See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/338563450

How people decide what they want to know

Article  in  Nature Human Behaviour · January 2020

DOI: 10.1038/s41562-019-0793-1

CITATIONS READS

6 480

2 authors:

Tali Sharot C. Sunstein

University College London Harvard University
66 PUBLICATIONS   3,758 CITATIONS    654 PUBLICATIONS   41,895 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Social Information Is Integrated into Value and Confidence Judgments According to Its Reliability View project

How People Update Beliefs about Climate Change: Good News and Bad News View project

All content following this page was uploaded by C. Sunstein on 23 June 2020.

The user has requested enhancement of the downloaded file.

 Perspective
https://doi.org/10.1038/s41562-019-0793-1

How people decide what they want to know

Tali Sharot   1* and Cass R. Sunstein2

Immense amounts of information are now accessible to people, including information that bears on their past, present and
future. An important research challenge is to determine how people decide to seek or avoid information. Here we propose a
framework of information-seeking that aims to integrate the diverse motives that drive information-seeking and its avoidance.
Our framework rests on the idea that information can alter people’s action, affect and cognition in both positive and negative
ways. The suggestion is that people assess these influences and integrate them into a calculation of the value of information
that leads to information-seeking or avoidance. The theory offers a framework for characterizing and quantifying individual
differences in information-seeking, which we hypothesize may also be diagnostic of mental health. We consider biases that can
lead to both insufficient and excessive information-seeking. We also discuss how the framework can help government agencies
to assess the welfare effects of mandatory information disclosure.

H
ow do people decide what they want to know? As massive It is apparent, however, that instrumental utility is not the sole
amounts of information are becoming available, this ques- factor guiding information-seeking, as agents often want informa-
tion is more pertinent today than ever. People must now tion that cannot be used to alter outcomes4–7. People might want
decide whether they want to receive highly personalized informa- information about the life of William Shakespeare or the origins of
tion about their health, genetic make-up and financial prospects, as the universe even if that information will have no effect on what
well as those of their parents and their children. At the same time, they do. This observation has led to the idea that people use a heu-
new laws and regulations are being issued that require disclosure of ristic according to which “knowledge is always valuable”4. Such an
information in diverse areas, including calories in restaurants, fuel approach may be adaptive, because information could turn out to be
economy of motor vehicles, energy efficiency of refrigerators and useful in the future even if it appears useless at present8,9. Consistent
genetic modification of food. with this claim, neuroscientists have shown that the opportunity to
An important research challenge is to determine how people gain knowledge is encoded using some of the same neural architec-
decide to seek or avoid information and how those decisions affect ture and algorithms as for primary rewards5,6,10–19, suggesting that
their welfare. By ‘information-seeking’ we mean the active pursuit of knowledge may have intrinsic value4,7 (Fig. 1). ‘Information predic-
knowledge, for example by asking questions, reading, running tests tion error’ signals (IPEs) have been identified in dopamine-rich
and conducting online searches. Understanding how people make brain regions6 (Fig. 1); analogous to reward prediction errors20, IPEs
such decisions is important for many fields, including the educa- are theorized to provide reinforcement for seeking information6.
tion and health sectors, for public policy and law, and for the design Human behaviour, however, is inconsistent with the proposi-
of intrinsically motivated, curious, artificial intelligence systems. tion that knowledge is always perceived to be valuable, as people
Research on information-seeking has been surprisingly limited in sometimes choose to remain ignorant. For instance, they often
comparison to other domains of human cognition and behaviour and reject medical screening21–23. Examples of this kind have led psy-
somewhat narrow in focus, but seems to be experiencing revitaliza- chologists, economists and neuroscientists to consider the hedon-
tion in recent years (for review see ref. 1). Below, we present a frame- ics of information2,5,24–28. It is theorized that when deciding whether
work for considering the diverse motives that lead people to seek or to seek information, people consider how knowledge will influence
avoid information, and we explore how this framework may be used their psychological well-being (i.e., ‘how will I feel if I learn I carry
to investigate individual differences and improve policy-making. the BRCA gene?’) and that, all else being equal, agents will be more
likely to seek information when they expect good news than when
An integrative framework of information-seeking motives they expect bad news2,5,24–26.
Deciding whether to seek information is a particularly difficult It remains unclear, however, how instrumental utility and
problem to solve, because knowledge and its avoidance can serve hedonic utility are integrated in the brain to guide information-
diverse and sometimes competing functions2. For instance, imagine seeking choices (for a computational proposal see ref. 29). It also
we had information on whether you have a genetic predisposition to remains unclear which other factors play a role. Here we propose
specific forms of cancer; would you want to know? When we posed a framework that aims to encompass the motives that drive infor-
this question to 400 individuals online just over half (58%) said yes mation-seeking and its avoidance. This framework also offers a
and the rest (42%) said no. What drove people’s decisions? structure for characterizing and quantifying individual differences
Classic theories of information-seeking suggest that agents in information-seeking.
seek information that can aid decisions to obtain reward and Our theory rests on the idea that information can alter people’s
avoid harm (i.e., that has ‘instrumental utility’)3. For example, action, affect and cognition in both positive and negative ways
knowing whether one carries the BRCA gene, which increases the (Fig.  2a). When deciding whether to seek information, people
likelihood of breast cancer, can inform decisions about whether to may estimate the expected impact of information on their action
undergo preventative surgery. (‘will the knowledge help, hinder or have no influence on my ability

Affective Brain Lab, Department of Experimental Psychology, University College London, London, UK. 2Harvard Law School, Harvard University,
1

Cambridge, MA, USA. *e-mail: [email protected]

Nature Human Behaviour | www.nature.com/nathumbehav

Perspective NATure HumAn BehAvIOur

1. Action (instrumental value). The ability to use information to

a
select actions that increase extrinsic rewards and help evade
losses is an important driver of information-seeking. This com-
ponent of our framework is found in most classic models of
VMPFC information-seeking3. What has often been overlooked, how-
and OFC ever, is that information can also have negative instrumental
value. That is, knowledge can at times cause individuals to
select actions that lead to worse outcomes, while deliberate
NAc ignorance can lead to better outcomes30. For example, not
knowing the gender of an applicant could improve hiring deci-
sions by reducing bias; not knowing whether a client is guilty
VTA
could improve a solicitor’s performance; not being aware of
one’s own limitations increases confidence and motivation,
Reward circuitry engaged
in information-seeking which at times may improve functioning31.
According to rational choice theory, information cannot have
negative instrumental value, because it could always be disre-
b garded after it is obtained. In practice, however, once a piece of
Informative outcome cue Large reward information is known, its impact on action is often beyond an
midbrain dopaminergic neurons

(signals large reward) (uncued)

agent’s control and/or awareness and thus cannot be reversed.
Firing rates (spikes s–1)

15
For example, once an applicant’s gender is known, one can-
10 Knowledge cue not undo this knowledge and reverse its unconscious impact
on hiring decisions. In law, the point is familiar in the context
5 of jury deliberations. If a judge directs a jury ‘not to consider’
0
evidence that has been wrongfully put before it (such as evi-
dence of prior crimes), there is a real question whether jurors
Ignorance cue Small reward will or can follow that direction.
(uncued)
When knowledge is predicted to influence one’s own action
Informative outcome cue
(signals small reward) negatively, information will have negative instrumental value,
and vice versa when the influence is predicted to be positive. If
Fig. 1 | Neural correlates of information-seeking. It is hypothesized knowledge is predicted not to influence action, its instrumental
that information-seeking is achieved via neural architecture and value will be zero.
computational rules similar to those used in reward-seeking. a, Past 2. Affect (hedonic value). Knowledge can induce both positive and
studies have implicated the reward circuitry (including mesolimbic and negative affect. Knowing that one has a predisposition to certain
mesocortical pathways) in information-seeking and curiosity (see Glossary cancers, for example, can generate sadness, despair or fear22. All
for the distinction between the two)5,6,10–17. For example, epistemic curiosity else being equal, individuals are motivated to avoid informa-
was associated with enhanced activation in nucleus accumbens (NAc) and tion that induces negative affect and to seek information that
ventral tegmental area (VTA)16, and the ventral medial prefrontal cortex evokes positive affect—using information to regulate emotion26.
(vmPFC) has been shown to signal instrumental value of information45. The Consistent with this proposition are observations that investors
orbital frontal cortex (OFC) has been shown to code for the opportunity to monitor their portfolio more frequently when they expect their
increase knowledge19 and for rewards and punishments15, and to respond worth has gone up rather than down32; that monkeys select to
to curiosity relief17. b, The firing rate of a midbrain dopamine neuron in know in advance the size of reward they are about to get6,18,19;
a non-human primate encodes both error in predicting the opportunity that some people refuse to receive results of medical tests they
to gain knowledge (IPEs; yellow oval) and reward (reward prediction have taken33 and prefer not to receive information about un-
errors or RPEs; red and blue peaks and valleys outside yellow oval). Image pleasant events, such as the year of their death. The latter find-
adapted with permission from ref. 6, Cell Press. The red line represents ing has been explained by some by reference to ‘a regret theory
firing rates when advance information is given to a monkey about the size of deliberate ignorance’, an effort to avoid the regret of choos-
of an upcoming reward. The blue line represents firing rates when advance ing to obtain information that one would have preferred not to
information is not given. Solid line represents a situation in which the learn34. A more parsimonious explanation is that people want
preferred (large) reward is given and dotted lines when the less preferred to avoid the negative affect evoked directly by unpleasant infor-
(small reward) is given. In humans, IPEs in the dopamine-rich VTA have mation. It has also been shown that humans are more likely to
been shown to be modulated by the hedonic utility of information5. pay for information when they expect good news (for example,
information indicating financial profit) and are even willing to
pay to avoid information when they expect bad news (for ex-
to make decisions to increase reward and avoid harm?’), affect (‘will ample, information indicating financial loss)5. Key subcortical
the information induce positive or negative feelings, or have no regions in the reward circuitry—the ventral tegmental area and
influence on my affect?’) and cognition (‘will information improve substantia nigra—selectively code for the opportunity to receive
my ability to comprehend and anticipate reality?’). Each of these information that is expected to convey good news, as if such
estimates can be positive (increasing information seeking), nega- information has utility in and of itself5. The code takes the form
tive (increasing information avoidance) or zero (inducing indiffer- of an IPE. Similar coding is not observed for the opportunity to
ence). These estimates will then be integrated into a computation gain information about losses, suggesting that the intrinsic util-
of the value of information, which will trigger information-seeking ity of knowledge is modulated by valence5.
(if the integrated value is sufficiently positive), its active avoidance We suspect, however, that hedonic utility considerations alone
(if the integrated value is sufficiently negative) or neither (i.e., indif- may at times cause individuals to seek information that is
ference). Each factor can be weighted differently, influencing the expected to induce negative affect. For example, uncertainty
decision to seek or avoid information to different degrees (Fig. 2a). about bad news may trigger prolonged anxiety that is worse
Below we expand on each of these three factors. than the expected reaction to such information. Thus, similarly

Nature Human Behaviour | www.nature.com/nathumbehav

NATure HumAn BehAvIOur Perspective
a The above examples all converge on the same principle: all else
Action being equal, people will choose to seek information when the
β1
‘Instrumental utility’ affective response to knowing at the present time is anticipated
β2
to be more positive than the affective response to remaining
Affect Information Information
‘Hedonic utility’ value seek/avoid ignorant. When the reverse is true, people will choose to avoid
β3 information.
Cognition 3. Cognition (cognitive value). Information can enhance or reduce
‘Cognitive utility’ people’s sense that they understand the world around them. In
particular, information alters people’s internal  mental models.
Mental models are representation of concepts (for example, ‘dog’,
Biases Individual
differences ‘Shakespeare’, ‘mom’, ‘alien’, ‘democracy’, ‘cancer’, ‘money’, ‘self ’)
and the relationships among them, which are used to compre-
b c
hend and anticipate reality36,37. Certain concepts (sometimes re-
2.5 *** ** ferred to as ‘nodes’ of the model) may be part of one person’s
mental model but absent in another’s. Moreover, certain con-
2
Coefficient predicting

1
cepts (for example, ‘dog’) can be frequently activated within
Probability of choosing

1.5 0.9 one person’s model and interconnected to other concepts (‘love’,
choice

‘daughter’, ‘fun’, ‘food’, ‘home’) but not in another person’s model.

knowledge

1 0.8 Gain
Loss
0.7
0.5 We suggest that people will be more likely to want information
0
0.6 relating to concepts that are frequently activated and highly inter-
0.5 connected to other concepts in their mental models (for example,
0 0.2 0.4 0.6 0.8 1
Participant
‘Hedonic
utility’
‘Cognitive
utility’ Indifference Probability of gain/loss
‘self ’, ’human’). Again, individual differences exist. For example, a
(β2) (β3) dog owner would be more interested in learning whether dogs are
related to wolves than a non-dog owner. This information will have
greater positive cognitive value for the former than the latter, even
Fig. 2 | Integrative framework of information-seeking motives.
if information has no clear practical use for either. This is because
a, Information-seeking or -avoidance is hypothesized to be driven by
‘dog’ is a central concept in the former’s mental model. If informa-
instrumental utility, hedonic utility and cognitive utility. These values reflect
tion is very weakly related to people’s mental model (for example,
the predicted impact of information on action (will the knowledge help,
knowing the middle name of someone passing you on the street), it
hinder or have no influence on my ability to make decisions to increase
will have zero cognitive value.
reward and avoid harm?), affect (will the information induce positive or
It has been suggested that people strive to minimize the differ-
negative feelings, or will it have no influence on my affect?) and cognition
ence between their mental models and external reality38. This can
(will information improve my ability to comprehend and anticipate
be achieved in two ways. The first is to improve existing models by
reality?), respectively. These estimates are integrated into a computation
seeking out information that will generate new connections among
of the value of information, with different weights (β1–3) assigned to each
concepts or that will strengthen connections that are suspected but
of the three factors. The integrated value can be positive (increasing
of which people are uncertain8,39. This approach will improve the
information-seeking), negative (increasing information-avoidance) or zero
fit of the model to reality by refining it, which will in turn increase
(leading to indifference). Biases in predicting the content of information
people’s ability to comprehend and anticipate reality. The second
and its influence on action, affect and cognition may result in suboptimal
approach, less intuitive, is to avoid information that people suspect
information-seeking behaviour. Individual differences in information-
will weaken strong existing interconnected ties within the model
seeking are hypothesized to be related to variations in these estimations as
(for example, disconfirming information40–42). In this approach,
well as to differences in the weights (β1–3) assigned to the three drivers.
people maintain a fit between internal representation and external
b, Data from our previous study5 showing that information-seeking choices
reality not by actively changing the model but by actively changing
are explained by hedonic utility (in this case, how likely is the information
the reality of which they are aware. Such avoidance can improve the
to be good versus bad) and cognitive utility (in this case, the amount of
sense of comprehension at present, because disconfirming informa-
uncertainty which is reduced by information). In this study, instrumental
tion can cause confusion and a need to rebuild large parts of the
utility was set to zero. Each dot represents a participant. For illustration
model. Thus, disconfirming information may be assigned nega-
purposes, dots are scattered along the x-axis such that each dot is visible.
tive cognitive value, despite the fact that avoiding such information
c, Figure from the same study5 highlighting the impact of hedonic utility
could create less accurate mental models.
in a non-instrumental information-seeking task. Specifically, the more
With the above we intend to outline a broad theoretical frame-
likely participants were to win a lottery, the more they wanted to know the
work that can help guide future work in the behavioural sciences
outcome; the more likely they were to lose, the less they wanted to know
and neuroscience. The framework builds on many past studies,
the outcome. Image adapted with permission from ref. 5, National Academy
each highlighting different aspects of information-seeking motives.
of Sciences. Error bars indicate s.e.m.
These include studies emphasizing the utility of beliefs26,27, the need
to “fill in gaps of knowledge”29 and the need for confirmation and
to cases in which people select to undergo painful proce- confidence40–43. Our framework is also related to economic models
dures (such as dental work) now rather than later to minimize outlining how motives may alter information-seeking. For example,
dread35, people may choose to obtain bad news now rather than Golman et al.28 present a specific model that might be fit into the
remain in the dark. somewhat more general framework being presented here. They
The opposite may also be true. People may prefer not to know highlight the importance of instrumental utility and, in addition,
in advance the outcome of a sporting event, the gender of their suggest that information-seeking is guided by motivated attention
unborn baby, or the location of a vacation being planned by and curiosity. These latter components of their model are nar-
their spouse, because they want to preserve the positive feeling rower than hedonic value and cognitive value in the sense that they
of surprise. Lack of knowledge can produce anxiety, but it can describe specific ways that motives can alter information seeking, for
also produce delight. example, by guiding attention or arousing a need to fill information

Nature Human Behaviour | www.nature.com/nathumbehav

Perspective NATure HumAn BehAvIOur

gaps, which is experienced as a feeling of curiosity. We view curios- predicting the likely content of information46,47. Second, there are
ity as an expression of information-seeking motive(s) rather than potential gaps between the utility expected at the time of decid-
the primary motive itself. It is analogous to hunger, which is the ing whether to seek information and the utility experienced dur-
feeling of needing food but not the reason food is needed (the latter ing consumption of information. Below are a few examples of such
may be, for example, low blood sugar). We speculate that the feeling biases and their potential influence on information-seeking.
of curiosity does not simply map on to one of our three motives, a 1. Illusion of control (biasing predictions of instrumental util-
speculation that requires future testing (see also Glossary for a dis- ity). People tend to overestimate the ability of their actions to
tinction between curiosity and information-seeking). With regard influence outcomes48. This bias will likely lead to an overestima-
to motivated attention, we suspect that it is one of several ways affect tion of the instrumental value of information. This is because
may modulate information-seeking. Our framework is also consis- individuals may believe that information about outcomes will
tent with the past suggestion that, when deciding whether to reveal enable them to take action to alter those outcomes, when in
information to others, agents consider the impact of information on fact outcomes are beyond their control. As a result, people may
the emotions of others and the impact of information on material overpay (in money, time or some other currency) for useless
outcomes for others44. information.
Our suggested framework could be used to explain informa- 2. Unrealistic optimism (biasing predictions of hedonic util-
tion-seeking choices in different contexts and domains, including ity). People tend to overestimate the probability of positive
health and finance, and to develop detailed models that predict such outcomes and underestimate the probability of negative out-
choices. We suggest that the three utilities could be measured and/ comes47,49. When deciding whether to seek information, people
or manipulated and that these measures could be used to explain may overestimate the probability of good news and underesti-
and/or predict choice (note, however, that future work is required mate the probability of bad news. This will lead to an overesti-
to identify the specific rules regarding what type of content under mation of positive hedonic utility and an underestimation of
which circumstances generate which type of utility). For example, negative hedonic utility, increasing information-seeking. The
a recent study found that participants’ decisions to know or remain result could be overexposure to unexpected negative informa-
ignorant about the outcome of a lottery game were explained by tion and disappointment. It has been suggested that optimis-
hedonic and cognitive utility (in this task, instrumental utility was tically biased agents may be better off with delayed revelation
set to zero by design)5 (Fig. 2b). In particular, the more likely par- of information, as this will allow them to maintain optimistic
ticipants were to receive good news (about monetary gain) and the beliefs, which have positive intrinsic utility, for longer27.
less likely they were to receive bad news (about monetary loss), the 3. Impact bias (biasing predictions of hedonic utility). People
more they wanted to know their outcomes. It was also found that tend to overestimate the duration and intensity of future affec-
the more uncertain the outcome was, the more people wanted to tive states50. This bias could lead to an overestimation of positive
know the outcome5 (Fig. 2c). This information facilitated the abil- hedonic utility (increasing information seeking for good news)
ity to anticipate future events, but not to alter them. Similar results and negative hedonic utility (increasing information avoidance
were observed in a separate study in which participants could use for bad news). The latter can cause individuals to avoid nega-
information to alter outcomes. In that study, instrumental utility tive information that could have otherwise had significant in-
was an additional orthogonal driver of information-seeking45. strumental utility (including information about their health or
The aforementioned studies5,45 were designed such that instru- finances and feedback about their performance), without the
mental, hedonic and cognitive utilities fluctuated independently expected benefit to their emotional well-being.
from each other. This made it possible to measure their separate 4. The illusion of knowledge (biasing predictions of cognitive
effects on information-seeking. These motives can often be inde- utility). People tend to underestimate how much new informa-
pendent ‘in the wild’. For example, a person may need to decide tion will influence their understanding51. Thus, people may un-
whether to undergo screening for a disease that she is quite certain derestimate the cognitive value of information.
or uncertain she carries (affecting cognitive utility) that is either 5. Overconfidence (biasing predictions of cognitive utility). Peo-
curable or not (affecting instrumental utility) such that the two ple tend to be overconfident in their knowledge52. Thus people
dimensions are orthogonal. But in some cases, the dimensions co- may underestimate how much new information may reduce
vary and can be difficult to disentangle. For example, all else being uncertainty, biasing their estimate of its cognitive value.
equal, the less curable a disease (instrumental utility) the worse it
would be to learn that one carries it (hedonic utility). The above is not meant to be an exhaustive list, but rather a
Moreover, in the aforementioned studies5,45 the expected influ- demonstration of how prediction biases may lead to suboptimal
ence of information on action, affect and cognition was quantified information-seeking by influencing estimations of each of the three
using objective cues that were available to participants. For example, factors. People’s motivation to seek information, and their willing-
the participants knew the exact probability of a gain or loss5,45 or ness to pay for information, will capture the expected welfare effects
how useful a piece of information was for attaining rewards45. Often, from information, including instrumental, hedonic and cognitive
however, these quantities need to be estimated by the decision- effects, but will often fail to capture the experienced welfare effects,
maker. This aspect of the decision process, as we explain below, can thus leading to suboptimal information-seeking choices. The dis-
result in information-seeking biases (Fig. 2a). junction between the expected utility of information and the actual
experienced utility of information will vary across individuals as a
Prediction biases influencing information-seeking function of individuals’ susceptibility to different biases.
According to the above framework, the first stage in deciding whether
to seek information involves solving a prediction problem. People Individual differences in information-seeking: a window
must predict the likely content of information and its influence on into mental health?
action, affect and cognition. Biases known to influence predictions Individual differences in information-seeking (Fig.  2a,b) may
will influence estimates of instrumental, hedonic and cognitive util- reflect differences in estimated hedonic, instrumental and cognitive
ity, producing both excessive and insufficient information-seeking. utilities (which could be partially due to susceptibility to the biases
Prediction problems can be extremely hard to solve and are described above), as well as differences in the weighting of those
notoriously vulnerable to biases. First, people have problems in utilities5. We hypothesize that such differences may be related to psy-
estimating probability and thus may commit systematic errors in chiatric symptoms and conditions. The reason for this hypothesis

Nature Human Behaviour | www.nature.com/nathumbehav

NATure HumAn BehAvIOur Perspective
Glossary behaviour, with potential consequences, positive or negative, to
patients’ well-being.
Information:Data previously not known with complete certainty. An established relationship between information-seeing patterns
Information-seeking:The active pursuit of knowledge, for and mental health would have important practical applications.
example by asking questions, reading and conducting online Because people constantly engage in information-seeking online, a
searches. stream of data could theoretically be used to facilitate clinical diag-
nosis, monitoring and treatment selection. Indeed, there is growing
Curiosity:The feeling of wanting to know. While curiosity interest in using online behaviour for diagnosis and stratification
is related to information-seeking the two concepts are of psychiatric and neurological conditions, as data can be obtained
distinct. In particular, it is possible to be curious but to avoid frequently and non-intrusively. For example, recent work has begun
information or to seek information despite a lack of curiosity. to link keystroke patterns to motor disorders67 and affective states68.
For example, a person may be curious about whether they Such endeavours could potentially be boosted for diagnosis if rich
have a predisposition for cancer, but decide not to pursuit such behaviourally relevant features could be extracted, such as mea-
information to avoid experiencing negative feelings. A person sured markers of information-seeking. Efforts in this vein should
may also seek financial information to make better financial protect personal privacy.
decisions, despite not being curious about such information.
Policy implications
Instrumental utility of information:A measure quantifying the Agencies in nations around the world are required to quantify the
amount by which information will enable achieving an end goal. costs and benefits of information disclosure mandates69. In the US,
the requirement of quantitative cost–benefit analysis applies to
Hedonic utility of information:A measure quantifying the a wide range of regulations, involving (for example) disclosure of
amount of pleasure (or other positive feeling) information fuel economy, graphic warnings for cigarettes, calorie labels, geneti-
would induce, minus the amount of pain (or other negative cally modified foods, minerals used to finance mass atrocities, and
feeling) it would induce, from which we subtract the amount country-of-origin labels70. Our framework could be useful in help-
of pleasure ignorance would induce plus the amount of pain ing officials do better in identifying the costs and benefits of disclo-
ignorance would induce. sure. It could also trigger ideas for how to make information more
attractive, thus increasing the likelihood that people will actually
Cognitive utility of information:A measure quantifying the read leaflets and labels and so benefit from them.
degree to which information would strengthen internal mental Currently, it is common for agencies to try to project the instrumen-
models. tal utility of disclosing (or not disclosing) information71. For example,
an agency might attempt to estimate the health benefits of disclosing
Mental models:Representation of concepts and the the caloric contents of food in restaurants72. The impact of informa-
relationships among them, which are used to comprehend and tion on non-instrumental utility, however, is often overlooked. This
anticipate reality. presents two problems. First, considering only instrumental utility
does not give a complete picture of the actual costs and benefits. As we
Affect:A physiological reaction that varies in valence (positive have seen, people might experience significant hedonic losses as well
or negative) and arousal. as gains as a result of receiving information or remaining ignorant73.
Second, without considering the effect of non-instrumental
utilities, policy-makers will systematically mispredict the impact
is that many conditions, including depression and anxiety, are char- of information on people’s action, because they will misjudge the
acterized by symptoms that could manifest as abnormalities in likelihood that people will ‘consume’ (i.e., read or hear about) the
information-seeking53–58. Anxiety, for example, is characterized by information in the first place. For example, in estimating the health
intolerance of uncertainty59, which may lead to increased weighting benefits of disclosing caloric content, policy-makers tend to assume
of cognitive utility. Depression is characterized by reduced sensitivity that consumers generally will read those labels72. Yet if the informa-
to valence60–62 and a reduction in the belief that one has control over tion is likely to induce negative affect, the consumer may decide not
outcomes63, which can lead to a reduction in the impact of hedonic to do so. Thus, the benefits of such labels will be overestimated. In
and instrumental utility on information-seeking, respectively. fact, it has been shown that some people select not to learn the caloric
It is likely that differences in information-seeking are not merely content of a tempting dessert, as they are aware that this knowledge
correlated with psychopathology symptoms, but rather that a bidi- will reduce the likelihood they would select that dessert74.
rectional causal relationship exists. That is, psychiatric disorders Considering the expected hedonic, cognitive and instrumental
may influence information-seeking behaviour and, at the same time, utility of information can also reveal how information should be
certain patterns of information-seeking may either protect individu- framed in order to maximize use. That is, information should be
als from psychiatric symptoms or make them more vulnerable. For framed in a way that highlights its ability to guide action and/or
example, if individuals fail to consider the expected valence of knowl- improve affect and/or cognition. For example, to encourage people
edge, they may be overexposed to information that induces negative to attend to pre-flight safety instructions, airlines have in recent
affect, thus increasing the likelihood of depressive symptoms. years created videos that induce positive affect using humour, music
It is interesting to consider whether certain psychotropic drugs and alluring images of vacation destinations. By doing so, they
reduce symptoms by altering such information-seeking patterns seek to overcome people’s tendency to avoid such information to
in patients. Dopamine function, in particular, is thought to play a suppress flying anxieties. This approach has been so successful that
role in information-seeking, perhaps in modulating the impact of millions now watch these videos even at home.
affect (Fig. 1). If that is true, conditions characterized by abnormal Finally, regulators should consider individual differences in
dopaminergic function (such as Parkinson’s disease, schizophrenia information-seeking patterns and the influence of information on
and depression)64–66 may be characterized by atypical information- welfare due to mental health issues (see above) or demographic char-
seeking. Indeed, there have been some reports of atypical infor- acteristics75. For example, older individuals put greater emphasis
mation-seeking in such patients54,55. Moreover, prescription drugs on positive feelings76, which could translate to greater weighting of
targeting dopamine function may alter patients’ information-seeking affect in seeking information. It is possible that what is a net benefit

Nature Human Behaviour | www.nature.com/nathumbehav

Perspective NATure HumAn BehAvIOur

of disclosure for some individuals and groups will be a net cost for 36. Craik, K.J.W. The Nature of Explanation. (CUP Archive, 1952).
others. Information may be targeted and/or framed according to 37. Johnson-Laird, P.N. Mental Models: Towards a Cognitive Science of Language,
Inference, and Consciousness. (Harvard University Press, 1983).
population characteristics such as age, education and mental health. 38. Friston, K. Nat. Rev. Neurosci. 11, 127–138 (2010).
Though much further research is needed in this domain, targeted 39. Kreps, D. M. & Porteus, E. L. Econometrica 46, 185–200 (1978).
disclosure could have far higher net welfare benefits than untar- 40. Nickerson, R. S. Rev. Gen. Psychol. 2, 175–220 (1998).
geted disclosure77. 41. Klayman, J. & Ha, Y.-W. Psychol. Rev. 94, 211–228 (1987).
42. Kappes, A., Harvey, A., Lohrenz, T., Montague, R. & Sharot, T. Nat. Neurosci.
https://doi.org/10.1038/s41593-019-0549-2 (2019).
Conclusion 43. Eliaz, K. & Schotter, A. Games Econ. Behav. 70, 304–324 (2010).
It is increasingly possible for people to obtain information that 44. Kőszegi, B. Q. J. Econ. 121, 121–155 (2006).
bears on their future prospects, in terms of health, finance and even 45. Kobayashi, K. & Hsu, M. Proc. Natl Acad. Sci. USA 116, 13061–13066 (2019).
romance. It is also increasingly possible for them to obtain informa- 46. Kahneman, D. & Tversky, A. Psychol. Rev. 80, 237–251 (1973).
47. Weinstein, N. D. J. Pers. Soc. Psychol. 39, 806–820 (1980).
tion about the past, the present and the future, whether or not that 48. Langer, E. J. J. Pers. Soc. Psychol. 32, 311–328 (1975).
information bears on their personal lives. In principle, people’s deci- 49. Sharot, T. Curr. Biol. 21, R941–R945 (2011).
sions about whether to seek or avoid information should depend on 50. Gilbert, D.T., Driver-Linn, E. & Wilson, T.D. in The Wisdom in Feeling:
some integration of instrumental value, hedonic value and cognitive Psychological Processes in Emotional Intelligence (eds Barrett, L. F. & Salovey,
value. But various biases can lead to both insufficient and excessive P.) 114–143 (Guilford Press, 2002).
51. Sloman, S. & Fernbach, P. The Knowledge Illusion: Why We Never Think
information-seeking. Individual differences in information-seeking Alone. (Riverhead Books, 2017).
may reflect different levels of susceptibility to those biases, as well 52. Pallier, G. et al. J. Gen. Psychol. 129, 257–299 (2002).
as varying emphasis on instrumental, hedonic and cognitive utility. 53. Aderka, I. M., Haker, A., Marom, S., Hermesh, H. & Gilboa-Schechtman, E.
Such differences may also be diagnostic of mental health. J. Abnorm. Psychol. 122, 7–12 (2013).
Whether positive or negative, the value of information bears 54. Hildebrand-Saints, L. & Weary, G. Pers. Soc. Psychol. Bull. 15, 150–160 (1989).
55. Camp, C. J. Educ. Gerontol. 12, 375–384 (1986).
directly on significant decisions of government agencies, which 56. Locander, W. B. & Hermann, P. W. The Effect of Self-confidence and Anxiety
are often charged with calculating the welfare effects of mandatory on Information Seeking in Consumer Risk Reduction. (SAGE Publications Sage
disclosure and which have long struggled with that task. Our hope CA, Los Angeles, CA, 1979).
is that the integrative framework of information-seeking motives 57. Gray, K. M. & Tonge, B. J. J. Paediatr. Child Health 37, 221–226 (2001).
offered here will facilitate these goals and promote future research 58. Jensen, P.S. et al. J. Am. Acad. Child Adolesc. Psychiatry 36, 1672–1679,
discussion 1679–1681 (1997).
in this important domain. 59. Holaway, R. M., Heimberg, R. G. & Coles, M. E. J. Anxiety Disord. 20,
158–174 (2006).
Received: 4 June 2019; Accepted: 19 November 2019; 60. Luking, K. R., Pagliaccio, D., Luby, J. L. & Barch, D. M. Trends Cogn. Sci. 20,
Published: xx xx xxxx 456–468 (2016).
61. McCabe, C., Cowen, P. J. & Harmer, C. J. Psychopharmacol. (Berl.) 205,
References 667–677 (2009).
1. Kidd, C. & Hayden, B. Y. Neuron 88, 449–460 (2015). 62. Pizzagalli, D. A., Iosifescu, D., Hallett, L. A., Ratner, K. G. & Fava, M.
2. Loewenstein, G. Psychol. Bull. 116, 75–98 (1994). J. Psychiatr. Res. 43, 76–87 (2008).
3. Stigler, G. J. J. Polit. Econ. 69, 213–225 (1961). 63. Alloy, L. B. & Abramson, L. Y. J. Exp. Psychol. 108, 441–485 (1979).
4. Grant, S., Kajii, A. & Polak, B. J. Econ. Theory 83, 233–259 (1998). 64. Nestler, E. J. & Carlezon, W. A. Jr. Biol. Psychiatry 59, 1151–1159 (2006).
5. Charpentier, C. J., Bromberg-Martin, E. S. & Sharot, T. Proc. Natl Acad. Sci. 65. Davie, C. A. Br. Med. Bull. 86, 109–127 (2008).
USA 115, E7255–E7264 (2018). 66. Grace, A. A. Nat. Rev. Neurosci. 17, 524–532 (2016).
6. Bromberg-Martin, E. S. & Hikosaka, O. Neuron 63, 119–126 (2009). 67. Giancardo, L. et al. Sci. Rep. 6, 34468 (2016).
7. Eliaz, K. & Schotter, A. Am. Econ. Rev. 97, 166–169 (2007). 68. Shukla, P. & Solanki, R. Web based keystroke dynamics application for
8. Berlyne, D. E. Psychol. Rev. 64, 329–339 (1957). identifying emotional state. Inst. Eng. Technol. DAVV, Indore, India (2013).
9. Kreps, D. M. & Porteus, E. L. Econom. J. 46, 185–200 (1978). 69. Sunstein, C.R. The Cost-benefit Revolution. (MIT Press, 2018).
10. Ligneul, R., Mermillod, M. & Morisseau, T. NeuroImage 181, 490–500 (2018). 70. Sunstein, C. R. J. Risk Uncertain. 58, 121–142 (2019).
11. Brydevall, M., Bennett, D., Murawski, C. & Bode, S. Sci. Rep. 8, 6134 (2018). 71. Food & Drug Administration. Fed. Regist. 76, 36628 (2011).
12. Kang, M. J. et al. Psychol. Sci. 20, 963–973 (2009). 72. Food & Drug Administration. Fed. Regist. 79, 71155 (2014).
13. Smith, D. V., Rigney, A. E. & Delgado, M. R. Sci. Rep. 6, 20093 (2016). 73. Thunström, L. Judgm. Decis. Mak. 14, 11–25 (2019).
14. Tricomi, E. & Fiez, J. A. Cogn. Affect. Behav. Neurosci. 12, 361–372 (2012). 74. Woolley, K. & Risen, J. L. J. Pers. Soc. Psychol. 114, 230–245 (2018).
15. Jessup, R. K. & O’Doherty, J. P. Eur. J. Neurosci. 39, 2014–2026 (2014). 75. Sakaki, M., Yagi, A. & Murayama, K. Neurosci. Biobehav. Rev. 88,
16. Gruber, M. J., Gelman, B. D. & Ranganath, C. Neuron 84, 486–496 (2014). 106–116 (2018).
17. van Lieshout, L. L. F., Vandenbroucke, A. R. E., Müller, N. C. J., Cools, R. & 76. Carstensen, L. L. Psychol. Aging 7, 331–338 (1992).
de Lange, F. P. J. Neurosci. 38, 2579–2588 (2018). 77. Allcott, H. & Kessler, J. B. Am. Econ. J. Appl. Econ. 11, 236–276 (2019).
18. Bromberg-Martin, E. S. & Hikosaka, O. Nat. Neurosci. 14, 1209–1216 (2011).
19. Blanchard, T. C., Hayden, B. Y. & Bromberg-Martin, E. S. Neuron 85, Acknowledgements
602–614 (2015). We thank L. Rozenkrantz, J. Marks, F. Gesiarz, C. Kelly, I. De Taranto, C. Charpentier
20. Schultz, W., Dayan, P. & Montague, P. R. Science 275, 1593–1599 (1997). and Y. Wang for comments on previous versions of this manuscript. Y. Wang also assisted
21. Thornton, R. L. Am. Econ. Rev. 98, 1829–1863 (2008). with generating brain images. T.S. is supported by a Wellcome Trust Senior Research
22. Persoskie, A., Ferrer, R. A. & Klein, W. M. J. Behav. Med. 37, 977–987 (2014). Fellowship (214268/Z/18/Z).
23. Dwyer, L. A., Shepperd, J. A. & Stock, M. L. Ann. Behav. Med. 49, 685–695
(2015).
24. Caplin, A. & Leahy, J. Q. J. Econ. 116, 55–79 (2001). Competing interests
25. Koszegi, B. Econ. Theory 44, 415–444 (2010). The authors declare no competing interests.
26. Golman, R., Hagman, D. & Loewenstein, G. J. Econ. Lit. 55, 96–135 (2017).
27. Brunnermeier, M. K. & Parker, J. A. Am. Econ. Rev. 95, 1092–1118 (2005).
28. Golman, R., Loewenstein, G., Molnar, A. & Saccardo, S. SSRN https://doi. Additional information
org/10.2139/ssrn.2149362 (2019). Correspondence should be addressed to T.S.
29. Golman, R. & Loewenstein, G. Decision 5, 143–164 (2018). Peer review information Primary Handling Editors: Mary Elizabeth Sutherland;
30. Hertwig, R. & Engel, C. Perspect. Psychol. Sci. 11, 359–372 (2016). Stavroula Kousta
31. Bénabou, R. & Tirole, J. Q. J. Econ. 117, 871–915 (2002). Reprints and permissions information is available at www.nature.com/reprints.
32. Karlsson, N., Loewenstein, G. & Seppi, D. J. Risk Uncertain. 38, 95–115 (2009).
33. Lerman, C. et al. J. Clin. Oncol. 16, 1650–1654 (1998). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
34. Gigerenzer, G. & Garcia-Retamero, R. Psychol. Rev. 124, 179–196 (2017). published maps and institutional affiliations.
35. Loewenstein, G. Econ. J. (Lond.) 97, 666–684 (1987). © Springer Nature Limited 2020

Nature Human Behaviour | www.nature.com/nathumbehav

View publication stats