond et al., 2008). Moreover, and even more worrying, the objectives
that are missed are not trivial, they are often as important as those identified at first (Bond et al.,
2008). Two possible reasons may explain this impediment: not thinking broadly enough about
the range of relevant objectives, and not thinking deeply enough to articulate every objective
within the range that is considered (Bond et al., 2010). Moreover, research on time-relevant
decisions and “temporal construal” shows that decision makers often focus only on either the
near future or distant future, to the exclusion of the other (e.g., Trope and Liberman, 2000).
The following paragraphs provide an account of the objectives’ generation challenge and
a proposed solution tested within an intervention aimed to support the Csányi Educational
Foundation for underprivileged children in Hungary.2 To achieve its mission, the Foundation
announces every year a tender for 15 applicants to enter the Foundation’s Educational pro-
gramme. Gifted children from underprivileged backgrounds who are completing their fourth
year in primary school are eligible to apply to the programme.
As the Foundation spends approximately $4700/year/child (operating costs excluded), a
crucial objective of the organization is to minimize dropout rates. However, mentors have
highlighted that a large number of children who were admitted into the Foundation do not fit
the profile and the focus of the programme for disparate reasons. Consequently, more than 25
per cent of the children quit the programme early.
After thoroughly examining these cases, the board of directors concluded that these problems
are caused by the lack of an adequate selection process. The board also acknowledged that
the selection criteria are not entirely reflective of the objectives of the Foundation and that
the fundamental objectives have not been thoroughly discussed. The Foundation’s board thus
expressed the desire to improve the selection process for the admissions season in May 2017.

Proposed solutions
To address the objectives’ generation challenge, we designed and deployed an intervention
which integrated Value Focused Thinking (Keeney, 1996; see Table 42.2) and the specific
Multicriteria Analysis technique called Multi Attribute Value Theory (Keeney & Raiffa, 1976).
The reason behind the integration of the two approaches is linked to the fact that MCDA
supports the user in the prioritization of objectives, i.e., in understanding which trade-offs are
more important in a complex decision-making problem, but without specific support during
the objectives’ identification phase. How then to be sure that the concerns we have in mind
represent a comprehensive list of relevant objectives? The ten categories of specific questions
illustrated in Table 42.2 have proved to be an effective solution to this challenge (e.g., Ferretti

615
616

Valentina Ferretti

Table 42.2 Value focused thinking techniques to be used for the identification of relevant objectives

Techniques Key questions to be asked

A wish list What do you want? What do you value? What should you want?
Alternatives What is a perfect alternative, a terrible alternative, some reasonable alternative?
What is good or bad about each?
Problems and What is wrong or right with your organization? What needs fixing?
shortcomings
Consequences What has occurred that was good or bad? What might occur that you care about?
Goals, constraints, What are your aspirations? What limitations are place upon you?
and guidelines
Different What would your competitor or your constituency be concerned about? At some
perspectives time in the future, what would concern you?
Strategic What are your ultimate objectives? What are your values that are absolutely
objectives fundamental?
Generic What objectives do you have for your costumers, your employees, your
objectives shareholders, yourself? What environmental, social, economic or health and
safety objectives are important?
Structuring Follow means-ends relationships: why is that objective important, how can you
objectives achieve it? Use specification: what do you mean by this objective?
Quantifying How would you measure achievement of this objective?
objectives

Source: Keeney (1996)

and Degioanni, 2017). We thus started the decision support process by organizing structured
interviews with the President of the Foundation Advisory Board and with the Foundation
Operational Director, in which we tested the list of questions from Table 42.2, adapted to fit
the context under analysis, i.e., underprivileged children selection for educational programmes.
Table 42.3 represents the result of the Value Focused Thinking stage. While some key
concerns and objectives were part of the initial mental model of the decision makers, those
objectives highlighted in italics in Table 42.3 represent the new objectives that were identified
after discussing the answers to the Value Focused Thinking questions. This now comprehensive
set of objectives became the input to the MCDA model where, again in a focus group setting,
we made the participants discuss the performance of the children under each objective and
learn about the trade-offs between them (column “weight” in Table 42.3).

Results and impacts of the multi-methodology approach

As shown in Table 42.3, the solution proposed in this intervention generated 40 per cent more
objectives with respect to the initial concerns that the decision makers had in mind.
Moreover, the newly identified objectives of “level of family support”, “emotional steadiness”,
“character” and “open-mindedness” were recognized as the 4th, 6th, 7th and 8th respectively
most important objectives among the 15 identified ones. These findings confirm that, when
confronted with important decisions, we fail to think about nearly half of the relevant objectives
that we later recognize as important and that the objectives that are missed are indeed important.

616
617

Layering, expanding, and visualizing

Table 42.3 Results of the combined value focused thinking and MCDA approach

Key concerns Fundamental objectives identified Weight (%)

Cognitive abilities Mathematical abilities 14.04

Reading comprehension 15.60
Socio-economic Underprivileged background 17.33
background Support from the children’s family 13.86
Social competences Communication skills 1.70
Empathy 6.93
Identifying the most Cooperative attitude 4.16
suitable candidates Leadership ability 2.23
for the Foundation Attention 0.90
programme Creativity 2.77
Personality Open-mindedness 4.85
Character 5.40
Emotional steadiness 6.07
Mentors’ perception Mentors’ approval 3.47
Mentors’ enthusiasm 0.70

This intervention and its findings supported the selection process for the academic year
2016/2017, when the Foundation opened applications in two cities in Hungary for a total of
17 children. The feedback provided by the President of the Foundation Advisory Board and
the Foundation Operational Director highlighted that the questions that were more effective in
helping them to identify new objectives were the questions about alternatives, about problems
and shortcomings and about goals, constraints and guidelines. This last category of questions in
particular was the one that allowed them to identify the new objective about the level of family
support, which emerged as a crucial indicator.
The key impacts observed after this multi-methodology boost can be summarized as follows:

1 Learning effect: the Foundation’s members indeed discovered new important objectives for
the programme, thanks to the combined value focused thinking and MCDA intervention.
2 Consensus building: the facilitated modelling approach employed in the intervention allowed
the board members to adopt an inclusive perspective throughout the process and thus
achieve consensus on both objectives to be achieved and the candidates to be selected.
3 Innovation: after the intervention, the Foundation board voted for the embedding of the
new tool in all subsequent selection processes and unanimously approved it.
4 Reduced drop-out rates: after one year monitoring of the Foundation programme, no children
have dropped-out for the academic year 2017–2018.

Visualizing preferences: how to support value functions’ elicitation

in decision making
The cognitive challenge and the context of the intervention
When decision makers have a clear set of objectives to achieve in a certain decision-making
context, it is helpful to measure the performance of each alternative on these objectives and
then aggregate performances and preferences into an overall score for each alternative. One of
the most consolidated approaches to do so is Multi-Attribute Value Theory (Keeney & Raiffa,

617
618

Valentina Ferretti

1976), where the crucial step of preference elicitation revolves around the definition of the
value functions and of the trade-offs coefficients.
Since people do not naturally express preferences and values using interval scales, value
functions have to be estimated through a specially designed interviewing process in which the
relevant judgements for the decision are organized and represented analytically.
Single-attribute value functions can be elicited with different methods (e.g., Von Winterfeldt
and Edwards, 1986). A particularly promising one in the context of environmental decision
making is the Mid-value Splitting Method (Bisection), which seems to lead to more reliable
results than direct rating (Schuwirth et al., 2012). However, a disadvantage of the protocol
consists in the generation of relatively abstract questions, which can dramatically increase the
cognitive load on the interviewee and delay the elicitation process or even make it fail (e.g.,
Schuwirth et al., 2012; Ferretti, 2016b).
The following paragraphs provide an account of the preference elicitation challenge and a
proposed solution tested within an abandoned railways requalification project in the Piedmont
Region in Italy.3 In 2012, in this region alone, 12 passenger railway lines, characterized by low
patronage, have been replaced by bus services.
Among these lines, the Pinerolo-Torre Pellice, which stretches for 16.5 km between the city
of Pinerolo and the Pellice Valley, crossing six municipalities, was selected as the most strategic
one to be studied for requalification purposes. Thanks to the development of stakeholders’
analysis, best practices’ analysis and a value focused thinking approach, we were able to ana-
lyse the regeneration opportunity through nine objectives (i.e., maximizing the square metres
of new green areas, maximizing the compatibility with the existing land use, minimizing the
construction works’ duration, minimizing the damage to the aesthetic impact on the landscape,
minimizing costs, maximizing new job opportunities, maximizing the positive impacts on the
touristic sector, maximizing the number of potential users, and maximizing the presence of
local attractions) and five possible solutions (creation of a greenway, rail banking, extension
of the urban railway service, old station recovery and “no action”). The reader interested in
knowing more about the deployment of this intervention can refer to Ferretti and Degioanni
(2017).
To allow for the aggregation of the alternatives’ performances across objectives, we elicited
value functions. Eliciting value functions means translating the performances of the alternatives
into a value score, which represents the degree to which an objective is achieved. The value
is a dimensionless score: 1 refers to a very good performance (i.e., full achievement of the
objective), while 0 refers to a poor performance (i.e., low objective achievement). To elicit
intermediate values in this intervention we used the mid-value splitting method and the
interviewing process was organized as a series of two half-day meetings with an expert in
the field of transportation planning and a consultant for different local authorities involved
in transportation-related projects (see Ferretti and Degioanni, 2017, for more details). In this
study two meetings were necessary for the value functions’ elicitation task because during the
elicitation of the value function for the attribute “new green areas” in the first meeting, the
expert being interviewed highlighted some drawbacks of the elicitation protocol, which will be
explained in the following paragraphs.
In particular, the elicitation protocol consists in interactively asking the interviewee to com-
pare two intervals (e.g., from a to b and from b to c, where a is the worst performance value
in the range and c is the best performance value in the range) in order to find the indifference
point, i.e. the performance value for which the interviewee is indifferent between an improve-
ment from a to b or from b to c.

618
619

Layering, expanding, and visualizing

In order to efficiently cope with time constraints inherently associated with real actors’ par-
ticipation, we elicited value functions by asking the expert for the mid-value of the intervals
[v=0, v=1], [v=0, v=0.5], and [v=0.5, v=1]. The mid-value of the interval [v=0.25, v=0.75]
was used as consistency check (see, e.g. Ferretti, 2016b, for a detailed explanation of the
protocol). To provide an example, the expert was asked the following question with reference
to the attribute “new green areas”:

Imagine two possible scenarios for this project: in the first scenario the square metres
of new green areas increase from 0 to 80,000, while in the second scenario the square
metres of new green areas increase from 80,000 to 160,000. Would you be equally
satisfied in these two scenarios?

During the first meeting the involved expert stated that the mid-value splitting protocol was
generating questions that were too abstract, making it difficult for him to provide a reliable
answer.

Proposed solutions
To improve the operability of the protocol, we enhanced it with visualization analytics. In
particular, each time we used a coordinate plane labelled with the attribute range and we inter-
actively modified the extension of the two intervals considered in each question (Ferretti and
Degioanni, 2017). Moreover, with reference to the attribute “new green areas”, in order to
make the questions more concrete and realistic, we linked each value of the subsequent intervals
to physical pictures of urban parks that are well known to the inhabitants of the region under
analysis. This helped the expert to clearly associate a quantity to the different extensions that
were compared in each question and thus to better complete the elicitation task.
Figure 42.1 shows the first step of the combined use of interactive visual analytics and the
mid-value splitting protocol that we experimented for the elicitation of the value function for

100
90
80
70
60
50
%

PICTURE OF A PICTURE OF A
KNOWN URBAN KNOWN URBAN
40 PARK IN TURIN WITH PARK IN TURIN WITH
AN EXTENSION OF AN EXTENSION OF
30
80000 M2 (PARCO 160000 M2 (PIAZZA
20 SEMPIONE) D’ARMI)

10 ΔV1 ΔV2
0
0 20000 40000 60000 80000 100000 120000 140000 160000
m2 of new green areas

Figure 42.1 Combined use of interactive visual analytics and the mid-value splitting protocol for the
attribute “m2 of new green areas”

619
620

Valentina Ferretti

the attribute “m2 of new green areas”. Using this visual approach, the new question that we
asked to the expert became:

Imagine two possible scenarios for this project: in the first scenario the square
metres of new green areas increase from nothing to something that is as big as Parco
Sempione, while in the second scenario the square metres of new green areas increase
from something that is as big as Parco Sempione to something that is as big as Piazza
d’Armi. Would you be equally satisfied?

The expert was now able to relate to the question and provided coherent answers to all subse-
quent questions.
A similar procedure was repeated for the other similar attributes. The elicited midpoints
were marked on the coordinate plane and were finally interpolated to a value function which
was further discussed with the interviewee to stimulate the learning effect arising from graph-
ical awareness and real-time visualization of results (Ferretti, 2016b).

Results and impacts of the multi-methodology boost

As a result, during the second workshop, the expert was able to understand the bisection
questions and easily state his preference for some of the improvements proposed by the elicit-
ation protocol. The combined use of visualization analytics and sound preference elicitation
protocols thus reduced the scope insensitivity (e.g., Carson and Mitchell, 1993) linked to the
quite abstract questions generated by the bisection elicitation protocol.
Figure 42.2 shows the elicited value function for the attribute “m2 of new green areas”.
For example, the answer to the first question of the new type mentioned above was “I am not
indifferent between these two scenarios, I actually prefer the first one.” During the second iter-
ation of the protocol we thus had to shorten the interval that satisfied most the expert being
interviewed, i.e., the 0–80,000 one. The second question that we asked was then

value
100
90
80
70
60
50
%

40
30
20
10
0
0 20000 40000 60000 80000 100000 120000 140000 160000
sq meters of new green areas

Figure 42.2 Final value function for the attribute “m2 of new green areas”

620
621

Layering, expanding, and visualizing

Third
indifference 100
100 point
90 First
80 indifference 75
point
70 Second
60 indifference
50
point
%

50
40
30 25
20
10
0
0
0 20000 40000 60000 80000 100000 120000 140000 160000
PICTURE OF A PICTURE OF A PICTURE OF A PICTURE OF A
KNOWN URBAN KNOWN URBAN KNOWN URBAN KNOWN URBAN
PARK IN TURIN WITH PARK IN TURIN WITH PARK IN TURIN WITH PARK IN TURIN WITH
AN EXTENSION OF AN EXTENSION OF AN EXTENSION OF AN EXTENSION OF
30000 M2 70000 M2 100000 M2 1600000 M2

Figure 42.3 Visual supports used to interactively elicit three indifference points for the value function
of the attribute “m2 of new green areas”

Imagine two possible scenarios for this project: in the first scenario the square metres
of new green areas increase from nothing to something that is as big as Parco della
Tesoriera4 while in the second scenario the square metres of new green areas increase
from something that is as big as Parco della Tesoriera to something that is as big as
Piazza d’Armi. Would you be equally satisfied with these two scenarios?

The answer was now yes and we were thus able to identify the first indifference point of
the graph (Figure 42.3). We used the same approach to identify the other points of the
function.
The key impacts observed after this multi-methodology intervention can be summarized as
follows:

1 Enhanced operability of MCDA thanks to the combined use of visualization analytics and
consolidated preference elicitation protocols. Such an integrated approach is expected to
increase the use of MCDA in strategic decisions.
2 Development of an online software package which implements the integration of the
proposed tools and approaches (Lopez, 2015), thus allowing for replicability of this frame-
work in other decision contexts.5
3 Opening of a new research avenue on tools to mitigate scope insensitivity.

Conclusion
This chapter proposed three examples of multi-methodology boosts designed to support three
key stages of the decision-making process, i.e., the framing of the problem/opportunity to be
addressed, the expansion of the original decision makers’ mental model about the objectives to
be achieved and, finally, the elicitation of preferences about the worthiness of the alternatives’
performances.

621
622

Valentina Ferretti

In the first intervention, the use of a spatially weighted SWOT analysis provided com-
prehensive support early in the life cycle of the project by highlighting the spatial density of
some critical factors (e.g., abandoned farms in the World Heritage site) and thus significantly
improving the components of purpose and scope of the decision frame.
In the second intervention, the combined use of value focused thinking and MCDA allowed
the decision makers to update their initial mental models with new relevant objectives that
promoted consensus, learning and greater commitment for the subsequent convergent phase of
the decision-making process.
In the third intervention, the combination of the bisection elicitation protocol and visu-
alization methods (i.e. a coordinate plane labelled with the attribute range where the exten-
sion of the two intervals is interactively modified and corresponding graphical pictures are
associated to each quantity) led to a more user-friendly elicitation protocol for the construc-
tion of value functions within MCDA and to reduced scope insensitivity in the decision
maker’s judgements.
There are three common denominators to the behavioural decision analysis interventions
discussed in this chapter. First, multi-methodology boosts require time and facilitators skilled in
multiple methods. Second, a key advantage of the proposed boosts is their applicability to many
different domains. Third, they stretch the bounds that the uncertain and often constrained
external environment places on human rationality, thus mitigating some of the associated cogni-
tive consequences such as poor framing of decisions, insufficient thinking about relevant object-
ives and scope insensitivity. In particular, the presence of limited information on alternatives and
their consequences can be tackled through informed value judgements supported by the inte-
gration of sound preference elicitation protocols and visualization analytics, as well as through
thought-provoking questions to expand the set of objectives and create better alternatives. To
cope instead with the constraints on the time available to make the decision, it seems beneficial
to use a structured approach like the one proposed in the three projects, as it has been shown to
lead to decision quality in both the divergent and subsequent convergent phase of the strategic
decision-making process, giving confidence in the decision and commitment for action, thus
avoiding decision paralysis and decision postponing.
The techniques proposed in this chapter should thus be considered as “process boosts” and
in particular as “framing boosts” and as “preference elicitation boosts”, which could be added
to first taxonomy of long-term boosts proposed by Hertwig and Grüne-Yanoff (2017), i.e.,
risk literacy boosts, uncertainty management boosts, and motivational boosts. A final high-
light concerns the new competences developed or boosted by the proposed techniques, as
summarized in Table 42.4. Current developments of this research are testing how long-lasting
the developed competences are.

Table 42.4 New competences boosted by the proposed techniques

Stage of the decision- Boosting techniques used New or boosted competences

making process

Framing the decision Spatial SWOT Analysis and MCDA Boosted capacity to identify spatial
context correlations
Framing objectives Value-focused thinking devices Boosted capacity to think wider and
deeper about objectives
Preference elicitation Visual analytics and sound preference Boosted scope sensitivity
elicitation protocol

622
623

Layering, expanding, and visualizing

Notes
1 This project won the runner-up position in the INFORMS Decision Analysis Practice Award 2016.
2 This project won the INFORMS Decision Analysis Practice Award 2017.
3 This project was awarded the second position in the INFORMS 2017 Innovative Applications in
Analytics Award.
4 The selected park covers a surface of 70000 m2.
5 The work developed in this master’s thesis won the title of Best Master Thesis 2015 for the IT4BI
Master, Ecole Centrale, Paris).

References
Baumol, W. J., and Quandt, R. E. 1964. Rules of thumb and optimally imperfect decisions. American
Economic Review, 54, 23–46.
Bazerman, M. 2008. Judgment in managerial decision making. Hoboken, NJ: John Wiley & Sons.
Belton V., and Stewart T. J. 2002. Multiple criteria decision analysis: An integrated approach. Boston: Kluwer
Academic Publishers.
Bendoly, E., Van Wezel, W., and Bachrach, D. G. (Eds.) 2015. The handbook of behavioural operations manage-
ment: Social and psychological dynamics in production and service settings. New York: Oxford University Press.
Birnberg, J. G., Luft, J., and Shields, M. D. 2007. Psychology theory in management accounting research.
In C. S. Chapman and A.G. Hopwood (Eds.), Handbook of management accounting research (pp. 113–135).
Oxford: Elsevier.
Bond, S., Carlson, K., and Keeney, R. 2008. Generating objectives: Can decision makers articulate what
they want? Management Science, 54(1): 56–70.
Bond, S. D., Carlson, K. A., and Keeney, R. L. 2010. Improving the generation of decision objectives.
Decision Analysis, 7(3): 238–255.
Bruce, B. (Ed.) 2010. Handbook of behavioural finance. Northampton: Edward Elgar Publishing.
Camerer, C. F. 2003. Behavioral game theory: Experiments in strategic interaction. Princeton, NJ: Princeton
University Press.
Camerer, C. F., and Lowenstein, G. (Eds.) 2003. Advances in behavioural economics. Princeton, NJ: Princeton
University Press.
Carson, R. T., and Mitchell, R. C. 1993. The issue of scope in contingent valuation studies. American
Journal of Agricultural Economics, 75(5).
Comino, E., and Ferretti, V. 2016. Indicators-based spatial SWOT analysis: Supporting the strategic
planning and management of complex territorial systems. Ecological Indicators, 60: 1104–1117.
Cronin, M.A., and Weingart, L.R. 2007. Representational gaps, information processing, and conflict in
functionally diverse teams. Academy of Management Review, 32: 761–773.
Ferretti, V. 2016a. Exploring multi-methodology approaches for policy support: Case studies and lessons
learned. OR58. Paper presented at The OR Society Annual Conference.
Ferretti, V. 2016b. From stakeholders analysis to cognitive mapping and Multi-Attribute Value Theory: An
integrated approach for policy support. European Journal of Operational Research, 252(2): 524–541.
Ferretti, V., and Degioanni, A. 2017. How to support the design and evaluation of redevelopment projects
for disused railways? A methodological proposal and key lessons learned. Transportation Research Part
D, 52: 29–48.
Ferretti, V., and Gandino, E. 2018. Co-designing the solution space for rural regeneration in a new
World Heritage site: A Choice Experiments approach. European Journal of Operational Research,
268(3): 1077–1091.
Franco, L. A., and Hämäläinen, R. P. 2016. Behavioural operational research: Returning to the roots of
the OR profession. European Journal of Operational Research, 249: 791–795.
Galizzi, M. 2014. What is really behavioral in behavioral health policy? And does it work? Applied Economic
Perspective, 36(1): 25–60.
Gigerenzer, G., and Selten, R. (Eds.) 2002. Bounded rationality: The adaptive toolbox. Cambridge,
MA: MIT Press.
Gigerenzer, G., Hertwig, R., and Pachur, T. (Eds.). 2011. Heuristics: The foundations of adaptive behaviour.
Oxford: Oxford University Press.
Hämäläinen, R. P. 2015. Behavioural issues in environmental modelling: The missing perspective.
Environmental Modelling & Software, 73: 244–253.

623
624

Valentina Ferretti

Henao, F., and Franco, L. A. 2016. Unpacking multimethodology: Impacts of a community development
intervention. European Journal of Operational Research, 253(3): 681–696.
Hertwig, R., and Grüne-Yanoff, T. 2017. Nudging and boosting: Steering or empowering good decisions.
Perspectives on Psychological Science, 12(6): 973–986.
Johnson, M. 2012. Community-based operations research: Introduction, theory, and applications. In
M. Johnson (Ed.), Community-Based Operations Research: Decision Modeling for Local Impact and Diverse
Populations (pp. 3–36). Boston: Springer.
Keeney, R. L. 1982. Decision analysis: An overview. Operations Research, 30(5): 803–838.
Keeney, R. L. 1996. Value-focused thinking: Identifying decision opportunities and creating alternatives.
European Journal of Operational Research, 92(3): 537–549.
Keeney, R. L., and Raiffa, H. 1976. Decisions with multiple objectives: Preferences and value trade-offs.
New York: Wiley.
Larrick, R. P. 2009. Broaden the decision frame to make effective decisions. In E. A. Locke (Ed.), Handbook
of principles of organizational behavior (2nd ed., pp. 461–480). Chichester: John Wiley & Sons, Ltd.
Lopez, A. M. 2015. ElectioVis. master’s thesis. Ecole Centrale, Paris.
Morton, A., and Fasolo, B. 2009. Behavioural decision theory for multi-criteria decision analysis: A
guided tour. Journal of the Operational Research Society, 60: 268–275.
Myllyviita, T., Hujala, T., Kangas, A., Eyvindson, K., Sironen, S., ... Kurttila, M. 2014. Mixing
methods: Assessment of potential benefits for natural resources planning. Scandinavian Journal of Forest
Research, 29(1): 20–29.
Powell, T. C., Lovallo, D., and Fox, C. R. 2011. Behavioural strategy. Strategic Management Journal,
32: 1369–1386.
Schuwirth, N., Reichert, P., and Lienert, J. 2012. Methodological aspects of multi-criteria decision ana-
lysis for policy support: A case study on pharmaceutical removal from hospital wastewater. European
Journal of Operational Research, 220: 472–483.
Shogren, J. F., and Taylor, L. O. 2008. On behavioural environmental economics. Review of Environmental
Economics and Policy, 2(1): 26–44.
Simon, H. A. 1955. A behavioral model of rational choice. The Quarterly Journal of Economics, 5: 99–118.
Simon, H. A. 1982. Models of bounded rationality. Cambridge, MA: MIT Press.
SiTI (Istituto Superiore sui Sistemi Territoriali per l’Innovazione) 2013. Dossier di Candidatura UNESCO
per il sito piemontese I Paesaggi Vitivinicoli di Langhe-Roero e Monferrato, Turin.
Slovic, P. 1987. Perception of risk. Science, 236(4799): 280–285.
Spetzler, C., Winter, H., and Meyer, J. 2016. Decision quality, value creation from better business decisions.
Hoboken, NJ: John Wiley and Sons.
Tani, S. N., and Parnell, G. S. 2012. Frame the decision opportunity. In G. S. Parnell, T. Bresnick, S. N.
Tani, and E. R. Johnson (Eds.), Handbook of decision analysis. New York: Wiley.
Thaler, R. H. 1999. The end of behavioural finance. Financial Analysts Journal, 56(6): 12–17.
Thaler, R. H., and Sunstein, C. R. 2008. Nudge: Improving decisions about health, wealth, and happiness. New
Haven, CT: Yale University Press.
Trope, Y., and Liberman, N. 2000. Temporal construal and time-dependent changes in preference. Journal
of Personality and Social Psychology, 79(6): 876–889.
Tversky, A., and Kahneman, D. 1974. Judgment under uncertainty: Heuristics and biases. Science,
185(4157): 1124–1131.
Viale, R. 2012. Methodological cognitivism. Vol. I, Mind, rationality and society. Heidelberg: Springer.
Viale, R. 2019. The normative and descriptive weaknesses of behavioral economics-informed
nudge: Depowered paternalism and unjustified libertarianism. Mind & Society, 17(1–2): 53–69.
Von Winterfeldt, D., and Edwards, W. 1986. Decision analysis and behavioural research. Cambridge: Cambridge
University Press.

624
625

43
COGNITIVE AND
AFFECTIVE CONSEQUENCES
OF INFORMATION AND
CHOICE OVERLOAD
Elena Reutskaja, Sheena Iyengar, Barbara Fasolo, and
Raffaella Misuraca

Introduction
When interviewed in 1992 in Pittsburgh, Pennsylvania, the Nobel laureate Herbert Simon
described a paradox at the heart of living in an economy that made every effort to design and
produce ever more “choice alternatives” but that simultaneously allocated very little energy to
encouraging people to devote the attention and time actually required to choose. He gave the
example of a decision to buy a new house, commenting: “Before you even start the choice pro-
cess, somebody has presented you with this, and this, and this house” (UBS, 1992).
The overabundance of alternatives was lamented by Simon in 1992, when computing power
was slower. It is all the more alarming in the modern and constantly connected world, which
now has the internet, smartphones, apps, and tablets—all used to make a plethora of decisions
every day.
Nowadays, people receive information from ever-increasing and often simultaneous sources.
The average US resident consumer views about 3,000 advertisements every day (Kardes, Cline,
& Cronley, 2011) and while in the 1970s to the 1990s grocery stores in the United States
carried around 7,000–8,000 items, the variety has increased to 40,000–50,000 items nowadays
(Jacoby et al., 1974a; Malito, 2017), including around 285 varieties of cookies and 275 types
of cereal (Schwartz & Ward, 2004). The explosion of choice is not limited to retail either and
has begun to permeate even people’s personal lives. A speed-dating event organized by China’s
Communist Youth League in 2017 was attended by about 5,000 young single people (Shim,
2017). And in a world where the internet increases an individual’s access to information, it
seems being surrounded by an overwhelming amount of information every day has become
the new norm.
The aim of this chapter is to examine this state of affairs. What is information and choice
overload, and what are the cognitive and emotional consequences of this overload?
Classical economics and psychology have argued that increased information and choice are
often desirable and lead to better outcomes (Steiner, 1970; Zuckerman, Porac, Lathin & Deci,
1978; Walton & Berkowitz, 1979; Rolls et al., 1981; Deci & Ryan, 1985; Loewenstein, 1999;
Ryan & Deci, 2000; Kahn & Wansink, 2004). However, theories of bounded and adaptive

625
626

Elena Reutskaja et al.

rationality posit the opposite (Simon, 1957 1991;Gigerenzer & Selten, 2002) and have been
supported by a large body of research. Extensive information and choice can be costly, demo-
tivating, and unsatisfying (Miller, 1956; Newell & Simon, 1972; Jacoby, 1974; Malhotra, 1982;
Iyengar & Lepper, 2000; Schwartz, 2004; Reutskaja & Hogarth, 2009; Grant & Schwartz,
2011; Reutskaja, Nagel, Camerer, Rangel, 2018). They can result in what Schwartz (2000)
has called a “tyranny of freedom.” However, using less information and making decisions based
on less information can lead to higher-quality outcomes. (For a review, see Gigerenzer &
Gaissmaier, 2011.)

Information and choice overload: a theoretical background

Definition of information and choice overload
Information load is closely connected to Herbert Simon’s behavioral model of rational choice,
or “a kind of rational behavior that is compatible with the access to information and the com-
putational capacities that are actually possessed by organisms, including man, in the kinds of
environments in which such organisms exist” (Simon, 1955, p. 99). In line with Simon, later
research has sought to differentiate information “load” from “overload.” The result has been a
long-standing debate in the relevant literature about how “information load” and “information
overload” can and should be measured. Jacoby (1977, p. 569) provides the following definitions:

Information load refers to the variety of stimuli (in type and number) to which the
receiver must attend. Information overload refers to the fact that there are finite limits
to the ability of human beings to assimilate and process information during any given
unit of time. Once these limits are surpassed, the system is said to be “overloaded” and
human performance (including decision making) becomes confused, less accurate,
and less effective.

Many researchers use these definitions and manipulate information load by varying both
the number of items and the number of attributes describing those items (Jacoby, Speller &
Berning, 1974; Jacoby, Speller & Kohn, 1974, Lee & Lee, 2004). Others argue that infor-
mation load cannot be measured using the product of the number of items and number of
attributes (Malhotra, Jain & Lagakos, 1982). As yet, there is no consensus on how to appro-
priately measure information load and, as a result, any effects driven by the number of choice
alternatives and choice attributes remain controversial.
In addition to information overload, it is necessary to define a closely related concept,
that of choice overload. Choice overload occurs when an increase in the number of options
to choose from has detrimental consequences. For example, choice overload may manifest
itself as lower intrinsic motivation to choose, decreased satisfaction with the option finally
chosen, decreased satisfaction with the process of choosing, stress, anxiety and choice paralysis
(Iyengar & Lepper, 2000; Schwartz, 2004; Shah & Wolford, 2007; Reutskaja & Hogarth, 2009;
Grant & Schwartz, 2011; Chernev, Böckenholt, & Goodman, 2015; Nagar & Gandotra, 2016;
Reutskaja et al., 2018).
Both information and choice overload can arise for both cognitive and affective reasons but
they differ in one important way: choice overload occurs only when there is a need to choose
an item from a set of alternatives. In contrast, information overload may occur regardless of
whether there is a need to make a choice or not. In the latter case, information overload is
driven not by choice per se but by the overwhelming amount of information to process.

626
627

Information and choice overload

Though it may be possible for choice overload to occur without information overload, the
two often go hand in hand and, for this reason, this chapter will use the terms interchangeably.
Despite the different approaches in methodology, most scholars do not dispute the premises of
information or choice overload (with some exceptions, such as Scheibehenne, Greifeneder, &
Todd, 2010).

Processes underlying the information-overload phenomenon

Next to be discussed are the underlying mechanisms of choice overload, which are connected
to the benefits and costs of choice and information provision.
The benefits of increased access to information are nearly always self-evident. Generally,
there is an increased likelihood that people’s diverse needs and wants will be satisfied. Full and
transparent information should also promote market competition by driving prices down and
quality up (Loewenstein, 1999). Moreover, actual variety and perceived variety often have the
potential to increase consumption (Kahn & Wansink, 2004; Rolls et al., 1981), while access
to more choice offerings can improve consumers’ welfare (Brynjolfsson, Hu, & Smith, 2003).
Full information and an abundance of choice can also be beneficial from a psychological
point of view. As a thought experiment, contrast the emotional feelings experienced when
entering a grocery store offering only a few options with what you might feel when entering
a well-stocked competitor. In general, individuals are attracted to larger choice sets (see, for
example, Iyengar & Lepper, 2000). This can be explained, at least in part, by research demon-
strating the association between more choice alternatives and greater perceived decision freedom
(Steiner, 1970; Reibstein, Youngblood, & Fromkin, 1975; Walton & Berkowitz, 1979).
On the other hand, information may also impose costs on an individual. Costs come in
the form of extra time spent assimilating new information, the potential for error, the greater
cognitive burden, and the potential to create an affectively taxing experience, such as feeling
greater regret as the number of forgone options increases (Miller, 1956; Loewenstein, 1999;
Botti & Iyengar, 2006; Reutskaja, 2008). Those are discussed in more detail below.
A theoretical model describing how the benefits and costs of choice can be integrated has
been proposed in previous research (Reutskaja & Hogarth, 2009). This model suggests that,
while choice implies both benefits (“goods”) and costs (“bads”), which may both increase
with the number of available alternatives (see also Coombs & Avrunin, 1977), the benefits
increase more slowly than the costs. This proposition was also supported later by Grant and
Schwartz (2011). As a result, their model’s first implication is that an intermediate amount
of choice leads almost always to a more optimal choice offering than a small or large number
of choices. That is, the provision of choice is beneficial up to a certain point, after which it
becomes detrimental and choice overload occurs. The net benefit to this outcome varies as a
function of the number of alternatives, which can be graphed as an inverted U shape, as shown
in Figure 43.1.
Another important consequence of this model is that changes in actual and perceived costs
and benefits shift the location of the function’s peak. For example, with benefits being held
constant, lower costs will shift the peak to the right, allowing people to deal with more choice
offerings. The costs and benefits of choice (and therefore, the location of the peak of the
resulting function) can be moderated by several significant factors: features of information and
the environment (such as perceptual attributes of information, organization of the informa-
tion set, whether a decision is made with or without time pressure, the presence or absence of
brand names) and individual differences between the decision makers (such as gender and age).
A discussion of the moderators of choice overload is beyond the scope of this chapter but for

627
628

Elena Reutskaja et al.

benefits benefits

benefits net benefits

0 0
number of choice number of choice
alternatives alternatives

costs

costs costs

Figure 43.1 Satisfaction as a function of the number of alternatives in the choice set
Source: Adapted from similar figures originally published in Reutskaja and Hogarth (2009).

more on this subject, see Chapter 44 in this volume, “How much choice is ‘good enough’?
Moderators of information and choice overload.”
Overall, the underlying mechanism described above can explain many findings from previous
research. Empirical evidence indeed suggests that variables such as choice quality, motivation
for choice and buying, and satisfaction with both process and outcome exhibit an inverted-U
relationship with the number of choice alternatives and/or the complexity of each alternative.

Empirical evidence: effect of the provision of information and choice

on the decision-making process and the outcomes
Next, this section considers empirical evidence of the effects of information and choice pro-
vision on the various steps of a decision-making process: information processing and usage,
decision-making outcomes (quality and accuracy), motivation for choosing, consumption and
buying, and subjective perceptions (individuals’ subjective and emotional states).

Information processing and usage

The provision of information can impose cognitive costs on individual decision-makers.
Cognitive costs include attention costs and the effort required to process information and
make trade-offs. In his classic 1956 paper, “The magical number seven, plus or minus two”
(Miller, 1956), George A. Miller argued that the amount of input information was not equal
to the amount of information transmitted. This is due to a human’s limited ability to receive,
process, and remember information. In particular, Miller demonstrated that, as the quantity of
input information increased, the amount of transmitted information would quickly plateau.
The “channel capacity”—or the greatest amount of information that a human can process
without error—was shown to be surprisingly low for unidimensional variables (around four
for tastes, six for sounds or tones, and from 10 to 15 for visual positions). Miller suggested that
the total “channel capacity” could be improved by increasing the number of dimensions in
which given objects differed. However, the accuracy of judgment on each particular attribute

628
629

Information and choice overload

would decline. Subsequently, Chase & Simon (1973) and Simon (1974) argued that the brain
possessed a “chunk” capacity and that short-term memory was capable of retaining a range of
five to seven pieces of information. It was shown that it took time to consciously identify an
object—sometimes more than half a second. In addition, the amount of information that the
brain can simultaneously maintain in its short-term memory is quite small: the brain is able
to remember or monitor only about four objects at once, regardless of how many objects are
shown to a person at a time (Marois & Ivanoff, 2005).
In addition, at any given time, the mind normally selects only a subset of relevant infor-
mation from a scene to process in detail. When full attention is paid to a certain location or
item, the attention given to other locations and objects is suppressed (Wedel & Pieters, 2007).
This inattentional blindness has been well demonstrated in one of the most famous studies in
psychology, known as the Invisible Gorilla Test (Simons & Chabris, 1999). In this experiment,
participants were asked to watch a short video in which students, wearing either a black or a
white T-shirt, passed a basketball between themselves. The participants’ task was to count the
number of times that players wearing a white T-shirt passed the ball. During the video a person
in a gorilla suit walked through the center of the scene. However, about half of the participants
did not actually see the gorilla, which suggests that the relationship between what is in our
visual field and what we actually perceive is based mostly on attention.
Attention costs are important costs to consider when dealing with lots of choice or infor-
mation because attention is limited and selective. However, it is hard to measure many covert
processes behind the decisions. Therefore, recent research has used modern tools, including eye
tracking and brain imaging to open the black box behind the phenomenon of choice overload.
By tracking people’s eye movements, previous research has found that attention costs increase
when people make choices from large rather than small sets of alternatives under time pressure
(Reutskaja et al., 2011). When the number of snacks that people were choosing from increased,
the number of eye fixations1 and the time spent on choosing a snack increased significantly.
The increased number of eye fixations likely reflects higher attention and processing costs (or
the time necessary to perceive and understand the information), which suggests that a higher
number of alternatives in the set imposes higher processing and attention costs on decision
makers.
The functional magnetic resonance imaging (fMRI) method, which tracks activity in the
brain, was also used to examine the processes behind the choice-overload phenomenon and
revealed that choices from large sets of alternatives were associated with a greater cognitive load.
Subjects choosing from larger (rather than smaller) choice sets showed increased eye movement
and stronger activity in visual and sensorimotor brain areas—areas previously associated with
the planning and control of movement such as reaches and saccades, the processing of visual
scenes or both (Andersen & Buneo, 2002; Grill-Spector & Malach, 2004; Orban, Van Essen,
& Vanduffel, 2004).
The amount of processed information that people use for their decision making (infor-
mation usage) is also constrained by bounded human processing abilities (Simon, 1976) and
imposes cognitive and time costs on decision-makers. Empirical evidence suggests that infor-
mation usage follows an inverted U-shaped curve with the increase in information load (Miller,
1956; Schroder, Driver & Streufert, 1967; Newell & Simon, 1972; Chewning & Harrell,
1990; Tuttle & Burton, 1999). Research has found that input-cue usage increases initially
with the number of cues provided but does not show any further increase after a particular
point (Chewning & Harrell, 1990; Stocks & Harrell, 1995). In addition, people consider and
process smaller percentages of information when the number of options and amount of rele-
vant information increase (Hauser & Wernerfelt, 1990). For example, a study tracking the eye

629
630

Elena Reutskaja et al.

movements of decision-makers has shown that, when participants chose a snack to consume
out of different-sized sets, the average number of items seen in a larger set was greater compared
to the smaller set but the percentage of items seen actually decreased with the set size (Reutskaja
at al., 2011). So, when confronted with few items (two or three alternatives), people often pro-
cess and use all the relevant information (Payne, Bettman, Coupey, & Johnson, 1992) but, in
an environment with many alternatives, people shift to simpler choice strategies (Rolls et al.,
1981). A more recent and large body of research found that the cognitive costs are not neces-
sarily as high as the classical theory suggested. Research on adaptive decision-making (Payne,
Bettman, & Johnson, 1993) and fast and frugal heuristics (such as Gigerenzer & Goldstein,
1996; Gigerenzer & Gaissmaier, 2011) established that humans could be good information
processors when using simple heuristics. Simple heuristics use little of the large amounts of
data available when the information is valid and the heuristic fits the choice environment well.
Simple heuristics such as “take the best” (Gigerenzer & Goldstein, 1999) can also be used as
techniques to simplify decisions, just like mnemonics and mnemotechnics can be used to over-
come memory bottlenecks. To overcome overload, summaries, the prioritization of informa-
tion, and mnemonics can be useful and have been used in medical education practice to help
cope with feelings of information-caused stress (Smith, 2010).
Another cost of too much information and choice is the time cost. Over the decades, evi-
dence has been accumulated that time is an increasing function of the amount of information
and choice: There is a positive linear relationship between the total amount of information
available and the time spent processing and evaluating information (Jacoby, 1974); people take
longer to decide when confronted with larger as opposed to smaller choice sets (Iyengar &
Lepper, 2000). Decision time rises over the entire range of task complexity as the number of
alternatives in a choice set increases (Hogarth, 1975; Loewenstein, 2000).
However, there is some biological evidence (eye movements) showing that people can
somewhat shorten processing times per item when the choice set increases (Reutskaja et al.,
2011). Ironically, even if people actually take longer to choose from larger sets, the time that is
perceived to pass often feels shorter (Fasolo, Carmeci, & Misuraca, 2009). This faulty perception
can be harmful, leading to delays, further increases in time pressure, and time mismanagement.
Indeed, many people tend to underestimate the time spent making a decision when facing an
extensive array (say, 24 options) and, inversely, overestimate the time spent when facing a small
choice set (say, only six options).

Motivation for choosing and consumption

Motivation to act refers to the intention to select one or more alternatives from a given set in
order to then act upon them in one of many ways, including choosing, buying, or consuming
them. More choice often means greater variety, and the provision of some variety has a posi-
tive effect on the motivation for choosing, purchasing and consumption. For example, in their
study on food intake, Rolls et al. (1981) found that the participants consumed more yogurt
when presented with three options—each differing in taste, color, and texture—compared to
when just one yogurt flavor was offered. Interestingly, when the three yogurts were different
flavors but visually indistinguishable, no increase in food consumption was observed. Moreover,
consumption increases not only when the actual variety increases but also when the perceived
variety increases (Kahn & Wansink, 2004).
However, too much information and choice may also lower the motivation to choose
and cause people to defer choosing or simply to select a default option (Dhar, 1997; Shafir,
Simonson, & Tversky, 1993). Probably the best-known evidence of choice overload is the “jam

630
631

Information and choice overload

study” (Iyengar & Lepper, 2000), where a tasting booth offering either six or 24 different flavors
of jam was set up in an “upscale grocery store” in California. Although more customers stopped
at the larger booth (the one with 24 rather than six jam flavors), they purchased less when faced
with the larger set of jam. A lab study carried out by the same authors (Iyengar & Lepper, 2000)
found that participants who sampled from a limited rather than an extensive assortment of
chocolate bars were more likely to choose chocolate instead of cash as compensation for partici-
pating. Large sets are more attractive at first sight but they are demotivating for decision makers.
So, is it better to have a larger or smaller choice set available when it comes to consump-
tion, buying, or choosing? As this review has shown, research suggests that neither of these
options is optimal, and it is best to offer a “golden mean” to decision makers. Purchasing
behavior follows an inverted-U pattern according to the number of alternatives in the set, and
the peak for buying is reached when people choose from sets of an intermediate size. In one
experiment, people could purchase pens from sets of varying sizes. More people purchased
pens from intermediate sets (10 alternatives) rather than from small sets or large sets (two or 20
alternatives). Intermediate sets were more “motivating” for people than larger or smaller sets
(Shah & Wolford, 2007). There is also biological evidence of too little and too much choice
being demotivating. A brain-imaging study in which participants had to choose images from
different-sized sets demonstrated that the activity in several brain areas also follows an inverted-U
pattern (Reutskaja et al., 2018). Activity in the striatum (the area that has been shown to reflect
reward or value processing) and the anterior cingulate cortex (ACC, the area that integrates
cost and benefits) was the highest when subjects actively chose from sets with an intermediate
number of alternatives, which were considered to have the right number of alternatives. The
fMRI activity in those areas was lower for sets that were considered by participants to be too
small or too large. Reutskaja et al. (2018) suggested that the activity in those areas reflected a
motivational signal that maintained cognitive and behavioral engagement in decision-making.
Additionally, when the choice set is large, people have a stronger preference for simple,
easy-to-understand options (Iyengar & Kamenica, 2010). More importantly, when the choice
becomes too difficult, people often defer that choice, even when it negatively affects their
future well-being. Iyengar, Huberman, and Jiang (2004) studied the influence of “too much”
choice on a life-changing decision: participation in the US 401(k) retirement plan. The plans
offered anywhere from 2 to 59 investment options. For every 10-option increase, the predicted
individual participation probabilities declined by 1.5–2 percent. If there were only two funds
offered, participation rates peaked at 75 percent, while the participation rate dropped to 61 per-
cent when the number of plans amounted to 59.

Decision accuracy and quality

Decision accuracy (defined as selecting the best item out of a set), like many of outcome
variables discussed earlier, exhibits an inverted-U-shaped relationship with information load
(Jacoby et al., 1974a). For example, in a laboratory experiment, subjects were asked to choose a
detergent from among 4, 8, or 12 different brands, each of which was described using two, four
or six items of information. As the information load increased, the decision accuracy initially
increased but ultimately leveled off. The number of brands and the information available about
each brand (which both constitute information load) had opposing effects on decision accuracy.
The number of brands was negatively related to decision accuracy, while the information per
brand had a positive effect on decision accuracy. The results were confirmed in the follow-up
experiment by the same authors (Jacoby, Speller, & Berning, 1974), which used larger amounts
of information. (See also Malhorta, 1982.)

631
632

Elena Reutskaja et al.

However, an eye-tracking study demonstrated that people were able to make choices that
were better than random from sets of all tested sizes (up to 16 alternatives) even under extreme
time pressure of three seconds, although the probability of picking the best item from the set
of seen alternatives decreased slightly when the choice set increased (Reutskaja et al., 2011).
Detrimental effects of extensive information load have been further confirmed in other dis-
ciplines and contexts. For example, one study on information overload in an online environ-
ment found that the probability of a correct choice decreased as the number of attributes where
the products differed increased, although the number of alternatives had no effect on accuracy
(Lee & Lee, 2004).

Feelings and subjective states

The provision of information and choice can also affect the subjective states of individuals.
Having access to full information and choice can be beneficial from a psychological point of
view, as people like to feel “informed” and defend their “right” to access relevant information
(Jacoby, Speller, & Kohn, 1974). Research has also shown that the ability to make choices by
themselves gives people a feeling of autonomy and self-control (Zuckerman et al., 1978; Deci
& Ryan, 1985). For example, people in nursing homes feel happier and more satisfied when
provided with choices, even when the decisions are relatively unimportant and inconsequential
(Langer & Rodin, 1976). These positive psychological states have the potential to enhance life
satisfaction, general well-being, and thus social welfare overall.
However, extensive information and choice offerings can impose high psychological costs
on individuals. These costs may be emotional (rather than cognitive) and stem from discomfort
caused by uncertainty about preferences, lack of expertise, concern or regret about making an
incorrect decision, and the presence of trade-offs (see, for example, Loewenstein, 2000). The
mere fact of making a decision naturally involves some degree of emotional conflict because
the selection of one option always accompanies the rejection of other alternatives (Botti &
Iyengar, 2006). Freedom to choose can thus turn into a “tyranny” (Schwartz, 2000). A choice
offering may also induce attachment to the options in the choice set, and people can feel they
have lost the items they have not chosen. In addition to this, as the choice increases and the
number of forgone options increases, the nonchosen options appear more attractive and feelings
of loss increase (Carmon, Wertenbroch, & Zeelenberg, 2003). As the amount of information
increases, not only will more options necessarily be rejected but people’s standards in relation
to the outcomes will rise (Schwartz, 2000). If the result of the decision is unsatisfactory, people
may feel personally responsible for choosing a “wrong” alternative. When people realize that
their choices are not ideal, they fall prey to the “tyranny of freedom.” This outcome may be
exacerbated by a modern society that fosters high expectations of achieving perfection in every
aspect of life.
Selection from larger choice sets is also associated with greater perceived decision difficulty
(Iyengar & Lepper, 2000; Reutskaja & Hogarth, 2009), less desire to have additional infor-
mation (Jacoby, Speller, & Kohn, 1974; Jacoby, Speller, & Berning, 1974), less confidence in
one’s choice, and greater confusion (Malhotra, 1982; Lee & Lee, 2004). Moreover, individuals
experience less satisfaction with the task and choice when selecting from a large rather than
a small sample of choices (Malhotra, 1982; Iyengar & Lepper, 2000). However, as with many
other previously described variables, satisfaction with both the chosen item and the decision-
making process itself are inverted U-shaped functions of the number of alternatives, with
people being the most satisfied with intermediate rather than small or large sets of alternatives
(Reutskaja & Hogarth, 2009). In addition, having many alternatives may lead to less satisfaction

632
633

Information and choice overload

with decisions even when more choices yield better objective outcomes (Schwartz, 2000, 2004;
Iyengar, Wells & Schwartz, 2006).
When the choice leads to painful ethical dilemmas, the emotional impact of having to
choose is very strong, especially when the decision maker is personally responsible for making
the choice. For instance, people who personally make psychologically painful decisions (e.g.,
ending a life-sustaining treatment) feel more intense negative emotions (such as anger, depres-
sion, guilt, and regret) compared to those who face the same outcome (e.g., the death of a loved
one) as the result of a decision made by someone else. Personal responsibility adds to the feeling
of loss (Botti, Orfali, & Iyengar, 2009).
The opportunity to choose, the amount of information, and the size of a choice set all
influence not only actual post-choice feelings but also expectations about how one will feel
about the choice, which can be quite different from the actual experience. For example, one
study found that people anticipated experiencing greater satisfaction with their choice and
with the selection process, as well as less regret over their choice, when choosing from a larger
list of 20–50 potential dating candidates rather than from a smaller list of four. However, when
participants actually chose a date from a mock website, they experienced greater memory
confusion regarding their choices and showed no improvement in their affect when choosing
from 20 rather than four profiles of potential partners. So, their expectation did not match their
experience (Lenton, Fasolo, & Todd, 2008).
Overall, while the provision of information and choice can positively influence the feelings
of individuals, extensive information and choice have a detrimental effect on the subjective
states of decision makers, making intermediate sets the most satisfying. When the choice
concerns painful outcomes, the mere act of choosing can trigger negative emotions.

Conclusion
This chapter has summarized evidence collected by researchers for more than half a cen-
tury on the topic of information and choice overload, exploring how people deal with large
amounts of information, how they make choices from sets with multiple alternatives, and
what consequences choice provision has on decision-making process and choice outcomes.
Traditionally, economics and psychology have emphasized the benefits of more information and
more choice but a more recent body of research has demonstrated the negative consequences
of too much choice. Too much information and choice hinder information processing and
usage, the motivation, and the quality and accuracy of decisions, and detrimentally impact the
affective states of decision-makers. All in all, empirical evidence suggests that both too little and
too much choice and information are bad and there is a golden mean of how much choice is
enough but not too much.

Note
1 A fixation occurs when a subject looks at an item for a continuous period of time: at least 100 ms, and
usually for 200–400 ms (Salvucci & Goldberg, 2000).

References
Andersen, R. A., & Buneo, C. A. (2002). Intentional maps in posterior parietal cortex. Annual Review of
Neuroscience, 25(1), 189–220.
Botti, S., & Iyengar, S. S. (2006). The dark side of choice: When choice impairs social welfare. Journal of
Public Policy & Marketing, 25(1), 24–38.

633
634

Elena Reutskaja et al.

Botti, S., Orfali, K., & Iyengar, S. S. (2009). Tragic choices: Autonomy and emotional responses to med-
ical decisions. Journal of Consumer Research, 36(3), 337–352.
Brynjolfsson, E., Hu, Y., & Smith, M. D. (2003). Consumer surplus in the digital economy: Estimating
the value of increased product variety at online booksellers. Management Science, 49(11), 1580–1596.
Carmon, Z., Wertenbroch, K., & Zeelenberg, M. (2003). Option attachment: When deliberating makes
choosing feel like losing. Journal of Consumer Research, 30(1), 15–29.
Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Pscyhology, 4, 55–81.
Chernev, A., Böckenholt, U., & Goodman, J. (2015). Choice overload: A conceptual review and meta-
analysis. Journal of Consumer Psychology, 25(2), 333–358.
Chewning Jr, E. G., & Harrell, A. M. (1990). The effect of information load on decision makers' cue
utilization levels and decision quality in a financial distress decision task. Accounting, Organizations and
Society, 15(6), 527–542.
Coombs, C. H., & Avrunin, G. S. (1977). Single-peaked functions and the theory of preference.
Psychological Review, 84(2), 216.
Deci, E. L., & Ryan, R. M. (1985). Cognitive evaluation theory. In E. L. Deci, & R. M. Ryan, Intrinsic
motivation and self-determination in human behavior (pp. 43–85). Boston: Springer.
Dhar, R. (1997). Consumer preference for a no-choice option. Journal of Consumer Research, 24(2),
215–231.
Fasolo, B., Carmeci, F. A., & Misuraca, R. (2009). The effect of choice complexity on perception of time
spent choosing: When choice takes longer but feels shorter. Psychology & Marketing, 26(3), 213–228.
Gigerenzer, G., & Gaissmaier, W. (2011). Heuristic decision making. Annual Review of Psychology, 62,
451–482.
Gigerenzer, G., & Goldstein, D. G. (1996). Reasoning the fast and frugal way: models of bounded ration-
ality. Psychological Review, 103(4), 650.
Gigerenzer, G., & Goldstein, D. G. (1999). Betting on one good reason: The take the best heuristic. In
G. Gigerenzer, P. M. Todd, & the ABC Research Group (Eds.), Simple heuristics that make us smart (pp.
75–95). New York: Oxford University Press.
Gigerenzer, G., & Selten, R. (Eds.) (2002). Bounded rationality: The adaptive toolbox. Cambridge,
MA: MIT Press.
Grant, A. M., & Schwartz, B. (2011). Too much of a good thing: The challenge and opportunity of the
inverted U. Perspectives on Psychological Science, 6(1), 61–76.
Grill-Spector, K., & Malach, R. (2004). The human visual cortex. Annual Review Neuroscience, 27,
649–677.
Hauser, J. R., & Wernerfelt, B. (1990). An evaluation cost model of consideration sets. Journal of Consumer
Research, 16(4), 393–408.
Hogarth, R. M. (1975). Decision time as a function of task complexity. In D. Wendt, & C. A. Vliet (Eds.),
Utility, probability, and human decision making (pp. 321–338). Dordrecht: Springer.
Iyengar, S. S., Huberman, G., & Jiang, W. (2004). How much choice is too much? Contributions to
401(k) retirement plans. In O. S. Mitchell & S. P. Utkus (Eds.), Pension design and structure: New lessons
from behavioral finance (pp. 83–87). Oxford: Oxford University Press.
Iyengar, S. S., & Kamenica, E. (2010). Choice proliferation, simplicity seeking, and asset allocation. Journal
of Public Economics, 94(7–8), 530–539.
Iyengar, S. S., & Lepper, M. R. (2000). When choice is demotivating: Can one desire too much of a good
thing? Journal of Personality and Social Psychology, 79(6), 995.
Iyengar, S. S., Wells, R. E., & Schwartz, B. (2006). Doing better but feeling worse: Looking for the “best”
job undermines satisfaction. Psychological Science, 17(2), 143–150.
Jacoby, J. (1974). Consumer reaction to information displays: Packaging and advertising. Advertising and
the Public Interest, 11, 101–118.
Jacoby, J. (1977). Information load and decision quality: Some contested issues. Journal of Marketing
Research, 14(4), 569–573.
Jacoby, J., Speller, D. E., & Berning, C. K. (1974). Brand choice behavior as a function of information
load: Replication and extension. Journal of Consumer Research, 1(1), 33–42.
Jacoby, J., Speller, D. E., & Kohn, C. A. (1974). Brand choice behavior as a function of information load.
Journal of Marketing Research, 11(1), 63–69.
Kahn, B. E., & Wansink, B. (2004). The influence of assortment structure on perceived variety and con-
sumption quantities. Journal of Consumer Research, 30(4), 519–533.

634
635

Information and choice overload

Kardes, F. R., Cline, T. W., & Cronley, M. L. (2011). Consumer behavior: Science and practice. Boston: Cengage
Learning, Incorporated.
Langer, E. J., & Rodin, J. (1976). The effects of choice and enhanced personal responsibility for the
aged: A field experiment in an institutional setting. Journal of Personality and Social Psychology, 34(2), 191.
Lee, B. K., & Lee, W. N. (2004). The effect of information overload on consumer choice quality in an
on-line environment. Psychology & Marketing, 21(3), 159–183.
Lenton, A. P., Fasolo, B., & Todd, P. M. (2008). “Shopping” for a mate: Expected versus experienced
preferences in online mate choice. IEEE Transactions on Professional Communication, 51(2), 169–182.
Loewenstein, G. (1999). Is more choice always better? Social Security Brief, 7(1), 7.
Loewenstein, G. (2000). Emotions in economic theory and economic behavior. American Economic Review,
90(2), 426–432.
Malhotra, N. K. (1982). Information load and consumer decision making. Journal of Consumer Research,
8(4), 419–430.
Malhotra, N. K., Jain, A. K., & Lagakos, S. W. (1982). The information overload controversy: An alter-
native viewpoint. Journal of Marketing, 46(2), 27–37.
Malito, A. (2017). Grocery stores carry 40,000 more items than they did in the 1990s. Available at: www.
marketwatch.com/story/grocery-stores-carry-40000-more-items-than-they-did-in-the-1990s-2017–
06–07 (accessed July 31, 2019).
Marois, R., & Ivanoff, J. (2005). Capacity limits of information processing in the brain. Trends in Cognitive
Sciences, 9(6), 296–305.
Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for pro-
cessing information. Psychological Review, 63(2), 81.
Nagar, K., & Gandotra, P. (2016). Exploring choice overload, internet shopping anxiety, variety seeking
and online shopping adoption relationship: Evidence from online fashion stores. Global Business
Review, 17(4), 851–869.
Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall.
Orban, G. A., Van Essen, D., & Vanduffel, W. (2004). Comparative mapping of higher visual areas in
monkeys and humans. Trends in Cognitive Sciences, 8(7), 315–324.
Payne, J. W., Bettman, J. R., Coupey, E., & Johnson, E. J. (1992). A constructive process view of decision
making: Multiple strategies in judgment and choice. Acta Psychologica, 80(1–3), 107–141.
Payne, J. W., Bettman, J. R., & Johnson, E. J. (1993). The adaptive decision maker. Cambridge: Cambridge
University Press.
Reibstein, D. J., Youngblood, S. A., & Fromkin, H. L. (1975). Number of choices and perceived deci-
sion freedom as a determinant of satisfaction and consumer behavior. Journal of Applied Psychology,
60(4), 434.
Reutskaja, E. (2008). Experiments on the role of the number of alternatives in choice. PhD thesis.
Pompeu Fabra University.
Reutskaja, E., & Hogarth, R. M. (2009). Satisfaction in choice as a function of the number of
alternatives: When “goods satiate.” Psychology & Marketing, 26(3), 197–203.
Reutskaja, E., Lindner, A., Nagel, R., Andersen, R. A., & Camerer, C. F. (2018). Choice overload
reduces neural signatures of choice set value in dorsal striatum and anterior cingulate cortex. Nature
Human Behaviour, 2(12), 925.
Reutskaja, E., Nagel, R., Camerer, C. F., & Rangel, A. (2011). Search dynamics in consumer choice
under time pressure: An eye-tracking study. American Economic Review, 101(2), 900–926.
Rolls, B. J., Rowe, E. A., Rolls, E. T., Kingston, B., Megson, A., & Gunary, R. (1981). Variety in a meal
enhances food intake in man. Physiology & Behavior, 26(2), 215–221.
Ryan, R. M., & Deci, E. L. (2000). Self-determination theory and the facilitation of intrinsic motivation,
social development, and well-being. American Psychologist, 55(1), 68.
Salvucci, D. D., & Goldberg, J. H. (2000). Identifying fixations and saccades in eye-tracking protocols.
Proceedings of the 2000 Symposium on Eye tracking Research & Applications, November, 71–78.
Scheibehenne, B., Greifeneder, R., & Todd, P. M. (2010). Can there ever be too many options? A meta-
analytic review of choice overload. Journal of Consumer Research, 37(3), 409–425.
Schroder, H. M., Driver, M. J., & Streufert, S. (1967). Human information processing: Individuals and groups
functioning in complex social situations. New York: Holt, Rinehart and Winston.
Schwartz, B. (2000). Self-determination: The tyranny of freedom. American Psychologist, 55(1), 79.
Schwartz, B. (2004). The paradox of choice: Why more is less. New York: Ecco.

635
636

Elena Reutskaja et al.

Schwartz, B., & Ward, A. (2004). Doing better but feeling worse: The paradox of choice. In P. A. Linley,
& S. Joseph (Eds.), Positive Psychology in Practice (pp. 86–104). New York: John Wiley & Sons.
Shafir, E., Simonson, I., & Tversky, A. (1993). Reason-based choice. Cognition, 49(1–2), 11–36.
Shah, A. M., & Wolford, G. (2007). Buying behavior as a function of parametric variation of number of
choices. Psychological Science, 18(5), 369.
Shim, E. (2017). China launches state-sponsored dating event. Available at: www.upi.com/China-
launches-state-sponsored-dating-event/9821498584045/ (accessed July 31, 2019).
Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics, 69(1),
99–118.
Simon, H. A. (1957). Models of man; social and rational. Chichester: John Wiley & Sons, Ltd.
Simon, H. A. (1974). How big is a chunk? Science, 183(4124), 482–488.
Simon, H. A. (1976). From substantive to procedural rationality. In T. J. Kastelein, S. K. Kulpers, W. A.
Nijenhuis, & G. R. Wagenaar (Eds.), 25 years of economic theory (pp. 65–86.). Leiden: Martinus Nijhoff.
Simon, H. A. (1991). Bounded rationality and organizational learning. Organization Science, 2(1), 125–134.
Simons, D. J., & Chabris, C. F. (1999). Gorillas in our midst: Sustained inattentional blindness for dynamic
events. Perception, 28(9), 1059–1074.
Smith, R. (2010). Dealing with the information overload: What about medical students? Available
at: www.bmj.com/rapid-response/2011/11/03/dealing-information-overload-what-about-medical-
students (accessed July 31, 2019).
Steiner, I. D. (1970). Perceived freedom. Advances in Experimental Social Psychology, 5, 187–248.
Stocks, M. H., & Harrell, A. (1995). The impact of an increase in accounting information level on the
judgment quality of individuals and groups. Accounting, Organizations and Society, 20(7–8), 685–700.
Tuttle, B., & Burton, F. G. (1999). The effects of a modest incentive on information overload in an invest-
ment analysis task. Accounting, Organizations and Society, 24(8), 673–687.
UBS, 1992. Video recording. Available at: www.ubs.com/HerbertSimon
Walton, J. R., & Berkowitz, E. N. (1979). The effects of choice complexity and decision freedom on con-
sumer choice behavior. ACR North American Advances, 6, 206–208.
Wedel, M., & Pieters, R. (Eds.) (2007). Visual marketing: From attention to action. Hove: Psychology Press.
Zuckerman, M., Porac, J., Lathin, D., & Deci, E. L. (1978). On the importance of self-determination for
intrinsically-motivated behavior. Personality and Social Psychology Bulletin, 4(3), 443–446.

636
637

44
HOW MUCH CHOICE IS
“GOOD ENOUGH”?
Moderators of information and choice overload

Raffaella Misuraca, Elena Reutskaja, Barbara Fasolo,

and Sheena Iyengar

Introduction
In today’s world, people face an abundance of information and a great number of choices both
in important domains, such as health care, retirement, and education, and in less important
domains, such as the choice of breakfast cereal or chocolate. Choice overload and information
overload have strong negative effects on many important decision-making aspects such as pro-
cessing and using information, the motivation to act, the quality of choices, and post-choice
feelings, which are discussed in Chapter 43 in this volume in more detail. However, small
choice and information sets are not always optimal either. Several variables––such as informa-
tion usage, decision accuracy, motivation to choose, and satisfaction with choice are “inverted-
U” functions of the amount of information and the number of choice alternatives available.
In other words, choosing from sets of an intermediate size usually brings more net benefits to
the decision-maker than choosing from large or small choice sets (Grant & Schwartz, 2011;
Reutskaja & Hogarth, 2009; Shah & Wolford, 2007). Indeed, in line with the assumptions
of bounded rationality, intermediate sizes are preferable when they do not entail the same
high, cognitively unmanageable load that large sets do, and simultaneously possess the benefits
of variety that small sets lack. However, exactly how much choice is enough, or, as Herbert
A. Simon would say, “good enough”? The size of the intermediate set is not always clear or
universal, and is often influenced by a variety of factors. The aim of this chapter is to set out
the factors which most affect and moderate the experience of too much choice, influen-
cing feelings of how much is “enough.” Broadly, there are two categories of moderators: one
pertaining to the choice environment and one pertaining to the characteristics of the decision-
making actor. This division is in line with Simon’s scissors analogy (Simon, 1990), which
views bounded rationality as the interplay between the two blades: the context or choice
environment, on the one hand, and the capabilities and characteristics of decision-makers, on
the other.

Context and choice environment

Within this first category, pertaining to the choice environment, we review the most critical
moderators: perceptual attributes of the information, the complexity of the set of alternatives,

637
638

Raffaella Misuraca et al.

decision accountability, the physical arrangement of the options, and general contextual
characteristics, such as the time and money involved.

Perceptual characteristics of the information presented

The first important moderator pertains to the perceptual nature of the choice presented, that
is, the perceptual attributes. Miller (1956) suggests that the “channel capacity” for information
processing is different for different stimuli: for tastes, it is four; for tones, it is six; and for visual
stimuli, it reaches 10–15 items.
Visual presentation is one of the most important perceptual characteristics. Regardless of
assortment size, consumers tend to prefer a visual rather than verbal representation of choice
options (the “visual preference heuristic”, Townsend & Kahn, 2013). Despite being preferred
by consumers, visual depictions of large assortments lead to suboptimal decisions compared
to verbal presentations, since visual presentation activates a less systematic approach. Visual
depictions in large choice sets also result in greater perceptions of complexity and in a reduction
of the likelihood to choose. With small assortments, however, visual representations of options
seem to be preferable, as they increase consumers’ perception of variety, improve the likelihood
of making a choice, and speed up the time spent examining options.

Choice set complexity, decision accountability, and the presence of a brand

The notion of choice complexity directly pertains to choice overload. It comes from the
important meta-analysis by Chernev, Böckenholt, and Goodman (2015) of choice overload
studies and concerns all the aspects of a decision task that affect the value of the available choice
options (Payne, Bettman, & Johnson, 1993). Choice set complexity, therefore, is not about
the structural characteristics of the decision problem, such as the number of options, number
of attributes of each option, or format in which the information is presented. Rather, choice
complexity involves the following four factors:

1 the level of attractiveness of the options;

2 the presence or the absence of a dominant (or ideal) option;
3 the alignability of the options’ attributes;
4 the complementarity of the choice options.

In terms of the first factor, choice-set complexity is higher when the assortment includes
higher-quality, more attractive options (for example, an assortment of sandwiches made with
premium, instead of average, ingredients; Chernev & Hamilton, 2009). Overall, when the
variability in the relative attractiveness of the choice alternatives increases, the probability of a
correct choice (if possible), the certainty about the choice, and the satisfaction with the task
increase (Malhotra, 1982). More choice leads to a decline in consumer satisfaction if the number
of attractive options is increased but it leads to an improvement in satisfaction if the number of
unattractive options is increased. This effect occurs because more choice generally highlights
the weaknesses of attractive choices and the strengths of unattractive choices (Chan, 2015).
In addition, when the attribute levels of a good are distributed unequally (i.e., the choice-
set contains alternatives that are not equally attractive), the probability of choosing the correct
option typically increases (Lee & Lee, 2004), whereas, people are less confident when attribute
levels are distributed equally. Items with similar attractiveness may lead people to defer choice
or simply choose a default option (Dhar, 1997).

638
639

How much choice is “good enough”?

Turning to the second factor of complexity, the presence of the ideal point simplifies large sets
and therefore leads to a stronger preference for the chosen alternative. However, the presence of
the ideal alternative in smaller sets leads to weaker preferences for the selected item (Chernev,
2003). Moreover, a brain-imaging experiment, where subjects chose from different-sized sets
of landscape images, demonstrated that large sets were not always “bad” or “overwhelming”
(Reutskaja, Lindner, Nagel, Andersen, & Camerer, 2018). Choosing from the sets containing
an “ideal” item (e.g., those containing an image most preferred by participants) was associated
with greater brain activity in the striatum and the anterior cingulate cortex (the areas involved
in reward and value processing as well as in the integration of costs and benefits) compared to
choosing from those sets with no item regarded as ideal. That is, the benefits of having an ideal
item in the set might compensate for the costs of overwhelming set size in the bounded rational
mind of humans.
The fact that large sets are not always more complex than small sets has been shown in other
research. For example, Fasolo, Hertwig, Huber, and Ludwig (2009) measured the level of com-
plexity due to the attractiveness of options in terms of assortment entropy, or the number of
attribute levels and the distribution of products on the attribute levels within the assortment
(Hoch, Bradlow, & Wansink, 2002; Lurie, 2004) and found that consumers considered it more
difficult to choose from an assortment with higher entropy. For example, in an assortment of
jams, for the same attribute (such as carb content), each jar can have different levels (such as
12 g, 15 g, or 7 g). Furthermore, in the same assortment, the distribution of jams on a given
attribute level can be even or uneven. If the number of attribute levels is large and products are
evenly distributed in attractiveness across the attribute levels, the assortment entropy is high.
Entropy can be higher in a smaller set than in a larger set, which suggests that a small set can
be more complex and difficult to choose from and can lead to poorer decisions than a large
choice set.
Third, choice complexity is also a function of the alignability and complementarity of
the attributes that differentiate the options available to the chooser (Chernev et al., 2015).
Complexity and choice overload increase when the options have attributes that cannot be
aligned (meaning that not all of options have attribute values for all attributes, as some options
have unique attributes). For instance, a choice between a shirt that “keeps in heat” and a
jacket that is “waterproof ” might be said to contain non-aligned features. The shirt and jacket
do different things in different ways and therefore a comparison of value becomes difficult
(e.g., apples and oranges). Similarly, “complementary” (meaning that they have additive utility
and need to be co-present to fully satisfy the consumer’s need) can increase complexity. For
example, gloves and scarves have complementary features, in that they provide warmth to
different parts of the body.
Besides choice complexity, another important factor that affects the extent to which a wide
choice causes overload is decision accountability, which can be defined as the requirement
(often due to the context) for decision-makers to justify their choices. With greater account-
ability, the preference for larger sets increases (Chernev et al., 2015).
Finally, the presence or absence of brand names in the choice sets can strongly influence
the level of satisfaction with the chosen option. One study has shown that choice overload
disappeared when the choice options contained brand names. Subjects showed the same level
of satisfaction when choosing from small and large sets of branded cellphones. However,
when the same cellphones were presented without brand names, a higher level of dissatis-
faction was observed for larger sets compared to smaller sets (Misuraca, Ceresia, Teuscher, &
Faraci, 2019).

639
640

Raffaella Misuraca et al.

Physical arrangement of assortment and option organization

The physical arrangement of information and the presentation format of options affect infor-
mation perception, processing, and decision-making and are factors of choice difficulty, which
is known to moderate the effect of choice overload (Chernev et al., 2015). The location of any
alternatives in a space and the information structure embedded in their display allow people
to retrieve additional information about the options, affecting choosers’ abilities to distinguish
among options, and helping in the evaluation of each option’s attributes (see, for example,
Chandon, Hutchinson, Bradlow, & Young, 2009). There is a vast amount of evidence that
the order in which information is presented results in strong primacy and recency effects
(Reutskaja, Nagel, Camerer, & Rangel, 2011). The order also affects the attention paid to the
products in the store: products on the top and middle shelves attract more attention than those
on lower shelves (Chandon et al., 2009). However, greater attention does not necessarily trans-
late directly into more sales.
The order of attributes also affects the perception of choice overload by changing people’s
preferences about diverse goods (suits, cars, etc.). For example, when attributes are presented
starting with the attribute for which there are the most options (such as 56 car interior colors)
and ending with the one for which there are the least number of options (such as four gear-
shift knob styles), participants are more likely to accept default options and to be less satisfied
with their final products than when participants face the opposite order of attributes (Levav,
Heitmann, Herrmann, & Iyengar, 2010). Overall, the organization and presentation of infor-
mation can be used as a tool to simplify information processing and therefore to let decision-
makers deal with a greater information load without too much cost (see Anderson & Misuraca,
2017). For example, the organization of information into “chunks” or sequences facilitates
information processing (Miller, 1956). In addition, the perceived variety is greater if the large
sets are organized, and the smaller sets are disorganized (Kahn & Wansink, 2004). For highly
varied sets, consumers are more satisfied (in terms of learning their own preferences), perceive
less complexity, and are more willing to make choices when information about the product
category is presented by attribute (e.g., consumers are asked how expensive or comfortable they
want their sofa), compared to presentation by alternative (e.g., consumers are shown many sofas
next to each other in a showroom) (Huffman & Kahn, 1998).
The alignment of the external organization of the information (the way the products are
displayed by retailers) with decision-makers’ internal schemes (that is, how decision-makers cat-
egorize those products in their mind) is extremely important for the perception of variety. In
particular, for familiar product categories, consumers are likely to perceive more variety and be
more satisfied when the external organization of an assortment matches their internal organiza-
tional schemas. However, for unfamiliar product categories, consumers feel more satisfied and
perceive more variety if the assortment is arranged in a way that makes it easier to satisfy spe-
cific shopping goals (such as buying a backpack to carry a laptop) (Morales, Kahn, McAlister,
& Broniarczyk, 2005).
Finally, presenting options either simultaneously (all at once) or sequentially (one at a time)
strongly affects individuals’ decisions and their subsequent satisfaction. Specifically, consumers
are less satisfied with their choice when the options are presented sequentially rather than
simultaneously (Mogilner, Shiv, & Iyengar, 2012). This happens because, in the simultaneous
format, decision-makers tend to stay focused on the given set of options while, in the sequential
format, decision-makers tend to evaluate each option by comparing it with an internal refer-
ence point, such as an imagined better option. This hope of finding the ideal option translates
into a lower level of satisfaction with the chosen option.

640
641

How much choice is “good enough”?

Context specificity
Preferences are constructed by the context (Lichtenstein & Slovic, 2006; Payne et al., 1993;
Tversky & Kahneman, 1974, 1981). The context, or domain, in which the decision is to be
made plays an important role in the experience of choice overload. While people tend to
like choice in a consumer context, they might not like choice in the context of making an
unpleasant and stressful medical decision or choosing from a set of undesirable alternatives. In
unpleasant choice domains, people often feel increased negative affect when they are personally
responsible for the choice and decreased negative emotions when they are allowed to delegate
the choice to someone else (Botti & Iyengar, 2006; Botti, Orfali, & Iyengar, 2009).
Whether the choice is presented online or offline is another important consideration. For
instance, satisfaction with the choice of chocolate was not affected by the size of the assortment
when the choice was presented online because e-commerce users expect larger choice sets
compared to shoppers in physical retail spaces (Moser, Phelan, Resnick, Schoenebeck, &
Reinecke, 2017).
Another important contextual factor is time pressure, which has been shown to affect the
quality of decisions and the strategies people utilize when making decisions in two possible
ways (Maule & Edland, 1997; Payne et al., 1993). First, people respond to time pressure by
attempting to speed up processing and/or by eliminating breaks (Payne et al., 1993; Pieters &
Warlop, 1999; Reutskaja et al., 2011). Second, decision-makers become more selective about
the type of information they choose to process and use. This may be reflected in filtering
(giving greater priority to the important information) or in omission (ignoring part of the
information entirely and looking at lower proportion of the items in the choice set) (Payne
et al., 1993; Reutskaja et al., 2011). People may also react to time pressure by choosing at
random or avoiding making choices. One study found that people defer choices less often
when high-conflict decisions were being made under time pressure, compared to the absence of
time pressure. In low-conflict decisions, however, time pressure has no effect on choice deferral.
Furthermore, under time pressure, people use more non-compensatory strategies, which par-
tially mediate the influence of time pressure on choice deferral (Dhar & Nowlis, 1999).
Among the “negative” consequences of time stress that have been mentioned are forget-
ting important data, the neglect and denial of important data, and inaccurate judgments and
evaluations (Zakay, 1993). However, time stress also has adaptive benefits (Gigerenzer & Garcia-
Retamero, 2017). There is also an inverted-U relationship between information load and deci-
sion quality under conditions of time pressure, and no such relationship when the time pressure
is removed (Hahn, Lawson, & Lee, 1992).
Monetary incentives have also been shown to affect the amount of information used and
the response times. When the information load increases, individuals provided with monetary
incentives use more information and take more time than those who are not offered such
incentives. However, there is a limit to the amount of information that can be processed per
time unit, and incentives do not affect information usage for decisions constrained by time
(Tuttle & Burton, 1999).

Individual characteristics of the decision-maker

Within this second category pertaining to the decision-maker’s characteristics, we review the
most critical moderators: decision goal, knowledge and experience, preference uncertainty
and mindset, affective state, decision style, and demographic variables such as age, gender, and
culture.

641
642

Raffaella Misuraca et al.

Decision goal
The term “decision goal” refers to the extent to which a decision-maker seeks to minimize
the amount of cognitive resources being spent on making a decision (Chernev et al., 2015).
This has been operationalized in the form of decision intent––buying (or the goal of making
a decision among the available options), versus browsing (or the goal of learning more about
the options)––and decision focus (choosing an option from an assortment versus choosing
an assortment). Concerning the decision intent, when consumers approach assortments
with the goal of browsing, cognitive overload is less likely to occur than when consumers
approach the assortments with the goal of buying. In the latter case, consumers need to
make trade-offs among the pros and cons of the options, something that demands more
cognitive resources. Accordingly, consumers whose goal is browsing, rather than buying,
are less likely to experience cognitive overload when facing large assortments (Chernev &
Hamilton, 2009).
The difference between browsing and choosing is also reflected in the brain activity of
decision-makers who are choosing from different size of sets. When they were choosing, the
activity in the striatum and the anterior cingulate cortex reflected the inverted-U-shaped
function of the number of alternatives people chose from. This suggests that neither too much
nor too little choice provides optimal cognitive net benefits. Whereas activity associated with
browsing intent has been observed as an increasing function of the set size in those areas because
the costs of choice were removed when subjects simply browse rather than when subjects were
faced with the effort of choosing (Reutskaja et al., 2018).
With regard to the decision focus (Chernev et al., 2015), when consumers approach the
assortments with the goal of choosing one of those assortments, rather than choosing an
item from a given assortment, cognitive overload is less likely to occur because the task does
not involve any process of evaluating the individual options or any trade-off among those
options. As a consequence, consumers focusing their attention on choosing an assortment
tend to prefer larger assortments, since they gain the benefit of variety without paying the
cognitive costs associated with the difficult trade-offs involved in choosing an item. In con-
trast, consumers focusing their attention on choosing an option from one assortment experi-
ence greater decision difficulty and, as a consequence, tend to prefer smaller assortments
(Chernev, 2006).
In addition, the order in which consumers decide whether to buy and which option to
choose moderates the purchasing likelihood under choice overload (Scheibehenne, Greifeneder,
& Todd, 2010). Large assortments are associated with a greater purchase likelihood when con-
sumers first decide whether to buy from an assortment, rather than choosing an option from
the set.

Knowledge and experience

Knowledge and experience play an important role when someone chooses from sets with mul-
tiple alternatives. For example, greater knowledge and experience are associated with increased
brand processing (Bettman & Park, 1980), while moderate prior knowledge of a product is
associated with processing more available information than either low or high levels of previous
knowledge or experience of the subject. Low-knowledge individuals tend to give up when
facing complex data, because it is hard for them to make sense of the data. People with high
prior knowledge can process the information but have no motivation to do so, preferring to use

642
643

How much choice is “good enough”?

the information they already possess. Moderate prior knowledge on the subject gives people
both the ability and the motivation to engage in further information processing.
In addition, decision-makers with a greater need for cognition (Cacioppo & Petty, 1982)
are less affected by choice overload and defer choice less than decision-makers with a lower
need for cognition (Pilli & Mazzon, 2016). Consumers who are more cognitively complex
(those who refer to a larger number of dimensions to interpret and evaluate the environment)
also use more information. However, such consumers are significantly less likely to experience
information overload with an optimizing goal than those who are more cognitively simple
(Malhotra, 1982).

Preference uncertainty and assessment orientation

Bounded rationality can mean that individuals do not know what they prefer, before they
choose, something Chernev et al. (2015) called “preference uncertainty.” In these cases, articu-
lating preferences and making trade-offs between alternatives before choosing can lead to
stronger preferences when choosing from large sets but weaker preferences when choosing from
small sets. Similarly, articulating one’s ideal product can simplify the choice from large sets if the
set contains this ideal product.
Assessment orientation or the motivation to evaluate and compare all the available options,
in order to choose the one with the best attributes, is another factor that influences choice
from large sets (Mathmann, Chylinski, de Ruyter, & Higgins, 2017). Customers with a high
assessment orientation perceive greater value in products chosen from large assortments,
compared to those who feel comfortable without many comparisons to make among the
options.

Positive affect
Positive affect has a strong influence on consumer satisfaction when people are choosing
from different sizes of sets. For example, one study discovered that individuals experiencing
positive affect did not also experience dissatisfaction when choosing from larger choice sets
(as though momentarily inoculating them), whereas individuals in neutral affect were more
satisfied when choosing from a smaller choice set (Spassova & Isen, 2013). Positive affect,
then, is likely to shift attention away from the difficulty of the task toward the quality of
the assortment. The role of positive affect in choice satisfaction is in line with research on
the affect heuristic, which is a mental shortcut that enables quick and efficient decisions,
based on the immediate emotional response to a stimulus (Slovic, Finucane, Peters, &
MacGregor, 2007).

Decision-making tendencies
Drawing on Simon’s terminology, Schwartz et al. (2002) have argued that satisfaction with an
extensive choice depends on whether one is a “maximizer,” who actively seeks the best possible
result, or a “satisficer,” who is content with the first result that is “good enough.” The authors
observed that maximizers reported less satisfaction, happiness, and optimism with life in gen-
eral, and, when facing choices, they engaged in more social comparisons, experienced more
regret, and were less satisfied with their choices. Even while doing better (e.g., obtaining a
higher salary for a job), maximizers may feel worse because of them “not always wanting what

643
644

Raffaella Misuraca et al.

they get” (Iyengar, Wells, & Schwartz, 2006). However, it is important to note, that the litera-
ture on maximizing is controversial on account of the proliferation and use of several different
maximization scales (e.g., Diab, Gillespie, & Highhouse, 2008; Misuraca Faraci, Gangemi,
Carmeci, & Miceli, 2015; Turner, Rim, Betz, & Nygren, 2012), each of which is based on
different definitions of the core maximizing construct (Misuraca & Fasolo, 2018).

Choosing for others versus oneself

The negative effects of choice overload are not replicated when individuals make choices for
others rather than for themselves (Polman, 2012). Individuals making choices for others (about
wines, ice-cream flavors, school courses, etc.) reported greater satisfaction when choosing
from larger rather than smaller assortments. Conversely, when choosing for themselves, people
reported higher satisfaction levels after choosing from smaller rather than larger assortments.
This may occur because, when choosing for others, people are typically oriented toward posi-
tive outcomes and positive information whereas, when they are choosing for themselves, indi-
viduals’ attention is directed to negative information and they are oriented away from negative
outcomes (see regulatory focus theory for details, Higgins, 1997).

Gender
There are gender differences in reaction to choice overload, in part, because men and women
may often employ different information-processing strategies. For example, one study has
demonstrated that, while ad information is encoded and is equally available to both women
and men, females are more likely to pay attention to the details, whereas males are less likely
to access or use this information (Meyers-Levy & Maheswaran, 1991). Advertisements with
many images are often more effective at targeting females, while male customers prefer simple
images and information that will lead to quick decisions. However, gender differences in
the desire for variety can depend on the type of choice. When choosing from different sizes
of sets of gift boxes, women were shown to be more satisfied with their choices over the
entire range of alternatives than men (Reutskaja, 2008). However, when choosing a poten-
tial mate from a set of online date options, women perceived 20 profiles as being close to
the ideal set size, whereas men perceived this as being too limited (Lenton, Fasolo, & Todd,
2008). In addition, women were generally more selective than men when they searched for
a potential mate during speed-dating events, which varied in size of potential mates (Fisman,
Iyengar, Kamenica, & Simonson, 2006). The selectivity of males was similar between groups
of different sizes, while women became much more selective when the speed-dating group
size was increased to more than 15.

Age
The choice overload experience depends greatly on the age of the decision-maker. For
example, when choosing from an extensive array of options, adolescents and adults suffer
similar negative consequences (i.e., greater difficulty and dissatisfaction), while children
and seniors suffer fewer negative consequences (i.e., less difficulty and dissatisfaction than
adolescents and adults) (Misuraca, Teuscher, & Faraci, 2016). In domains where risk is not
involved, adults and adolescents seem to adopt very similar decision-making processes (Furby
& Beyth-Marom, 1992): a maximizing approach. This would explain their greater perceived

644
645

How much choice is “good enough”?

difficulty and post-choice dissatisfaction when facing a high number of options (see Iyengar
et al., 2006). Children, on the other hand, tend to approach decisions in a more intuitive
manner and quickly develop strong preferences (Schlottmann & Wilkening, 2011). This
mitigates the negative consequences of choice overload for this age group. Seniors tend to be
overconfident in their judgments (Stankov & Crawford, 1996), demonstrating a pronounced
focus on positive information (Mather & Carstensen, 2005), and they adopt a satisficing
approach when making decisions (Tanius, Wood, Hanoch, & Rice, 2009). These tenden-
cies would explain why the negative consequences of too many choices were milder among
seniors.

Cultural background
People of different cultures have different preferences for variety (for further discussion, see
Iyengar, 2010). For example, Anglo-Americans were shown to be more motivated by choice,
especially by personal choice, and rated having choice as more important than Asian Americans
did (Iyengar & Lepper, 1999). People from Eastern Europe were more satisfied with larger
choice sets than their Western European counterparts, which could be explained by the fact
that choice was limited in Eastern European countries for a long time (Reutskaja, 2008).
Choice provision might have different effect on people from different cultures, because freedom
and choice might not have the same meaning for Westerners and non-Westerners (Markus
& Schwartz, 2010). In addition, Western and non-Western cultures seem to have different
patterns of perception: while Asians tend to focus more on contextual information, Americans
tend to focus on salient foreground objects (Miyamoto, Nisbett, & Masuda, 2006). This diffe-
rence is due to the distinctive characteristics of each culture’s perceptual environment, which
afford distinctive ways to perceive information (see also Viale, 2012). Though there are cultural
differences in choice perception, most of the research on choice overload to date has focused
on university-educated samples from Western societies and should be taken with a grain of salt
as freedom of choice may not be a universal aspiration.

Conclusion
In an interview in Pittsburgh, Herbert Simon was asked whether simple decision-making could
be achieved by presenting a smaller number of alternatives. His response was: “Partly. I think
the difficulty of decision-making centers very much around the degree of uncertainty and the
gaps in our knowledge” (UBS, 1992). We agree with his statement: attempting to gird ourselves
against the effects of choice overload, either by pursuing or presenting an “ideal” number of
options, is an admirable goal but an incredibly challenging one. Previous research has shown
that both too much and too little information and choice are bad. For more than half a cen-
tury, researchers have tried to answer the question of how much information and choice are
enough and what is the “ideal” number of alternatives to present to consumers and the public.
As this chapter has demonstrated, the ideal number of choices depends greatly on many con-
textual and demographic factors, such as the availability of an ideal alternative in the choice set,
the existence of time constraints, knowledge and expertise, and the gender, age, and culture
of the decision-makers. Nevertheless, finding the ideal choice set may ultimately be worth the
trouble when decisions are recurrent or strategic, such as pension schemes, health plans, or
career options. This review has covered a substantial amount of research that can be utilized to
mitigate these distinct challenges.

645
646

Raffaella Misuraca et al.

We believe that the most promising directions for offering long-lasting solutions against
choice overload lie in understanding the roles of preference certainty, focused attention, and
generated self-knowledge. As Simon (1971, p. 40) put it:

In an information-rich world, the wealth of information means a dearth of some-

thing else: a scarcity of whatever it is that information consumes. What information
consumes is rather obvious: it consumes the attention of its recipients. Hence a wealth
of information creates a poverty of attention and a need to allocate that attention
efficiently among the overabundance of information sources that might consume it.

Smart and ethical choice architecture should be designed based on goals, needs, and personal
preferences by using apps, reminders, checklists, websites, buddy systems, and alerts (Johnson
et al., 2012; Thaler & Sunstein, 2008) that direct and hold decision-makers’ attention and respect
decision-makers’ freedom. How people’s attention is managed in the ever more information-
rich world will ultimately dictate their future.

References
Anderson, B. F., & Misuraca, R. (2017). Perceptual commensuration in decision tables. Quarterly Journal
of Experimental Psychology, 70(3), 544-553.
Bettman, J. R., & Park, C. W. (1980). Effects of prior knowledge and experience and phase of the
choice process on consumer decision processes: A protocol analysis. Journal of Consumer Research, 7(3),
234–248.
Botti, S., & Iyengar, S. S. (2006). The dark side of choice: When choice impairs social welfare. Journal of
Public Policy & Marketing, 25(1), 24–38.
Botti, S., Orfali, K., & Iyengar, S. S. (2009). Tragic choices: Autonomy and emotional responses to med-
ical decisions. Journal of Consumer Research, 36, 337–352.
Cacioppo, J. T., & Petty, R. E. (1982). The need for cognition. Journal of Personality and Social Psychology,
42(1), 116–131.
Chan, E. Y. (2015). Attractiveness of options moderates the effect of choice overload. International Journal
of Research in Marketing, 32(4), 425–427.
Chandon, P., Hutchinson, J. W., Bradlow, E. T., & Young, S. H. (2009). Does in-store marketing work?
Effects of the number and position of shelf facings on brand attention and evaluation at the point of
purchase. Journal of Marketing, 73(6), 1–17.
Chernev, A. (2003). When more is less and less is more: The role of ideal point availability and assortment
in consumer choice. Journal of Consumer Research, 30(2), 170–183.
Chernev, A. (2006). Decision focus and consumer choice among assortments. Journal of Consumer Research,
33(1), 50–59.
Chernev, A., Böckenholt, U., & Goodman, J. (2015). Choice overload: A conceptual review and meta-
analysis. Journal of Consumer Psychology, 25(2), 333–358.
Chernev, A., & Hamilton, R. (2009). Assortment size and option attractiveness in consumer choice
among retailers. Journal of Marketing Research, 46(3), 410–420.
Dhar, R. (1997). Consumer preference for a no-choice option. Journal of Consumer Research, 24(2),
215–231.
Dhar, R., & Nowlis, S. M. (1999). The effect of time pressure on consumer choice deferral. Journal of
Consumer Research, 25(4), 369–384.
Diab, D. L., Gillespie, M. A., & Highhouse, S. E. (2008). Are maximizers really unhappy? The measure-
ment of maximizing tendency. Judgment and Decision Making, 3(5), 364–370.
Fasolo, B., Hertwig, R., Huber, M., & Ludwig, M. (2009). Size, entropy, and density: What is the diffe-
rence that makes the difference between small and large real world assortments? Psychology & Marketing,
26(3), 254–279.
Fisman, R., Iyengar, S. S., Kamenica, E., & Simonson, I. (2006). Gender differences in mate selec-
tion: Evidence from a speed dating experiment. The Quarterly Journal of Economics, 121(2), 673–697.

646
647

How much choice is “good enough”?

Furby, L., & Beyth-Marom, R. (1992). Risk taking in adolescence: A decision-making perspective.
Developmental Review, 12(1), 1–44.
Gigerenzer, G., & García-Retamero, R. (2017). Cassandra's regret: The psychology of not wanting to
know. Psychological Review, 124(2), 179–196.
Grant, A. M., & Schwartz, B. (2011). Too much of a good thing: The challenge and opportunity of the
inverted U. Perspectives on Psychological Science, 6(1), 61–76.
Hahn, M., Lawson, R., & Lee, Y. G. (1992). The effects of time pressure and information load on decision
quality. Psychology & Marketing, 9(5), 365–378.
Higgins, E. T. (1997). Beyond pleasure and pain. American Psychologist, 52(12), 1280–1300.
Hoch, S. J., Bradlow, E., & Wansink, B. (2002). Rejoinder to “The variety of an assortment: An extension
to the attribute-based approach”. Marketing Science, 21(3), 342–346.
Huffman, C., & Kahn, B. E. (1998). Variety for sale: Mass customization or mass confusion? Journal of
Retailing, 74(4), 491–513.
Iyengar, S. S. (2010). The art of choosing. London: Little Brown.
Iyengar, S. S., & Lepper, M. R. (1999). Rethinking the value of choice: A cultural perspective on intrinsic
motivation. Journal of Personality and Social Psychology, 76(3), 349–366.
Iyengar, S. S., Wells, R. E., & Schwartz, B. (2006). Doing better but feeling worse: Looking for the “best”
job undermines satisfaction. Psychological Science, 17(2), 143–150.
Johnson, E. J., Shu, S. B., Dellaert, B. G. C., Fox, C., Goldstein, D., Häubl, G., ... Weber, E.U., 2012.
Beyond nudges: Tools of a choice architecture. Marketing Letters, 23(2), 487–504.
Kahn, B. E., & Wansink, B. (2004). The influence of assortment structure on perceived variety and con-
sumption quantities. Journal of Consumer Research, 30(4), 519–533.
Lee, B. K., & Lee, W. N. (2004). The effect of information overload on consumer choice quality in an
on-line environment. Psychology & Marketing, 21(3), 159–183.
Lenton, A. P., Fasolo, B., & Todd, P. M. (2008). “Shopping” for a mate: Expected versus experienced
preferences in online mate choice. IEEE Transactions on Professional Communication, 51(2), 169–182.
Levav, J., Heitmann, M., Herrmann, A., & Iyengar, S. S. (2010). Order in product customization
decisions: Evidence from field experiments. Journal of Political Economy, 118(2), 274–299.
Lichtenstein, S., & Slovic, P. (Eds.) (2006). The construction of preference. New York: Cambridge
University Press.
Lurie, N. H. (2004). Decision making in information-rich environments: The role of information struc-
ture. Journal of Consumer Research, 30(4), 473–486.
Malhotra, N. K. (1982). Information load and consumer decision-making. Journal of Consumer Research,
8(4), 419–430.
Markus, H. R., & Schwartz, B. (2010). Does choice mean freedom and well-being? Journal of Consumer
Research, 37(2), 344–355.
Mather, M., & Carstensen, L. L. (2005). Aging and motivated cognition: The positivity effect in attention
and memory. Trends in Cognitive Sciences, 9(10), 496–502.
Mathmann, F., Chylinski, M., de Ruyter, K., & Higgins, E. T. (2017). When plentiful platforms pay
off: Assessment orientation moderates the effect of assortment size on choice engagement and product
valuation. Journal of Retailing, 93(2), 212–227.
Maule, A. J., & Edland, A. C. (1997). The effects of time pressure on human judgment and decision
making. In R. Ranyard, W. R. Crozier, & O. Svenson (Eds.), Decision making: Cognitive models and
explanations (pp. 189–204). London: Routledge.
Meyers-Levy, J., & Maheswaran, D. (1991). Exploring differences in males' and females' processing strat-
egies. Journal of Consumer Research, 18(1), 63–70.
Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for pro-
cessing information. Psychological Review, 63(2), 81–97.
Misuraca, R., Ceresia, F., Teuscher, U., & Faraci, P. (2019). The role of the brand on choice overload.
Mind & Society, 18(1), 57–76
Misuraca, R., Faraci, P., Gangemi, A., Carmeci, F.A., & Miceli, S. (2015). The Decision Making Tendency
Inventory: A new measure to assess maximizing, satisficing, and minimizing. Personality and Individual
Differences, 85, 111–116.
Misuraca, R., & Fasolo, B. (2018). Maximizing versus satisficing in the digital age: disjoint scales and the
case for “construct consensus”. Personality and Individual Differences, 121, 152–160.
Misuraca, R., Teuscher, U., & Faraci, P. (2016). Is more choice always worse? Age differences in the
overchoice effect. Journal of Cognitive Psychology, 28(2), 242–255.

647
648

Raffaella Misuraca et al.

Miyamoto, Y., Nisbett, R. E., & Masuda, T. (2006). Culture and the physical environment. Holistic versus
analytic perceptual affordances. Psychological Science, 17(2), 113–119.
Mogilner, C., Shiv, B., & Iyengar, S. S. (2012). Eternal quest for the best: Sequential (vs. simultan-
eous) option presentation undermines choice commitment. Journal of Consumer Research, 39(6),
1300–1312.
Morales, A., Kahn, B. E., McAlister, L., & Broniarczyk, S. M. (2005). Perceptions of assortment var-
iety: The effects of congruency between consumers’ internal and retailers’ external organization.
Journal of Retailing, 81(2), 159–169.
Moser, C., Phelan, C., Resnick, P., Schoenebeck, S. Y., & Reinecke, K. (2017). No such thing as too
much chocolate: Evidence against choice overload in e-commerce. In Proceedings of the 2017 CHI
Conference on Human Factors in Computing Systems (pp. 4358–4369). ACM.
Payne, J. W., Bettman, J. R., & Johnson, E. J. (Eds.) (1993). The adaptive decision-maker. New York: Cambridge
University Press.
Pieters, R., & Warlop, L. (1999). Visual attention during brand choice: The impact of time pressure and
task motivation. International Journal of Research in Marketing, 16(1), 1–16.
Pilli, L. E., & Mazzon, J. A. (2016). Information overload, choice deferral, and moderating role of need
for cognition. Empirica: Revista de Administração, 51(1), 36–55.
Polman, E. (2012). Self–other decision making and loss aversion. Organizational Behavior and Human
Decision Processes, 119(2), 141–150.
Reutskaja, E. (2008). Experiments on the role of the number of alternatives in choice. thesis. Pompeu
Fabra University, Spain.
Reutskaja, E., & Hogarth, R. M. (2009). Satisfaction in choice as a function of the number of
alternatives: When “goods satiate”. Psychology & Marketing, 26(3), 197–203.
Reutskaja, E., Lindner, A., Nagel, R., Andersen, R. A., & Camerer, C. F. (2018). Choice overload
reduces neural signatures of choice set value in dorsal striatum and anterior cingulate cortex. Nature
Human Behaviour, 2, 925–935.
Reutskaja, E., Nagel, R., Camerer, C. F., & Rangel, A. (2011). Search dynamics in consumer choice
under time pressure: An eye-tracking study. American Economic Review, 101(2), 900–926.
Scheibehenne, B., Greifeneder, R., & Todd, P. M. (2010). Can there ever be too many options? A meta-
analytic review of choice overload. Journal of Consumer Research, 37(3), 409–425.
Schlottmann, A., & Wilkening, F. (2011). Judgment and decision making in young children. In
M. Dhami, A. Schlottmann, & M. Waldmann (Eds.), Judgment and decision making as a skill (pp. 55–83).
New York: Cambridge University Press.
Schwartz, B., Ward, A., Monterosso, J., Lyubomirsky, S., White, K., & Lehman, D. R. (2002). Maximizing
versus satisficing: Happiness is a matter of choice. Journal of Personality and Social Psychology, 83(5),
1178–1197.
Shah, A. M., & Wolford, G. (2007). Buying as a function of parametric variation of number of choices.
Psychological Science, 18(5), 369–370.
Simon, H. A. (1971). Designing organizations for an information-rich world. In M. Greenberger
(Ed.), Computers, communication, and the public interest (pp. 37–72). Baltimore: MD: Johns Hopkins
University Press.
Simon, H. A. (1990). Invariants of human behavior. Annual Review of Psychology, 41, 1–19.
Slovic, P., Finucane, M. L., Peters, E., & MacGregor, D. G. (2007). The affect heuristic. European Journal
of Operational Research, 177, 1333–1352.
Spassova, G., & Isen, A. M. (2013). Positive affect moderates the impact of assortment size on choice sat-
isfaction. Journal of Retailing, 89(4), 397–408.
Stankov, L., & Crawford, J. D. (1996). Confidence judgments in studies of individual differences. Personality
and Individual Differences, 21(6), 971–986.
Tanius, B. E., Wood, S., Hanoch, Y., & Rice, T. (2009). Aging and choice: Applications to Medicare
Part 500. Judgment and Decision Making, 4(1), 92–101.
Thaler, R. H., & Sunstein, C. R. (2008). Nudge: Improving decisions about health, wealth, and happiness.
New York: Penguin Books.
Townsend, C., & Kahn, B. E. (2013). The “visual preference heuristic”: The influence of visual versus
verbal depiction on assortment processing, perceived variety, and choice overload. Journal of Consumer
Research, 40(5), 993–1015.
Turner, B. M., Rim, H. B., Betz, N. E., & Nygren, T. E. (2012). The maximization inventory. Judgment
and Decision Making, 7(1), 48–60.

648
649

How much choice is “good enough”?

Tuttle, B, & Burton, F. G. (1999). The effects of a modest incentive on information overload in an invest-
ment analysis task. Accounting, Organizations and Society, 24(8), 673–687.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science,
185(4157), 1124–1131.
Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science,
211(4481), 453–458.
UBS (1992). Video recording. Available at: www.ubs.com/HerbertSimon
Viale, R. (2012). Cognitive diversity in Western and Eastern thinking. In R. Viale (Ed.), Methodological
Cognitivism, vol. I: Mind, Rationality and Society (pp. 295–308). Berlin: Springer.
Zakay, D. (1993). The impact of time perception processes on decision making under time stress. In
O. Svenson & A. J. Maule (Eds.), Time pressure and stress in human judgment and decision making (pp.
59–72). New York: Plenum Press.

649
650

INDEX

Note: Page numbers in italic denote figures; those in bold denote tables.

abductive inference 208, 209, 212, 406; defeasible analogical inference 402
232, 233 analogy-based equilibrium 435
Abreu, Dilip 431–432 Analyses of Competing Hypotheses 553
accessibility of knowledge 170–178 anchoring bias 105, 497
accuracy–effort trade-off 82 Anderson, J. R. 270, 272
ACT-R theory of cognition 282 Anderson, Michael 213, 413
act-state independence 473 Ando, Albert 459–460, 463
adaptive toolbox 61, 62–63, 282, 317–318, 357 animal spirits 57, 114, 438
adaptively responsive routines 487 antecedent preferences, nudging and 563,
additivity 245, 246 564–565, 566–567, 568
adjacency heuristic 393, 393 anterior cingulate cortex (ACC) 631
adversarial communication 168 anticipatory reason 261
adverse selection problem 471–472 anti-naturalism 74, 80
affect heuristic 643 anti-psychologism 74–76, 80
affirming the consequent 232–237, 233, 235, Antonelli, G. A. 177
236, 351 Apple 525
affordances 282, 286, 380–382 Arieli, A. 249
Aimone, J. A. 249 Arieti, S. 349
Akerlof, George 57, 471–472 Aristotle 2, 398, 399–400
algorithmic level of analysis 232, 269–270, Arkes, H. R. 355–356, 357, 548
275–276 Arrow, K. J. 59
algorithmic modeling 272–276, 272 artificial intelligence 156–157, 338–339, 346, 518;
algorithmic models of heuristics 62, 64 branching factor and 339, 346; causal Bayes
Allais, Maurice 334, 511, 512 nets (CBN) 231; chess programs 156,
Allais paradox 325, 326 185–186; frame problem 209; machine
Alós-Ferrer, Carlos 367, 384–385 learning 64, 186, 273, 274–275, 338; musical
AlphaZero chess program 186 composition program 156; reinforcement
Alter, A. L. 359 learning 338; risks of digitalization for children
alternative-based choice 242, 243–244, 243, 304–306; sequential decision problems
248–250 339, 344–346, 345; Tetris program 339,
ambiguity 56 344–346, 345
ambiguity aversion 352–353 Artinger, F. M. 62–63
American Association for Artificial Ashton, R. H. 554
Intelligence 174 as-if expected utility theories 57, 93, 94, 448, 494
amygdala 368, 402, 404 as-if models of decision making 544–545

650
651

Index

aspiration levels 57–58, 62, 104, 170 behavioral social choice framework 543–544
assessment orientation 643 behavioural public policy 570–576, 572, 575;
attention: brain limitations on 404–405; see also nudging
inattentional blindness 105, 108, 530, 629; belief bias 191, 550, 551, 552, 554
information overload and 629; selective 171, belief fixation 208, 209, 210, 212
211, 528, 530; see also organizations and beliefs: ecological rationality of 65–66, 66; false
attention 82, 319–320; mathematical representation of
attention-based view of the firm 524, 525 172–174, 172; partial 476; religious 441; social
attribute commonality 541, 542 benefits of heterogeneity in 578–579, 583–584
attribute-based choice 242, 243, 244–250; Bendor, J. 442, 443
difference-ratio-interest-finance-time (DRIFT) Ben-Porath, Elhanan 432–433
model 246; elimination-by-aspects heuristic Bentham, Jeremy 406, 501, 572
90, 94, 245, 318; fuzzy-trace theory 247–248; Benthamite utility maximisation 572–574
information acquisition 248–250, 248, 249; Benussi, Vittorio 133
intertemporal choice heuristic (ITCH) model Berg, N. 58
246; lexicographic heuristics 63, 165, 244–245, Bettelheim, Bruno 368
339, 340, 341–343, 345; proportional Bewley, T. F. 450
difference model 245, 325; similarity model bias bias 66
247; tradeoff model 246 bias–variance dilemma 65
Augier, Mie 92 Binmore, K. 271
autism 352, 353, 363, 366, 368 Birkinshaw, Julian 525
availability heuristic 61, 257 Blaug, Mark 463
aversive change 280–306; aversive digital blood donation 527
environments 289–306; environments and board games 338; see also chess
heuristics 286–289; evolutionary perspective Bond, Michael 99
302–306; fast-and-frugal heuristics framework Bonhoeffer, Dietrich 306
282–286; heuristics to manage aversive change Bonnefon, J.-F. 232
295–302; risks of digitalization for children boosting 98, 99, 610–622, 622; educational
304–306; scoring systems 295, 299–301; foundation for underprivileged children in
Sioux example 280–281, 283, 286, 288 Hungary 615–617, 616, 617; eliciting value
awareness logics 176 functions in decision making 617–621, 619,
620, 621; rural regeneration project in new
Bacdayan, P. 499, 516 Italian World Heritage Site 612–614, 614
backgammon 338 borderline personality disorder (BPD) 362, 363
backwards induction 177 boundary rules 288
Bacon, Francis 398, 400 bounded knowledge 170–178
Bandara, R. 441 bounded-resource models in biology 391–396;
bargaining games 94–95 brainwiring optimization 391–394, 393, 394;
Bargh, J. A. 109 genome connection-minimization model
Barnard, Chester I. 522 394–395, 395, 396
basal ganglia 320–321, 386–387 Bouquet, Cyril 525
base rate fallacy 77 Boyd, R. 196, 600
Baumol, William J. 438, 442, 459, 610 brain 398–406; basal ganglia 320–321, 386–387;
Bayes, Thomas 159 choice overload and 631; coherence judgments
Bayes rule 159, 228, 231–232, 362 and 402, 406; default network 413, 414;
Bayesian decision theory 55, 56–57, 271, 277; ecologically rational behavior and 320–321,
see also Good’s principle 386–387; emotions and 382–384, 385–386;
Bayesian reasoning 98, 187; icon arrays and Venn integration of cognition and emotion
diagrams 160–162, 161; natural frequency and 402–404, 403; intuition and 143; limitations
probability 158–162, 160, 161, 300; optimality on attention and consciousness 404–405;
268–269, 270, 272, 272, 273–275, 276–277 mirror neurons 368, 381, 383–384, 385–386;
Bayesian updating of probabilities 55, 56, 60 motor neurons 381, 383; neural reuse 378;
Bayesian views of perception 107 neural wiring optimization 391–394, 393, 394;
Beatles 535, 545 Neuron Doctrine 414; neurotransmitters 401;
Becker, Gary 448, 451–452 perception and 380–381; psychopathological
behavioral ecology 313, 315–317 irrationality and 367–369; Semantic Pointer
behavioral game theory 410 Architecture 402–403, 405; size and speed

651
652

Index

400–402; social 368; switch from routinized cognition: distributed 121, 124–126; expertise
to exploratory behavior 516–517; time and 186, 416–417; integration with emotion
discounting 404; versus computers 157; see also in brain 402–404, 403; limitations and learning
neuroscience; System 1 and System 2 thinking 464–465; metacognition 191–192; rational
brain imaging 143, 368, 516, 629, 631, 639 analysis of 270, 272, 282; see also embodied
branch method of decision making 598–599, cognition
600, 602 cognitive biases 186, 188, 190, 191, 402,
branching factor 339, 346 548–549; debiasing interventions 552–553,
brand names 639 554–555, 555–556; in intelligence analysis
Brandstätter, Eduard 92, 93, 96, 333, 334 550–556, 554–555; see also heuristics-and-
Breiman, Leo 90, 271, 272, 272, 275 biases program
Brooks, David 97 cognitive categories 496–499
Brooks, Rodney 377 cognitive coalitions 530
Brothers, Leslie 368 cognitive continuum theory 549
Brunswik, E. 282 cognitive development theory 77
budges 575–576, 575 cognitive libertarian paternalism 48, 49–50
budgets, public 602–603 cognitive memory 499, 500
Burke, R. J. 150 cognitive miser hypothesis 190, 196–197, 415–416
Busetti, F. 464 cognitive neuroscience 409–418
cognitive success 354–355
Cacioppo, J. T. 384 cognitivism 135, 139, 377–378, 385, 454–455
Caenorhabditis elegans worms 315, 392, 393 Cohen, A. L. 231
Carnap, Rudolf 3, 4, 76, 474 Cohen, Jonathan 186
Carnegie School 523 Cohen, M. D. 499, 516
Carruthers, P. 211–212 coherence judgments 402, 406
causal attribution 229, 229, 237–238 coin flipping 65–66, 66
causal Bayes nets (CBN) 231 collective mental models 499
causal diagnostic reasoning 228–238; causal common difference effect 246, 247
attribution 229, 229, 237–238; causal models common ratio effect 325, 326, 334
229, 230–231, 230; defeasible inference in communication: adversarial 168; cooperative 168;
232–237, 233, 235, 236; inference to the best decentralized 529; in organizations 525, 529,
explanation 229, 229, 237; sub-optimal use of 530; social communication rules 540
Bayes’s rule 231 comparison problem 339–344, 344
cause-effect isomorphism fallacy 415 competence gaps 498–499, 517
certainty effect 325, 326, 334 competitive testing 62, 64
Chabris, C. F. 108 computational efficiency 455; see also cognitive
Challenge 449, 451–452, 453 miser hypothesis
change blindness 105 computational intractability 56, 57, 77–78,
Chapman, J. P. 351 139–140, 391
Chapman, L. J. 351 computational level of analysis 232, 269–270, 275,
Chartrand, T. L. 109 276, 277
Chase, W. G. 417 computational modeling in systems biology
Chater, N. 273 120–129; bounded rationality of 126–128;
Chen, Lijie 433 building-out strategies 123–124, 126–127;
Chernev, A. 638, 643 distributed model-based reasoning 121,
Cherniak, Christopher 207, 356 124–126; explorations 123; mental models and
chess 56–57, 156, 185–186, 338, 416–417, 123, 124, 125–126, 127; mesoscopic modeling
512–513, 517, 597 122–124, 126–128; modeling tasks 121–122;
Choi, Syngjoo 424 modifications 123–124, 126–127
choice architects 563, 567, 568, 583–584, 606 computers: Simon’s predictions on future of 156;
choice architecture 553, 563, 565, 567, 571, 646 versus brain 157; see also artificial intelligence
choice complexity 638–639 concept attainment 454
Church, A. 391, 392 conceptual blending 411–412
Clark, A. 213–214 Condorcet paradox 543
classifier systems 500 confirmation bias 258–259, 351, 550, 551,
coefficient of variation 331 553, 554
coercive paternalism 564, 572, 585 conjunction errors 77

652
653

Index

conjunctive rule 245 sequential decision problems 339, 344–346,

consider-the-opposite strategies 554 345; strategy selection 318, 320, 386–387;
Consoli, Davide 487 time pressures 443, 641; under uncertainty 56,
constructivist rationality 56, 441 57–58, 66–67; see also attribute-based choice;
contagious laughter 386 groups and teams; heuristics; information and
control systems 525 choice overload; satisficing
Cooper, W. S. 203 deductive reasoning 186, 187, 277; defeasible 233;
cooperative communication 168 deliberation and 223; intuition and 222–224;
cooperative naturalism 81–83 kinematic inferences 226; knowledge and
coordination rules 288 221–222; mental models and 217–226;
corporate governance 526 minimalism 221; spatial and temporal
correspondence, principle of 357 inferences 224–225
Corsi block-tapping test 211 Deep Blue chess program 185–186
cost-plus pricing 438 default-interventionist theories 188–191, 189, 192
critical naturalism 84–85 defeasible affirming the consequent 232–237, 233,
crowding-out problem 527 235, 236
Csányi Educational Foundation, Hungary defeasible logic: in causal diagnostic reasoning
615–617, 616, 617 232–237, 233, 235, 236; ecological rationality
Cummins, Denise 167, 232 of 167–168; mathematics education and
cumulative dominance 341–342, 343–344, 344, 167–168
345–346 defeasible Modus Ponens 232–237, 233, 235, 236
cumulative prospect theory 92, 96, 324–327, 327, defensive decision making 293–294
328–335, 330, 332, 333 Dekel, E. 176
Cyert, Richard M. 487, 511–512, 523–524, delay effect 246
528, 530 deliberation 223
Czerlinski, J. 288 deliberative democracy 264
DellaVigna, S. 460
dash cameras 294 Dennett, Daniel 1
data modeling 271–272, 272, 275 denying the antecedent fallacy 74–75
Davidson, P. 462 Dessein, Wouter 529
Davis, O. A. 602 Dewey, John 486
Dawes’ rule 243 Diagnostic and Statistical Manual of Mental Disorders
Dawkins, Richard 197 (DSM) 361
de Clippel, Geoffroy 426 diagnostic reasoning 228–238; causal attribution
de Finetti, B. 476 229, 229, 237–238; causal models 229,
De Martino, B. 353, 366 230–231, 230; defeasible inference in
Dearborn, DeWitt C. 523 232–237, 233, 235, 236; inference to the best
debiasing interventions 552–553, 554–555, explanation 229, 229, 237; sub-optimal use of
555–556 Bayes’s rule 231
decentralized communication 529 Diamond, Jared 441
decision field theory 325 Dicke, W. 305
decision making: accountability 261–263, 639; difference-ratio-interest-finance-time (DRIFT)
alternative-based choice 242, 243–244, 243, model 246
248–250; body states and 378; branch method differential accessibility of knowledge 170–178
598–599, 600, 602; branching factor 339, digital environments, aversive 289–306;
346; cognitive success 354–355; comparison deceleration principle 297; disconnectedness
problem 339–344, 344; decision goal 642; principle 296; heuristics to manage 295–302;
defensive 293–294; delaying 443; discounting information loss principle 297; non-sharing
models 244, 245, 246, 247; emotions and principle 297; risks of digitalization for children
352–353, 364–365, 366, 382, 404; 304–306; scoring systems 295, 299–301
environment-strategy mismatch 319–320; dilation of probabilities 477–481, 479, 480
evolutionary perspective 313–321; expected discounting models 244, 245, 246, 247, 404
value approach 244, 248; information disruptive change see aversive change
acquisition 248–250, 248, 249; logical distributed cognition 121, 124–126
omniscience 173, 175, 482; optimally distributed model-based reasoning 121, 124–126
imperfect decisions 438; organizational division of labor, in organizations 523
509–518, 597; rules versus discretion 439; dominant cue condition 64–65

653
654

Index

dopamine 401 Einstein, Albert 112

dorsolateral prefrontal cortex 352 Eisenhardt, K. M. 288
Dosher, B. A. 248 elections and sampling equilibrium 426–428
dual process theories 60, 105, 185–193, 207–215, Eliasmith, Chris 402
404–405, 512–513, 548–549; accessibility of Eliaz, Kfir 433
knowledge 171; confirmation bias and elimination-by-aspects heuristic 90, 94, 245, 318
258–259; default-interventionist theories Ellis, Rob 381
188–191, 189, 192; flexibility of System 1 Ellsberg paradox 352–353
212–214; Fodor’s model 208–210, 212, 213; Elqayam, S. 190
genetic and vehicle goals 199–201, 201, Ely, R. T. 452
202; intuition and 188–191, 189, 257–259; embodied cognition 377–387; heuristics 380–383;
limitations of System 2 211–212; metacognitive representations and simulations 381, 383–384,
feelings 191–192; parallel-competitive 385–386
theories 188 embodied emotions 382–383, 385–386
Duncker, Karl 135–137, 138, 140–141 embodied simulations 381, 383–384
Dutton, J. E. 525 emotions: decision making and 352–353,
dynamic stochastic general equilibrium (DSGE) 364–365, 366, 382, 404; embodied 382–384,
models 462 385–386; information and choice overload and
632–633; integration with cognition in brain
Easter Island (Rapa Nui) 441 402–404, 403; psychiatric disorders and
eat sweet food strategy 319 352–353, 364–365, 366; social contagions
ecological rationality 61–66, 83, 92–93, 107, 386
268–277, 313–321, 357–360, 379, 440–441, endowment effect 441, 460
581–582; adaptive toolbox 61, 62–63, 282, Engel, C. 417
317–318, 357; algorithmic modeling and Enlightenment 349, 353
272–276, 272; Bayesian optimality and environment-strategy mismatch 319–320
268–269, 270, 272, 272, 273–275, 276–277; of epistemic logic 172–174, 172, 175, 176
beliefs 65–66, 66; of defeasible logic 167–168; epistemic rationality 79, 81–83, 185, 186;
environment-strategy mismatch 319–320; of awareness logics 176; differential accessibility of
heuristics 63–65; Hindu-Arabic number system knowledge 170–178
165–167, 166; icon arrays and Venn diagrams Epstein, Brian 535, 545
160–162, 161; of lexicographic strategies Epstein, R. A. 582, 586
162–164, 162, 163, 165, 360–361; Marr’s equal weight rule 243
levels of analysis and 269–270, 275–276, 277; European Commission 290–291
natural frequency and probability 158–162, Evans, George 464
160, 161, 300; nudging and 581–582, 583, Evans, Jonathan St. B. T. 554
586–587, 588–589; psychiatric disorders and evolutionary economics 484–490
361–365; representation and problem solving expectations, economic 459–466
157–158; social 360–361; statistical foundations expected utility maximization 55, 56, 57, 59, 60,
of 270–275, 272; strategy selection 318, 320, 76, 78, 283, 287, 355, 406, 448, 474, 475, 480,
386–387; see also fast-and-frugal heuristics 509, 510, 511; see also idealistic and pragmatic
ecological validity 354 cultures of bounded rationality
economics: animal spirits 57, 114, 438; expected value approach 244, 248
bounded rationality in 437–445, 448–456; expertise, cognition and 186, 416–417
cognitive limitations and learning 464–465; expertise bias 550
evolutionary 484–490; expectations 459–466; explanation bias 550, 554
innovation 488–490, 510, 517; learning in eye tracking studies 248–249, 248, 250,
492–501; monetary policy 439, 462, 464, 629–630, 632
591; neoclassical economic theory 55–56, Eyster, Erik 435
57, 59, 60, 61, 90, 93, 94, 437, 438, 439,
484–485, 487, 510–511; routines 486–488; Fagin, R. 176
see also modeling bounded rationality in fallacies: base rate 77; cause-effect isomorphism
economic theory 415; denying the antecedent 74–75; hot hand
Eddy, David 167 66, 105; Is–Ought 75, 78; planning 551, 555;
Edelman, S. 276–277 sunk-cost 404, 460, 602
educational libertarian paternalism 48–49, 50–51 false beliefs 82, 319–320
E-evidence 290–291 falsificationism 76

654
655

Index

fast-and-frugal heuristics 63, 83–84, 164, 281, Prisoner’s Dilemma 429–433; voting models
282–286, 316–317, 444, 498; environment and 427, 428
286–289; fast-and-frugal trees 163, 163, 164, Gangemi, A. 365–366
298; information overload and 630; to manage garbage can model 523–524
aversive change 295–302; research questions Gardner, Howard 452
284–286; scoring systems and 299–301 Garrouste, Pierre 523
Fauconnier, G. 412 gaze heuristic 380
fear-driven inference 319–320, 403, 403 Gazzaniga, Michael 365
feelings of rightness 189, 191, 192 Geanakoplos, J. 176
Fehr, Ernst 90, 94–95 Geertz, C. 417
Fermat, Pierre 4 geese, greylag 316
Ferrero, G. 464 genetic and vehicle goals: in dual-process
Festré, Agnes 523 organisms 199–201, 201, 202; logic of
Fibonacci 166–167 197–199, 198, 199
figure-ground organization 133, 134 genome connection-minimization model
finite automata 429–433 394–395, 395, 396
Fischhoff, B. 548 gestalt characteristics 572–574
Fischman, M. G. 554 Gestalt theory 132–137, 138–139, 140–141,
fixed action patterns 316 146, 148
fixed parameters 91, 94–95 Gibbard, Alan 203
fluency effects 550, 551, 554 Gibson, James 282, 286, 380
fluency heuristic 78, 359 Gigerenzer, Gerd 78, 83, 84, 92, 93, 94, 99, 105,
Fodor, Jerry 208–210, 212, 213 107, 108, 109, 158, 164, 208, 257, 281, 287,
folk psychology 188 379, 382, 438, 439, 442–443, 444, 493, 582,
food choices: environment-strategy mismatch 319; 586–587
nudging and 98, 564–565, 580, 584, 590, 606 Glazer, Jacob 425–426
food pyramids 581, 590 global warming 441
Foucault, Michel 349 glutamate 401
Four Color Theorem 156 Go (game) 338
Fourier Transform 276 Gödel, Kurt 518
frame problem 209 Goldman, Alvin 81–83, 356
framing effect 171, 353, 356, 366, 460, 497–498, Goldstein, D. G. 164, 287
550–551, 554 Gonzalez, Richard 94
Frederick, S. 214 Good, I. J. 471
free parameters 91, 94–95 Good’s principle 471–482; basic and derived
Frege, Gottlob 74, 75, 400, 511 decision problems 473–475, 474, 480–481;
frequentism 76 dilation of probabilities 477–481, 479, 480;
Friedman, Benjamin 463 lower probability 476; Savage’s version
Friedman, Milton 57, 90, 439, 494 473–475, 474, 476; strategic decision problems
Frings, C. 358 471–473, 472; uncertainty and imprecision
full rationality 56 475–477
functional magnetic resonance imaging (fMRI) Google 290–291
368, 629, 631 Gotts, Stephen 368
fundamental analytic bias 190 Gould, Stephen J. 97
fundamental heuristic bias 190 governance, corporate 526
fundamental uncertainty 56 Grant, A. M. 627
fuzzy-trace theory 247–248 Graybiel, A. M. 320
Great Rationality Debate 196–203
gamma aminobutyric acid (GABA) 401 Greve, Henrich R. 528
Gabaix, Xavier 528, 529 group rationality 441
Gaissmaier, W. 107, 109 groups and teams 535–545; group adaptivity
Galbreath, Jeremy 526 542–544; hidden-profile tasks 536–542, 538;
game theory 409–410, 411, 412, 418, 448, 494; as-if models vs. process models 544–545;
adverse selection problem 471–472; asymmetric information environments 540–542; strategies
information in strategic games 471–473, 472; 539–540
backwards induction 177; behavioral 410; Grüne-Yanoff, Till 96, 622
long interactions and finite automata 429–433; Güth, Werner 92

655
656

Index

habit formation 316 344, 345–346; social 63, 360–361, 363, 364,
Hadamard, J. 276 368; strategy selection 318, 320, 386–387;
Hall, R. 438 take-the-best heuristic 63, 78, 84, 164,
hallucinations 362, 365 270–271, 284, 287, 317–318, 382, 630;
halo effect 550 take-the-first heuristic 358, 381–382; see
Halpern, J. 176 also attribute-based choice; fast-and-frugal
Hammond, Kenneth R. 549 heuristics; satisficing
harm principle 572, 574, 575 heuristics-and-biases program 60–61, 77, 80–83,
Hassrick, R. B. 288 138, 257, 351, 357, 599–600; diagnostic
Hausman, Daniel 573 probability estimation 229–230, 231; Great
Hayek, F. A. 498, 579–580 Rationality Debate 202–203; idealistic
healthy lifestyles: environment-strategy mismatch culture of bounded rationality and 93, 96–97;
319; nudging and 98, 564–566, 580, 584, perception/judgement boundary 214; visual
590, 606 perception 104–105; see also dual process
Hegarty, M. 127 theories
Heiner, R. A. 499 hiatus heuristic 63–64, 358, 359
Helbing, D. 290, 296, 305 hidden-profile tasks 536–542, 538
Hertwig, R. 333, 352, 360–361, 622 Hindu-Arabic number system 165–167, 166
Herzog, M. S. 360–361 Hintikka, Jaakko 172, 173–174
heterogeneity in beliefs and behavior, social Hitch, C. 438
benefits of 578–579, 583–584 Hoffman, Andrew J. 527
Hetwig, G. 354 Hoffrage, U. 164
heuristics 170, 242, 268, 357–360, 377–378, Hogarth, R. M. 627, 628
460, 599–600; adjacency heuristic 393, 393; Holmstrom, Bengt 526
affect heuristic 643; affordances and 380–382; Honkapohja, Seppo 464
algorithmic models of 62, 64; availability horse-trading problem 142, 150
heuristic 61, 257; comparison problem hot hand fallacy 66, 105
340–344, 344; cumulative dominance 341–342, household shopping routines 487
343–344, 344, 345–346; cumulative prospect how-to rules 288
theory and 324–325, 328–335, 330, 332, 333; Hull, D. L. 196
ecological rationality of 63–65; elimination-by- human information processing approach 135, 136,
aspects heuristic 90, 94, 245, 318; embodied 138, 139–140, 148, 283
cognition and 380–383; embodied emotions Hume, David 75, 450
and 382–383; environment and 286–289, Hutchins, E. 124
331–332, 332; fluency heuristic 78, 359;
imitate the successful heuristic 360–361; icon arrays 160–162, 161
imitation heuristics 293, 605; inferential 287; idealistic and pragmatic cultures of bounded
insight problem solving 137–142; intelligence rationality 90–99; free and fixed parameters
analysis and 550; intertemporal choice heuristic 91, 94–95; model testing 91, 94–96; modeling
(ITCH) model 246; least-likely heuristic 327, differences 91–96, 91; optimization 91, 93–94;
328, 329, 330; lexicographic 63, 165, 244–245, storytelling differences 96–99, 99
339, 340, 341–343, 345; libertarian paternalism illicit conversion 351
and 50, 571; to manage aversive change ill-structured problems 157, 355
295–302; maximax heuristic 328, 329, 330, imitate the successful heuristic 360–361
331; minimalist heuristic 317–318; minimax imitation heuristics 293, 605
heuristic 327, 328, 329, 330, 331; most-likely implementation level of analysis 269–270
heuristic 327–328, 328, 329, 330, 331–332, implicit learning 186, 188
332; multiple-cue 318; noncompensatoriness inattentional blindness 105, 108, 530, 629
340, 342, 343, 344, 344; notice invariants incentives: attention and 526–527; information
heuristic 140, 141; one-reason 63–65, overload and 641
317–318, 377, 382; psychiatric disorders and inductive reasoning 186–187, 399, 400
359, 363, 364, 365, 367, 368; risk attitudes of Infante, G. 578
331–332, 332; of risky choice 327–328, 328, inference machines 158
329, 330, 331–332, 332, 333–334; scoring inference to the best explanation 229, 229, 237
systems and 299–301; search, stopping, and inferential heuristics 287
decision rules 287, 288, 317, 377; sequential information and choice overload 625–633,
search 318; simple dominance 341, 342–344, 637–646; age differences 644–645; attention

656
657

Index

costs 629; benefits and costs model 627–628, Joseph, John 524, 525
628; characteristics of decision-makers and Juhos, C. A. 221–222
641–645; context and choice environment Jung-Beeman, M. 143
637–641; cultural differences 645; decision justifications and arguments: anticipating need for
accuracy and quality 631–632; definitions 261; choice overload and 639; evaluating
626–627; emotional and subjective states 261–262; production of 262–263
632–633; fast-and-frugal heuristics 630; gender
differences 644; information processing and Kaemmer, J. 543
usage 628–630; motivation for choosing and Kahneman, Daniel 59, 60, 77, 82, 91–92, 97, 99,
consumption 630–631; time costs 630 104–105, 107, 108, 109, 112, 138, 171, 214,
information sharedness 541 257, 351, 357, 400, 404, 441, 486, 513, 549,
innovation 488–490; competence gap 517; 599–600
organizational 510, 517 Kandel, Eric 368
insight problem solving 131–149; business as usual Kanizsa, G. 133
approach 146; Gestalt theory 132–137, Kanizsa Triangle 132
138–139, 140–141, 146, 148; heuristics Kant, Immanuel 2, 75, 78, 84, 147
137–142; horse-trading problem 142, 150; Kaplan, C. A. 140, 143–144
human information processing approach 135, Kaplan, J. T 414
136, 138, 139–140, 148; incubation period Kapur, Shitij 365
141, 142–144, 146; interpretative heuristic Kareev, Yaakov 528
147–149; mutilated checkerboard (MC) Karni, E. 113
problem 141, 149–150, 150; optical illusions Kasparov, Gary 156, 185–186
133, 133, 134, 148–149; special process Katsikopolous, K. V. 548
views 146; study window problem 143, 143, Kennedy, J. 554
144–145, 150–151, 151; thinking aloud 136; Keynes, John Maynard 57, 438, 462, 477
unconscious analytic thought in 142–147 Keynesian theories 461, 464
institutional totalitarianism 295 kinematic mental models 225
instrumental rationality 185–186, 197 Kingdon, J. W. 603
insula 382–383 Klein, Gary 93
intelligence analysis, cognitive biases and debiasing Knight, Frank 56, 57, 287, 477
strategies 549–556, 554–555 knowledge: awareness logics 176; bounded
interactionist view of reason 259–265 170–178; differential accessibility of
internet of things 304–306 170–178; knowledge gap 496; mathematical
intertemporal choice 244, 245, 246, 247–248, representation of 172–174, 172; veridicality
249, 250 axiom 172–173
intertemporal choice heuristic (ITCH) model 246 knowledge level of analysis 174–175
intractability 56, 57, 77–78, 139–140, 391 Knudsen, Thorbjørn 528
intraparietal sulcus 381 Kolmogorov, Andrej 76, 80
intuition 257–259, 382; about chance 65–66, Konstantinidis, E. 250
66; confirmation bias and 258–259; creative Koriat, A. 555
143; default-interventionist theories 188–191, Kornblith, Hilary 80–81, 83
189, 192; feelings of rightness 189, 191, 192; Kornblith, Steven 356
interactionist view of reason 259–265; logical Kropotkin, Peter 579
intuitions 192; mental models and 222–224; Krynski, T. R. 231
metacognitive feelings 191–192
Invisible Gorilla Test 629 Lamentation of Christ, The (Mantegna) 149, 149
irrationality 60–61, 81, 93, 96–97, 398, 399–400, Lan, C. 548
399; see also psychopathological irrationality Laplace, P. S. de 511
isotropic processes 209, 212, 214 law of small numbers 65, 77, 81
Is–Ought fallacy 75, 78 learning: in economics 492–501; implicit
186, 188; machine 64, 186, 273, 274–275,
Jacoby, J. 626 338; mental models and 496–497, 499;
James, William 113, 221, 522, 524 organizational 499–501, 517–518, 530;
Jehiel, Philippe 435 reinforcement 318, 320, 338, 362, 369,
Johnson, E. J. 250 386–387; social 360–361, 362–363, 364–365;
Johnson-Laird, P. N. 351 trial-and-error 318
Jones, M. 273 learning algorithms 64, 186, 273, 274–275, 338

657
658

Index

least cognitive effort principle 415–416 March, James G. 92, 450, 487, 497, 513–514,
least-likely heuristic 327, 328, 329, 330 515, 517–518, 523–524, 525, 527, 528, 530
Lee, Cassey 93 Marengo, L. 515
Lehrer, Ehud 433 Markowitz, Harry 359
Leland, J. W. 247 Marr, David 106, 232, 268, 269–270, 277
Leonardo da Pisa 166–167 Martignon, L. 163, 163, 164
Leptothorax albipennis ants 315, 316 matching bias 191
“less-is-more” principle 270, 358, 359, 528; mathematics education 155–168; defeasible logic
see also Good’s principle and 167–168; Hindu-Arabic number system
levels of analysis: algorithmic 232, 269–270, 165–167, 166; icon arrays and Venn diagrams
275–276; computational 232, 269–270, 275, 160–162, 161; lexicographic strategies
276, 277; implementation 269–270; knowledge 162–164, 162, 163, 165; natural frequency and
174–175; symbolic 174 probability 158–162, 160, 161; representation
Lewin, K. 321 and problem solving 157–158
Lewin, Kurt 58 Matte Blanco, I. 350, 351
Lewisohn, M. 535, 545 maximax heuristic 328, 329, 330, 331
lexicographic heuristics 63, 165, 244–245, 339, Mayan civilisation 441
340, 341–343, 345 Mayberg, Helen 368
lexicographic strategies 162–164, 162, 163, 165, means paternalism 564
360–361 mean-variance portfolio algorithm 359
Li, Shengwu 426 mechanism design 424–426
libertarian paternalism 48–51, 357, 564, 570–576, Meder, B. 231, 238
571, 585, 606; budging as alternative medial prefrontal cortex 516–517
575–576, 575; cognitive 48, 49–50; educational Megiddo, Nimrod 433
48–49, 50–51 Meliorists 202–203
Linda problem 77, 97 mental models: collective 499; computational
Lindblom, Charles 598–599, 600, 601, 602 simulation and 123, 124, 125–126, 127;
Lindley, D. V. 471 deductive reasoning and 217–226; intuition and
linear algorithms 64–65 222–224; kinematic inferences 226; learning
Loasby, B. J. 498 and 496–497, 499; spatial and temporal
Logan, S. W. 554 inferences 224–225
logic of awareness 176 Mercier, H. 232
Logic Theorist program 74, 75 mesoscopic modeling 122–124, 126–128
logical intuitions 192 metacognition 191–192
logical omniscience 173, 175, 482 metarepresentational cognitive mechanisms
Lohse, G. L. 250 259–261
Lopes, Lola 60, 97 methodological naturalism 79, 80, 81
Lorenz, Konrad 110 Milgrom, Paul 526
loss aversion 56, 326, 327, 352, 353, 366, Mill, John Stuart 570, 572, 574, 575
404, 441 Miller, George A. 405, 628–629, 638
lotteries 56, 328, 333 minimalism, in deductive reasoning 221
Love, B. 273 minimalist heuristic 317–318
Luan, S. 543 minimax heuristic 327, 328, 329, 330, 331
Lucas, Robert E. 461, 463, 464 mirror neurons 368, 381, 383–384, 385–386
Luce, R. Duncan 94 miserly cognitive processing 190, 196–197,
Luchins, A. S. 512 415–416
Luchins, E. H. 512 Mitchell, Wesley Clair 489
Lupia, A. 410 modeling bounded rationality in economic theory
423–435; agents with different models in mind
machine learning 64, 186, 273, 274–275, 338 433–435; elections and sampling equilibrium
magnitude effect 245, 246, 247 426–428; long interactions and finite automata
Maidique, Modesto 288 429–433; mechanism design 424–426
Maier, N. R. F. 150 Modica, S. 176
majority rule 537, 539, 544–545 Modus Ponens 232–237, 233, 235, 236
Manches, A. 304 monetary incentives: attention and 526–527;
Manski, C. F. 463 information overload and 641
Mantegna, Andrea 149, 149 monetary policy 439, 462, 464, 591

658
659

Index

“more-is-more” principle see Good’s principle neuroscience: cognitive 409–418; social 384–386;
Morgan, Mary S. 91, 96 see also brain
Morgenstern, Oskar 76, 97, 438, 476, 509 neuroscientifically relevant human psychological
Mosconi, G. 139 factors (NRPs) 414–417
most-likely heuristic 327–328, 328, 329, 330, neurotransmitters 401
331–332, 332 New York Times 97
motivated inference 319–320, 403, 403 Newell, Allen 74, 75, 137–138, 139–140, 156,
motor cognition hypothesis 383 174–176, 339, 377, 379, 448, 518
motor neurons 381, 383 Neyman, Abraham 432–433
Motorola 525 NK fitness landscape models 500
mouse tracking studies 249–250, 249 Nobel Prize for economics 59, 63, 105, 171, 398,
Müller-Lyer illusion 133, 148–149 448, 459
multi-attribute choice 243–244, 248, 250 noncompensatoriness 340, 342, 343, 344, 344
Multicriteria Decision Analysis (MCDA) 613, non-compensatory weights 164, 165
614, 615–616, 617, 617, 621, 622, 622 Norman, D. A. 125
Multifaceted Empathy test 364 normative systems of rationality 74–78
multi-methodology interventions 610–622, 622; notice invariants heuristic 140, 141
educational foundation for underprivileged Nozick, R. 79
children in Hungary 615–617, 616, 617; nudging 98, 99, 362, 553, 570–572, 571,
eliciting value functions in decision making 578–591; antecedent preferences and 563,
617–621, 619, 620, 621; rural regeneration 564–565, 566–567, 568; “as judged by
project in new Italian World Heritage Site themselves” criterion 563–568, 580, 587;
612–614, 614 authoritarian turn 587–588; budging as
multiple-cue decision heuristics 318 alternative 575–576, 575; choice architects 563,
multitasking 211, 526 567, 568, 583–584, 606; choice architecture
Mumma, G. H. 554 553, 563, 565, 567, 571, 646; coercive
Musgrave, Alan 95 paternalism 564, 572, 585; costs and risks of
musical composition program 156 584–586, 589–591; disagreements on expert
Muth, John F. 439, 461, 462, 463, 464, 465 advice 580–581, 590; ecological rationality and
mutilated checkerboard (MC) problem 141, 581–582, 583, 586–587, 588–589; education as
149–150, 150 alternative 98, 574, 586–587; healthy lifestyles
98, 564–566, 580, 584, 590, 606; lobbying and
naïve Bayes classifiers 274 regulatory capture 581, 584, 586; organ-donor
Narduzzo, A. 516 status 98, 585; political uses of scientific claims
Nash equilibrium strategies 409–410, 428, 432, 581; retirement savings 580, 581, 589–590,
472–473 591; self-control problems 565–566, 590;
Native American nations 280–281, 283, social benefits of heterogeneity in beliefs and
286, 288 behavior 578–579, 583–584; see also libertarian
natural frequency and probability 158–162, 160, paternalism
161, 300 numeric systems 165–167, 166, 378
natural selection 196–197, 400 Nussbaum, A. D. 320
naturalism 73, 78–85
Naturalizing Epistemology programme 2–4 Oaksford, M. 273
Nature 99 obesity 319
negative introspection 173, 176 Ocasio, William 524, 525, 531
Nelson, R. R. 495, 514–515 Ohly, S. 516
neoclassical economic theory 55–56, 57, 59, 60, Olympian rationality 170, 173, 257, 492; see also
61, 90, 93, 94, 437, 438, 439, 484–485, 487, neoclassical economic theory
510–511 O’Neil, C. 300
Nerlove, M. 464 one-reason heuristics 63–65, 317–318, 377, 382
Nersessian, N. J. 125 online dating 644
neural reuse 378 ontological naturalism 79, 80, 84
neural wiring optimization 391–394, operational memory 499, 500
393, 394 Oppenheimer, D. M. 359
Neurath, Otto 160 optical illusions 105, 133, 133, 134, 148–149
neurocognitive models 384–386 optimality, Bayesian 268–269, 270, 272, 272,
Neuron Doctrine 414 273–275, 276–277

659
660

Index

optimally imperfect decisions 438 112–114; search images 111; visual illusions
optimization: under constraints 59, 62, 443; 105, 133, 133, 134, 148–149
idealistic and pragmatic cultures 91, 93–94; perceptual priming 105
neural wiring 391–394, 393, 394; see also perfect foresight 56, 58, 59
expected utility maximization personal construct psychology 498
orbitofrontal cortex 352–353, 382, 404 personality disorders 362–363, 365, 366–367
order effects 551, 554 Pesaran, H. 462
organ-donor status 98, 585 Pfeiffer, J. 385
organism-environment relationship 109–110 physical symbol system hypothesis 377
organizational communication 525, 529, 530 Piaget, Jean 77
organizational decision making 509–518, 597; Piccione, Michele 434–435
see also organizations and attention planning fallacy 551, 555
organizational innovation 510, 517 Platonic solids 155
organizational learning 499–501, 517–518, 530 PMM-algorithm 287
organizational memory 499, 500 Poggio, Tomaso 269
organizational politics 530 Polanyi, M. 111
organizational procedures 513–514, 529 political science 597–607; information processing
organizational routines 500–501, 513–517 604–605, 607; institutions 599–600, 601, 607;
organizations and attention 522–531; adaptive policy bubbles 605, 606–607; policy process
value of attention 527–528; attention failures 601–606; public budgets 602–603; punctuation
530–531; attention-based view of the firm 524, in policy change 603–605, 607
525; decision hierarchies 523; division of labor Politzer, G. 232
523; experimental paradigms 531; garbage Popper, Karl 5, 76, 112, 579
can model 523–524; inattention 528–529; positive affect 527, 643
incentives 526–527; organizational mechanisms positive introspection 173
524–527 possibility effect 334
Ortmann, A. 358 possible worlds 172, 172, 173
Osborne, Martin 428, 472–473, 472 pragmatic culture see idealistic and pragmatic
Ostrom, Elinor 599 cultures of bounded rationality
Otto, P. E. 318, 320, 386 precautionary principle 443–444
overconfidence 551, 552, 555 precision weighting 213–214
prediction 57, 61, 62, 63–65
paleologic reasoning 349–351 predictive processing theories 213
Panglossians 202–203 preference uncertainty 643
parable of the ant 121 prefrontal cortex 352, 382–383, 402, 404,
parallel-competitive theories 188 516–517
paranoid personality 363, 365, 366–367 Prendergast, Canice 526–527
Pareto optimality 438 Prescott, Edward 463
Pareto/NBD model 63–64, 358 present bias 523
partial belief 476 pricing routines of firms 438, 487
Pascal, Blaise 4 principle of correspondence 357
paternalism 570–576; budging as alternative principle of least cognitive effort 415–416
575–576, 575; coercive 564, 572, 585; means prioritizing rules 288
564; see also libertarian paternalism; priority heuristic 92, 93, 94, 96, 328, 328, 329,
nudging 330, 331, 333–334
Patokorpi, E. 377 Prisoner’s Dilemma 429–433
pattern recognition 156, 186, 188 privacy 290–291
peak-end evaluation 573 probabilistic inference errors 351
Peirce, Charles S. 232, 400 problem solving 454; artificial intelligence and
Penrose, Edith 113 518; body states and 379; cognitive categories
perception 103–115; )500 bill axiom 103, 114; and 496–499; cognitive success 354–355; in
affordances 380–382; brain and 380–381; organizations 514; representation and 157–158;
embodied nature of 380–382; Gestalt theory see also groups and teams; insight problem
132–135; “looking for” rationality 110–112; solving
organism-environment relationship and procedural economics 456
109–110; perception/judgement boundary procedural rationality 77, 126–127, 451, 454–455,
214; rationality and perception of value 456, 493, 494–495

660
661

Index

propensity theory 76 Reeck, C. 250

proportional difference model 245, 325 reflection effect 245, 247, 325, 334
prospect theory 90, 92, 93, 94, 96, 171, 324; Regenwetter, M. 543–544
cumulative 92, 96, 324–327, 327, 328–335, Reichenbach, Hans 76
330, 332, 333 reinforcement learning 318, 320, 338, 362, 369,
psychologism 74–76, 80 386–387
psychopathological irrationality 349–369; autism reliabilism 82
352, 353, 363, 366, 368; brain and 367–369; religious beliefs 441
cognitive success 354–355; distorted emotions resource rationality 270
and 352–353, 364–365, 366; ecological retirement savings 580, 581, 589–590, 591
irrationality of psychiatric disorders 361–365; Richerson, P. J. 196, 600
greater logical rationality in psychiatric patients Rieskamp, J. 318, 320, 386
365–366; heuristics and 359, 363, 364, 365, Riksbank 448
367, 368; impermeability to environmental risk 56, 58
feedbacks 361–362; loss aversion 352, 353, 366; risk attitudes 56, 325, 326–327, 327, 329,
paleologic reasoning 349–351; paradoxes of 331–333, 333, 404, 550–551
365–367; paranoid personality 363, 365, risky choice 244, 245, 247–248, 249, 250, 324,
366–367; personality disorders 362–363, 365, 325, 327, 455, 509; cumulative prospect theory
366–367; principle of generalisation 350; 326–327, 327; heuristics of 327–328, 328, 329,
principle of symmetry 350; schizophrenia 330, 331–332, 332, 333–334
349–351, 361, 362, 363, 365, 367, 369; Rolls, B. J. 630
social learning 360–361, 362–363, 364–365; Roman numeral system 165–166
social rationality 360–361, 363; violation of Rosen, L. D. 248
coherence 355–357 Rothschild, Emma 113
psychophysics 107, 109 routines 486–488, 500–501, 513–517
public budgets 602–603 Rowe, Dick 535
public policy: behavioural 570–576, 572, 575; Rubinstein, Ariel 59, 90, 247
see also nudging; political science Russell, Bertrand 74, 75
punctuation in policy change 603–605, 607 Russo, J. E. 248
Rustichini, A. 176
Quandt, R. E. 438, 610
quasi-hyperbolic discounting 529 Sahlins, M. 442
Quine, Willard O. 2–4, 356 sampling equilibrium 426–428
Quineian processes 209, 212, 214 Santos, Tano 529
Sargent, Thomas 59, 463, 464
radical uncertainty 56 satisficing 57–58, 59, 78, 83, 84, 93, 94, 170,
Raiffa, H. 471 243–244, 263, 283, 318, 437, 438, 442,
Ramsey, F. P. 471 450–452, 459, 464, 498, 548, 643–644
Rapa Nui (Easter Island) 441 satisficing with adaptation 62–63
rational analysis of cognition 270, 272, 282 Savage, Leonard J. 55–57, 91, 271, 277, 287, 471,
rational choice theory 60, 76, 456, 509, 511, 473–475, 476, 482, 492, 509
572, 598 Save More Tomorrow programs 589–590, 591
rational expectations theory 103, 108, 114, 439, savings, retirement 580, 581, 589–590, 591
461–462, 463, 464–465, 493 Scherrer, B. 344–346, 345
rationality: concept of 55–56, 399, 399, 440; schizophrenia 349–351, 361, 362, 363, 365,
philosophical tradition of 2 367, 369
rationalization 403, 404 Schlaiffer, R. 471
Read, D. 246 Schmidt, Klaus 90, 94–95
Read, Stephen 402 Schuck, N. W. 516–517
reason, interactionist view of 259–265 Schumpeter, J. A. 488, 510, 517
reasons/justifications: anticipating need for 261; Schurtz, G. 354
choice overload and 639; evaluating 261–262; Schutt, R. K. 384
production of 262–263 Schwartz, B. 108, 626, 627, 643
Rebonato, R. 585 Science 97
recognition heuristic 83, 270, 316–317, scissors analogy 58, 61, 282–283, 313, 353, 379,
358–359, 382 548, 637
Redelmeier, D. 573 score cards 525

661
662

Index

scoring systems 295, 299–301 simulations, embodied 381, 383–384

search, stopping, and decision rules 287, 288, Sioux 280–281, 283, 286, 288
317, 377 small worlds 56–57, 271, 277, 287, 492, 597
search images 111 Smith, Adam 93, 113, 410, 412, 485, 501
search theory 59 Smith, Vernon 63, 440–441
security-potential aspiration model 325 Snow, C. P. 99
selective attention 171, 211, 528, 530 social brain 368
self-control problems, nudging and 565–566, 590 social combination rules 537, 539
self-fulfilling prophecies 114, 367 social communication rules 540
Selten, Reinhard 78, 90, 92, 109, 275, 438, 439, social conformity 293
442–443, 510–511 social embeddedness 498
Semantic Pointer Architecture 402–403, 405 social heuristics 63, 360–361, 363, 364, 368
Sen, Amartya 356 social learning 360–361, 362–363, 364–365
Seneca 2 social neuroscience research 384–386
sensory order 498 social norms 441, 600
sequential decision problems 339, 344–346, 345 social rationality 360–361, 363
sequential search heuristics 318 sour grapes 319, 403–404, 403
serotonin 401 spatial inferences 224–225
Serwe, S. 358 speed-dating events 625, 644
Shafer, Glen 579 Spence, Michael 434
Shahbazi, R. 276–277 Sperber, D. 232
Sherlock Holmes 187 Spiegler, Rani 424, 428, 433, 435
Shiller, R. J. 57 Spivey, M. 415
similarity bias 550 standard operating procedures 513–514, 529
similarity model 247 “standard picture” of rationality 74–78
Simon, Dan 402 Stanovich, Keith 188, 189–190, 210–211,
Simon, Herbert A. 5, 55–60, 92–93, 104, 108, 548–549
109, 112, 113–114, 155, 257, 354–355, 358, statistical learning methods 340–341
398, 400, 486, 549, 599; artificial intelligence Staub, A. 231
156–157, 518; Challenge interview 449, Stein, E. 74
451–452, 453; chess 156, 185, 416–417, Stenning, K. 167–168
512–513; cognitivism 135, 139, 377–378, Stich, Steven 356
379, 454–455; computational efficiency 455; Stigler, G. J. 59, 62, 438
economic man 210; expectations in economics stochastic customer base models 63–64
459–460, 463–464; expertise and cognition stochastic dominance violations 245, 247
186, 416–417; Hindu-Arabic number system stopping rules 287, 288, 317, 377, 438, 448
165; human information processing approach strategic decision problems 471–473, 472
135, 136, 138, 139–140, 148, 283; information strategy selection 318, 320, 386–387
and choice overload 625, 626, 629, 637, strategy selection learning procedure 318,
645, 646; insight problem solving 135, 136, 320, 386
137–138, 139–140, 141, 142, 143–144; Logic striatum 404, 631
Theorist program 74, 75; multiple logics 167; Strotz, R. H. 591
Nobel Prize 59, 448, 459; organizational structure induction model 231, 238
decision making 509, 510, 511–512, 513–514, structured analytic techniques (SATs)
515, 517–518, 597; organizations and attention 552–553
522, 523, 524, 525, 526, 527; parable of the ant study window problem 143, 143, 144–145,
121; predictions on future of computers 156; 150–152, 151
problem solving 518; procedural rationality 77, subjective expected utility see expected utility
126, 451, 454–455, 456, 495; representation maximization
and problem solving 157; routinized behaviors Sugden, Robert 564, 574, 578
513–514, 515; satisficing 57–58, 59, 78, 83, 84, Sull, D. N. 288
93, 94, 170, 283, 437, 438, 442, 450–452, 459, sunk-cost fallacies 404, 460, 602
464, 548; scissors analogy 58, 61, 282–283, Sunstein, Cass R. 98, 357, 571, 585, 590
313, 353, 379, 548, 637 Svizzero, S. 442
simple dominance 341, 342–344, 344, 345–346 SWOT analysis 613, 614, 614, 622, 622
simple heuristics approach see fast-and-frugal syllogistic reasoning 399–400
heuristics symbolic level of analysis 174

662
663

Index

System 1 and System 2 thinking 60, 105, Uexküll, Jakob von 110
185–193, 207–208, 210–215, 404–405, ultimatum game 94–95
548–549, 604; accessibility of knowledge unbounded rationality 56, 59, 98–99
171; confirmation bias and 258–259; default- uncertainty 56–58, 66–67, 450
interventionist theories 188–191, 189, 192; undecidability of first-order logic 391, 392
flexibility of System 1 212–214; genetic and UNESCO World Heritage Site, Italy
vehicle goals 199–201, 201, 202; intuition 612–614, 614
and 188–191, 189, 257–259; limitations of utilitarianism 572–574
System 2 211–212; nudging and 583–584, 591; utility maximisation, Benthamite 572–574
parallel-competitive theories 188 utility maximization, expected see expected utility
maximization
take-the-best heuristic 63, 78, 84, 164, 270–271,
284, 287, 317–318, 382, 630 Value Focused Thinking 615–616, 616, 617, 617,
take-the-first heuristic 358, 381–382 618, 622, 622
tallying heuristic 164, 339, 340, 341–343 van den Steen, Eric 113
Target the Two (game) 516 van Fraassen, Bas 1
Tarski, Alfred 74 van Os, Jim 365
teams see groups and teams Vandenbosch, Betty 525
temporal inferences 225 Venn diagrams 160
Tenenbaum, J. B. 231 ventral striatum 404
Tetris (video game) 339, 344–346, 345 ventromedial prefrontal cortex 352, 382–383
Thagard, Paul 319, 320, 598 Viale, Riccardo 451, 549, 578, 581, 583–584, 585
Thaler, Richard H. 98, 99, 104, 357, 398, 400, Vienna Circle 76
441, 495, 563, 564, 566, 567, 571, 585 Vierø, M. L. 113
theory of mind 188, 368, 385, 411, 416 vision 269
Thiery, C. 344–346, 345 visual perception 104–105, 282; gaze heuristic
Thog problem 207 380; Gestalt theory 132–135; optical illusions
time discounting 404 105, 133, 133, 134, 148–149
Tinbergen, Niko 110, 313 Voit, E. O. 127, 128
Tipper, S. P. 381 Volz, K. G. 352
Tisdell, C. A. 439, 441, 442, 443 von Domarus, E. 350
Todd, P. M. 63 von Neumann, John 76, 97, 157, 438, 476, 509
Tolman, Edward C. 522 voting and sampling equilibrium 426–428
total evidence principle 474
totalitarianism 295 Wangenheim, F. 358
tradeoff model 246 Wason, Peter 75, 77, 207, 351
Trafton, J. G. 125–126 Wason selection task 170, 171, 207, 351, 402
transaction costs 439 wayfinding 412–413
transfer-of-attention exchange model Weckstein, R. S. 442
325 Weick, Karl E. 527
Transiently Assembled Local Neural Systems weighted additive (WADD) rule 243
(TALoNS) 213 Weisberg, R. W. 146
transitivity of preferences 91, 245, 246, 247, Wertheimer, Michael 139, 145, 145
424–426 Whitehead, Alfred N. 74, 75
transparency 295 Wickens, C. 554
traveling salesman problem 77 Wigderson, Avi 433
trial-and-error learning 318 Wildavsky, A. B. 599, 601
Trickett, S. B. 125–126 Williams, George 203
Tucker, Mike 381 Williamson, J. 229
Tuggle, Chris S. 526 Wilson, Alex J. 524
Turing, Alan 74, 75, 518 Wilson, S. B. 554
Turing Award 448 Winter, Sidney 486, 487, 494, 495, 514–515
Turing machine 414, 433 Wittgenstein, L. 417
Tversky, Amos 59, 60, 61, 77, 82, 90, 91–92, working memory 259–260; insight problem
97, 138, 257, 351, 357, 400, 404, 441, 566, solving and 142, 146; limitations of 405; mental
599–600 modeling and 125, 127; System 2 thinking 188,
tyranny of freedom 626, 632 189, 190, 192, 193, 210, 211–212

663
664

Index

Wu, George 94 Zegart, Amy B. 530

Wübben, M. 358 Zenger, T. R. 113
Zhang, J. 125
young girl-old woman illusion 133, 134 Zywicki, T. J. 589–590

664