Centaur: a foundation model of human cognition

Marcel Binz1*, Elif Akata1 , Matthias Bethge2 ,

Franziska Brändle3,5 , Fred Callaway4 , Julian Coda-Forno1 ,
Peter Dayan2,5 , Can Demircan1 , Maria K. Eckstein6 ,
Noémi Éltető5 , Thomas L. Griffiths7 , Susanne Haridi1,13 ,
Akshay K. Jagadish1,2,5 , Li Ji-An8 , Alexander Kipnis1 ,
Sreejan Kumar7 , Tobias Ludwig2,5 , Marvin Mathony1 ,
Marcelo Mattar4 , Alireza Modirshanechi1 , Surabhi S. Nath2,5,13 ,
arXiv:2410.20268v3 [cs.LG] 28 Apr 2025

Joshua C. Peterson9 , Milena Rmus1 , Evan M. Russek7 ,

Tankred Saanum5 , Johannes A. Schubert5 ,
Luca M. Schulze Buschoff1 , Nishad Singhi14 , Xin Sui2,5 ,
Mirko Thalmann1 , Fabian Theis1 , Vuong Truong5 ,
Vishaal Udandarao2,15 , Konstantinos Voudouris1 ,
Robert Wilson10 , Kristin Witte1 , Shuchen Wu1 ,
Dirk U. Wulff11,12 , Huadong Xiong10 , Eric Schulz1
1 Helmholtz Munich.
2 University of Tuebingen.
3 University of Oxford.
4 New York University.
5 Max Planck Institute for Biological Cybernetics.
6 Google DeepMind.
7 Princeton University.
8 University of California San Diego.
9 Boston University.
10 Georgia Institute of Technology.
11 University of Basel.
12 Max Planck Institute for Human Development.
13 Max Planck School of Cognition.
14 TU Darmstadt.
15 University of Cambridge.

*Corresponding author(s). E-mail(s): [email protected];

1
 Abstract
Establishing a unified theory of cognition has been a major goal of psychology
[1, 2]. While there have been previous attempts to instantiate such theories by
building computational models [1, 2], we currently do not have one model that
captures the human mind in its entirety. A first step in this direction is to create
a model that can predict human behavior in a wide range of settings. Here we
introduce Centaur, a computational model that can predict and simulate human
behavior in any experiment expressible in natural language. We derived Centaur
by finetuning a state-of-the-art language model on a novel, large-scale data set
called Psych-101. Psych-101 reaches an unprecedented scale, covering trial-by-
trial data from over 60,000 participants performing over 10,000,000 choices in
160 experiments. Centaur not only captures the behavior of held-out participants
better than existing cognitive models, but also generalizes to new cover stories,
structural task modifications, and entirely new domains. Furthermore, we find
that the model’s internal representations become more aligned with human neu-
ral activity after finetuning. Taken together, our results demonstrate that it is
possible to discover computational models that capture human behavior across a
wide range of domains. We believe that such models provide tremendous poten-
tial for guiding the development of cognitive theories and present a case study to
demonstrate this.

Keywords: cognitive science, cognitive modeling, unified theory of cognition, large

language models

Introduction
The human mind is remarkably general [3–5]. Not only do we routinely make mundane
decisions, like choosing a breakfast cereal or selecting an outfit, but we also tackle
complex challenges, such as figuring out how to cure cancer or explore outer space.
We learn new skills from only a few demonstrations [6], reason causally [7], and fuel
our actions through curiosity [8]. Whether we are climbing mountains, playing video
games, or creating captivating art, our versatility defines what it means to be human.
In contrast to this, most contemporary computational models – whether in machine
learning or the cognitive sciences – are domain-specific. They are designed to excel
at one particular problem and that problem alone. Take, for instance, AlphaGo – a
computer system created by Google DeepMind to master the game of Go [9]. Even
though the system can play this particular game at an impressive level, it can do
not much beyond that. A similar pattern can be observed in the cognitive sciences.
Prospect theory, one of the most influential accounts of human cognition, for instance,
offers valuable insights into how people make choices [10], but it tells us nothing about
how we learn, plan, or explore.
If we want to understand the human mind in its entirety, we must move from
domain-specific theories to an integrated one. The importance of such a unified
approach has already been recognized by the pioneers of our field. For example, in
1990, Newell stated that “unified theories of cognition are the only way to bring [our]

2
a
160 psychological experiments, 60,092 individual participants, 10,681,650 human choices, 253,597,411 text tokens
Multi-armed bandits Decision-making Memory
In this task, you have to repeatedly choose between You will choose from two monetary lotteries by You will view a stream of letters on the screen, one
two slot machines labeled B and C. When you select pressing N or U. Your choice will trigger a random letter at a time. You have to remember the last two
one of the machines, you will win or lose points. draw from the chosen lottery that will be added to letters you saw since the beginning of the block. If
Your goal is to choose the slot machines that will your bonus. the letter you see matches the letter two trials ago,
give you the most points. Lottery N offers 4.0 points with 80.0% or 0.0 points press E, otherwise press K.
You press <<C>> and get -8 points. with 20.0%. You see the letter V and press <<K>>.
You press <<B>> and get 0 points. Lottery U offers 3.0 points with 100.0%. You see the letter X and press <<K>>.
You press <<B>> and get 1 points. You press <<U>>. You see the letter V and press <<E>>.

Supervised learning Markov decision processes Miscellaneous

In each trial, you will see between one and three You will be taking one of the spaceships F or V to You will be presented with triplets of objects, which
tarot cards. Your task is to decide if the combina- one of the planets M or S. When you arrive at each will be assigned to the keys E, Z, and B. In each
tion of cards presented predicts rainy weather (by planet, you will ask one of the aliens for space trea- trial, please indicate which object you think is the
pressing P) or fine weather (by pressing L). sure. odd one out by pressing the corresponding key.
You are seeing the following: card 3, card 4. You You are presented with spaceships V and F. E: tablet, Z: fox, and B: vent. You press <<Z>>.
press <<L>>. You are wrong, the weather is rainy. You press <<V>>. You end up on planet M and E: ivy, Z: coop, and B: drink. You press <<B>>.
You are seeing the following: card 1, card 4. You see aliens G and W. You press <<G>>. E: kite, Z: flan, and B: jar. You press <<E>>.
press <<P>>. You are right, the weather is rainy. You find 1 pieces of space treasure. E: wand, Z: flag, and B: globe. You press <<Z>>.

b
Centaur: a foundation model of human cognition

In this task, you

embedding

have to repeatedly
...

OUTPUT
Token

choose between two Self- Feedforward Self- Feedforward

C
slot machines labeled attention network attention network
B and C. [. . .]
You press <<

Low-rank Low-rank Low-rank Low-rank

adapter adapter adapter adapter

Fig. 1 Psych-101 and Centaur overview. a, Psych-101 comprises of trial-by-trial data from 160
psychological experiments and 60,092 participants, making 10,681,650 choices in total. It contains
domains such as multi-armed bandits, decision-making, memory, supervised learning, Markov decision
processes, and others (shown examples are stylized and abbreviated for readability). b, Centaur is a
foundation of model human cognition that is obtained by adding low-rank adapters to a state-of-the-
art language model and finetuning it on Psych-101.

wonderful, increasing fund of knowledge under intellectual control” [2]. How can we
make meaningful progress toward such theories?
An important step towards a unified theory of cognition is to build a computa-
tional model that can predict and simulate human behavior in any domain [2, 11].
The present paper takes up this challenge and introduces Centaur – the first founda-
tion model of human cognition [12]. Centaur was designed in a data-driven manner
by finetuning a state-of-the-art large language model [13] on a large corpus of human
behavior. For this purpose, we curated a novel, large-scale data set called Psych-101,
covering trial-by-trial data from 160 psychological experiments. We transcribed each of
these experiments into natural language, which provides a common format for express-
ing vastly different experimental paradigms [14, 15]. The resulting data set reaches an
unprecedented scale, containing over 10,000,000 human choices and including many
canonical studies from domains such as multi-armed bandits, decision-making, mem-
ory, supervised learning, Markov decision processes, and others (see Figure 1a for an
overview and examples).
We subject Centaur to a series of rigorous tests and demonstrate that it captures
human behavior at several levels of generalization. First, we show that Centaur predicts
behavior of held-out participants (i.e., participants that are not part of the training
data) better than existing cognitive models in almost every single experiment. We then

3
demonstrate that its ability to capture human behavior also generalizes to held-out
experiments. In this context, we find that Centaur accurately predicts human behavior
under modified cover stories, problem structures, and even in entirely novel domains.
Finally, we show that Centaur’s internal representations become more human-aligned,
even though it was never explicitly trained to capture human neural activity.
Taken together, our results demonstrate that it is possible to discover computa-
tional models that capture human behavior across a wide range of domains. We believe
that such a predictive model offers many direct opportunities to obtain a better under-
standing of the human mind [16, 17] and present a case study that demonstrates this
potential.

Results
Model overview
We built Centaur on top of the open-source language model Llama 3.1 70B – a state-
of-the-art model pre-trained by Meta AI [13] (hereafter, we refer to this model simply
as Llama). Having a large language model as the backbone allowed us to rely on the
vast amounts of knowledge that is present in these models [18]. The training pro-
cess involved finetuning on Psych-101 using a parameter-efficient finetuning technique
known as quantized low-rank adaptation (QLoRA) [19]. QLoRA relies on a frozen four-
bit quantized language model as a base model. While the parameters of the base model
are left unchanged, it adds so-called low-rank adapters, which contain only a few addi-
tional, trainable parameters (typically represented in a half-precision floating-point
format). In our case, we added low-rank adapters of rank r = 8 to all non-embedding
layers (i.e., all linear layers of the self-attention mechanisms and the feedforward net-
works) as illustrated in Figure 1b. With these settings, the newly added parameters
amount to 0.15% of the base model’s parameters. We then trained the model for one
epoch on the entire data set using a standard cross-entropy loss. We masked out the
loss for all tokens that do not correspond to human responses, thereby ensuring that
the model focuses on capturing human behavior and not on completing experimen-
tal instructions. The entire training process took approximately five days on an A100
80GB GPU. Further details on the finetuning procedure are provided in the Methods
section.

Centaur predicts human behavior better than domain-specific

cognitive models
We evaluated Centaur on different types of held-out data to demonstrate that it
robustly captures human behavior. In our first analysis, we tested whether it can pre-
dict behavior of participants that were not part of the training data. For this, we
split each transcribed experiment into two parts, and used 90% of participants for
training and retained 10% for testing. We measured goodness-of-fit to human choices
using negative log-likelihoods averaged across responses. Figure 2a presents the result
of this analysis, comparing Centaur against the base model without finetuning and

4
 a b Centaur Humans
Overall 2

Information bonus
Spatially correlated multi-armed bandit 1
Multiple-cue judgment
Gardening task 0
Multi-task reinforcement learning
Structured bandit 1
Digit span
CPC18 45 50 55 60
Conditional associative learning Reward
Iowa gambling task
choices13k c
Intertemporal choice
Drifting four-armed bandit 1.5

Model-basedness
N-back
Decisions from experience 1.0
Changing bandit 0.5
Multi-attribute decision-making
Recent probes 0
Two-step task -0.5
Two-armed bandit
Zoopermarket 0.2 0.4 0.6 0.8
Horizon task
Decisions from description Reward
Columbia card task

Accuracy (artificial strategies)

Go/no-go d
Weather prediction task 1.0
Shepard categorization
Medin categorization
Balloon analog risk task 0.5
Serial reaction time task Centaur
THINGS odd-one-out Llama
Probabilistic instrumental learning Cognitive model
Grammar judgement 0
0.6 0.4 0.2 0.0 0.2 0.4 0.6 0.8 1.0 0 0.5 1.0
log-likelihood Accuracy (human strategies)

Fig. 2 Goodness-of-fit on Psych-101. a, Difference in log-likelihood of Centaur and Llama relative to

a domain-specific cognitive model for each experiment. A value of zero corresponds to the goodness-of-
fit of the domain-specific cognitive model while a value above zero indicates improved goodness-of-fit
to human responses. Error bars correspond to the standard error of the mean, taken over responses.
Centaur outperforms both Llama and a collection of domain-specific cognitive models in almost every
experiment. Note that we only included experiments for which we have implemented a domain-specific
cognitive model in this graphic and merged different studies using the same paradigm. A full table for
all experiments can be found in the Supplementary Information. b, Model simulations on the horizon
task. The plot visualizes probability densities over reward and an information bonus parameter for
both people and simulated runs of Centaur. c, Model simulations on the two-step task. The plot
visualizes probability densities over reward and a parameter indicating how model-based learning was
for both people and simulated runs of Centaur. d, Model simulations on a social prediction game.
The plot visualizes probability densities over accuracies of predicting human strategies and strategies
of an artificial agent with matched statistics for both people and simulated runs of Centaur.

collection of domain-specific models that represent the state-of-the-art in the cogni-

tive science literature. While there was substantial variance in predictability between
the experiments, finetuning always improved goodness-of-fit. The average difference
in log-likelihoods across experiments after finetuning was 0.14 (Centaur negative
log-likelihood: 0.44; Llama negative log-likelihood: 0.58; t = −144.22, p ≤ 0.0001).
Furthermore, we compared Centaur against the previously mentioned collection
of domain-specific cognitive models. These models include, amongst others, the gen-
eralized context model [20], a prospect theory model [21], and various reinforcement
learning models [22, 23]. Further technical details can be found in the Supplementary
Information. We observed that Centaur outperforms domain-specific cognitive models

5
in all but one experiment. The average difference in predicting human behavior to the
domain-specific cognitive models was 0.13 (cognitive models negative log-likelihood:
0.56; t = −178.7, p ≤ 0.0001).
The previous analyses have focused on predicting human responses conditioned
on previously executed behavior. We may ask whether Centaur can also generate
human-like behavior when simulated in an open-loop fashion (i.e., when feeding its
own responses back into the model). This setting arguably provides a much stronger
test for the model’s capabilities and is sometimes also referred to as model falsifica-
tion [24]. To check whether Centaur survives this test, we ran open-loop simulations
in three different experimental paradigms and inspected the distributions of statis-
tics that resulted from these simulations. First, we simulated Centaur on the horizon
task paradigm, a two-armed bandit task used to detect different types of exploration
strategies [23]. We found that Centaur achieved a performance comparable to human
participants, as well that it engaged in a similar level of uncertainty-guided, directed
exploration (see Figure 2b), a pattern that is absent in many contemporary language
models [14]. We also observed that Centaur’s behavior is well-aligned with the human
population, demonstrating that it does not merely model the behavior of the average
participant but rather the distribution over trajectories produced by the entire pop-
ulation. For example, in the two-step task – a well-known paradigm to tease apart
model-free and model-based reinforcement learning [22] – Centaur, like human sub-
jects, produced trajectories in which learning is purely model-free, purely model-based,
and mixtures thereof (see Figure 2c). Lastly, we verified that Centaur fails at predict-
ing non-human behavior. For this, we considered a study that required participants to
predict either human responses or responses of an artificial agent with matched statis-
tics in four canonical economic games [25]. Mirroring the results of the original human
study, Centaur accurately predicted human responses (64% accuracy) but struggled
to predict artificial responses (35% accuracy; see Figure 2d). Taken together, these
results demonstrate that Centaur exhibits human-like characteristics across various
settings, confirming that it can generate meaningful open-loop behavior.

Probing increasingly complex generalization abilities

Thus far, we have shown that Centaur generalizes to previously unseen participants
performing experiments that were part of the training data. A true foundation model
of human cognition, however, must also capture behavior in any arbitrary experiment,
even if that experiment was not part of the training data. To probe whether Centaur
has this ability, we exposed it to a series of increasingly complex out-of-distribution
evaluations.
First, we investigated whether Centaur is robust in the face of changes to the cover
story. For this analysis, we relied on data collected by Feher da Silva and Hare [26],
who conducted a study using the aforementioned two-step task. In addition to the
canonical cover story (spaceships traveling to foreign planets in search of treasures),
their study introduced a novel cover story involving magical carpets. Importantly,
Psych-101 includes experiments using the canonical spaceship cover story [29, 30] but
no experiments with the magical carpet cover story. Yet, we still found that Centaur
captures human behavior in the magical carpet experiment of Feher da Silva and

6
 a Modified cover story b Modified problem structure c Entirely novel domain
51 51

39 39

40 40

26 26

1.2 2.2
Random guessing
Negative log-likelihood

0.7 Random guessing Random guessing

1.0 2.0

0.6 0.8
1.8
0.6
0.5 1.6
N/A
0.4
Centaur Llama Cognitive Centaur Llama Cognitive Centaur Llama Cognitive
model model model
Fig. 3 Evaluation in different held-out settings. a, Negative log-likelihoods for the two-step task
with a modified cover story [26]. b, Negative log-likelihoods for a three-armed bandit experiment [27].
c, Negative log-likelihoods for an experiment probing logical reasoning [28] with items based on the
Law School Admission Test (LSAT). Centaur outperforms both Llama and domain-specific cognitive
models when faced with modified cover stories, problem structures, and entirely novel domains. Error
bars correspond to the standard error of the mean, taken over responses.

Hare (see Figure 3a). Like in our previous analysis, we observed an improvement after
finetuning, as well as a favorable goodness-of-fit when compared to a domain-specific
cognitive model (Centaur negative log-likelihood: 0.51; Llama negative log-likelihood:
0.63; cognitive model negative log-likelihood: 0.61; all pairwise comparisons significant
with p ≤ 0.0001).
In a second out-of-distribution evaluation, we probed whether Centaur is robust
to modifications in task structure. To test this, we exposed it to a paradigm known
as Maggie’s farm [27]. Maggie’s farm extends the horizon task paradigm by adding
a third choice option. Psych-101 encompasses several two-armed bandit experiments
(including the horizon task) but not Maggie’s farm or any other three-armed bandit
experiments.1 Thus, this analysis provides a test of Centaur’s robustness to structural
task modifications. We found that Centaur captures human behavior on Maggie’s
farm as shown in Figure 3b. We again observed a benefit of finetuning, as well as a
favorable goodness-of-fit compared to a domain-specific cognitive model, which did
not generalize well to this setting (Centaur negative log-likelihood: 0.42; Llama neg-
ative log-likelihood: 0.62; cognitive model negative log-likelihood: 0.98; all pairwise
comparisons significant with p ≤ 0.0001).
Finally, we investigated whether Centaur can capture human behavior even in
entirely novel domains. In this context, we considered a study investigating logical rea-
soning [28]. While Psych-101 includes probabilistic and causal reasoning problems, we
purposefully excluded any studies involving logical reasoning. Like in the previous anal-
yses, there was again a positive effect of finetuning (Centaur negative log-likelihood:

1
It does, however, contain multi-armed bandit experiments with more than three choice options.

7
1.65; Llama negative log-likelihood: 1.92; t = −50.39, p ≤ 0.0001; see Figure 3c). Note
that we did not compare to any domain-specific cognitive model in this setting, as it
is unclear how to construct a model that would make any meaningful transfer from
training data that does not include any related problems.
We consolidated these results by analyzing Centaur on six additional out-of-
distribution experimental paradigms that were not part of the training data in any
shape or form (including moral decision-making [31], economic games [32], naturalistic
category and reward learning [33], behavioral propensities [34], and a deep sequential
decision task [35]). Centaur robustly captured human behavior in all of these settings,
while smaller and non-finetuned models did not do so consistently. We present the
corresponding results in the Supplementary Information.
In addition to analyzing human choice data, we also examined whether Centaur
can predict human response times. Hick’s law [36] suggests that individual response
times are a linear function of the response entropy. Therefore, we extracted nearly
4,000,000 response times for a subset of experiments in Psych-101 and fitted three lin-
ear mixed effects models each predicting response time based on the response entropy
derived from a different model. We found that the response entropy derived from Cen-
taur captured a larger proportion of the variance in response times (conditional R2 :
0.58) than those derived from Llama (conditional R2 : 0.4) and the cognitive mod-
els (conditional R2 : 0.38), thereby highlighting Centaur’s ability to predict measures
beyond pure choice data.
To demonstrate that the model does not degrade on problems it was pretrained
for, we furthermore verified it on a collection of benchmarks from the machine learn-
ing literature [37, 38]. We found that Centaur remains stable in performance-based
benchmarks, even improving over the base model in some of them [38]. Finally, in
benchmarks that measure human alignment, we observed a shift towards human-
like characteristics. Figure 4a depicts this improved alignment on a low-dimensional
embedding derived from ten behavioral metrics in CogBench, a benchmark to test the
cognitive abilities of large language models [37]. A more detailed description of these
additional benchmarking results can be found in the Supplementary Information.

Internal representations become more aligned to human neural

activity
Despite only being trained to match human behavior, we also wondered whether Cen-
taur’s internal representations become more aligned with human neural activity. To
check whether this is the case, we conducted two analyses in which we predicted human
neural activity using the model’s internal representations [39, 40]. We first conducted
a whole-brain analysis, in which we predicted fMRI measurements of people perform-
ing the two-step task [41]. For this, we relied on data collected in a previous study [41],
involving 94 participants each making 300 choices. Participants were tested on either
the magical carpet cover story (that we already used in one of our earlier generaliza-
tion analyses) or an abstract cover story. Neither of these two cover stories were part of
Centaur’s training data. We extracted recordings from models’ residual stream before
each choice and after feedback. We then aggregated human neural activity in each
region and regressed the aggregated activity on Centaur’s internal representations.

8
 a 6 b Centaur Llama Control c Centaur Llama Control
0.6 Noise ceiling
0.40
Embedding dimension 2 4 Claude 2 0.5

Pearson correlation

Pearson correlation
Claude 3 0.35
2 0.4
GPT-3.5 0.30
GPT-3 Tuckute et al. (2024)
0 Claude 1 0.3
GPT-4 0.25
Human 0.2
2 Centaur 0.20 Cognitive model
Llama 0.1
4
4 2 0 2 4 6 10 20 30 40 10 20 30 40
Embedding dimension 1 Layer Layer

Fig. 4 Human alignment. a, Multidimensional scaling embedding of the ten behavioral metrics in
CogBench [37] for different models. b, Pearson correlation coefficients indicating how well human
neural activity in the two-step task [41] can be decoded using Centaur’s internal representations
extracted from different layers. c, Pearson correlation coefficients indicating how well human neu-
ral activity in a sentence-reading task [42] can be decoded using Centaur’s internal representations
extracted from different layers. Control refers to a model that uses representations extracted from a
randomly-initialized transformer model with matched architecture.

This procedure was then repeated separately for each participant and region. Further
details can be found in the Methods section. Figure 4b shows the resulting Pearson
correlation coefficients across layers for both Centaur and Llama. We found that Cen-
taur’s representations consistently outperform Llama’s representations in predicting
human neural activity (all p ≤ 0.001), suggesting that finetuning a model on large-
scale behavioral data aligned its internal representations to human neural activity. It
is worth noting that this type of analysis was only possible due to the expressivity
of Centaur’s representations, and that using representations of a traditional cognitive
model led to a substantially drop in performance (dashed line in Figure 4b).
We expanded these results in a second analysis, for which we relied on a previously
collected data set involving fMRI measurements of people reading simple, six-word
sentences, such as “That is such a beautiful picture!”[42]. We focused our analysis on
a subset of five participants who each passively read 1,000 sentences, spread across 20
experimental runs and two scanning sessions. The presented sentences were extracted
from nine corpora and selected to maximize semantic diversity. We closely followed
the protocol of the original study and predicted aggregated neural activity across
participants in the language network. We repeated this procedure for representations
extracted from different layers in both Centaur and Llama. Predictability peaked at
around layer 20 as shown in Figure 4c. This peak is consistent with the hypothesis that
the intermediate layers of such models contain the most amount of information [43].
We performed an inverse-weighted meta-analysis [44] on the difference in correlations
between Centaur and Llama which indicated that there was a significant benefit of
finetuning when pooling across layers (β = 0.007, 95% CI [0.0002, 0.013], p = 0.045).
While this effect was consistent across layers, it was not statistically significant for any
individual layer. Yet, this analysis demonstrates that Centaur’s alignment to human
neural activity increases (although only marginally) even for stimuli far away from its
training data, thereby highlighting the generalization abilities of the model.

9
 Original study DeepSeek-R1 explanation Scientific regret minimization

The participant employed a two- Product A Product B Response

step decision-making strategy. First,
1 0 0 1 0 1 1 1 A
they determined which product had
0 1 1 0 1 0 0 0 B
Expert 1 (90%) 1 0 the majority of positive ratings (1s)
0 1 1 0 1 0 0 0 B
Expert 2 (80%) 0 1 across all experts. If the products
1 0 0 0 0 1 1 0 A
Expert 3 (70%) 0 1 were tied in the number of positive
0 1 1 1 1 0 0 1 B
Expert 4 (60%) 1 1 ratings, the participant then con-
1 0 0 1 0 1 1 1 A
sidered the rating from the highest
0 1 1 0 1 0 0 0 B
Which product is superior in quality? validity expert to break the tie.

Fig. 5 Model-guided scientific discovery. We used Psych-101 and Centaur to guide the development
of a cognitive model for a multi-attribute decision-making study. Each panel shows the Akaike infor-
mation criterion (AIC) for the set of models considered at the given stage, starting with the models
considered in the original study. We first asked DeepSeek-R1 to generate an explanation for human
responses and formalized the resulting verbal strategy into a formal computational model. We then
further refined this model through scientific regret minimization using Centaur as a reference model.
Eight data points are visualized for which Centaur makes accurate predictions but the DeepSeek-
R1-discovered model does not. We then used this information to design a domain-specific cognitive
model that is as predictive as Centaur yet still interpretable.

Model-guided scientific discovery using Psych-101 and Centaur

Psych-101 and Centaur both constitute valuable tools for scientific discovery. In the
following section, we present an example for how each of them can be used to improve
our understanding of human decision-making. The individual steps of this process are
illustrated in Figure 5.
Psych-101 contains human behavioral data in a natural language format, which
means that it can be readily processed and analyzed by language-based reasoning
model, such as DeepSeek-R1 [45]. To demonstrate this use case, we asked DeepSeek-R1
to generate an explanation of participants’ behavior in a multi-attribute decision-
making experiment [46]. The model produced several explanations, one of which caught
our attention: “The participant employed a two-step decision-making strategy. First,
they determined which product had the majority of positive ratings (1s) across all
experts. If the products were tied in the number of positive ratings, the participant
then considered the rating from the highest validity expert to break the tie.” This strat-
egy combines two well-known heuristic decision-making strategies that – best to our
knowledge – have not been considered in this particular combination before. We then
took this verbal strategy, implemented it as a formal computational model, and found
that it indeed explained human response behavior more accurately than the three
strategies considered in the original study (a weighted-additive strategy, equal weight-
ing, and the take-the-best heuristic). However, the DeepSeek-R1-discovered model

10
(AIC: 181.7) still fell short of the goodness-of-fit of Centaur (AIC: 72.5), indicating that
there is still room for improvement. We, therefore, turned towards a method known as
scientific regret minimization, which uses a black-box predictive model as a reference
to identify responses that are in principle predictable but are not captured by a given
model [47]. Typically, scientific regret minimization requires the collection of a large-
scale experiment-specific data set to train this predictive model. Centaur, on the other
hand, can be used out-of-the-box and without the need to collect any domain-specific
data, thereby circumventing this step and broadening the scope of scientific regret
minimization considerably (indeed, the multi-attribute decision-making data set under
consideration contains less than a hundred participants, placing it far out of reach
for traditional scientific regret minimization). When inspecting the responses that are
well-predicted by Centaur but not the DeepSeek-R1-discovered model, we observed
that they all involve choices in which participants choose the option with fewer pos-
itive ratings overall but which was rated positively by a higher validity expert. This
pattern suggests that the switch between the two heuristics is likely not as strict as ini-
tially suggested by the DeepSeek-R1-discovered strategy. To capture this, we replaced
the either-or rule with a weighted average of both heuristics. We found that the model
that resulted from this process matched Centaur in terms of its goodness-of-fit (AIC:
71.7) while still being interpretable. Notably, the final result of this model compari-
son stands in contrast to the one that was conducted with the original set of models
and indicates that people rely on a combination of heuristics when making decisions
as opposed to following a weighted-additive strategy [48].

Discussion
The present paper introduced Centaur, the first foundation model of human cognition.
We obtained Centaur by finetuning a state-of-the-art language model on Psych-101 – a
novel, large-scale data set of human behavior. This approach allowed us to leverage the
vast knowledge embedded in large language models, while, at the same time, aligning
them with human behavior [15]. Centaur successfully captures human behavior and
passes a wide range of out-of-distribution checks. It generalizes not only to unseen
participants but also to new cover stories, structural variations, and entirely novel
domains. In addition to analyzing the model on a behavioral level, we also conducted
a series of analyses on its internal representations, in which we found an increased
alignment with human neural activity.
We furthermore conducted a case study demonstrating how both Psych-101 and
Centaur can be used for guiding the development of predictive, yet interpretable,
cognitive models. Looking beyond this example, Centaur finds many additional appli-
cations in the context of automated cognitive science [49–51]. It may, for instance, be
used for in-silico prototyping of experimental studies [52]. In this context, one could
use the model to figure out which designs lead to the largest effect sizes, how to design
a study to reduce the number of required participants, or to estimate the power of an
effect.

11
 While the present paper takes initial steps in leveraging Centaur to gain deeper
insights into human cognition, it also opens up exciting new avenues for future explo-
ration. First, one could further probe Centaur’s internal representations to understand
how it represents knowledge and processes information. The resulting insights could,
in turn, be used to generate new hypotheses about knowledge representation and infor-
mation processing in humans that could be validated in future experimental studies.
We believe that tools like sparse autoencoders [53] and attention map visualization
[54] provide promising avenues towards accomplishing this goal and hope to explore
them in future studies.
In addition, it might also be possible to train models with different architectures
from scratch using the data set that we have created in the process of this paper.
Doing so would enable us to investigate the neural architecture of human cognition at
a scale that could not have been done before. We might, for example, ask questions
such as whether human information processing is better described by attention-based
architectures [55] or architectures with a vector-based memory [56], or how much we
can improve by incorporating theories from the neuroscience literature [57, 58]. We
expect an eventual outcome of such an approach to contain both domain-specific and
domain-general modules, thereby allowing us to investigate the interplay between the
two.
Even though Psych-101 is already the broadest and largest data set of human
behavior out there, we view its development as an ongoing process and plan to further
develop it. The focus in its current state is largely on learning and decision-making,
but eventually, we intend to include additional domains such as psycholinguistics,
social psychology, and economic games. Experiments with information about indi-
vidual differences are another source of neglected data in the current iteration of
Psych-101. Ideally, we want to include all types of relevant information about sub-
jects (e.g., age, personality traits, or socioeconomic status) in the prompt, such
that a model trained on this data can capture individual differences. Experiments
from developmental psychology or computational psychiatry provide an ideal source
for this purpose. Finally, while we have already included some cross-cultural and
meta-studies [59–62], the current iteration has still a strong bias toward a Western,
educated, industrialized, rich, and democratic (WEIRD) population [63]. To address
all of these shortcomings, we have created an open-source repository and invite
everyone to contribute to the next iteration of Psych-101 in an open research collabora-
tion (https://github.com/marcelbinz/Psych-201). The goal of this effort is to provide
psychological data in a standardized format that facilitates benchmarking, thereby
complementing existing efforts from the neuroscience community [64–67]. While the
natural language format (together with quite a bit of reverse-engineering) used in this
work allows us to express a vast range of experimental paradigms, it introduces a
selection bias against experiments that cannot be expressed in natural language. The
long-term objective should be to move towards a multi-modal data format instead [68].

12
Conclusion
When the idea of a unified model of cognition was first proposed, researchers expressed
concern that established areas of cognitive science might react negatively to such a
model. In particular, they feared that the new approach might be seen as unfamiliar or
incompatible with existing theories, just like an “intruder with improper pheromones”
[69]. This could lead to an “attack of the killer bees”, where researchers in tradi-
tional fields would fiercely critique or reject the new model to defend their established
approaches. To mitigate these concerns, the concept of a cognitive decathlon was
proposed: a rigorous evaluation framework in which competing models of cognition
are tested across ten experiments and judged based on their cumulative performance
in them. In the current work, we applied Centaur to the equivalent of sixteen such
cognitive decathlons, where it was tested against numerous established models and
consistently won every competition. This outcome suggests that the data-driven dis-
covery of domain-general models of cognition is a promising research direction. The
next step for future research should be to translate this domain-general computational
model into a unified theory of human cognition as envisioned by Newell [2].

13
Methods
Data collection
We constructed Psych-101 by transcribing data from 160 psychological experiments
into natural language. Each prompt was designed to include the entire trial-by-trial
history of a complete session from a single participant. The included experiments were
selected using the following criteria: (a) publicly available data on a trial-by-trial level,
(b) possibility of transcription into text without a significant loss of information, and
(c) coverage of a broad spectrum of domains. The transcription of each experiment
was done manually by the authors. We designed our natural language prompts using
the following principles: (a) instructions follow the original study as closely as possible,
(b) simplifications were made where appropriate, and (c) a maximum prompt length
of roughly 32,768 tokens. Full information about all included experiments is provided
in the Supplementary Information.

Finetuning procedure
We used Llama 3.1 70B as the base model for our finetuning procedure. We relied
on a parameter-efficient finetuning technique known as QLoRA [19]. QLoRA adds
so-called low-rank adapters to each layer of a four-bit quantized base model. The
base model was kept fixed during finetuning and only the parameters of the low-rank
adapters were adjusted. We added low-rank adapters to all linear layers of the self-
attention mechanisms and the feedforward networks. Each low-rank adapter modifies
the forward pass as follows:

Y = XW + αXL1 L2 W ∈ Rh×o ; L1 ∈ Rh×r ; L2 ∈ Rr×o

where XW is the (quantized) linear transformation of the base model and XL1 L2
the low-rank adapter. The hyperparameter α controls the trade-off between the two.
Low-rank adapter computations are performed in half-precision floating-point format.
For further details on this technique, we refer the reader to [19].
We finetuned the model for one epoch on the entire data set using a standard
cross-entropy loss (we experimented with prolonged training but found that this led
to overfitting). We only backpropagated the loss at human responses and masked out
the loss for all other tokens. The effective batch size was set to 32, the learning rate
to 0.00005, and the weight decay to 0.01. We used an 8-bit AdamW optimizer [70]
with a linearly increasing warm up over the first 100 gradient steps. The finetuning
procedure was implemented using the unsloth library (https://unsloth.ai/).
We have also trained a smaller version of Centaur – called Minitaur – that uses
Llama 3.1 8B as base model following the same recipe. Minitaur captures human
behavior close to its training distribution but generalizes less robustly than the larger
model to out-of-distribution experiments. Nevertheless, we believe that Mintaur is
useful for prototyping as it does not require access to any specific hardware (it runs,
for instance, on the free GPU instances in Google Colab). More detailed results for
this model are provided in the Supplementary Information.

14
Evaluation metric
We use (negative) log-likelihoods averaged over responses as our evaluation metric. For
experiments with multi-token responses, we summed log-likelihoods within a response
and averaged across responses.

Domain-specific cognitive models

We selected 14 cognitive and statistical models that together cover most of the exper-
iments in Psych-101 as our baseline models. Further details regarding the included
models and their specifications are provided in the Supplementary Information.
For our main analysis, we were interested in predicting the behavior of held-out
participants. Therefore, we fitted a joint set of parameters for all participants in
the training data and evaluated how well a model with these parameters predicts
responses of held-out participants. Mirroring the evaluation metric for the language-
based models, we evaluated goodness-of-fit using (negative) log-likelihoods averaged
over responses.
For the out-of-distribution evaluations, we fitted model parameters using the most
similar experiment in the training set, and then evaluated how well a model with
the resulting parameters predicts human responses in the unseen setting. The most
similar experiment for the magical carpet version of the two-step task was a two-step
task experiment with the default spaceship cover story. The most similar experiment
for Maggie’s farm was the horizon task. We included no baseline model for the logical
reasoning task, as none of the experiments in the training data were similar to it.

Neural alignment
The neural alignment analysis on the two-step task was conducted using data col-
lected in a previous study [41]. We used a regularized linear regression model to
predict fMRI data from internal representations of Centaur and Llama (a separate
model was used for each participant and region). We fitted each of these models on
data from two scanning blocks and evaluated them on data from the third. The reg-
ularization strength was selected using a nested cross-validation procedure. For each
run, we split the beta maps into cortical and subcortical regions of interest (ROI)
using the Schaefer 2018 atlas [71]. We averaged the betas within each ROI, reducing
the number of betas from the number of voxels to the number of ROIs. All cortical
and subcortical ROIs from the atlas were evaluated. Reported Pearson correlation
coefficients correspond to the average across all ROIs.
Internal representations were extracted from the models’ residual stream and
transformed using a principal component analysis. We set the number of retained
components such that they explain 95% of the variance.
The fMRI data were preprocessed using fMRIPrep 24.0 [72]. We used the default
settings of fMRIPrep, and all the scans were aligned to the MNI152NLin2009cAsym
atlas [73]. To extract effect estimates for each subtrial of the task (e.g., second step
of the fifth trial, feedback of the tenth trial), we built separate general linear models
(GLMs). Each GLM included the subtrial of interest as a separate regressor, whose
z-scored beta estimates were used for the alignment analysis. This part of the data

15
was not modeled using other regressors. Additionally, we included different regressors
capturing all the first steps, all the second steps, and all the feedback steps. Lastly, we
used 6 rotation and translation estimates as well as framewise displacement as noise
regressors. The hemodynamic response was modeled using the spm [74] model. A high
pass filter of 0.01 Hz and a Gaussian kernel with 6mm full-width at half-maximum
was applied. The GLMs were built using nilearn [75].

The neural alignment analysis on the sentence-reading task was conducted using
publicly available code from the original study [42]. No other changes were made
besides replacing GPT2-XL with Centaur and Llama. We refer the reader to [42] for
further details.

Model-guided scientific discovery

In our model-guided scientific discovery analysis, we focused on participants in the
test set to avoid any potential contamination issues. We fitted parameters of all cogni-
tive models individually for each participant using a maximum likelihood estimation.
Models were compared against each other using the Akaike information criterion
(AIC). The three models from the original study were implemented via the following
equations:
⊤


p(a = A|xA , xB , WADD) ∝ exp β · wWADD xA
wWADD = [0.9, 0.8, 0.7, 0.6]
⊤


p(a = A|xA , xB , EW) ∝ exp β · wEW xA
wEW = [1, 1, 1, 1]
⊤


p(a = A|xA , xB , TTB) ∝ exp β · wTTB xA
wTTB = [1, 0.5, 0.25, 0.125]

where xA and xB are vectors containing four expert ratings (either 0 or 1) and β is a
free parameter controlling the noise level.
We prompted DeepSeek-R1 (in the Distill-Llama-70B variant) to generate expla-
nations of human decision-making; the corresponding prompt is provided in the
Supplementary Information. We then formalized the explanation shown in Figure 5
into the following computational model:
(
⊤

P P
exp β · wTTB xA if i xA,i = i xB,i
p(a = A|xA , xB , DeepSeek-R1) ∝ ⊤


exp β · wEW xA otherwise

For the scientific regret minimization pipeline, we computed the difference in log-
likelihoods between Centaur and the DeepSeek-R1-discovered model. We visualized
and inspected the ten data points with the highest difference. This process resulted in
the following computational model:
⊤ ⊤



p(a = A|xA , xB , SRM) ∝ exp β · σ · wTTB xA + (1 − σ) · wEW xA

16
where σ is a free parameter constrained between 0 and 1 controlling the trade-off
between the two strategies.

Data availability
Psych-101 is publicly available on the Huggingface platform: https://huggingface.
co/datasets/marcelbinz/Psych-101. The test set is accessible under a CC-BY-
ND-4.0 license via a gated repository: https://huggingface.co/datasets/marcelbinz/
Psych-101-test.

Code availability
Centaur is available on the Huggingface platform: https://huggingface.co/marcelbinz/
Llama-3.1-Centaur-70B-adapter. We provide additional code to reproduce our results
under https://github.com/marcelbinz/Llama-3.1-Centaur-70B.
Acknowledgements. Funding was from the Max Planck Society (PD), the Hum-
boldt Foundation (PD), and the NOMIS Foundation (TLG). PD is a member of
the Machine Learning Cluster of Excellence, EXC number 2064/1 – Project num-
ber 39072764 and of the Else Kröner Medical Scientist Kolleg ”ClinbrAIn: Artificial
Intelligence for Clinical Brain Research”. SK is supported by a Google PhD Fellowship.
Author contributions.
Project lead: Marcel Binz
Data curation: Elif Akata, Franziska Brändle, Marcel Binz, Fred Callaway, Julian
Coda-Forno, Can Demircan, Maria Eckstein, Noemi Elteto, Susanne Haridi, Akshay
Jagadish, Li Ji-An, Alexander Kipnis, Sreejan Kumar, Tobias Ludwig, Surabhi Nath,
Joshua Peterson, Evan Russek, Tankred Saanum, Johannes Schubert, Luca Schulze
Buschoff, Nishad Singhi, Xin Sui, Mirko Thalmann, Vuong Truong, Kristin Witte,
Shuchen Wu, Dirk Wulff, Huadong Xiong
Data quality control: Elif Akata, Marcel Binz, Julian Coda-Forno, Can Demircan,
Susanne Haridi, Luca Schulze Buschoff
Model training: Marcel Binz, Vishaal Udandarao
Model evaluation: Marcel Binz, Julian Coda-Forno, Alexander Kipnis, Mirko
Thalmann, Konstantinos Voudouris
Domain-specific models: Marcel Binz, Julian Coda-Forno, Can Demircan, Akshay
Jagadish, Marvin Mathony, Alireza Modirshanechi, Milena Rmus, Tobias Ludwig
Neural analyses: Marcel Binz, Can Demircan, Sreejan Kumar, Marcelo Mattar,
Evan Russek
First draft: Marcel Binz, Eric Schulz
Review and editing: All authors
No researchers at Google DeepMind used Llama for this research. We thank Natalia
Scharfenberg for her contributions to the data collection.

17
Supplementary Information
Psych-101

Decision-making

Memory Multi-armed bandits

Supervised learning
Miscellaneous
Markov decision processes

Fig. 6 Visual summary of Psych-101. The left panel shows the proportion of included domains. The
right panel shows a word cloud with the included experimental paradigms.

Data contamination analysis

We performed a data contamination analysis using the LogProber method proposed in
[76]. LogProber fits the following two-parameter exponential model to the cumulative
log-likelihood of each sequence being checked for contamination:

f (x) = −A 1 − e−Bx

Prompts that are memorized from the pretraining data will show a high accel-
eration (B). Following the results presented in [76], we set a threshold for possible
contamination to log B ≥ 1. We repeat this analysis for each experiment in Psych-101.
For this, we used prompts up to the point of the first human choice. This analysis
indicated no evidence of contamination as shown in Figure 7.

potentially contaminated
1.0 not contaminated
0.5
0.0
logB

0.5
1.0
1.5
0 10 20 30 40 50 60 70
Experiment

Fig. 7 Data contamination analysis using the LogProber method [76]. LogProber fits an exponen-
tial model to the cumulative log-likelihood of each sequence being checked for contamination. High
acceleration (log B) suggests that a prompt is memorized from the pretraining data.

18
Experiment embeddings
Figure 8 shows a two-dimensional embedding of the experiments used in this paper. To
obtain this embedding, we took the corresponding natural language prompts up to the
point of the first human choice, extracted a vector-based representation for them using
ModernBERT [77], and finally projected these representations onto two dimensions
using multidimensional scaling. We observed that most of our out-of-distribution eval-
uation experiments are placed on the boundaries of the embedding space, confirming
that they indeed probe the generalization abilities of Centaur.

10
Entirely novel domain (Figure 3c)

5 Behavioral propensities Deep sequential decision task

Embedding dimension 2

Naturalistic reward learning

Moral decision-making Modified problem structure (Figure 3b)
0
Naturalistic category learning
Economic games
5
Modified cover story (Figure 3a)

10

6 4 2 0 2 4 6
Embedding dimension 1

Fig. 8 Multidimensional scaling embedding of the experiments used in this paper. Purple dots
correspond to experiments from Psych-101, whereas the colored dots correspond to the indicated
evaluation experiment.

19
Full goodness-of-fit results

Experiment Centaur Llama Cognitive model

Shepard categorization 0.5394 0.5818 0.6108
Drifting four-armed bandit 0.7029 0.8810 0.9043
N-back 0.3954 0.5209 0.5787
Digit span 0.5520 0.6618 0.9359
Go/no-go 0.0000 0.0062 0.0757
Recent probes 0.2572 0.3433 0.3868
Horizon task 0.4032 0.5237 0.3595
Gardening task 0.3783 0.5040 0.9105
Columbia card task 0.1867 0.2261 0.2629
Balloon analog risk task 0.0593 0.0753 0.0922
Two-armed bandit 0.2963 0.3829 0.4187
Conditional associative learning 0.5380 0.6373 0.8575
THINGS odd-one-out 0.8068 1.1386 0.8253
Multi-attribute decision-making 0.0619 0.1502 0.1922
Two-step task 0.4998 0.6075 0.6043
Probabilistic instrumental learning 0.4937 0.5382 0.5047
Medin categorization 0.4967 0.5772 0.5313
Zoopermarket 0.4850 0.6026 0.6047
choices13k 0.4274 0.5342 0.6563
CPC18 0.3390 0.4118 0.6607
Intertemporal choice 0.4340 0.7336 0.6591
Structured bandit 0.6410 0.8114 1.0530
Weather prediction task 0.5514 0.5749 0.6267
Iowa gambling task 0.8890 0.9880 1.1555
Virtual subway network 1.1271 1.5347 nan
Multi-task reinforcement learning 0.5672 0.6604 1.0424
Serial reaction time task 0.1718 0.1900 0.1962
Decisions from description 0.5336 0.7569 0.6120
Decisions from experience 0.3686 0.4339 0.5404
Changing bandit 0.3025 0.3824 0.4378
Multiple-cue judgment 1.1236 1.2818 1.9157
Recall and recognition 1.0591 1.3759 nan
Experiential-symbolic task 0.4536 0.6983 nan
Grammar judgement 1.4355 1.9949 1.4127
Risky choice 0.4281 0.6475 nan
Tile-revealing task 1.8713 2.7380 nan
Episodic long-term memory 0.8684 1.1344 nan
Aversive learning 4.0733 5.1066 nan
Spatially correlated multi-armed bandit 1.8319 2.4479 2.7635
Probabilistic reasoning 2.3731 2.6406 nan

Table 1 Full negative log-likelihoods on held-out participants.

Minitaur
Figure 9 compares Centaur and to Minitaur on the analyses from the main text.

20
 Modified cover story

Negative log-likelihood
Overall Random guessing
0.7
Spatially correlated multi-armed bandit
Multiple-cue judgment 0.6
Gardening task
Multi-task reinforcement learning
Structured bandit 0.5
Digit span
CPC18 Centaur Minitaur Cognitive
Conditional associative learning model
Iowa gambling task
choices13k Modified problem structure
Intertemporal choice 1.2

Negative log-likelihood
Drifting four-armed bandit Random guessing
N-back 1.0
Decisions from experience
Changing bandit 0.8
Multi-attribute decision-making
Recent probes 0.6
Two-step task
Two-armed bandit 0.4
Zoopermarket Centaur Minitaur Cognitive
Horizon task model
Decisions from description
Columbia card task Entirely novel domain
Go/no-go 2.2

Negative log-likelihood
Weather prediction task
Shepard categorization Random guessing
Medin categorization 2.0
Balloon analog risk task
Serial reaction time task Centaur 1.8
THINGS odd-one-out Minitaur
Probabilistic instrumental learning Cognitive model 1.6 N/A
Grammar judgement
0.0 0.2 0.4 0.6 0.8 Centaur Minitaur Cognitive
log-likelihood model

Fig. 9 Log-likelihood comparison between Centaur and Minitaur on the analyses from the main
text. The left subplot contains results from Psych-101, while the right subplot shows contains results
for different held-out settings. Error bars correspond to the standard error of the mean, taken over
responses.

Further out-of-distribution evaluations

Figure 10 contains results for six additional out-of-distribution experimental
paradigms, including moral decision-making [31], economic games [32], naturalistic
category and reward learning [33], behavioral propensities [34], and a deep sequential
decision task [35]. None of these paradigms were included in Psych-101, hence they
provide a stress test for a model’s generalization capabilities. Centaur robustly cap-
tured human behavior in all of these settings, while smaller and non-finetuned models
did not do so consistently.

21
 Moral decision-making Naturalistic category learning Naturalistic reward learning
1.0

Negative log-likelihood
0.7 Random guessing
0.9 0.75
0.8 0.70 Random guessing
0.6
0.7 Random guessing 0.65
0.5
0.6 0.60
Centaur Minitaur Llama Llama Centaur Minitaur Llama Llama Centaur Minitaur Llama Llama
(70B) (8B) (70B) (8B) (70B) (8B)
Economic games Behavioral propensities Deep sequential decision task
1.2
Negative log-likelihood

0.8 2.0 Random guessing

Random guessing
1.1
0.7 Random guessing 1.8
1.6 1.0
0.6
1.4 0.9
Centaur Minitaur Llama Llama Centaur Minitaur Llama Llama Centaur Minitaur Llama Llama
(70B) (8B) (70B) (8B) (70B) (8B)

Fig. 10 Further out-of-distribution evaluations. Each subplot shows negative log-likelihoods for a
different experiment. Error bars correspond to the standard error of the mean, taken over responses.

Further model comparisons

To rule out the hypothesis that finetuning on any data aligns a model with human
behavior, we compared Centaur to various Llama variants finetuned for other purposes
(i.e. non-cognitive tasks). These models include the following:
• Nemotron [78]: finetuned for instruction-following.
• Hermes [79]: finetuned for various purposes, including agentic capabilities, roleplay-
ing, reasoning, multi-turn conversation, and long context coherence.
• Reflection: finetuned for reasoning.
Figure 11 illustrates the results of this analysis. We observed that none of the
Llama variants captures human behavior better than the base model, ruling out the
hypothesis that finetuning generally leads to models that are better at predicting
human behavior.

22
 0.80
0.75

Negative log-likelihood
0.70
0.65
0.60
0.55
0.50
Centaur Llama Nemotron Hermes Reflection

Fig. 11 Negative log-likelihoods of Centaur and alternative Llama variants on Psych-101. Error bars
correspond to the standard error of the mean, taken over responses.

Noise ceiling analysis

We conducted a noise ceiling analysis to better understand the capabilities of Centaur.
It is not straightforward to estimate the noise ceiling for experiments with sequential
dependencies, which includes the majority of Psych-101. Hence, we focused on two
experiments for which such an analysis is possible: the choices13k data set [21] and
an intertemporal choice experiment [59]. In both cases, we found that Centaur sub-
stantially exceeds the estimated noise ceiling. This is possible because Centaur can
pick up on context-dependent patterns that are not captured by standard noise ceil-
ing analyses. Therefore, we have performed an additional analysis testing how well
Centaur can predict human responses if we prompt it to predict each response inde-
pendently. We use the suffix “ind.” to indicate this way of prompting the model.
Centaur still matches the performance of domain-specific cognitive models when
context-independent prompts are used, amounting to roughly half of the estimated
noise ceiling. Figure 12 visualizes these results.

a choices13k b Intertemporal choice

2.7
2.0
Negative log-likelihood

0.7 Random guessing 0.7 Random guessing

0.6 0.6

0.5 0.5

0.4 0.4
Centaur Llama Cog. Noise Centaur Llama Centaur Llama Cog. Noise Centaur Llama
model ceiling (ind.) (ind.) model ceiling (ind.) (ind.)

Fig. 12 Noise ceiling analysis. We use the suffix ”ind.” to indicate models that are prompted with
to predict each response independently.

23
Benchmarks
metabench
Figure 13 shows additional results in metabench, a sparse benchmark containing sev-
eral canonical benchmarks from the machine learning literature [38]. We find that
Centaur maintains the level of performance of Llama, indicating that finetuning on
human behavior did not lead to deterioration in other tasks. Performance on Truth-
fulQA [80] – which measures how models mimic human falsehoods – even improved
significantly with finetuning. We refer the reader to [38] for further details.

Centaur Llama

1.0
0.8
Performance

0.6
0.4
0.2
0.0
ARC

GSM8K

MMLU
Mean

HellaSwag

Winogrande
TruthfulQA
Fig. 13 metabench [38] results.

CogBench
Figure 14 shows additional results in CogBench, a benchmark that includes ten
behavioral metrics derived from seven cognitive psychology experiments [37]. We find
that – relative to Llama – Centaur’s performance improves in all experiments (see
Figure 14a). Furthermore, Centaur becomes more similar to human subjects in all ten
behavioral metrics (see Figure 14b). We refer the reader to [37] for further details.

24
 a Centaur Llama b Centaur Llama
2 2

Parameter value
Performance 1 Humans 1 Humans

0 Random 0 Random

Restless bandit

Temporal discounting
Prior weighting

Random exploration
Probabilistic reasoning

Horizon task

Instrumental learning

Learning rate
Optimism bias
Model-basedness
Balloon analog risk task

Likelihood weighting
Directed exploration
Two-step task

Meta-cognition

Risk taking
Fig. 14 CogBench [37] results. a, Performance-based metrics. b, Behavioral metrics. All metrics are
human-normalized: a value of zero corresponds to a random agent, while a value of one corresponds
to the average human subject.

Fine-grained neural analysis results

Figure 15 shows fine-grained neural alignment results on the two-step task data. Cen-
taur achieves the most accurate predictions in the left motor cortex (see Figure 15a).
As participants performed the task with their right hand in the scanner, this effect
may be explained by Centaur’s strong performance in predicting choices.
Figure 15b contains numerical results for ROIs that have been identified as behav-
iorally relevant in previous work. Centaur outperformed Llama and the cognitive
model in predicting activity in accumbens, the ROI from the original study that
showed a reward prediction error effect [22, 41]. We found a similar pattern in the
medial PFC, another region that showed an effect in the original article [41], as well
as in the sensory and motor cortices.

a Pearson correlation b Centaur Llama Cognitive model Control

0.8 0.5
0.6 0.4
Pearson correlation

0.3
0.4
0.2
0.2
0.1
0.0 0.0
Accumbens Medial PFC Motor Cortex Visual Cortex

Fig. 15 Neural alignment results in the two-step task. a, Pearson correlation coefficients between
the predicted activity from Centaur’s representations and the BOLD data shown on a surface brain.
b, Predictive performance of Centaur’s representations against alternatives for specific ROIs. Cortical
scores are averaged over the corresponding bilateral parcels in the Schaefer atlas. The accumbens is
defined based on the Harvard-Oxford atlas. Pearson correlation coefficients are shown for layer 20
but exhibit a similar pattern across all layers.

25
Criterion Fulfilled by Centaur
Behave as an (almost) arbitrary function of the environment ✓
Operate in real time ✓
Exhibit rational, that is, effective adaptive behavior ✓
Use vast amounts of knowledge about the environment ✓
Behave robustly in the face of error, the unexpected, and the unknown ✓
Integrate diverse knowledge ✓
Use (natural) language ✓
Exhibit self-awareness and a sense of self •
Learn from its environment ✓
Acquire capabilities through development ✗
Arise through evolution ✗
Be realizable within the brain ✓

Table 2 Newell test for a theory of cognition.

Model-guided scientific discovery

We generated an explanation for the behavior of every participant in the multi-
attribute decision-making study of [46] using DeepSeek-R1 based on the following
prompt template:

In the following, you will see a transcript of a psychological experiment. Participant

responses are highlighted within “<<” and “>>” characters. Your job will be
to explain what strategies that participant used to solve the given task. Your
explanation should be targeted at an expert cognitive scientist.
### START OF TRANSCRIPT ###
[TRANSCRIPT]
### END OF TRANSCRIPT ###
Please explain what strategies that participant used to solve the given task in less
than two hundred words:

Newell test
Together with his call for unified theories of cognition [2, 81], Newell outlined a set of
criteria that a unified computational model should fulfill. Centaur is the first model
to satisfy the majority of these criteria (see Table 2). Most importantly, it (1) behaves
as an almost arbitrary function of the environment, (2) operates in real time, and
(3) relies on vast amounts of knowledge about the world. We provide an extended
discussion on Newell’s criteria in the following.

Behave as an (almost) arbitrary function of the environment

This is the most important criterion according to Newell. Centaur fulfills it more than
any previous model as shown by our extensive analysis. Yet, Centaur’s scope is still
limited to psychological experiments that can be expressed in natural language. It

26
will be an important avenue for future research to transfer this ability to real-world
applications.

Operate in real time

Centaur can simulate human behavior in (almost) real-time. For example, running an
open-loop simulation of a typical two-step task experiment takes around 30 minutes,
while it takes around 20 minutes for the average human participant. We believe that
inference time could be further optimized to fully close this gap.

Exhibit rational, that is, effective adaptive behavior

Bayesian inference is the gold standard for rational and adaptive behavior [82]. Pre-
vious work has shown that systems that engage in in-context learning implement
Bayesian inference implicitly [83]. In-context learning is at the heart of Centaur,
thereby making it a rational and adaptive system.

Use vast amounts of knowledge about the environment

Large language models are the biggest knowledge bases we have to date. As Centaur
is built on top of a state-of-the-art language model, it fulfills this criterion by design.

Behave robustly in the face of error, the unexpected, and the

unknown
Our extensive out-of-distribution evaluations clearly demonstrate that Centaur has
this ability.

Integrate diverse knowledge

This was originally a criterion on symbols and abstractions [2]. At the basic level,
Centaur is a system that processes language. Language is a symbolic system, meaning
that Centaur fulfills this criterion.

Use (natural) language

No further elaboration is required.

Exhibit self-awareness and a sense of self

This is a tricky criterion as Centaur is trained on a population of individuals. Yet,
to make good predictions about the future, Centaur needs to make inferences about
the person who has produced a given trajectory, and in order to do that it requires a
representation of that person [84].

Learn from its environment

Psych-101 contains many experiments that require learning from an environment.
Centaur does well in those experiments, thereby clearly satisfying this criterion.

27
Acquire capabilities through development
Centaur makes no claims about how human behavior might arise through development.

Arise through evolution

Centaur makes no claims about how human behavior might arise through evolution.

Be realizable within the brain

We have shown that Centaur’s internal representations are robust predictors of human
neural activity. Even though there is clearly a gap between the transformer architecture
that Centaur is based on and the human brain, Centaur still represents the current
state-of-the-art when looking at neural alignment to human subjects.

Modeling details
In the following, we list the domain-specific cognitive and statistical models used in
our comparison. Each model was implemented in PyTorch [85]. We optimized a joint
set of parameters for all participants in the training data by maximizing log-likelihoods
of their choices and then evaluated how well a model with these parameters predicts
choices of held-out participants. The optimization procedure involved 1000 iterations
over the entire training set, and relied on a gradient-based algorithm [86] with an
initial learning rate of 0.1. We use 1 to denote the indicator function that takes a
value of one if the argument is true and zero otherwise.

Generalized context model

Reference: [20]
This model was used for the following experiments:
• Shepard categorization
• Medin categorization
• Weather prediction task
It uses the following log-likelihood:

t−1
!
exp (−||xk − xt ||2 ) · 1 [yk = i]
X
p(ct = i|xt = xt ) ∝ exp β
k=1

where xt are the features of the item observed at trial t and yt is the corresponding
class label. β is a free parameter of the model.

Prospect theory model

Reference: [21]
This model was used for the following experiments:
• CPC18

28
• choices13k
• Decisions from description
It uses the following log-likelihood:
  
⊤
p(ct = i|pi = pi , xi = xi ) ∝ exp exp (β) π (pi ) u (xi )
π (pi ) = sigmoid (a) + sigmoid (b) · pi
(
sigmoid(d)
sigmoid (c) · xi where xi ≥ 0
u (xi ) = sigmoid(g)
−sigmoid (e) · (−sigmoid (f ) xi ) where xi < 0

where pi is the vector of probabilities and xi is the vector of values for each possible
outcome in option i. β, a, b, c, d, e, f , and g are free parameters of the model.

Hyperbolic discounting model

Reference: [87]
This model was used for the following experiments:
• Intertemporal choice
It uses the following log-likelihood:
  
1
p(ct = i|xi = xi , γi = γi ) ∝ exp β xi ·
1 + (a · γi )

where xi is the reward and γi is the delay of delivery for option i. β and a are free
parameters of the model.

Dual-systems model
Reference: [22]
This model was used for the following experiments:
• Two-step task
It uses the following log-likelihood:
(


exp β sigmoid (τ ) QMB MF
s,i + (1 − sigmoid (τ )) Qs,i if s = 0
p(ct = i|st = s) ∝

exp βQMF s,i if s > 0

where QMB MF
s,i and Qs,i are model-based and model-free value estimates that are com-
puted as described in [22]. β and τ are free parameters of the model. We also included a
stickiness term for the first stage choices, which is omitted for brevity in the equations
above.

Rescorla-Wagner model
Reference: [23]

29
This model was used for the following experiments:
• Drifting four-armed bandit
• Horizon task
• Two-armed bandit
• Probabilistic instrumental learning
• Iowa gambling task
• Changing bandit
• Decisions from experience
It uses the following log-likelihood:

p(ct = i) ∝ exp (aVi,t + bSi,t + cIi,t )


+
Vi,t−1 + sigmoid (α ) (rt−1 − Vi,t−1 ) if ct−1 = i and rt−1 − Vi,t−1 ≥ 0

Vi,t = Vi,t−1 + sigmoid (α− ) (rt−1 − Vi,t−1 ) if ct−1 = i and rt−1 − Vi,t−1 < 0

Vi,t−1 otherwise


Si,t = 1 [ct−1 = i]
t−1
1 [ck = i]
X
Ii,t =
k=1
Vi,1 = d
Si,1 = 0
Ii,1 = 0

where rt is the reward obtained in trial t. α+ , α− , a, b, c, and d are free parameters

of the model.

Rescorla-Wagner model with context

Reference: [88]
This model was used for the following experiments:
• Conditional associative learning
It uses the following log-likelihood:

p(ct = i|st = s) ∝ exp (βVs,i,t )

(
Vs,i,t−1 + sigmoid (α) (rt−1 − Vs,i,t−1 ) if ct−1 = i and st−1 = s
Vs,i,t =
Vs,i,t−1 otherwise
Vs,i,1 = d

where rt is the reward obtained in trial t. α, β, and d are free parameters of the model.

Linear regression model

Reference: [89]

30
This model was used for the following experiments:
• Multiple-cue judgment
• Gardening task
It uses the following log-likelihood for multiple-cue judgment:

2 
p(ct = i|xt = xt ) ∝ exp β wt⊤ xt − i + γ

It uses the following log-likelihood for the gardening task:

p(ct = accept|xt = xt ) ∝ exp βwt⊤ xt

p(ct = reject|xt = xt ) ∝ exp (0)

and the following learning rule for both tasks:

⊤


wt = wt−1 + α rt−1 − wt−1 xt−1 xt−1
w1 = d

where rt is the reward obtained in trial t and xt are the observed features. α, β, γ,
and d are free parameters of the model.

Weighted-additive model
Reference: [90]
This model was used for the following experiments:
• Multi-attribute decision-making
It uses the following log-likelihood:

p(ct = i|xi = xi ) ∝ exp w⊤ xi

where xi is the vector of features for option i. w are free parameters of the model.

Decision-updated reference point model

Reference: [91]
This model was used for the following experiments:
• Columbia card task
It uses the following log-likelihood:

p(ct = sample|xwin , xloss , pwin , ploss ) ∝ exp (h · (xwin · pwin + xloss · ploss ) + i)
p(ct = stop|xwin , xloss , pwin , ploss ) ∝ exp (j)
π (p) = sigmoid (a) + sigmoid (b) · p

31
 (
sigmoid (c) · v sigmoid(d) where v ≥ 0
u (v) = sigmoid(g)
−sigmoid (e) · (−sigmoid (f ) v) where v < 0

where xwin and xloss are the values that can be won or lost respectively, and pwin
and ploss are the corresponding probabilities. a, b, c, d, e, f , g, h, i, and j are free
parameters of the model.

Odd-one-out model
Reference: [92]
This model was used for the following experiments:
• THINGS odd-one-out
It uses the following log-likelihood:

p(ct = i|xi , xj , xk ) ∝ exp x⊤

j xk

where xi , xj , and xk are the observed objects with their corresponding embeddings
xi , xj , and xk ∈ R16 that are free parameters of the model.

Multi-task reinforcement learning model

Reference: [93]
This model was used for the following experiments:
• Multi-task reinforcement learning
• Zoopermarket

GP-UCB model
Reference: [94]
This model was used for the following experiments:
• Spatially correlated multi-armed bandit
• Structured bandit
It uses the following log-likelihood:

p(ct = i) ∝ exp (β (mi,t + exp (γ) si,t ))

where mi,t and si,t are obtained via Gaussian Process regression with a radial basis
function kernel as described in [94]. β and γ are free parameters of the model.

Rational model
Reference: N/A
This model was used for the following experiments:
• Balloon analog risk task

32
• N-back
• Digit span
• Go/no-go
• Recent probes
• Serial reaction time task
It uses the following log-likelihood:

p(ct = i|ot = j) ∝ exp (Θj,i )

where j is the optimal choice at trial t. Θ ∈ RNc ×Nc are free parameters of the model.

Lookup table model

Reference: N/A
This model was used for the following experiments:
• Grammar judgement
It uses the following log-likelihood:

p(ct = i) ∝ exp (Θt,i )

where Θ ∈ RT ×Nc are free parameters of the model.

Example prompts
The following sections present the experiments contained in Psych-101 in detail.
Example prompts are truncated to 4096 characters but otherwise shown as is.

Shepard categorization
Data source: [95]

Number of experiments: 1
Number of participants: 85
Number of choices: 29776

Example prompt:
You will be shown several examples of geometric objects.
Your task is to learn a rule that allows you to tell whether an object belongs to the E
or K category.
For each presented object, you will be asked to make a category judgment by pressing
the corresponding key and then you will receive feedback.
You will encounter four different problems with different rules.

You encounter a new problem with a new rule determining which objects belong to
each category:

33
You see a big black square. You press <<K>>. The correct category is K.
You see a small black triangle. You press <<K>>. The correct category is E.
You see a big white triangle. You press <<E>>. The correct category is K.
You see a small white triangle. You press <<E>>. The correct category is E.
You see a small white square. You press <<E>>. The correct category is E.
You see a small black square. You press <<K>>. The correct category is E.
You see a big white square. You press <<E>>. The correct category is K.
You see a big black triangle. You press <<E>>. The correct category is K.
You see a big white square. You press <<E>>. The correct category is K.
You see a big black square. You press <<E>>. The correct category is K.
You see a small white triangle. You press <<K>>. The correct category is E.
You see a small black triangle. You press <<K>>. The correct category is E.
You see a big white triangle. You press <<E>>. The correct category is K.
You see a small white square. You press <<K>>. The correct category is E.
You see a small black square. You press <<K>>. The correct category is E.
You see a big black triangle. You press <<E>>. The correct category is K.
You see a small white square. You press <<E>>. The correct category is E.
You see a small black square. You press <<E>>. The correct category is E.
You see a big white triangle. You press <<K>>. The correct category is K.
You see a small black triangle. You press <<K>>. The correct category is E.
You see a small white triangle. You press <<E>>. The correct category is E.
You see a big white square. You press <<K>>. The correct category is K.
You see a big black triangle. You press <<E>>. The correct category is K.
You see a big black square. You press <<E>>. The correct category is K.
You see a small black triangle. You press <<E>>. The correct category is E.
You see a small white square. You press <<K>>. The correct category is E.
You see a small black square. You press <<E>>. The correct category is E.
You see a big black square. You press <<K>>. The correct category is K.
You see a big white square. You press <<K>>. The correct category is K.
You see a big white triangle. You press <<K>>. The correct category is K.
You see a small white triangle. You press <<E>>. The correct category is E.
You see a big black triangle. You press <<K>>. The correct category is K.
You see a small white square. You press <<E>>. The correct category is E.
You see a big black triangle. You press <<K>>. The correct category is K.
You see a big black triangle. You press <<K>>. The correct category is K.
You see a small white triangle. You press <<E>>. The correct category is E.
You see a big white triangle. You press <<K>>. The correct category is K.
You see a small white triangle. You press <<E>>. The correct category is E.
You see a small black triangle. You press <<E>>. The correct category is E.
You see a small black square. You press <<E>>. The correct category is E.
You see a small black triangle. You press <<E>>. The correct category is E.
You see a big white square. You press <<K>>. The correct category is K.
You see a big black square. You press <<K>>. The correct category is K.
You see a big white square. You press <<K>>. The correct category is K.
You see a big black square. You press <<K>>. The correct category is K.

34
You see a small black square. You press <<E>>. The correct category is E.
You see a big white triangle. You press <<E>>. The correct category is K.
You see a small white square. You press <<E>>. The correct category is E.
You see a big white square. You press <<K>>. The correct category is K.
You see

Drifting four-armed bandit

Data source: [96]

Number of experiments: 1
Number of participants: 869
Number of choices: 125952

Example prompt:
You will be asked to repeatedly choose between four different options labeled L, G,
O, and U.
You select an option by pressing the corresponding key on your keyboard.
Each time you select an option, you will get a different number of points.
Your goal is to win as many points as possible.

You press <<L>> and get 84.0 points.

You press <<G>> and get 90.0 points.
You press <<O>> and get 53.0 points.
You press <<U>> and get 24.0 points.
You press <<G>> and get 92.0 points.
You press <<G>> and get 78.0 points.
You press <<L>> and get 71.0 points.
You press <<L>> and get 75.0 points.
You press <<G>> and get 80.0 points.
You press <<G>> and get 80.0 points.
You press <<G>> and get 91.0 points.
You press <<G>> and get 90.0 points.
You press <<U>> and get 29.0 points.
You press <<O>> and get 45.0 points.
You press <<G>> and get 81.0 points.
You press <<G>> and get 75.0 points.
You press <<G>> and get 82.0 points.
You press <<G>> and get 82.0 points.
You press <<G>> and get 87.0 points.
You press <<G>> and get 85.0 points.
You press <<G>> and get 87.0 points.
You press <<G>> and get 87.0 points.
You press <<G>> and get 79.0 points.
You press <<G>> and get 75.0 points.
You press <<L>> and get 61.0 points.

35
You press <<O>> and get 40.0 points.
You press <<U>> and get 37.0 points.
You press <<G>> and get 72.0 points.
You press <<G>> and get 73.0 points.
You press <<L>> and get 66.0 points.
You press <<G>> and get 57.0 points.
You press <<L>> and get 64.0 points.
You press <<L>> and get 63.0 points.
You press <<L>> and get 61.0 points.
You press <<O>> and get 54.0 points.
You press <<U>> and get 30.0 points.
You press <<L>> and get 59.0 points.
You press <<O>> and get 56.0 points.
You press <<O>> and get 46.0 points.
You press <<L>> and get 59.0 points.
You press <<G>> and get 63.0 points.
You press <<G>> and get 63.0 points.
You press <<G>> and get 58.0 points.
You press <<L>> and get 53.0 points.
You press <<O>> and get 60.0 points.
You press <<O>> and get 59.0 points.
You press <<L>> and get 52.0 points.
You press <<O>> and get 54.0 points.
You press <<U>> and get 23.0 points.
You press <<G>> and get 52.0 points.
You press <<O>> and get 52.0 points.
You press <<L>> and get 67.0 points.
You press <<L>> and get 71.0 points.
You press <<L>> and get 71.0 points.
You press <<L>> and get 69.0 points.
You press <<G>> and get 46.0 points.
You press <<O>> and get 47.0 points.
You press <<U>> and get 19.0 points.
You press <<L>> and get 63.0 points.
You press <<L>> and get 58.0 points.
You press <<L>> and get 62.0 points.
You press <<L>> and get 53.0 points.
You press <<L>> and get 59.0 points.
You press <<L>> and get 65.0 points.
You press <<G>> and get 58.0 points.
You press <<O>> and get 54.0 points.
You press <<L>> and get 61.0 points.
You press <<G>> and get 66.0 points.
You press <<O>> and get 62.0 points.
You press <<L>> and get 61.0 points.

36
You press <<G>> and get 56.0 points.
You press <<O>> and get 58.0 points.
You press <<L>> and get 49.0 points.
You press <<O>> and get 50.0 points.
You press <<G>> and get 63.0 points.
You press <<G>> and get 68.0 points.
You press <<G>> and get 56.0 points.
You press <<G>> and get 59.0 points.
You press <<G>> and get 66.0 points.
You press <<G>> and get 56.0 points.
You press <<G>> and get 56.0 points.
You press <<L>> and get 48.0 points.
You press <<G>> and get 59.0 points.
You press <<G>> and get 55.0 points.
You press <<O>> and get 51.0 points.
You press <<O>> and get 58.0 points.
You press <<O>> and get 51.0 points.
You press <<O>> and get 62.0 points.
You press <<O>> and get 64.0 points.
You press <<O>> and get 60.0 points.
You press <<O>> and get 62.0 points.
You press <<O>> and get 64.0 points.
You press <<O>> and get 50.0 points.
You press <<U>> and get 40.0 points.
You press <<O>> and get 50.0 points.
You press <<G>> and get 49.0 points.
You press <<L>> and get 33.0 points.
You press <<G>> and get 55.0 points.
You press <<O>> and get 78.0 points.
You press <<O>> and get 79.0 points.
You press <<O>> and get 85.0 points.
You press <<O>> and get 82.0 points.
You press <<O>> and get 83.0 po

Multiple-cue judgment
Data source: [97]

Number of experiments: 3
Number of participants: 232
Number of choices: 52464

Example prompt:
Your task is to estimate the blood concentration of the hormone Caldionine based on
information about the amount of two other hormones, Progladine and Amalydine, in
multiple individuals’ urine.

37
Both Progladine and Amalydine can take five values (very little, a little, average, a
lot, very much).
Caldionine can take nine values (extremely low, very low, low, somewhat low, normal,
somewhat high, high, very high, extremely high).
Your goal is to estimate the concentration of Caldionine correctly.
You will receive feedback about the actual concentration after making your estimate.
This feedback will stop at some point.

Progladine: a lot. Amalydine: very much. You say that the Caldionine concentration
is <<high>>. That is incorrect. The correct concentration of Caldionine is somewhat
low.
Progladine: average. Amalydine: a lot. You say that the Caldionine concentration
is <<somewhat low>>. That is correct. The correct concentration of Caldionine is
indeed somewhat low.
Progladine: a lot. Amalydine: average. You say that the Caldionine concentration is
<<normal>>. That is incorrect. The correct concentration of Caldionine is somewhat
high.
Progladine: average. Amalydine: a little. You say that the Caldionine concentration
is <<low>>. That is incorrect. The correct concentration of Caldionine is somewhat
high.
Progladine: a little. Amalydine: average. You say that the Caldionine concentration
is <<normal>>. That is incorrect. The correct concentration of Caldionine is some-
what low.
Progladine: a lot. Amalydine: a little. You say that the Caldionine concentration is
<<very high>>. That is incorrect. The correct concentration of Caldionine is high.
Progladine: very little. Amalydine: very little. You say that the Caldionine concen-
tration is <<normal>>. That is correct. The correct concentration of Caldionine is
indeed normal.
Progladine: very much. Amalydine: a little. You say that the Caldionine concentration
is <<very high>>. That is correct. The correct concentration of Caldionine is indeed
very high.
Progladine: a lot. Amalydine: very little. You say that the Caldionine concentration
is <<very high>>. That is correct. The correct concentration of Caldionine is indeed
very high.
Progladine: a little. Amalydine: a little. You say that the Caldionine concentration is
<<normal>>. That is correct. The correct concentration of Caldionine is indeed nor-
mal.
Progladine: a little. Amalydine: very little. You say that the Caldionine concentration
is <<somewhat high>>. That is correct. The correct concentration of Caldionine is
indeed somewhat high.
Progladine: very little. Amalydine: a little. You say that the Caldionine concentration
is <<somewhat low>>. That is correct. The correct concentration of Caldionine is
indeed somewhat low.
Progladine: very much. Amalydine: a lot. You say that the Caldionine concentration is

38
<<very high>>. That is incorrect. The correct concentration of Caldionine is some-
what high.
Progladine: a little. Amalydine: very much. You say that the Caldionine concentration
is <<very low>>. That is correct. The correct concentration of Caldionine is indeed
very low.
Progladine: average. Amalydine: very much. You say that the Caldionine concentra-
tion is <<low>>. That is correct. The correct concentration of Caldionine is indeed
low.
Progladine: very much. Amalydine: average. You say that the Caldionine concentration
is <<somewhat high>>. That is incorrect. The correct concentration of Caldionine
is high.
Progladine: average. Amalydine: very little. You say that the Caldionine concentra-
tion is <<high>>. That is correct. The correct concentration of Caldionine is indeed
high.
Progladine: a little. Amalydine: a lot. You say that the Caldionine concentration is
<<normal>>. That is incorrect. The correct concentration of Caldionine is low.
Progladine: very much. Amalydine: very much. You say that the Caldionine concen-
tration is <<normal>>. That is correct. The correct concentration of Caldionine is
indeed normal.
Progladine: a lot. Amalydine: a lot. You say that the Caldionine concentration is
<<normal>>. That is correct.

Recall and recognition

Data source: [98]

Number of experiments: 1
Number of participants: 424
Number of choices: 109201

Example prompt:
You study the following 20 word pairs:
FILE, GERMAN
STANDS, RISES
OFFICER, FUEL
CLASSROOM, JOURNEY
TERRITORY, EDUCATIONAL
TRANSPORTATION, AGREEMENT
SIGNIFICANT, SPECIALIZED
TUBE, ENEMY
CLUB, RAPID
LEGAL, CONCEPT
TOWARDS, VICTORY
RARELY, BAY
ELECTION, ROYAL
BELONG, SUPREME

39
FRIGHTENED, PLAINS
SLAVES, FILM
STOMACH, WHISPERED
NURSE, GOLDEN
POLITICS, WINGS
BOXES, TOM

You will now view a single pair of words.

Your task is to indicate if the pair of words you see on the screen was studied as a
pair on the list you just studied (by pressing D) or was not a pair (by pressing N).

You view the word pair SPECIALIZED, POLITICS. You press <<N>>.
You view the word pair TOWARDS, VICTORY. You press <<D>>.
You view the word pair SUPREME, BELONG. You press <<D>>.
You view the word pair WINGS, OFFICER. You press <<D>>.
You view the word pair CONCEPT, FRIGHTENED. You press <<N>>.
You view the word pair GERMAN, NURSE. You press <<N>>.
You view the word pair BAY, RARELY. You press <<D>>.
You view the word pair FILE, GOLDEN. You press <<N>>.
You view the word pair JOURNEY, LEGAL. You press <<N>>.
You view the word pair CLASSROOM, ENEMY. You press <<N>>.
You view the word pair ROYAL, ELECTION. You press <<D>>.
You view the word pair BOXES, TOM. You press <<D>>.
You view the word pair TERRITORY, EDUCATIONAL. You press <<D>>.
You view the word pair WHISPERED, STOMACH. You press <<D>>.
You view the word pair STANDS, RISES. You press <<D>>.
You view the word pair AGREEMENT, TUBE. You press <<N>>.
You view the word pair FUEL, TRANSPORTATION. You press <<N>>.
You view the word pair CLUB, RAPID. You press <<D>>.
You view the word pair SLAVES, FILM. You press <<D>>.
You view the word pair PLAINS, SIGNIFICANT. You press <<N>>.

You study the following 20 word pairs:

METALS, BEAUTY
SHORE, COLONISTS
HIGHEST, FASHION
INFLUENCED, FLEW
SHADOW, GRANDFATHER
MOTOR, DISAPPEARED
DETAIL, SHOULDERS
REPORTS, MALE
PAINT, COMMUNITIES
STORM, RESULTING
SELECTION, NIGHTS
APPLICATION, FELLOW

40
DESIGN, PRINCIPLE
POSSIBILITY, AFRICAN
REMOVE, TRAIL
STAFF, JUSTICE
SECTIONS, TRADITIONAL
CLAY, MIXED
TEA, AGRICULTURE
PATIENT, LIMIT

You will now view a single word item.

Your task is to type the other word in the pair.
For example if you studied BRICK, BRACK and you now see BRICK, your response
should be BRACK.
The spelling does not matter; focus on providing as many responses as possible.
If you cannot recall the word, answer DON’T REMEMBER.

You view the word STAFF. You answer <<JUSTICE>>.

You view the word MOTOR. You answer <<DON’T REMEMBER>>.
You view the word INFLUENCED. You answer <<DON’T REMEMBER>>.
You view the word DETAIL. You answer <<SHOULDER>>.
You view the word REMOVE. You answer <<DON’T REMEMBER>>.
You view the word PATIENT. You answer <<LIMITED>>.
You view the word PAINT. You answer <<COMMUNITY>>.
You view the word STORM. You answer <<RESULTS>>.
You view the word TEA. You answer <<ARIGCULTURE>>.
You view the word SHADOW. You answer <<GRANDFATHER>>.
You view the word APPLICATION. You answer <<FELLOW>>.
You view the word METALS. You answer <<BEAUTY>>.
You view the word POSSIBILITY. You answer <<AFRICANS>>.
You view the word DESIGN. You answer <<PRINCPLE>>.
You view the word SHORE. You answer <<DON’T REMEMBER>>.
You view the word HIGHEST. You answer <<DON’T REMEMBER>>.
You view the word CLAY. You answer <<MIXED>>.
You view the word REPORTS. You answer <<DON’T REMEMBER>>.
You view the word SECTIONS. You answer <<DON’T REMEMBER>>.
You view the word SELECTION. You answer <<DON’T REMEMBER>>.

You study the following 20 word pairs:

DEFENSE, EXPANSION
PROGRESS, PROMISE
EGG, ATTRACTIVE
CRYING, RECOGNIZE
PURE, AUNT
OBTAINED, FEATURE
TRAVELED, FLOWER

41
INCREASINGLY, FEMALE
POURED, VALUABLE
TIGHT, HOLDS
BIGGER, HONOR
DRUG, SPENDING
DOUBLE, OUTPUT
ESTABLISH, CRITICAL
DRIED, CHRISTMAS
PROCESSING, NUMEROUS
MEASURES, PARAGRAPH
AFFAIRS, MOVEMENTS
DAMAGE, FED
WARS, CONTRACT

You will now view a single word item.

Your task is to indicate if the item you view was on the list you just studied (by
pressing D) or not on the list (by pressing N).

You view the word EXPANSION. You press <<D>>.

N-back
Data source: [62]

Number of experiments: 1
Number of participants: 470
Number of choices: 200821

Example prompt:
You will view a stream of letters on the screen, one letter at a time.
At the beginning of a block, you are told a number N.
You have to remember the last N letters you saw since the beginning of the block.
If the letter you see matches the letter N trials ago, press W, otherwise press D.
The case of the letter is irrelevant, so ”t” matches ”T” for example.
If you make more than 5 mistakes in a block, N is decreased by 1.
If you make fewer than 3 mistakes in a block, N is increased by 1.
You will go through 20 blocks with 20+N trials each.

Block 0, N = 2:
You see the letter d.
You see the letter g.
You see the letter D and press <<W>>.
You see the letter v.
You see the letter B.
You see the letter V.
You see the letter d.

42
You see the letter g.
You see the letter D and press <<W>>.
You see the letter t.
You see the letter G.
You see the letter B and press <<D>>.
You see the letter G.
You see the letter g and press <<W>>.
You see the letter g and press <<W>>.
You see the letter v.
You see the letter V.
You see the letter B.
You see the letter G.
You see the letter G.
You see the letter G and press <<W>>.
You see the letter G.

Block 1, N = 1:
You see the letter g.
You see the letter D and press <<D>>.
You see the letter D.
You see the letter d and press <<W>>.
You see the letter D.
You see the letter d and press <<W>>.
You see the letter v.
You see the letter b and press <<D>>.
You see the letter V.
You see the letter t and press <<D>>.
You see the letter d.
You see the letter b and press <<D>>.
You see the letter G.
You see the letter t and press <<D>>.
You see the letter D.
You see the letter t and press <<D>>.
You see the letter t.
You see the letter b and press <<D>>.
You see the letter B.
You see the letter B and press <<W>>.
You see the letter v.

Block 2, N = 1:
You see the letter g.
You see the letter g and press <<W>>.
You see the letter d.
You see the letter d and press <<W>>.
You see the letter T.
You see the letter v.

43
You see the letter V.
You see the letter D and press <<W>>.
You see the letter g.
You see the letter G and press <<W>>.
You see the letter t.
You see the letter D and press <<D>>.
You see the letter t.
You see the letter g and press <<D>>.
You see the letter g.
You see the letter D and press <<W>>.
You see the letter d.
You see the letter V.
You see the letter B.
You see the letter v and press <<D>>.
You see the letter V.

Block 3, N = 1:
You see the letter T.
You see the letter d and press <<W>>.
You see the letter t.
You see the letter V and press <<D>>.
You see the letter v.
You see the letter B and press <<D>>.
You see the letter b.
You see the letter D and press <<D>>.
You see the letter d.
You see the letter d and press <<W>>.
You see the letter G.
You see the letter B and press <<D>>.
You see the letter G.
You see the letter d.
You see the letter G.
You see the letter v and press <<D>>.
You see the letter B.
You see the letter t and press <<D>>.
You see the letter T.
You see the letter T and press <<W>>.
You see the letter T.

Block 4, N = 1:
You see the letter d.
You see the letter d and press <<W>>.
You see the letter d.
You see the letter V and press <<D>>.
You see the letter V.

44
You see the letter v and press <<D>>.
You see the letter B.
You see the letter B and press <<W>>.
You see the letter g.
You see the letter B.
You see the letter t.
You see the letter D.
You see the letter d.
You see the letter B.
You see the letter B.
You see the letter t.
You see the letter d.
You see the letter G.
You see the letter d.
You see the letter v.
You see the letter t.

Block 5, N = 1:
You see the letter v.
You see the letter t and press <<D>>.
You see the letter D.
You see the letter V and press <<D>>.
You see the letter v.
You see the letter D and press <<D>>.
You see the letter D.
You see the letter B and press <<D>>.
You see the letter g.
You see the letter v and press <<D>>.
You see the letter d.
You see the letter g.
You see the letter G.
You see the letter D and press <<D>>.
You see the letter b.
You see the letter v.
You see the letter V.
You see the letter D and press <<D>>.
You see the

Digit span
Data source: [62]

Number of experiments: 1
Number of participants: 472
Number of choices: 97012

45
Example prompt:
You will view a series of digits and are then asked to recall them in the order you have
seen them by pressing the corresponding keys.
After having recalled all digits, please press ’S’ to indicate the end of your recalled
sequence.

The digits are the following: [4, 8, 2]

You press <<4>>.
You press <<8>>.
You press <<2>>.
You press <<S>>.

You will view a series of digits and are then asked to recall them in the order you have
seen them by pressing the corresponding keys.
After having recalled all digits, please press ’S’ to indicate the end of your recalled
sequence.

The digits are the following: [5, 2, 8, 5]

You press <<5>>.
You press <<2>>.
You press <<8>>.
You press <<5>>.
You press <<S>>.

You will view a series of digits and are then asked to recall them in the order you have
seen them by pressing the corresponding keys.
After having recalled all digits, please press ’S’ to indicate the end of your recalled
sequence.

The digits are the following: [9, 2, 8, 2, 5]

You press <<9>>.
You press <<2>>.
You press <<8>>.
You press <<2>>.
You press <<5>>.
You press <<S>>.

You will view a series of digits and are then asked to recall them in the order you have
seen them by pressing the corresponding keys.
After having recalled all digits, please press ’S’ to indicate the end of your recalled
sequence.

The digits are the following: [8, 1, 4, 8, 4, 9]

You press <<8>>.
You press <<1>>.

46
You press <<4>>.
You press <<8>>.
You press <<8>>.
You press <<1>>.
You press <<S>>.

You will view a series of digits and are then asked to recall them in the order you have
seen them by pressing the corresponding keys.
After having recalled all digits, please press ’S’ to indicate the end of your recalled
sequence.

The digits are the following: [5, 1, 7, 1, 6, 3]

You press <<7>>.
You press <<1>>.
You press <<5>>.
You press <<1>>.
You press <<6>>.
You press <<3>>.
You press <<S>>.

You will view a series of digits and are then asked to recall them in the order you have
seen them by pressing the corresponding keys.
After having recalled all digits, please press ’S’ to indicate the end of your recalled
sequence.

The digits are the following: [5, 9, 5, 9, 1]

You press <<5>>.
You press <<9>>.
You press <<5>>.
You press <<9>>.
You press <<1>>.
You press <<S>>.

You will view a series of digits and are then asked to recall them in the order you have
seen them by pressing the corresponding keys.
After having recalled all digits, please press ’S’ to indicate the end of your recalled
sequence.

The digits are the following: [7, 2, 5, 8, 2, 6]

You press <<7>>.
You press <<7>>.
You press <<2>>.
You press <<8>>.
You press <<5>>.
You press <<2>>.

47
You press <<S>>.

You will view a series of digits and are then asked to recall them in the order you have
seen them by pressing the corresponding keys.
After having recalled all digits, please press ’S’ to indicate the end of your recalled
sequence.

The digits are the following: [8, 5, 8, 3, 9, 4]

You press <<8>>.
You press <<5>>.
You press <<8>>.
You press <<3>>.
You press <<9>>.
You press <<4>>.
You press <<S>>.

You will view a series of digits and are then asked to recall them in the order you have
seen them by pressing the corresponding keys.
After having recalled all digits, please press ’S’ to indicate the end of your recalled
sequence.

The digits are the following: [9, 6, 9, 5, 1, 7, 1]

You press <<9>>.
You press <<6>>.
You press <<9>>.
You press <<5>>.
You press <<7>>.
You press <<1>>.
You press <<7>>.
You press <<S>>.

You will view a series of digits and are then asked to recall them in the order you have
seen them by pressing the corresponding keys.
After having recalled all digits, please press ’S’ to indicate the end of your recalled
sequence.

The digits are the following: [7, 2, 5, 9, 1, 8]

You press <<7>>.
You press <<2>>.
You press <<5>>.
You press <<9>>.
You press <<1>>.
You press <<8>>.
You press <<S>>.

48
You will view a series of digits and are then asked to recall them in the order you have
seen them by pressing the corresponding keys.
After having recalled all digits, please

Go/no-go
Data source: [62]

Number of experiments: 1
Number of participants: 463
Number of choices: 150517

Example prompt:
In this task, you need to emit responses to certain stimuli and omit responses to others.

You will see one of two colours, colour1 or colour2, on the screen in each trial.
You need to press button X when you see colour1 and press nothing when you see
colour2.
You need to respond as quickly as possible.
You will be doing 10 practice trials followed by 350 test trials.

You see colour1 and press nothing.

You see colour2 and press <<X>> in 753.0ms.
You see colour2 and press <<X>> in 381.0ms.
You see colour2 and press nothing.
You see colour1 and press <<X>> in 473.0ms.
You see colour1 and press <<X>> in 713.0ms.
You see colour2 and press nothing.
You see colour1 and press <<X>> in 364.0ms.
You see colour2 and press nothing.
You see colour1 and press <<X>> in 378.0ms.
You see colour1 and press <<X>> in 794.0ms.
You see colour1 and press <<X>> in 436.0ms.
You see colour1 and press <<X>> in 427.0ms.
You see colour1 and press <<X>> in 337.0ms.
You see colour1 and press <<X>> in 269.0ms.
You see colour1 and press <<X>> in 312.0ms.
You see colour1 and press <<X>> in 273.0ms.
You see colour2 and press nothing.
You see colour1 and press <<X>> in 288.0ms.
You see colour1 and press <<X>> in 276.0ms.
You see colour1 and press <<X>> in 314.0ms.
You see colour1 and press <<X>> in 309.0ms.
You see colour1 and press <<X>> in 320.0ms.
You see colour1 and press <<X>> in 342.0ms.
You see colour1 and press <<X>> in 301.0ms.

49
You see colour1 and press <<X>> in 289.0ms.
You see colour2 and press nothing.
You see colour1 and press <<X>> in 360.0ms.
You see colour2 and press <<X>> in 424.0ms.
You see colour2 and press nothing.
You see colour1 and press <<X>> in 525.0ms.
You see colour1 and press <<X>> in 306.0ms.
You see colour1 and press <<X>> in 387.0ms.
You see colour1 and press <<X>> in 292.0ms.
You see colour1 and press <<X>> in 317.0ms.
You see colour1 and press <<X>> in 270.0ms.
You see colour1 and press <<X>> in 278.0ms.
You see colour2 and press nothing.
You see colour1 and press <<X>> in 277.0ms.
You see colour1 and press <<X>> in 311.0ms.
You see colour1 and press <<X>> in 338.0ms.
You see colour1 and press <<X>> in 323.0ms.
You see colour1 and press <<X>> in 304.0ms.
You see colour1 and press <<X>> in 323.0ms.
You see colour1 and press <<X>> in 354.0ms.
You see colour1 and press <<X>> in 292.0ms.
You see colour1 and press <<X>> in 302.0ms.
You see colour1 and press <<X>> in 309.0ms.
You see colour2 and press nothing.
You see colour1 and press <<X>> in 340.0ms.
You see colour1 and press <<X>> in 603.0ms.
You see colour1 and press <<X>> in 289.0ms.
You see colour1 and press <<X>> in 284.0ms.
You see colour1 and press <<X>> in 275.0ms.
You see colour1 and press <<X>> in 299.0ms.
You see colour2 and press nothing.
You see colour1 and press <<X>> in 265.0ms.
You see colour1 and press <<X>> in 267.0ms.
You see colour2 and press nothing.
You see colour1 and press <<X>> in 274.0ms.
You see colour1 and press <<X>> in 382.0ms.
You see colour1 and press <<X>> in 272.0ms.
You see colour2 and press nothing.
You see colour1 and press <<X>> in 258.0ms.
You see colour1 and press <<X>> in 305.0ms.
You see colour1 and press <<X>> in 320.0ms.
You see colour1 and press <<X>> in 261.0ms.
You see colour1 and press <<X>> in 275.0ms.
You see colour2 and press nothing.
You see colour1 and press <<X>> in 424.0ms.

50
You see colour1 and press <<X>> in 266.0ms.
You see colour1 and press <<X>> in 273.0ms.
You see colour1 and press <<X>> in 287.0ms.
You see colour1 and press <<X>> in 437.0ms.
You see colour1 and press <<X>> in 293.0ms.
You see colour1 and press <<X>> in 297.0ms.
You see colour1 and press <<X>> in 308.0ms.
You see colour1 and press <<X>> in 313.0ms.
You see colour1 and press <<X>> in 373.0ms.
You see colour1 and press <<X>> in 390.0ms.
You see colour1 and press <<X>> in 304.0ms.
You see colour1 and press <<X>> in 334.0ms.
You see colour1 and press <<X>> in 326.0ms.
You see colour1 and press <<X>> in 382.0ms.
You see colour1 and press <<X>> in 803.0ms.
You see colour1 and press <<X>> in 430.0ms.
You see colour1 and press <<X>> in 324.0ms.
You see colour

Recent probes
Data source: [62]

Number of experiments: 1
Number of participants: 471
Number of choices: 34714

Example prompt:
You will repeatedly observe sequences of six letters.
You have to remember these letters before they disappear.
Afterward, you will be prompted with one letter. You have to answer whether the
letter was part of the six previous letters.
If you think it was, you have to press C. If you think it was not, press Q.

You are shown the letters [’C’, ’I’, ’Q’, ’F’, ’W’, ’Z’]. You see the letter Y. You press
<<Q>>.
You are shown the letters [’I’, ’Q’, ’C’, ’D’, ’M’, ’V’]. You see the letter U. You press
<<Q>>.
You are shown the letters [’I’, ’O’, ’C’, ’X’, ’A’, ’Q’]. You see the letter M. You press
<<C>>.
You are shown the letters [’Z’, ’C’, ’W’, ’I’, ’J’, ’O’]. You see the letter C. You press
<<Q>>.
You are shown the letters [’Q’, ’M’, ’F’, ’V’, ’P’, ’E’]. You see the letter W. You press
<<C>>.
You are shown the letters [’W’, ’F’, ’U’, ’M’, ’B’, ’Q’]. You see the letter V. You press
<<Q>>.

51
You are shown the letters [’R’, ’U’, ’F’, ’J’, ’W’, ’D’]. You see the letter W. You press
<<C>>.
You are shown the letters [’X’, ’U’, ’R’, ’Y’, ’H’, ’F’]. You see the letter X. You press
<<Q>>.
You are shown the letters [’R’, ’Q’, ’M’, ’X’, ’V’, ’U’]. You see the letter W. You press
<<C>>.
You are shown the letters [’G’, ’Q’, ’M’, ’N’, ’R’, ’O’]. You see the letter V. You press
<<Q>>.
You are shown the letters [’T’, ’P’, ’Q’, ’M’, ’W’, ’G’]. You see the letter X. You press
<<Q>>.
You are shown the letters [’P’, ’J’, ’Q’, ’S’, ’D’, ’T’]. You see the letter J. You press
<<Q>>.
You are shown the letters [’J’, ’R’, ’H’, ’Q’, ’F’, ’P’]. You see the letter F. You press
<<C>>.
You are shown the letters [’B’, ’V’, ’J’, ’G’, ’R’, ’H’]. You see the letter R. You press
<<Q>>.
You are shown the letters [’X’, ’J’, ’V’, ’L’, ’B’, ’D’]. You see the letter B. You press
<<Q>>.
You are shown the letters [’N’, ’M’, ’J’, ’C’, ’V’, ’X’]. You see the letter C. You press
<<Q>>.
You are shown the letters [’N’, ’T’, ’J’, ’R’, ’M’, ’W’]. You see the letter C. You press
<<Q>>.
You are shown the letters [’N’, ’T’, ’J’, ’E’, ’I’, ’D’]. You see the letter N. You press
<<Q>>.
You are shown the letters [’J’, ’T’, ’N’, ’K’, ’C’, ’B’]. You see the letter T. You press
<<C>>.
You are shown the letters [’M’, ’O’, ’N’, ’T’, ’P’, ’J’]. You see the letter O. You press
<<Q>>.
You are shown the letters [’O’, ’Q’, ’W’, ’U’, ’M’, ’N’]. You see the letter Q. You press
<<C>>.
You are shown the letters [’Q’, ’Z’, ’Y’, ’O’, ’I’, ’W’]. You see the letter I. You press
<<C>>.
You are shown the letters [’Z’, ’A’, ’Y’, ’F’, ’Q’, ’G’]. You see the letter Q. You press
<<C>>.
You are shown the letters [’Z’, ’M’, ’Y’, ’P’, ’A’, ’B’]. You see the letter X. You press
<<Q>>.
You are shown the letters [’L’, ’X’, ’M’, ’Z’, ’Y’, ’N’]. You see the letter B. You press
<<Q>>.
You are shown the letters [’M’, ’J’, ’X’, ’C’, ’L’, ’U’]. You see the letter P. You press
<<Q>>.
You are shown the letters [’S’, ’J’, ’E’, ’H’, ’X’, ’M’]. You see the letter M. You press
<<C>>.
You are shown the letters [’F’, ’I’, ’E’, ’A’, ’J’, ’S’]. You see the letter X. You press
<<Q>>.
You are shown the letters [’D’, ’U’, ’I’, ’M’, ’O’, ’T’]. You see the letter Q. You press

52
<<Q>>.
You are shown the letters [’I’, ’D’, ’U’, ’G’, ’Q’, ’W’]. You see the letter B. You press
<<Q>>.
You are shown the letters [’U’, ’I’, ’B’, ’L’, ’D’, ’Z’]. You see the letter W. You press
<<Q>>.
You are shown the letters [’U’, ’E’, ’B’, ’I’, ’H’, ’R’]. You see the letter I. You press
<<C>>.
You are shown the letters [’S’, ’B’, ’U’, ’E’, ’F’, ’N’]. You see the letter I. You press
<<Q>>.
You are shown the letters [’Y’, ’S’, ’Q’, ’U’, ’B’, ’A’]. You see the letter Y. You press
<<C>>.
You are shown the letters [’Q’, ’V’, ’Y’, ’S’, ’H’, ’O’]. You see the letter H. You press
<<C>>.
You are shown the letters [’V’, ’Z’, ’P’, ’D’, ’Q’, ’Y’]. You see the letter H. You press
<<C>>.
You are shown the letters [’Z’, ’P’, ’V’, ’W’, ’L’, ’N’]. You see the letter H. You press
<<Q>>.
You are shown the letters [’U’, ’S’, ’Z’, ’V’, ’P’, ’M’]. You see the letter W. You press
<<Q>>.
You are shown the letters [’U’, ’Y’, ’Z’, ’S’, ’C’, ’G’]. You see the letter W. You press
<<Q>>.

Horizon task
Data source: [99]

Number of experiments: 1
Number of participants: 26
Number of choices: 29120

Example prompt:
You are participating in multiple games involving two slot machines, labeled I and H.
The two slot machines are different across different games.
Each time you choose a slot machine, you get some points.
You choose a slot machine by pressing the corresponding key.
Each slot machine tends to pay out about the same amount of points on average.
Your goal is to choose the slot machines that will give you the most points across the
experiment.
The first 4 trials in each game are instructed trials where you will be told which slot
machine to choose.
After these instructed trials, you will have the freedom to choose for either 1 or 6
trials.

Game 1. There are 10 trials in this game.

You are instructed to press I and get 73 points.
You are instructed to press H and get 91 points.

53
You are instructed to press I and get 68 points.
You are instructed to press H and get 95 points.
You press <<I>> and get 71 points.
You press <<H>> and get 96 points.
You press <<I>> and get 57 points.
You press <<I>> and get 30 points.
You press <<I>> and get 54 points.
You press <<H>> and get 81 points.

Game 2. There are 10 trials in this game.

You are instructed to press I and get 38 points.
You are instructed to press I and get 1 points.
You are instructed to press I and get 18 points.
You are instructed to press H and get 44 points.
You press <<I>> and get 5 points.
You press <<I>> and get 3 points.
You press <<H>> and get 53 points.
You press <<I>> and get 6 points.
You press <<H>> and get 37 points.
You press <<H>> and get 50 points.

Game 3. There are 5 trials in this game.

You are instructed to press H and get 24 points.
You are instructed to press H and get 34 points.
You are instructed to press I and get 68 points.
You are instructed to press I and get 53 points.
You press <<I>> and get 57 points.

Game 4. There are 10 trials in this game.

You are instructed to press H and get 53 points.
You are instructed to press I and get 57 points.
You are instructed to press H and get 45 points.
You are instructed to press H and get 51 points.
You press <<H>> and get 37 points.
You press <<I>> and get 35 points.
You press <<I>> and get 51 points.
You press <<I>> and get 37 points.
You press <<I>> and get 29 points.
You press <<H>> and get 53 points.

Game 5. There are 5 trials in this game.

You are instructed to press H and get 46 points.
You are instructed to press I and get 9 points.
You are instructed to press H and get 38 points.
You are instructed to press H and get 36 points.

54
You press <<H>> and get 37 points.

Game 6. There are 5 trials in this game.

You are instructed to press I and get 35 points.
You are instructed to press I and get 36 points.
You are instructed to press I and get 27 points.
You are instructed to press H and get 49 points.
You press <<H>> and get 41 points.

Game 7. There are 5 trials in this game.

You are instructed to press H and get 53 points.
You are instructed to press H and get 59 points.
You are instructed to press I and get 80 points.
You are instructed to press I and get 99 points.
You press <<H>> and get 54 points.

Game 8. There are 10 trials in this game.

You are instructed to press H and get 21 points.
You are instructed to press I and get 43 points.
You are instructed to press H and get 18 points.
You are instructed to press H and get 19 points.
You press <<H>> and get 26 points.
You press <<H>> and get 21 points.
You press <<I>> and get 47 points.
You press <<I>> and get 37 points.
You press <<I>> and get 37 points.
You press <<I>> and get 36 points.

Game 9. There are 5 trials in this game.

You are instructed to press H and get 47 points.
You are instructed to press H and get 41 points.
You are instructed to press I and get 40 points.
You are instructed to press I and get 35 points.
You press <<I>> and get 30 points.

Game 10. There are 5 trials in this game.

You are instructed to press I and get 69 points.
You are instructed to press I and get 69 points.
You are instructed to press H and get 65 points.
You are instructed to press I and get 68 points.
You press <<I>> and get 63 points.

Game 11. There are 10

55
Gardening task
Data source: [100]

Number of experiments: 1
Number of participants: 320
Number of choices: 192000

Example prompt:
You are going to plant trees in two different gardens labeled North and South.
The trees look different from each other regarding their leafiness and branchiness.
There are 5 levels of leafiness (0, 1, 2, 3, 4) and 5 levels of branchiness (0, 1, 2, 3, 4).
In each round, you get presented with a tree.
You can accept to plant the tree by pressing T and reject to plant it by pressing N.
If you accept to plant the tree and your answer is correct, you will be rewarded with
points, otherwise, you will lose some points.
If you reject to plant the tree, you will not be rewarded (0 points).
Your task is to learn which type of tree grows best in each garden.
During the training phase, there will be feedback on every trial about your decisions.
During the testing phase, there will be no feedback for your decision.

You get a tree with level 3 of leafiness and level 0 of branchiness in the South garden.
You press <<T>> and get -50 points. You would have gotten 0 points, had you
rejected to plant the tree.
You get a tree with level 4 of leafiness and level 0 of branchiness in the North garden.
You press <<T>> and get -50 points. You would have gotten 0 points, had you
rejected to plant the tree.
You get a tree with level 1 of leafiness and level 1 of branchiness in the South garden.
You press <<T>> and get -25 points. You would have gotten 0 points, had you
rejected to plant the tree.
You get a tree with level 3 of leafiness and level 1 of branchiness in the North garden.
You press <<T>> and get -25 points. You would have gotten 0 points, had you
rejected to plant the tree.
You get a tree with level 0 of leafiness and level 4 of branchiness in the North garden.
You press <<T>> and get 50 points. You would have gotten 0 points, had you rejected
to plant the tree.
You get a tree with level 2 of leafiness and level 2 of branchiness in the North garden.
You press <<N>> and get 0 points. You would have gotten 0 points, had you accepted
to plant the tree.
You get a tree with level 0 of leafiness and level 2 of branchiness in the South garden.
You press <<T>> and get 0 points. You would have gotten 0 points, had you rejected
to plant the tree.
You get a tree with level 4 of leafiness and level 3 of branchiness in the North garden.
You press <<N>> and get 0 points. You would have gotten -50 points, had you
accepted to plant the tree.
You get a tree with level 1 of leafiness and level 1 of branchiness in the North garden.

56
You press <<T>> and get 25 points. You would have gotten 0 points, had you rejected
to plant the tree.
You get a tree with level 2 of leafiness and level 1 of branchiness in the South garden.
You press <<N>> and get 0 points. You would have gotten -25 points, had you
accepted to plant the tree.
You get a tree with level 4 of leafiness and level 4 of branchiness in the North garden.
You press <<N>> and get 0 points. You would have gotten -50 points, had you
accepted to plant the tree.
You get a tree with level 1 of leafiness and level 3 of branchiness in the South garden.
You press <<N>> and get 0 points. You would have gotten 25 points, had you
accepted to plant the tree.
You get a tree with level 0 of leafiness and level 2 of branchiness in the South garden.
You press <<N>> and get 0 points. You would have gotten 0 points, had you accepted
to plant the tree.
You get a tree with level 4 of leafiness and level 1 of branchiness in the South garden.
You press <<T>> and get -25 points. You would have gotten 0 points, had you
rejected to plant the tree.
You get a tree with level 4 of leafiness and level 0 of branchiness in the North garden.
You press <<N>> and get 0 points. You would have gotten -50 points, had you
accepted to plant the tree.
You get a tree with level 1 of leafiness and level 1 of branchiness in the South garden.
You press <<T>> and get -25 points. You would have gotten 0 points, had you
rejected to plant the tree.
You get a tree with level 0 of leafiness and level 2 of branchiness in the South garden.
You press <<N>> and get 0 points. You would have gotten 0 points, had you accepted
to plant the tree.
You get a

Columbia card task

Data source: [61]

Number of experiments: 1
Number of participants: 1368
Number of choices: 613299

Example prompt:
You will play a games with 84 rounds.
In each round, you will be presented with 32 face-down cards.
Every card is either a gain card or a loss card.
If you turn over a gain card, the gain amount of that card (between 10 and 600 points)
will be added to your current game score.
If you turn over a loss card, the loss amount of that card (between 25 and 750 points)
will be subtracted from your game score.
In different rounds, between 1 and 28 cards are loss cards.
Loss and gain amounts also differ between rounds.

57
You may keep turning over cards as long as you keep encountering gain cards.
You may also stop the round at any point and claim your current payout.
If you encounter a loss card, the round ends immediately.
Your gains and losses will be summed up to give you your final score for each round.
Press E to turn a card over, or C to stop the round and claim your current payout.

Round 1:
You will be awarded 150 points for turning over a gain card.
You will lose 75 points for turning over a loss card.
There are 20 loss cards in this round.
You press <<E>> and turn over a loss card. Your current score is -75. The round
has now ended because you encountered a loss card.
Your final score for this round is -75.

Round 2:
You will be awarded 50 points for turning over a gain card.
You will lose 100 points for turning over a loss card.
There are 1 loss cards in this round.
You press <<E>> and turn over a gain card. Your current score is 50.
You press <<E>> and turn over a gain card. Your current score is 100.
You press <<E>> and turn over a gain card. Your current score is 150.
You press <<E>> and turn over a gain card. Your current score is 200.
You press <<E>> and turn over a loss card. Your current score is 100. The round
has now ended because you encountered a loss card.
Your final score for this round is 100.

Round 3:
You will be awarded 200 points for turning over a gain card.
You will lose 100 points for turning over a loss card.
There are 10 loss cards in this round.
You press <<E>> and turn over a gain card. Your current score is 200.
You press <<E>> and turn over a gain card. Your current score is 400.
You press <<E>> and turn over a gain card. Your current score is 600.
You press <<C>> and claim your payout.
Your final score for this round is 600.

Round 4:
You will be awarded 200 points for turning over a gain card.
You will lose 50 points for turning over a loss card.
There are 28 loss cards in this round.
You press <<C>> and claim your payout.
Your final score for this round is 0.

Round 5:
You will be awarded 20 points for turning over a gain card.

58
You will lose 750 points for turning over a loss card.
There are 1 loss cards in this round.
You press <<E>> and turn over a gain card. Your current score is 20.
You press <<E>> and turn over a gain card. Your current score is 40.
You press <<E>> and turn over a gain card. Your current score is 60.
You press <<E>> and turn over a gain card. Your current score is 80.
You press <<E>> and turn over a gain card. Your current score is 100.
You press <<C>> and claim your payout.
Your final score for this round is 100.

Round 6:
You will be awarded 300 points for turning over a gain card.
You will lose 100 points for turning over a loss card.
There are 16 loss cards in this round.
You press <<E>> and turn over a loss card. Your current score is -100. The round
has now ended because you encountered a loss card.
Your final score for this round is -100.

Round 7:
You will be awarded 10 points for turning over a gain card.
You will lose 500 points for turning over a loss card.
There are 3 loss cards in this round.
You press <<E>> and turn over a gain card. Your current score is 10.
You press <<E>> and turn over a gain card. Your current score is 20.
You press <<E>> and turn over a gain card. Your current score is 30.
You press <<E>> and turn over a gain card. Your current score is 40.
You press <<C>> and claim your payout.
Your final score for this round is 40.

Round 8:
You will be awarded 10 points for turning over a gain card.
You will lose 250 points for turning over a loss

Balloon analog risk task

Data source: [61]

Number of experiments: 1
Number of participants: 1331
Number of choices: 1496974

Example prompt:
Throughout the task, you will be presented with balloons, one at a time.
In each step, you can choose to pump up the balloon by pressing H and you will accu-
mulate 1 point for each pump.

59
At any point, you can stop pumping up the balloon by pressing W and you will col-
lect your accumulated points.
You will repeat this procedure on multiple different balloons.
It is your choice to determine how much to pump up the balloon, but be aware that
at some point the balloon will explode.
If the balloon explodes before you collect your accumulated points, then you move on
to the next balloon and the points are lost.

Balloon 1:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>>. The balloon was inflated too much
and explodes.

Balloon 2:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<W>>. You stop
inflating the balloon and get 60 points.

Balloon 3:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>>. The balloon was inflated too much
and explodes.

Balloon 4:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>>. The balloon was inflated too much and
explodes.

Balloon 5:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>>. The balloon was inflated too much and explodes.

Balloon 6:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>

60
<<H>> <<W>>. You stop inflating the balloon and get 18 points.

Balloon 7:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>>. The balloon was inflated too much and explodes.

Balloon 8:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<W>>. You stop inflat-
ing the balloon and get 23 points.

Balloon 9:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<W>>. You stop inflating the balloon and get 26 points.

Balloon 10:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<W>>. You stop inflating the balloon and get 35 points.

Balloon 11:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<W>>. You stop inflating the balloon and get 35 points.

Balloon 12:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<W>>. You stop inflat-
ing the balloon and get 50 points.

Balloon 13:
You press <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>

61
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>> <<H>>
<<H>> <<H>> <<H>> <<H>>

Experiential-symbolic task
Data source: [101]

Number of experiments: 4
Number of participants: 346
Number of choices: 70608

Example prompt:
This experiment is composed of three parts.
In each round of the first part, you have to choose between one of two options repre-
sented by letters.
In a given pair, one option is, on average, more advantageous compared to the other.
You can win or lose the following outcomes: 1.0 points and -1.0 points.
In the second part, there will be two types of options.
The first type of option is represented by the letters you already encountered during
the previous part.
Note that the options maintain the same odds of winning or losing a point as in the
first part.
The second type of option is represented by an explicit description of the odds of win-
ning or losing a point.
In each round of the third part, you will be presented with the options you met in the
first and the second part.
This is the occasion to test your knowledge of each options’s outcome.
You will be asked to indicate (in percentages), what are the odds that a given option
makes you win a point.
You can choose an option by pressing its corresponding key.
Your goal for the first two parts is to maximize the amount of received points.
Your goal in the third part is to guess as accurately as possible.

You can choose between option Q and option L. You press <<Q>> and get 1.0 points.
You would have gotten 1.0 points had you chosen option L instead.
You can choose between option Z and option H. You press <<Z>> and get 1.0 points.
You would have gotten 1.0 points had you chosen option H instead.
You can choose between option D and option X. You press <<X>> and get 1.0 points.
You would have gotten -1.0 points had you chosen option D instead.
You can choose between option U and option C. You press <<C>> and get 1.0 points.
You would have gotten 1.0 points had you chosen option U instead.
You can choose between option U and option C. You press <<C>> and get -1.0
points. You would have gotten 1.0 points had you chosen option U instead.
You can choose between option D and option X. You press <<X>> and get 1.0 points.

62
You would have gotten 1.0 points had you chosen option D instead.
You can choose between option Z and option H. You press <<Z>> and get 1.0 points.
You would have gotten -1.0 points had you chosen option H instead.
You can choose between option Z and option H. You press <<Z>> and get 1.0 points.
You would have gotten 1.0 points had you chosen option H instead.
You can choose between option D and option X. You press <<X>> and get -1.0
points. You would have gotten 1.0 points had you chosen option D instead.
You can choose between option D and option X. You press <<D>> and get 1.0 points.
You would have gotten 1.0 points had you chosen option X instead.
You can choose between option Q and option L. You press <<Q>> and get 1.0 points.
You would have gotten -1.0 points had you chosen option L instead.
You can choose between option D and option X. You press <<X>> and get -1.0
points. You would have gotten 1.0 points had you chosen option D instead.
You can choose between option D and option X. You press <<D>> and get 1.0 points.
You would have gotten -1.0 points had you chosen option X instead.
You can choose between option Q and option L. You press <<Q>> and get -1.0
points. You would have gotten -1.0 points had you chosen option L instead.
You can choose between option Q and option L. You press <<L>> and get -1.0
points. You would have gotten 1.0 points had you chosen option Q instead.
You can choose between option Q and option L. You press <<Q>> and get 1.0 points.
You would have gotten -1.0 points had you chosen option L instead.
You can choose between option Q and option L. You press <<Q>> and get -1.0
points. You would have gotten 1.0 points had you chosen option L instead.
You can choose between option U and option C. You press <<C>> and get -1.0
points. You would have gotten 1.0 points had you chosen option U instead.
You can choose between option U and option C. You press <<C>> and get -1.0
points. You would have gotten -1.0 points had you chosen option U instead.
You can choose between option Q and option L. You press <<Q>> and get 1.0 poi

Two-armed bandit
Data source: [102]

Number of experiments: 2
Number of participants: 80
Number of choices: 16000

Example prompt:
In this task, you have to repeatedly choose between two slot machines labeled U and
P.
You can choose a slot machine by pressing its corresponding key.
When you select one of the machines, you will win or lose points.
Machine U will not always give you the same points when you select it again, but
machine P will always give 0 points when you select it.
Your goal is to choose the slot machines that will give you the most points.
You will receive feedback about the outcome after making a choice.

63
You will play 20 games in total, each with a different pair of slot machines.
Each game will consist of 10 trials.

Game 1:
You press <<U>> and get -1 points.
You press <<U>> and get 0 points.
You press <<U>> and get 2 points.
You press <<U>> and get -1 points.
You press <<U>> and get 1 points.
You press <<U>> and get -1 points.
You press <<U>> and get -1 points.
You press <<U>> and get -1 points.
You press <<U>> and get 1 points.
You press <<U>> and get 1 points.

Game 2:
You press <<U>> and get -1 points.
You press <<U>> and get -1 points.
You press <<U>> and get -2 points.
You press <<U>> and get 0 points.
You press <<U>> and get -1 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.

Game 3:
You press <<U>> and get -2 points.
You press <<U>> and get -1 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<U>> and get -4 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.

Game 4:
You press <<U>> and get 2 points.
You press <<U>> and get 0 points.
You press <<U>> and get 2 points.
You press <<U>> and get 1 points.
You press <<U>> and get 2 points.

64
You press <<U>> and get -1 points.
You press <<U>> and get 0 points.
You press <<U>> and get 2 points.
You press <<U>> and get 1 points.
You press <<U>> and get 0 points.

Game 5:
You press <<U>> and get 0 points.
You press <<U>> and get 1 points.
You press <<U>> and get 1 points.
You press <<U>> and get 0 points.
You press <<U>> and get 2 points.
You press <<U>> and get 1 points.
You press <<U>> and get 2 points.
You press <<U>> and get 1 points.
You press <<U>> and get 2 points.
You press <<U>> and get 1 points.

Game 6:
You press <<U>> and get 3 points.
You press <<U>> and get 2 points.
You press <<U>> and get 0 points.
You press <<U>> and get 2 points.
You press <<U>> and get 2 points.
You press <<U>> and get 2 points.
You press <<U>> and get 1 points.
You press <<U>> and get 3 points.
You press <<U>> and get 2 points.
You press <<U>> and get 1 points.

Game 7:
You press <<U>> and get -1 points.
You press <<U>> and get -3 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.

Game 8:
You press <<U>> and get 0 points.
You press <<U>> and get -1 points.

65
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.
You press <<P>> and get 0 points.

Game 9:
You press <<U>> and get 2 points.
You press <<U>> and get 3 points.
You press <<U>> and get 4 points.
You press <<U>> and get 1 points.
You press <<U>> and get 1 points.
You press <<U>> and get 2 points.
You press <<U>> and get 2 points.
You press <<U>> and get 2 points.
You press <<U>> and get -1 points.
You press <<U>> and get 2 points.

Game 10:
You press <<U>> and get 1 points.
You press <<U>> and get 0 points.
You press <<U>> and get 0 points.
You press <<U>> and get 1 points.
You press <<U>> and get -1 points.
You press <<U>> and get -2 points.
You press <<U>> and get 1 points.
You press <<U>> and get 2 points.
You press <<U>> and get 1 points.
You pr

Conditional associative learning

Data source: [103]

Number of experiments: 1
Number of participants: 74
Number of choices: 40539

Example prompt:
You are presented with a series of stimuli, each associated with one of three possible
responses.
Your goal is to learn which response is the correct one for each stimulus.
When a stimulus is presented, you can press one of three keys to respond.

66
The three responses available are S, F, and A.
After your response, you will receive feedback: 1 point for a correct response, or 0
points for an incorrect response.
The correct response for one stimulus does not inform you about the correct response
for another stimulus.
You will play 13 games, each with a different mapping from stimuli to responses.

Game 1:
There are 6 different stimuli.
You see stimulus 1. You press <<S>> and get 0 points.
You see stimulus 0. You press <<F>> and get 0 points.
You see stimulus 4. You press <<A>> and get 1 points.
You see stimulus 5. You press <<S>> and get 0 points.
You see stimulus 3. You press <<F>> and get 1 points.
You see stimulus 3. You press <<F>> and get 1 points.
You see stimulus 4. You press <<A>> and get 1 points.
You see stimulus 2. You press <<S>> and get 0 points.
You see stimulus 1. You press <<S>> and get 0 points.
You see stimulus 5. You press <<S>> and get 0 points.
You see stimulus 2. You press <<S>> and get 0 points.
You see stimulus 0. You press <<A>> and get 0 points.
You see stimulus 3. You press <<F>> and get 1 points.
You see stimulus 3. You press <<F>> and get 1 points.
You see stimulus 5. You press <<A>> and get 1 points.
You see stimulus 1. You press <<A>> and get 0 points.
You see stimulus 4. You press <<A>> and get 1 points.
You see stimulus 1. You press <<S>> and get 0 points.
You see stimulus 0. You press <<S>> and get 1 points.
You see stimulus 5. You press <<S>> and get 0 points.
You see stimulus 4. You press <<A>> and get 1 points.
You see stimulus 2. You press <<S>> and get 0 points.
You see stimulus 0. You press <<S>> and get 1 points.
You see stimulus 2. You press <<S>> and get 0 points.
You see stimulus 3. You press <<F>> and get 1 points.
You see stimulus 1. You press <<A>> and get 0 points.
You see stimulus 5. You press <<A>> and get 1 points.
You see stimulus 4. You press <<A>> and get 1 points.
You see stimulus 0. You press <<S>> and get 1 points.
You see stimulus 1. You press <<S>> and get 0 points.
You see stimulus 3. You press <<F>> and get 1 points.
You see stimulus 4. You press <<A>> and get 1 points.
You see stimulus 0. You press <<S>> and get 1 points.
You see stimulus 2. You press <<S>> and get 0 points.
You see stimulus 2. You press <<S>> and get 0 points.
You see stimulus 3. You press <<F>> and get 1 points.

67
You see stimulus 5. You press <<S>> and get 0 points.
You see stimulus 1. You press <<A>> and get 0 points.
You see stimulus 0. You press <<F>> and get 0 points.
You see stimulus 5. You press <<F>> and get 0 points.
You see stimulus 4. You press <<A>> and get 1 points.
You see stimulus 4. You press <<A>> and get 1 points.
You see stimulus 1. You press <<A>> and get 0 points.
You see stimulus 3. You press <<F>> and get 1 points.
You see stimulus 2. You press <<F>> and get 1 points.
You see stimulus 2. You press <<F>> and get 1 points.
You see stimulus 3. You press <<F>> and get 1 points.
You see stimulus 0. You press <<F>> and get 0 points.
You see stimulus 1. You press <<F>> and get 1 points.
You see stimulus 5. You press <<F>> and get 0 points.
You see stimulus 0. You press <<F>> and get 0 points.
You see stimulus 5. You press <<F>> and get 0 points.
You see stimulus 4. You press <<A>> and get 1 points.
You see stimulus 2. You press <<A>> and get 0 points.
You see stimulus 0. You press <<F>> and get 0 points.
You see stimulus 0. You press <<F>> and get 0 points.
You see stimulus 1. You press <<A>> and get 0 points.
You see stimulus 2. You press <<A>> and get 0 points.
You see stimulus 1. You press <<A>> and get 0 points.
You see stimulus 4. You press <<A>> and get 1 points.
You see stimulus 4. You press <<A>> and get 1 points.
You see stimulus 2. You press <<S>> and get 0 points.
You see stimulus 5. You press <<S>> and get 0 points.
You see stimulus 5. You press <<F>> and get 0 points.

THINGS odd-one-out
Data source: [104]

Number of experiments: 1
Number of participants: 11122
Number of choices: 2611240

Example prompt:
You will be presented with triplets of objects, which will be assigned to the keys B, J,
and K.
In each trial, please indicate which object you think is the odd one out by pressing
the corresponding key.
In other words, please choose the object that is the least similar to the other two.

B: prune, J: nail polish, and K: diskette. You press <<K>>.

B: ladle, J: water bottle, and K: pug. You press <<K>>.

68
B: punch, J: hair, and K: lollipop. You press <<J>>.
B: oar, J: mug, and K: macaroni. You press <<B>>.
B: towel, J: hot tub, and K: mallet. You press <<K>>.
B: train set, J: hot-water bottle, and K: treasure. You press <<J>>.
B: shutter, J: cleaver, and K: toe. You press <<K>>.
B: straw, J: sweeper, and K: baton. You press <<J>>.
B: footprint, J: beehive, and K: skunk. You press <<B>>.
B: cash machine, J: thermostat, and K: mandolin. You press <<K>>.
B: orange, J: throne, and K: stir fry. You press <<J>>.
B: boy, J: burrito, and K: microscope. You press <<J>>.
B: pheasant, J: sponge, and K: orchid. You press <<J>>.
B: forklift, J: clipper, and K: hip. You press <<J>>.
B: candelabra, J: beard, and K: glue. You press <<J>>.
B: raccoon, J: hammer, and K: roulette wheel. You press <<B>>.
B: wing, J: beanie, and K: girl. You press <<B>>.
B: piggy bank, J: footrest, and K: sandal. You press <<B>>.
B: knee, J: cornhusk, and K: tuning fork. You press <<B>>.
B: anklet, J: bedpost, and K: ice cube. You press <<B>>.
B: mannequin, J: stove, and K: coin. You press <<K>>.
B: tortellini, J: cantaloupe, and K: sequin. You press <<K>>.
B: coffee filter, J: fingerprint, and K: rose. You press <<B>>.
B: porcupine, J: christmas tree, and K: corkscrew. You press <<B>>.
B: freezer, J: coat rack, and K: puffin. You press <<K>>.
B: maggot, J: mouth, and K: stockings. You press <<B>>.
B: soap, J: hot-water bottle, and K: knitting needle. You press <<K>>.
B: mosquito net, J: baklava, and K: beanbag. You press <<J>>.
B: skewer, J: baklava, and K: propeller. You press <<K>>.
B: nail polish, J: goose, and K: pizza. You press <<J>>.
B: face mask, J: cinnamon, and K: toilet paper. You press <<B>>.
B: bag, J: eel, and K: trampoline. You press <<J>>.
B: lightbulb, J: moose, and K: curling iron. You press <<J>>.
B: dumbwaiter, J: jigsaw puzzle, and K: lamb. You press <<K>>.
B: eyeliner, J: shopping basket, and K: flipper. You press <<K>>.
B: bowtie, J: wooden leg, and K: kangaroo. You press <<K>>.
B: puffin, J: wire cutters, and K: battery. You press <<B>>.
B: toilet, J: rack, and K: star fruit. You press <<K>>.
B: stingray, J: cork, and K: fire pit. You press <<B>>.
B: bun, J: snow, and K: tinsel. You press <<J>>.
B: almond, J: trailer, and K: paper. You press <<B>>.
B: anteater, J: chalice, and K: wedge. You press <<B>>.
B: scone, J: pie, and K: ant. You press <<K>>.
B: heater, J: aircraft carrier, and K: joystick. You press <<B>>.
B: comic book, J: playing card, and K: organ. You press <<K>>.
B: flag, J: measuring cup, and K: strawberry. You press <<B>>.
B: bunkbed, J: tractor, and K: windshield. You press <<B>>.

69
B: aircraft carrier, J: prism, and K: turban. You press <<K>>.
B: egg, J: scrambled egg, and K: doorknocker. You press <<K>>.
B: microscope, J: stained glass, and K: strainer. You press <<J>>.
B: polygraph, J: hairdryer, and K: harness. You press <<J>>.
B: chip, J: iguana, and K: hedge. You press <<B>>.
B: tick, J: binder, and K: shoelace. You press <<B>>.
B: nut, J: yogurt, and K: jug. You press <<K>>.
B: hairpin, J: giraffe, and K: fur coat. You press <<B>>.
B: lasagna, J: statue, and K: bookshelf. You press <<B>>.
B: grill, J: catapult, and K: sonogram. You press <<K>>.
B: fondue, J: pill, and K: firewood. You press <<K>>.
B: sunroof, J: onion, and K: flan. You press <<B>>.
B: wreck, J: bungee, and K: cockroach. You press <<K>>.

Multi-attribute decision-making
Data source: [46]

Number of experiments: 1
Number of participants: 73
Number of choices: 7008

Example prompt:
You are repeatedly presented with two options, labeled A and R.
Each option represents a fictitious product and you have to infer which product is
superior in terms of quality.
You select a product by pressing the corresponding key.
For each decision, you are provided with four expert ratings (with 1 representing a
positive and 0 representing a negative rating).
The four experts differ in their validity.
The ratings of experts are given in descending order of their validity (having validities
of 90%, 80%, 70%, and 60%).

Product A ratings: [0 1 1 1]. Product R ratings: [1 0 0 1]. You press <<A>>.

Product A ratings: [1 1 1 1]. Product R ratings: [0 0 1 1]. You press <<A>>.
Product A ratings: [1 0 0 0]. Product R ratings: [0 0 0 1]. You press <<A>>.
Product A ratings: [1 1 1 0]. Product R ratings: [0 0 1 0]. You press <<A>>.
Product A ratings: [0 1 1 1]. Product R ratings: [1 1 1 0]. You press <<R>>.
Product A ratings: [0 1 0 1]. Product R ratings: [1 1 0 0]. You press <<R>>.
Product A ratings: [0 0 1 1]. Product R ratings: [1 0 1 0]. You press <<R>>.
Product A ratings: [1 0 0 1]. Product R ratings: [0 1 1 1]. You press <<R>>.
Product A ratings: [0 1 1 1]. Product R ratings: [1 0 0 1]. You press <<A>>.
Product A ratings: [0 1 1 1]. Product R ratings: [1 1 1 0]. You press <<R>>.
Product A ratings: [1 0 1 0]. Product R ratings: [0 0 1 1]. You press <<A>>.
Product A ratings: [1 1 0 0]. Product R ratings: [0 1 0 1]. You press <<A>>.
Product A ratings: [1 0 0 0]. Product R ratings: [0 1 1 0]. You press <<R>>.

70
Product A ratings: [1 0 0 1]. Product R ratings: [0 1 1 1]. You press <<R>>.
Product A ratings: [1 1 1 1]. Product R ratings: [0 0 1 1]. You press <<A>>.
Product A ratings: [1 0 0 0]. Product R ratings: [0 0 0 1]. You press <<A>>.
Product A ratings: [1 1 0 1]. Product R ratings: [0 0 0 1]. You press <<A>>.
Product A ratings: [1 0 0 1]. Product R ratings: [0 1 1 1]. You press <<R>>.
Product A ratings: [0 1 0 1]. Product R ratings: [1 1 0 0]. You press <<R>>.
Product A ratings: [1 0 1 1]. Product R ratings: [0 0 0 0]. You press <<A>>.
Product A ratings: [1 0 0 0]. Product R ratings: [0 0 0 1]. You press <<A>>.
Product A ratings: [0 1 1 1]. Product R ratings: [1 1 1 0]. You press <<R>>.
Product A ratings: [0 1 1 0]. Product R ratings: [1 0 0 0]. You press <<A>>.
Product A ratings: [1 1 1 0]. Product R ratings: [0 1 1 1]. You press <<A>>.
Product A ratings: [0 0 0 0]. Product R ratings: [1 1 0 0]. You press <<R>>.
Product A ratings: [0 1 1 1]. Product R ratings: [1 0 0 1]. You press <<A>>.
Product A ratings: [0 0 0 0]. Product R ratings: [1 1 0 0]. You press <<R>>.
Product A ratings: [1 0 0 0]. Product R ratings: [0 1 1 0]. You press <<R>>.
Product A ratings: [0 0 0 0]. Product R ratings: [1 1 0 0]. You press <<R>>.
Product A ratings: [0 1 1 1]. Product R ratings: [1 0 0 1]. You press <<A>>.
Product A ratings: [0 0 0 1]. Product R ratings: [1 0 0 0]. You press <<R>>.
Product A ratings: [1 0 0 0]. Product R ratings: [0 1 1 0]. You press <<R>>.
Product A ratings: [1 1 0 0]. Product R ratings: [0 0 0 0]. You press <<A>>.
Product A ratings: [1 1 0 0]. Product R ratings: [0 1 0 1]. You press <<A>>.
Product A ratings: [1 0 0 0]. Product R ratings: [0 0 0 1]. You press <<A>>.
Product A ratings: [0 0 0 1]. Product R ratings: [1 0 0 0]. You press <<R>>.
Product A ratings: [1 0 1 0]. Product R ratings: [0 0 1 1]. You press <<A>>.
Product A ratings: [1 1 0 0]. Product R ratings: [0 0 0 0]. You press <<A>>.
Product A ratings: [0 0 1 1]. Product R ratings: [1 0 1 0]. You press <<R>>.
Product A ratings: [1 0 0 0]. Product R ratings: [0 1 1 0]. You press <<R>>.
Product A ratings: [1 1 1 0]. Product R ratings: [0 0 1 0]. You press <<A>>.
Product A ratings: [1 0 0 1]. Product R ratings: [0 1 1 1]. You press <<R>>.
Product A ratings: [1 0 1 0]. Product R ratings: [0 0 1 1]. You press <<A>>.
Product A ratings: [0 1 1 0]. Product R ratings: [1 0 0 0]. You press <<A>>.
Product A ratings: [1 1 0 0]. Product R ratings: [0 1 0 1]. You press <<A>>.
Product A ratings: [0 1 1 0]. Product R ratings: [1 0 0 0]. You press <<A>>.
Product A ratings: [1 0 0 0

Grammar judgement
Data source: [28]

Number of experiments: 1
Number of participants: 3192
Number of choices: 89376

Example prompt:
You’re about to answer a set of 20 questions about grammar. How many of the 20
questions do you think you will answer correctly?

71
You say <<4>>.
Compared to other participants in this study, how well do you think you will do?
Marking 90% means you will do better than 90% of participants, marking 10% means
you will do better than only 10%, and marking 50% means that you will perform bet-
ter than half of the participants.
You say <<50>>%.
On a scale of 0 to 10, how difficult is recognizing correct grammar for the average
participant?
You say <<6>>.
On a scale of 0 to 10, how difficult is recognizing correct grammar for you?
You say <<6>>.

You will now see twenty questions.

In each question, some part of each sentence is in square brackets.
Five choices for rephrasing that part follow each sentence; one choice repeats the orig-
inal, and the other four are different.
Your task is to use the buttons J, E, V, H, and G to select the grammatically correct
choice.

Q1. The school-age child faces a formidable task when during the first few years of
classroom experiences [he or she is expected to master the printed form of language.]
The choices are:
V: he or she expects to master the printed form of language.
E: he or she is expected to master the printed form of language.
H: he or she faces expectations of mastering the printed form of language.
J: mastery of the printed form of language is expected of him or her.
G: mastery of print is expected by his or her teacher.
You press <<E>>.

Q2. He came to the United States as a young [man, he found] a job as a coal miner.
The choices are:
H: man, he found
G: man and found
E: man and there he was able to find
V: man and then finding
J: man and had found
You press <<H>>.

Q3. To a large degree, [poetry, along with all the other arts, is] a form of imitation.
The choices are:
E: poetry, along with all the other arts, is
V: poetry along with all the other arts is
J: poetry, along with all the other arts, are
G: poetry, and other arts, is
H: poetry and art are

72
You press <<V>>.

Q4. Delegates to the political convention found [difficulty to choose] a candidate from
among the few nominated.
The choices are:
E: difficulty to choose
G: it difficult in making the choice of
H: it difficult to choose
V: choosing difficult when selecting
J: making a choice difficult in selecting
You press <<E>>.

Q5. Reading in any language can be viewed as a developmental task much the same
as learning to walk, to cross the street independently, to care for one’s possessions, or
[accepting responsibility for one’s own decisions.]
The choices are:
J: accepting responsibility for one’s own decisions.
V: accepting one’s own decisions responsibly.
H: to accept responsibility for one’s own decisions.
E: accepting responsibility and making one’s own decisions.
G: to make one’s own decisions.
You press <<G>>.

Q6. Sea forests of giant kelp, which fringe only one coastline in the Northern Hemi-
sphere, [is native to shores] throughout the Southern Hemisphere.
The choices are:
E: is native to shores
V: is native to most shores
G: are native only in shores
J: are native
H: are native to shores
You press <<J>>.

Q7. Taking an occasional respite between chapters or assignments is more desirable

[than a long, continuous period of study.
The choices are:
E: than a long, continuous period of study.
G: than a period of long, continuous study.
V: than a long period of continuous study.
J: than studying for a long, continuous period.
H: than a study period long and continuous.
You press <<V>>.

Q8. Like so many characters in Russian fiction, [Crime and Punishment exhibits] a
behavior so foreign to the American temperament that many readers find the story
rather incredible.

73
The choices are:
J: Crime and Punishment exhibits
H: those in Crime and Punishment exhibit
G: those in Crime and Punishment exhibits
V: often exhibiting
E: characterized by
You press <<G>>.

Q9. Don Quixote provides a cross section of Spanish life, thought, and [portrays the
feelings of many Spaniards] at the end of the chivalric age.
The choices are:
H: portrays th

Two-step task
Data source: [29]

Number of experiments: 2
Number of participants: 181
Number of choices: 52861

Example prompt:
Each day you will either be presented with spaceships P and F or with spaceships Z
and J.
These spaceships will take you to two different planets L and Q.
You can take a spaceship by pressing the corresponding key.
Each planet has one alien on it and each alien has its own space treasure mine.
When you arrive at a planet, you will ask the alien for space treasure from its mine.
When you ask the alien, you will find out whether you got space treasure.
However, sometimes the alien will not bring up any treasure.
The quality of each alien’s mine will change during the game.
Before you choose a spaceship, you will be told whether there is a treasure multiplier.
If there is a treasure multiplier, you will receive 5 times the amount of treasure you
will find.
Your goal is to get as much treasure as possible over the next 200 days.

There is no treasure multiplier. You are presented with spaceships F and P. You press
<<P>>. You end up on planet L. You find 7 pieces of space treasure. You receive 7
pieces of space treasure.
There is no treasure multiplier. You are presented with spaceships F and P. You press
<<F>>. You end up on planet Q. You find 0 pieces of space treasure. You receive 0
pieces of space treasure.
There is no treasure multiplier. You are presented with spaceships J and Z. You press
<<Z>>. You end up on planet L. You find 7 pieces of space treasure. You receive 7
pieces of space treasure.
There is no treasure multiplier. You are presented with spaceships Z and J. You press

74
<<Z>>. You end up on planet L. You find 7 pieces of space treasure. You receive 7
pieces of space treasure.
There is no treasure multiplier. You are presented with spaceships Z and J. You press
<<Z>>. You end up on planet L. You find 8 pieces of space treasure. You receive 8
pieces of space treasure.
There is a treasure multiplier. You are presented with spaceships F and P. You press
<<P>>. You end up on planet L. You find 6 pieces of space treasure. You receive 30
pieces of space treasure.
There is a treasure multiplier. You are presented with spaceships F and P. You press
<<P>>. You end up on planet L. You find 8 pieces of space treasure. You receive 40
pieces of space treasure.
There is no treasure multiplier. You are presented with spaceships J and Z. You press
<<Z>>. You end up on planet L. You find 9 pieces of space treasure. You receive 9
pieces of space treasure.
There is no treasure multiplier. You are presented with spaceships P and F. You press
<<P>>. You end up on planet L. You find 8 pieces of space treasure. You receive 8
pieces of space treasure.
There is a treasure multiplier. You are presented with spaceships J and Z. You press
<<Z>>. You end up on planet L. You find 9 pieces of space treasure. You receive 45
pieces of space treasure.
There is a treasure multiplier. You are presented with spaceships P and F. You press
<<P>>. You end up on planet L. You find 9 pieces of space treasure. You receive 45
pieces of space treasure.
There is no treasure multiplier. You are presented with spaceships J and Z. You press
<<Z>>. You end up on planet L. You find 8 pieces of space treasure. You receive 8
pieces of space treasure.
There is a treasure multiplier. You are presented with spaceships J and Z. You press
<<Z>>. You end up on planet L. You find 9 pieces of space treasure. You receive 45
pieces of space treasure.
There is no treasure multiplier. You are presented with spaceships Z and J. You press
<<Z>>. You end up on planet L. You find 5 pieces of space treasure. You receive 5
pieces of space treasure.
There is no treasure multiplier. You are presented with spaceships Z and J. You press
<<Z>>. You end up on planet L. You find 6 pieces of space treasure. You receive 6
pieces of space treasure.
There is a treasure multiplier. You are presented with spaceships P and F. You press
<<P>>. You end up on planet L. You find 8 pieces of space treasure. You receive 40
pieces of space treasure.
There is no treasure multiplier. You are presented with spaceships J and Z. You press
<<Z>>. You end up on planet L. You find 7 pieces of space treasure.

Two-step task
Data source: [30]

Number of experiments: 2

75
Number of participants: 367
Number of choices: 67041

Example prompt:
Each day you will either be presented with spaceships G and S or with spaceships T
and N.
These spaceships will take you to two different planets R and Z.
You can take a spaceship by pressing the corresponding key.
Each planet has one alien on it and each alien has its own space treasure mine.
When you arrive at a planet, you will ask the alien for space treasure from its mine.
When you ask the alien, you will find out whether you got space treasure.
However, sometimes the alien’s mine will dig up antimatter.
Antimatter is bad because each piece will destroy a piece of space treasure, reducing
the total amount of treasure that you have.
The quality of each alien’s mine will change during the game.
Your goal is to get as much treasure and as little antimatter as possible over the next
125 days.

You are presented with spaceships N and T. You press <<T>>. You end up on planet
R. You find 1 pieces of antimatter.
You are presented with spaceships T and N. You do not respond in time on this day.
You do not go to any planet. You find nothing.
You are presented with spaceships N and T. You press <<N>>. You end up on planet
Z. You find 4 pieces of space treasure.
You are presented with spaceships T and N. You press <<N>>. You end up on planet
Z. You find 4 pieces of space treasure.
You are presented with spaceships S and G. You press <<S>>. You end up on planet
Z. You find 5 pieces of space treasure.
You are presented with spaceships T and N. You press <<N>>. You end up on planet
Z. You find 3 pieces of space treasure.
You are presented with spaceships N and T. You press <<N>>. You end up on planet
Z. You find 4 pieces of space treasure.
You are presented with spaceships T and N. You press <<N>>. You end up on planet
Z. You find 2 pieces of space treasure.
You are presented with spaceships S and G. You press <<S>>. You end up on planet
Z. You find 1 pieces of space treasure.
You are presented with spaceships T and N. You press <<N>>. You end up on planet
Z. You find 3 pieces of antimatter.
You are presented with spaceships T and N. You press <<T>>. You end up on planet
R. You find 2 pieces of space treasure.
You are presented with spaceships G and S. You press <<G>>. You end up on planet
R. You find 4 pieces of space treasure.
You are presented with spaceships S and G. You press <<G>>. You end up on planet
R. You find 2 pieces of space treasure.
You are presented with spaceships T and N. You press <<T>>. You end up on planet

76
R. You find 3 pieces of space treasure.
You are presented with spaceships T and N. You press <<T>>. You end up on planet
R. You find 5 pieces of space treasure.
You are presented with spaceships N and T. You press <<T>>. You end up on planet
R. You find 4 pieces of space treasure.
You are presented with spaceships S and G. You press <<G>>. You end up on planet
R. You find 4 pieces of space treasure.
You are presented with spaceships S and G. You press <<G>>. You end up on planet
R. You find nothing.
You are presented with spaceships T and N. You press <<T>>. You end up on planet
R. You find 3 pieces of antimatter.
You are presented with spaceships T and N. You press <<N>>. You end up on planet
Z. You find 2 pieces of space treasure.
You are presented with spaceships S and G. You press <<S>>. You end up on planet
Z. You find 2 pieces of space treasure.
You are presented with spaceships S and G. You press <<S>>. You end up on planet
Z. You find 4 pieces of space treasure.
You are presented with spaceships S and G. You press <<S>>. You end up on planet
Z. You find 4 pieces of space treasure.
You are presented with spaceships N and T. You press <<N>>. You end up on planet
Z. You find 4 pieces of space treasure.
You are presented with spaceships N and T. You press <<N>>. You end up on planet
Z. You find 5 pieces of space treasure.
You are presented with spaceships S and G. You press <<S>>. You end up on planet
Z. You find 3 pieces of space treasure.
You are presented with spaceships G and S. You press <<S>>. You end up on planet
Z. You find 4 pieces of space treasure.
You are presented with spaceships S and G. Yo

Risky choice
Data source: [105]

Number of experiments: 1
Number of participants: 1755
Number of choices: 499728

Example prompt:
You will play multiple rounds of a gambling game.
In each round, you will be presented with 6 different gambles labeled: Q, N, E, S, H,
and K.
You will have to choose one of the gambles and receive a payoff for doing so.
The payoff you receive depends on both the gamble you choose and also the color of
a ball we pull out of a jar with 100 colored balls.
There are a different number of balls of each color on every round.
The colors with more balls are more likely to be chosen.

77
Before making your choice, you may check how much different gambles are worth for
different ball colors.
Each time you check a gamble will cost you 4 points.
To choose or check a gamble, first press the corresponding key, followed by typing
”stop” (for choosing) or the ball color you would like to check.

A new round begins.

There are 50 pink balls, 19 red balls, 15 black balls, and 16 maroon balls.
You press <<Q>> and then type <<pink>>. The payoff for this combination would
be -105 points.
You press <<E>> and then type <<pink>>. The payoff for this combination would
be 47 points.
You press <<H>> and then type <<pink>>. The payoff for this combination would
be 91 points.
You press <<H>> and then type <<stop>>. A maroon ball is chosen, and you earn
-16 points.

A new round begins.

There are 2 pink balls, 47 red balls, 14 black balls, and 37 maroon balls.
You press <<Q>> and then type <<red>>. The payoff for this combination would
be -168 points.
You press <<E>> and then type <<red>>. The payoff for this combination would
be 209 points.
You press <<H>> and then type <<red>>. The payoff for this combination would
be 22 points.
You press <<E>> and then type <<stop>>. A maroon ball is chosen, and you earn
25 points.

A new round begins.

There are 13 pink balls, 1 red balls, 47 black balls, and 39 maroon balls.
You press <<Q>> and then type <<black>>. The payoff for this combination would
be -101 points.
You press <<E>> and then type <<black>>. The payoff for this combination would
be -98 points.
You press <<H>> and then type <<black>>. The payoff for this combination would
be 32 points.
You press <<H>> and then type <<stop>>. A maroon ball is chosen, and you earn
-85 points.

A new round begins.

There are 7 pink balls, 15 red balls, 76 black balls, and 2 maroon balls.
You press <<N>> and then type <<black>>. The payoff for this combination would
be -108 points.
You press <<Q>> and then type <<black>>. The payoff for this combination would
be -191 points.

78
You press <<S>> and then type <<black>>. The payoff for this combination would
be -222 points.
You press <<K>> and then type <<stop>>. A black ball is chosen, and you earn
-90 points.

A new round begins.

There are 17 pink balls, 54 red balls, 16 black balls, and 13 maroon balls.
You press <<Q>> and then type <<red>>. The payoff for this combination would
be 138 points.
You press <<N>> and then type <<red>>. The payoff for this combination would
be 171 points.
You press <<E>> and then type <<red>>. The payoff for this combination would
be -197 points.
You press <<N>> and then type <<stop>>. A red ball is chosen, and you earn 171
points.

A new round begins.

There are 1 pink balls, 38 red balls, 25 black balls, and 36 maroon balls.
You press <<N>> and then type <<red>>. The payoff for this combination would
be -124 points.
You press <<E>> and then type <<red>>. The payoff for this combination would
be -158 points.
You press <<S>> and then type <<red>>. The payoff for this combination would
be -114 points.
You press <<H>> and then type <<stop>>. A black ball is chosen, and you earn
338 points.

A new round begins.

There are 37 pink balls, 44 red balls, 17 black balls, and 2 maroon balls.
You press <<Q>> and then type <<red>>. The payoff for this combination would
be -198 points.
You press <<S>> and then type <<red>>. The payoff for this combination would
be 39 points.
You press <<H>> and then type <<red>>. The payoff for this combination would
be -63 points.
You press <<E>> and then type <<stop>>. A red ball is chosen, and you earn 51
points.

A new round begins.

There are 12 pink balls, 32 red balls, 45

Tile-revealing task
Data source: [106]

Number of experiments: 1

79
Number of participants: 707
Number of choices: 321738

Example prompt:
You are playing a game where you are revealing patterns on a binary grid.
Each turn you will be presented with a 7x7 grid of tiles.
If you give the location on the grid with an ”X” on it, then it will either turn into a
”1” or a ”0”.
Please give the location as a [row, column] pair where both row and column take
values between 0 and 6.
Do not give the location on the grid where there is already a ”1” or a ”0”.
Your job is to reveal as many 1s as possible while revealing as little 0s as possible.
The game ends when you have revealed all 1s on the grid.
You will have 25 boards to complete, each with a different underlying pattern.

You are currently solving board 1.

The board state is now:
XXXXXXX
XXXXXXX
XXXXXXX
XXX1XXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[2, 2]>> and get a miss!

The board state is now:

XXXXXXX
XXXXXXX
XX0XXXX
XXX1XXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[2, 4]>> and get a miss!

The board state is now:

XXXXXXX
XXXXXXX
XX0X0XX
XXX1XXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[2, 3]>> and get a hit!

80
The board state is now:
XXXXXXX
XXXXXXX
XX010XX
XXX1XXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[1, 3]>> and get a miss!

The board state is now:

XXXXXXX
XXX0XXX
XX010XX
XXX1XXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[3, 4]>> and get a hit!

The board state is now:

XXXXXXX
XXX0XXX
XX010XX
XXX11XX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[3, 2]>> and get a hit!

The board state is now:

XXXXXXX
XXX0XXX
XX010XX
XX111XX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[3, 5]>> and get a hit!

The board state is now:

XXXXXXX
XXX0XXX
XX010XX
XX1111X
XXXXXXX

81
XXXXXXX
XXXXXXX
You choose <<[3, 6]>> and get a hit!

The board state is now:

XXXXXXX
XXX0XXX
XX010XX
XX11111
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[3, 1]>> and get a hit!

You are currently solving board 2.

The board state is now:
XXXXXXX
X1XXXXX
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[2, 1]>> and get a miss!

The board state is now:

XXXXXXX
X1XXXXX
X0XXXXX
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[0, 1]>> and get a miss!

The board state is now:

X0XXXXX
X1XXXXX
X0XXXXX
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[1, 2]>> and get a hit!

The board state is now:

82
X0XXXXX
X11XXXX
X0XXXXX
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[1, 0]>> and get a hit!

The board state is now:

X0XXXXX
111XXXX
X0XXXXX
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[1, 3]>> and get a hit!

The board state is now:

X0XXXXX
1111XXX
X0XXXXX
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[1, 4]>> and get a hit!

The board state is now:

X0XXXXX
11111XX
X0XXXXX
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[1, 5]>> and get a miss!

The board state is now:

X0XXXXX
111110X
X0XXXXX
XXXXXXX
XXXXXXX
XXXXXXX

83
XXXXXXX
You choose <<[0, 0]>> and get a miss!

The board state is now:

00XXXXX
111110X
X0XXXXX
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[2, 2]>> and get a miss!

The board state is now:

00XXXXX
111110X
X00XXXX
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[2, 3]>> and get a hit!

The board state is now:

00XXXXX
111110X
X001XXX
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[3, 3]>> and get a hit!

The board state is now:

00XXXXX
111110X
X001XXX
XXX1XXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[0, 3]>> and get a hit!

The board state is now:

00X1XXX
111110X

84
X001XXX
XXX1XXX
XXXXXXX
XXXXXXX
XXXXXXX
You choose <<[4, 3]>> and get a hit!

You are currently solving board 3.

The board state is now:
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
XXX1XXX
XXXXXXX
You choose <<[4, 2]>> and get a hit!

The board state is now:

XXXXXXX
XXXXXXX
XXXXXXX
XXXXXXX
XX1XXXX
XXX1XXX
XXXXXXX
You choose <<[3, 1]>> and get a hit!

The board state is now:

XXXXXXX
XXXXXXX
XXXXXXX
X1XXXXX
XX1XXXX
XXX1XXX
XXXXXXX
You choose <<[2, 0]>> and get a hit!

The board state is now:

XXXXXXX
XXXXXXX
1XXXXXX
X1XXXXX
XX1XXXX
XXX1XXX
XXXXXXX

85
You choose <<[1, 1]>> and get a miss!

The board state is now:

XXXXXXX
X0XXXXX
1XXXXXX
X1XXXXX
XX1XXXX
XXX1XXX
XXXXXXX
You choose <<[2, 2]>> and get a miss!

The board state is now:

XXXXXXX
X0XXXXX
1X0XXXX
X1XXXXX
XX1XXXX
XXX1XXX
XXXXXXX
You choose <<[3, 3]>> and get a hit!

The board state is now:

XXXXXXX
X0XXXXX

Probabilistic instrumental learning

Data source: [107]

Number of experiments: 2
Number of participants: 77
Number of choices: 7392

Example prompt:
You are going to visit four different casinos (named 1, 2, 3, and 4) 24 times each.
Each casino owns two slot machines that return either 0 or 0.5 points stochastically
with different probabilities.
You can play one of the machines in order to win points by pressing the corresponding
key.
Your goal is to maximize the sum of received points within all visits.

You go to casino 3. You can choose between machines B and C. You press <<B>>
and receive 0.0 points.
You go to casino 4. You can choose between machines P and T. You press <<T>>
and receive 0.5 points.

86
You go to casino 4. You can choose between machines P and T. You press <<T>>
and receive 0.5 points.
You go to casino 3. You can choose between machines B and C. You press <<B>>
and receive 0.0 points.
You go to casino 1. You can choose between machines F and I. You press <<I>> and
receive 0.0 points.
You go to casino 1. You can choose between machines F and I. You press <<F>>
and receive 0.0 points.
You go to casino 2. You can choose between machines L and J. You press <<L>>
and receive 0.5 points.
You go to casino 3. You can choose between machines B and C. You press <<B>>
and receive 0.0 points.
You go to casino 4. You can choose between machines P and T. You press <<T>>
and receive 0.0 points.
You go to casino 3. You can choose between machines B and C. You press <<C>>
and receive 0.0 points.
You go to casino 4. You can choose between machines P and T. You press <<P>>
and receive 0.5 points.
You go to casino 1. You can choose between machines F and I. You press <<F>>
and receive 0.5 points.
You go to casino 2. You can choose between machines L and J. You press <<L>>
and receive 0.5 points.
You go to casino 2. You can choose between machines L and J. You press <<L>>
and receive 0.5 points.
You go to casino 1. You can choose between machines F and I. You press <<F>>
and receive 0.0 points.
You go to casino 2. You can choose between machines L and J. You press <<J>>
and receive 0.5 points.
You go to casino 1. You can choose between machines F and I. You press <<I>> and
receive 0.0 points.
You go to casino 1. You can choose between machines F and I. You press <<F>>
and receive 0.0 points.
You go to casino 2. You can choose between machines L and J. You press <<J>>
and receive 0.0 points.
You go to casino 2. You can choose between machines L and J. You press <<L>>
and receive 0.5 points.
You go to casino 3. You can choose between machines B and C. You press <<B>>
and receive 0.5 points.
You go to casino 3. You can choose between machines B and C. You press <<B>>
and receive 0.0 points.
You go to casino 2. You can choose between machines L and J. You press <<L>>
and receive 0.5 points.
You go to casino 3. You can choose between machines B and C. You press <<C>>
and receive 0.5 points.
You go to casino 2. You can choose between machines L and J. You press <<L>>

87
and receive 0.0 points.
You go to casino 1. You can choose between machines F and I. You press <<I>> and
receive 0.0 points.
You go to casino 4. You can choose between machines P and T. You press <<P>>
and receive 0.0 points.
You go to casino 4. You can choose between machines P and T. You press <<T>>
and receive 0.5 points.
You go to casino 4. You can choose between machines P and T. You press <<T>>
and receive 0.5 points.
You go to casino 1. You can choose between machines F and I. You press <<I>> and
receive 0.5 points.
You go to casino 4. You can choose between machines P and T. You press <<P>>
and receive 0.5 points.
You go to casino 3. You can choose between machines B and C. You press <<C>>
and receive 0.5 points.
You go to casino 2. You can choose between machines L and J. You press <<L>>
and receive 0.5 points.
You go to casino 3. You can choose between machines B and C. You press <<C>>
and receive 0.5 points.
You go to casino 4. You can choose between machines P and T. You press <<P>>
and receive 0.5 points.
You go to casino 3. You can choose between machines B and C. You press <<C>>
and receive 0.5 points.
You go to casino 4. You can choose between machines P and T. You press <<P>>
and receive 0.5 point

Medin categorization
Data source: [108]

Number of experiments: 2
Number of participants: 228
Number of choices: 37848

Example prompt:
You will observe a series of objects, one at a time.
The objects differ along three binary dimensions: shape (square vs. triangle), size (1.50
inch vs. 0.75 inch), and shading (black vs. white).
Each dimension is indicated by the three digits, for example, ’121’ means a square,
0.75 inch, black object.
Based on some combination of the three dimensions, each object belongs to one of two
categories, W or N.
You have to assign each object to one of the two categories by pressing the corre-
sponding key.
If your choice is correct, you get a point, otherwise you lose a point.
Your goal is to get as many points as possible.

88
At some point, you begin a ’test block’ in which you will see eight objects.
Here, you have to assign each object to one of the two categories as before.
Furthermore, you have to rate how typical the object is for the category you chose,
on a scale from 1 to 9.
1 means ’not at all typical’, and 9 means ’most typical’.

You see the image 112, press <<W>> and get 1 points.
You see the image 121, press <<N>> and get 0 points.
You see the image 212, press <<W>> and get 0 points.
You see the image 211, press <<N>> and get 0 points.
You see the image 221, press <<W>> and get 0 points.
You see the image 122, press <<W>> and get 0 points.
You see the image 122, press <<N>> and get 1 points.
You see the image 221, press <<W>> and get 0 points.
You see the image 121, press <<N>> and get 0 points.
You see the image 212, press <<N>> and get 1 points.
You see the image 211, press <<N>> and get 0 points.
You see the image 112, press <<W>> and get 1 points.
You see the image 112, press <<W>> and get 1 points.
You see the image 212, press <<N>> and get 1 points.
You see the image 221, press <<W>> and get 0 points.
You see the image 211, press <<N>> and get 0 points.
You see the image 121, press <<N>> and get 0 points.
You see the image 122, press <<W>> and get 0 points.
You see the image 212, press <<N>> and get 1 points.
You see the image 221, press <<W>> and get 0 points.
You see the image 211, press <<N>> and get 0 points.
You see the image 121, press <<N>> and get 0 points.
You see the image 112, press <<W>> and get 1 points.
You see the image 122, press <<W>> and get 0 points.
You see the image 122, press <<N>> and get 1 points.
You see the image 112, press <<W>> and get 1 points.
You see the image 121, press <<N>> and get 0 points.
You see the image 221, press <<N>> and get 1 points.
You see the image 212, press <<N>> and get 1 points.
You see the image 211, press <<N>> and get 0 points.
You see the image 112, press <<W>> and get 1 points.
You see the image 212, press <<N>> and get 1 points.
You see the image 122, press <<W>> and get 0 points.
You see the image 211, press <<N>> and get 0 points.
You see the image 221, press <<W>> and get 0 points.
You see the image 121, press <<N>> and get 0 points.
You see the image 122, press <<N>> and get 1 points.
You see the image 212, press <<W>> and get 0 points.
You see the image 211, press <<N>> and get 0 points.

89
You see the image 221, press <<N>> and get 1 points.
You see the image 112, press <<W>> and get 1 points.
You see the image 121, press <<W>> and get 1 points.
You see the image 121, press <<W>> and get 1 points.
You see the image 211, press <<N>> and get 0 points.
You see the image 221, press <<N>> and get 1 points.
You see the image 212, press <<N>> and get 1 points.
You see the image 122, press <<W>> and get 0 points.
You see the image 112, press <<W>> and get 1 points.
You see the image 212, press <<N>> and get 1 points.
You see the image 112, press <<W>> and get 1 points.
You see the image 121, press <<N>> and get 0 points.
You see the image 122, press <<W>> and get 0 points.
You see the image 221, press <<N>> and get 1 points.
You see the image 211, press <<N>> and get 0 points.
You see the image 112, press <<W>> and get 1 points.
You see the image 122, press <<W>> and get 0 points.
You see the image 121, press <<N>> and get 0 points.
You see the image 221, press <<N>> and get 1 points.
You see the image 212, press <<W>> and get 0 points.
You see the image 211, p

Zoopermarket
Data source: [109]

Number of experiments: 3
Number of participants: 96
Number of choices: 34442

Example prompt:
You will have to repeatedly feed animals with fruits.
Each fruit contains two vitamins.
Every animal has a different preference for the vitamins.
The vitamin contents and the preferences are both given as vectors with two entries.
Your points are calculated as the dot product of the vitamin content with the prefer-
ence of the current animal.
For example, let us assume that you have to feed the elephant who has a preference
[-1 1].
Then, if you feed the elephant a fruit with vitamin content [-1 1], this would yield 2
points.
If you feed it a fruit with vitamins [1 0], this would yield -1 points.
You have to buy the fruits in a market, in which you can go left or right for two steps.
You can press I to go left, and V to go right.
Per round, you always collect two fruits.
There are eight animals in total and you have to feed one, two, or three of them in

90
each block.
In each block, there are twelve trials with different animals in random order.
After these twelve trials, there are three more in which you have to feed new animals.
The fruits in the market are rearranged after each block, meaning that you have to
relearn the positions.
Your goal is to maximize the points obtained.

A new block starts. The locations of the fruits in the market got scrambled.
You have to feed the crocodile. It has the preference [1 1].
You press <<V>> and find the apple which has the vitamins [-1 -1]. You get -2 points.

You press <<V>> and find the orange which has the vitamins [0 1]. You get 1 points.
You have to feed the crocodile. It has the preference [1 1].
You press <<V>> and find the apple which has the vitamins [-1 -1]. You get -2 points.

You press <<I>> and find the blueberry which has the vitamins [-1 1]. You get 0
points.
You have to feed the crocodile. It has the preference [1 1].
You press <<I>> and find the strawberry which has the vitamins [1 1]. You get 2
points.
You press <<V>> and find the grapes which has the vitamins [ 1 -1]. You get 0
points.
You have to feed the crocodile. It has the preference [1 1].
You press <<I>> and find the strawberry which has the vitamins [1 1]. You get 2
points.
You press <<V>> and find the grapes which has the vitamins [ 1 -1]. You get 0
points.
You have to feed the kangaroo. It has the preference [-1 0].
You press <<V>> and find the apple which has the vitamins [-1 -1]. You get 1 points.

You press <<I>> and find the blueberry which has the vitamins [-1 1]. You get 1
points.
You have to feed the kangaroo. It has the preference [-1 0].
You press <<V>> and find the apple which has the vitamins [-1 -1]. You get 1 points.

You press <<I>> and find the blueberry which has the vitamins [-1 1]. You get 1
points.
You have to feed the crocodile. It has the preference [1 1].
You press <<I>> and find the strawberry which has the vitamins [1 1]. You get 2
points.
You press <<V>> and find the grapes which has the vitamins [ 1 -1]. You get 0
points.
You have to feed the kangaroo. It has the preference [-1 0].
You press <<V>> and find the apple which has the vitamins [-1 -1]. You get 1 points.

91
You press <<I>> and find the blueberry which has the vitamins [-1 1]. You get 1
points.
You have to feed the kangaroo. It has the preference [-1 0].
You press <<V>> and find the apple which has the vitamins [-1 -1]. You get 1 points.

You press <<I>> and find the blueberry which has the vitamins [-1 1]. You get 1
points.
You have to feed the kangaroo. It has the preference [-1 0].
You press <<V>> and find the apple which has the vitamins [-1 -1]. You get 1 points.

You press <<I>> and find the blueberry which has the vitamins [-1 1]. You get 1
points.
You have to feed the kangaroo. It has the preference [-1 0].
You press <<V>> and find the apple which has the vitamins [-1 -1]. You get 1 points.

You press <<I>> and find the blueberry which has the vitamins [-1 1]. You get 1
points.
You have to feed the crocodile. It has the preference [1 1].
You press <<V>> and find the apple which has the vitamins [-1 -1]. You get -2 points.

You press <<V>> and find the orange which has the vitamins [0 1]. You get 1 points.
You

choices13k
Data source: [21]

Number of experiments: 1
Number of participants: 13735
Number of choices: 1097375

Example prompt:
You will encounter a series of gambling problems where you have to select between
two options.
You can select an option by pressing the corresponding key.
For some problems, you are told the points you received and missed out on after each
selection, while for others this information is suppressed.
In cases where the probabilities are unknown, they sum up to one and remain constant
within a problem.

Option L delivers 10.0 points with 80.0% chance, or -25.0 points with 20.0% chance.
Option B delivers 0.0 points with 20.0% chance, or 5.0 points with 80.0% chance.
You press <<B>>. You receive 5.0 points by selecting this option. You would have
received 10.0 points had you chosen the other option.
You press <<B>>. You receive 5.0 points by selecting this option. You would have
received -25.0 points had you chosen the other option.

92
You press <<B>>. You receive 5.0 points by selecting this option. You would have
received 10.0 points had you chosen the other option.
You press <<B>>. You receive 5.0 points by selecting this option. You would have
received -25.0 points had you chosen the other option.
You press <<B>>. You receive 0.0 points by selecting this option. You would have
received -25.0 points had you chosen the other option.

Option L delivers either 30.0 points with 100.0% chance, or 30.0 points with 0.0%
chance.
Option B delivers either 0.0 points with unknown chance, or 42.0 points with unknown
chance.
You press <<L>>. You receive 30.0 points by selecting this option. You would have
received 42.0 points had you chosen the other option.
You press <<L>>. You receive 30.0 points by selecting this option. You would have
received 0.0 points had you chosen the other option.
You press <<L>>. You receive 30.0 points by selecting this option. You would have
received 42.0 points had you chosen the other option.
You press <<L>>. You receive 30.0 points by selecting this option. You would have
received 42.0 points had you chosen the other option.
You press <<B>>. You receive 42.0 points by selecting this option. You would have
received 30.0 points had you chosen the other option.

Option L delivers 8.0 points with 100.0% chance, or 8.0 points with 0.0% chance.
Option B delivers 5.0 points with 95.0% chance, or 54.0 points with 5.0% chance.
You press <<B>>. You receive 5.0 points by selecting this option. You would have
received 8.0 points had you chosen the other option.
You press <<B>>. You receive 5.0 points by selecting this option. You would have
received 8.0 points had you chosen the other option.
You press <<B>>. You receive 5.0 points by selecting this option. You would have
received 8.0 points had you chosen the other option.
You press <<B>>. You receive 5.0 points by selecting this option. You would have
received 8.0 points had you chosen the other option.
You press <<B>>. You receive 5.0 points by selecting this option. You would have
received 8.0 points had you chosen the other option.

Option L delivers 20.0 points with 100.0% chance, or 20.0 points with 0.0% chance.
Option B delivers 15.0 points with 99.0% chance, 47.5 points with 0.0078% chance,
48.5 points with 0.0547% chance, 49.5 points with 0.1641% chance, 50.5 points with
0.2734% chance, 51.5 points with 0.2734% chance, 52.5 points with 0.1641% chance,
53.5 points with 0.0547% chance, or 54.5 points with 0.0078% chance.
You press <<B>>. You receive 15.0 points by selecting this option. You would have
received 20.0 points had you chosen the other option.
You press <<B>>. You receive 15.0 points by selecting this option. You would have
received 20.0 points had you chosen the other option.
You press <<B>>. You receive 15.0 points by selecting this option. You would have

93
received 20.0 points had you chosen the other option.
You press <<B>>. You receive 15.0 points by selecting this option. You would have
received 20.0 points had you chosen the other option.
You press <<B>>. You receive 15.0 points by selecting this option. You would have
received 20.0 points had you chosen the other option.

Option L delivers 15.0 points with 5.0% chance, or 9.0 points with 95.0%

CPC18
Data source: [110]

Number of experiments: 1
Number of participants: 216
Number of choices: 162000

Example prompt:
You will encounter a series of gambling problems where you have to select between
two options.
You can select an option by pressing the corresponding key.
You will encounter each problem 25 times.
In the first five encounters, you will not receive feedback.
In the remaining 20 encounters, you will receive feedback about the outcomes of both
options.
In cases where the probabilities are stated to be unknown, they sum up to one and
remain constant within a problem.

Option F delivers 3 points with 80.0% chance, 94 points with 1.25% chance, 95 points
with 5.0% chance, 96 points with 7.5% chance, 97 points with 5.0% chance, 98 points
with 1.25% chance.
Option X delivers -19 points with 50.0% chance, 59 points with 50.0% chance.
You press <<X>>.
You press <<X>>.
You press <<X>>.
You press <<X>>.
You press <<X>>.
You press <<X>> and gain 59 points. You would have gained 3 points had you chosen
option F.
You press <<X>> and gain 59 points. You would have gained 98 points had you
chosen option F.
You press <<X>> and gain 59 points. You would have gained 3 points had you chosen
option F.
You press <<X>> and lose 19 points. You would have gained 95 points had you
chosen option F.
You press <<X>> and lose 19 points. You would have gained 3 points had you chosen
option F.

94
You press <<X>> and gain 59 points. You would have gained 3 points had you chosen
option F.
You press <<X>> and lose 19 points. You would have gained 3 points had you chosen
option F.
You press <<X>> and lose 19 points. You would have gained 3 points had you chosen
option F.
You press <<F>> and gain 3 points. You would have lost 19 points had you chosen
option X.
You press <<X>> and lose 19 points. You would have gained 3 points had you chosen
option F.
You press <<X>> and lose 19 points. You would have gained 3 points had you chosen
option F.
You press <<X>> and gain 59 points. You would have gained 3 points had you chosen
option F.
You press <<X>> and gain 59 points. You would have gained 3 points had you chosen
option F.
You press <<X>> and lose 19 points. You would have gained 3 points had you chosen
option F.
You press <<X>> and lose 19 points. You would have gained 96 points had you
chosen option F.
You press <<X>> and gain 59 points. You would have gained 3 points had you chosen
option F.
You press <<X>> and lose 19 points. You would have gained 3 points had you chosen
option F.
You press <<X>> and gain 59 points. You would have gained 98 points had you
chosen option F.
You press <<X>> and gain 59 points. You would have gained 3 points had you chosen
option F.
You press <<X>> and lose 19 points. You would have gained 3 points had you chosen
option F.

Option F delivers -12 points with 95.0% chance, 47 points with 2.5% chance, 49 points
with 1.25% chance, 53 points with 0.625% chance, 61 points with 0.3125% chance,
77 points with 0.15625% chance, 109 points with 0.078125% chance, 173 points with
0.078125% chance.
Option X delivers -9 points with 100.0% chance.
You press <<F>>.
You press <<F>>.
You press <<F>>.
You press <<F>>.
You press <<F>>.
You press <<F>> and lose 12 points. You would have lost 9 points had you chosen
option X.
You press <<F>> and lose 12 points. You would have lost 9 points had you chosen
option X.

95
You press <<F>> and lose 12 points. You would have lost 9 points had you chosen
option X.
You press <<F>> and lose 12 points. You would have lost 9 points had you chosen
option X.
You press <<F>> and lose 12 points. You would have lost 9 points had you chosen
option X.
You press <<F>> and lose 12 points. You would have lost 9 points had you chosen
option X.
You press <<F>> and lose 12 points. You would have lost 9 points had you chosen
option X.
You press <<F>> and lose 12 points. You would have lost 9 points had you chosen
option X.
You press <<F>> and gain 49 points. You would have lost 9 points had you chosen
option X.
You press <<F>> and gain 53 points. You would have lost 9 points had you chosen
option X.
You press <<F>> and lose 12 points. You would have lost 9 points had you chosen
option X.
You press <<F>> and lose 12 points. You would

Episodic long-term memory

Data source: [111]

Number of experiments: 3
Number of participants: 132
Number of choices: 18649

Example prompt:
In this experiment, you will go through three cycles of three tasks.
Each cycle concerns itself with a list of 30 words for you to study.
In the first task of each cycle, you will go through the list.
You need to remember the words indicated by a red border.
For each word, please make a judgement on whether the object is larger or smaller
than a football, and press the key ”O” if the object is larger than a football, and the
key ”M” if it is smaller instead.
In the second task of each cycle, you will try to solve as many arithmetic equations
as you can in one minute.
In the third task of each cycle, you will recall the words that you memorized in the
first task of that cycle.

List 1, task 1:
You see the word ”church”, surrounded by a blue border. You press <<O>>. The
word disappears but the blue border stays for another 3 seconds before disappearing.
You see the word ”robot”, surrounded by a blue border. You press <<O>>. The word
disappears but the blue border stays for another 3 seconds before disappearing.

96
You see the word ”jewelry”, surrounded by a red border. You press <<M>>. The
word disappears but the red border stays for another 3 seconds before disappearing.
You see the word ”skull”, surrounded by a blue border. You press <<O>>. The word
disappears but the blue border stays for another 3 seconds before disappearing.
You see the word ”apple”, surrounded by a blue border. You press <<M>>. The
word disappears but the blue border stays for another 3 seconds before disappearing.
You see the word ”garden”, surrounded by a blue border. You press <<O>>. The
word disappears but the blue border stays for another 3 seconds before disappearing.
You see the word ”pipe”, surrounded by a red border. You press <<M>>. The word
disappears but the red border stays for another 3 seconds before disappearing.
You see the word ”needle”, surrounded by a red border. You press <<M>>. The
word disappears but the red border stays for another 3 seconds before disappearing.
You see the word ”circus”, surrounded by a red border. You press <<O>>. The word
disappears but the red border stays for another 3 seconds before disappearing.
You see the word ”towel”, surrounded by a blue border. You press <<M>>. The word
disappears but the blue border stays for another 3 seconds before disappearing.
You see the word ”rabbit”, surrounded by a blue border. You press <<M>>. The
word disappears but the blue border stays for another 3 seconds before disappearing.
You see the word ”diamond”, surrounded by a red border. You press <<M>>. The
word disappears but the red border stays for another 3 seconds before disappearing.
You see the word ”cocktail”, surrounded by a red border. You press <<M>>. The
word disappears but the red border stays for another 3 seconds before disappearing.
You see the word ”satellite”, surrounded by a red border. You press <<O>>. The
word disappears but the red border stays for another 3 seconds before disappearing.
You see the word ”sweater”, surrounded by a blue border. You press <<O>>. The
word disappears but the blue border stays for another 3 seconds before disappearing.
You see the word ”planet”, surrounded by a red border. You press <<O>>. The word
disappears but the red border stays for another 3 seconds before disappearing.
You see the word ”pizza”, surrounded by a blue border. You press <<O>>. The word
disappears but the blue border stays for another 3 seconds before disappearing.
You see the word ”forest”, surrounded by a blue border. You press <<O>>. The word
disappears but the blue border stays for another 3 seconds before disappearing.
You see the word ”carpet”, surrounded by a blue border. You press <<O>>. The
word disappears but the blue border stays for another 3 seconds before disappearing.
You see the word ”highway”, surrounded by a red border. You press <<O>>. The
word disappears but the red border stays for another 3 seconds before disappearing.
You see the word ”jeep”, surrounded by a blue border. You press <<O>>. The word
disappears but the blue border stays for another 3 seconds before disappearing.
You see t

Intertemporal choice
Data source: [59]

Number of experiments: 1
Number of participants: 11937

97
Number of choices: 142236

Example prompt:
In the following you will be presented with multiple choices between two options G
and C.
Please name which option you would prefer by pressing the corresponding key.

You have the choice between receiving 500$ immediately (press G) or receiving 550$
in one year (press C). You press <<G>>.
You have the choice between receiving 500$ immediately (press G) or receiving 600$
in one year (press C). You press <<C>>.
You have the choice between paying 500$ immediately (press G) or paying 550$ in
one year (press C). You press <<G>>.
You have the choice between paying 500$ immediately (press G) or paying 510$ in
one year (press C). You press <<C>>.
You have the choice between receiving 5000$ immediately (press G) or receiving 5500$
in one year (press C). You press <<G>>.
You have the choice between receiving 5000$ immediately (press G) or receiving 6000$
in one year (press C). You press <<C>>.
You have the choice between receiving 500$ in one year (press G) or receiving 600$ in
two years (press C). You press <<C>>.
You have the choice between receiving 500$ immediately (press G) or receiving 700$
in two years (press C). You press <<C>>.
You have the choice between wait 500$ immediately (press G) or wait 600$ in one
year (press C). You press <<C>>.
You have the choice between receive 500$ immediately (press G) or receive 600$ in
one year (press C). You press <<C>>.

Horizon task
Data source: [112]

Number of experiments: 1
Number of participants: 78
Number of choices: 25336

Example prompt:
You are participating in multiple games involving two slot machines, labeled J and R.
The two slot machines are different across different games.
Each time you choose a slot machine, you get some points.
You choose a slot machine by pressing the corresponding key.
Each slot machine tends to pay out about the same amount of points on average.
Your goal is to choose the slot machines that will give you the most points across the
experiment.
The first 4 trials in each game are instructed trials where you will be told which slot
machine to choose.

98
After these instructed trials, you will have the freedom to choose for either 1 or 6
trials.

Game 1. There are 10 trials in this game.

You are instructed to press J and get 57 points.
You are instructed to press R and get 29 points.
You are instructed to press J and get 66 points.
You are instructed to press R and get 38 points.
You press <<R>> and get 45 points.
You press <<R>> and get 38 points.
You press <<R>> and get 49 points.
You press <<J>> and get 59 points.
You press <<R>> and get 28 points.
You press <<R>> and get 51 points.

Game 2. There are 10 trials in this game.

You are instructed to press J and get 76 points.
You are instructed to press J and get 89 points.
You are instructed to press R and get 61 points.
You are instructed to press J and get 74 points.
You press <<R>> and get 59 points.
You press <<J>> and get 72 points.
You press <<J>> and get 61 points.
You press <<R>> and get 70 points.
You press <<R>> and get 73 points.
You press <<J>> and get 65 points.

Game 3. There are 10 trials in this game.

You are instructed to press R and get 60 points.
You are instructed to press R and get 43 points.
You are instructed to press J and get 54 points.
You are instructed to press J and get 65 points.
You press <<R>> and get 56 points.
You press <<J>> and get 38 points.
You press <<R>> and get 56 points.
You press <<J>> and get 61 points.
You press <<R>> and get 59 points.
You press <<J>> and get 56 points.

Game 4. There are 5 trials in this game.

You are instructed to press J and get 36 points.
You are instructed to press J and get 43 points.
You are instructed to press R and get 57 points.
You are instructed to press R and get 49 points.
You press <<R>> and get 63 points.

99
Game 5. There are 5 trials in this game.
You are instructed to press R and get 44 points.
You are instructed to press J and get 45 points.
You are instructed to press R and get 29 points.
You are instructed to press J and get 38 points.
You press <<R>> and get 43 points.

Game 6. There are 5 trials in this game.

You are instructed to press R and get 74 points.
You are instructed to press R and get 70 points.
You are instructed to press R and get 61 points.
You are instructed to press J and get 77 points.
You press <<J>> and get 74 points.

Game 7. There are 5 trials in this game.

You are instructed to press J and get 36 points.
You are instructed to press R and get 50 points.
You are instructed to press J and get 49 points.
You are instructed to press J and get 34 points.
You press <<R>> and get 48 points.

Game 8. There are 10 trials in this game.

You are instructed to press J and get 54 points.
You are instructed to press R and get 64 points.
You are instructed to press R and get 63 points.
You are instructed to press R and get 52 points.
You press <<J>> and get 63 points.
You press <<J>> and get 65 points.
You press <<R>> and get 70 points.
You press <<J>> and get 69 points.
You press <<R>> and get 64 points.
You press <<J>> and get 64 points.

Game 9. There are 5 trials in this game.

You are instructed to press J and get 57 points.
You are instructed to press R and get 50 points.
You are instructed to press R and get 57 points.
You are instructed to press J and get 72 points.
You press <<R>> and get 47 points.

Game 10. There are 10 trials in this game.

You are instructed to press R and get 21 points.
You are instructed to press J and get 26 points.
You are instructed to press J and get 52 points.
You are instructed to press J and get 27 points.

100
You press <<R>> and get 21 points.
You press <<J>> a

Structured bandit
Data source: [113]

Number of experiments: 5
Number of participants: 534
Number of choices: 160200

Example prompt:
You will be playing a game for 30 rounds.
Each round contains 10 trials.
In each trial, you have to select one option that will generate a reward between 0 and
50 points.
You can choose between options 1, 2, 3, 4, 5, 6, 7 and 8 by pressing the corresponding
key.
After each round the options reset and each option can produce different rewards in
the following round.
Your goal is to maximize your reward.

You are playing round 1:

You press <<1>> and get 13.927462234 points.
You press <<2>> and get 36.7688570508 points.
You press <<1>> and get 14.2022179045 points.
You press <<1>> and get 14.255711791 points.
You press <<1>> and get 14.0630012349 points.
You press <<1>> and get 13.7662251776 points.
You press <<1>> and get 14.0950976416 points.
You press <<1>> and get 13.9059322374 points.
You press <<1>> and get 13.7876455405 points.
You press <<1>> and get 14.0791620504 points.

You are playing round 2:

You press <<1>> and get 7.9614376644 points.
You press <<1>> and get 8.0581019194 points.
You press <<1>> and get 7.8981838872 points.
You press <<1>> and get 7.6801851393 points.
You press <<1>> and get 7.8750440099 points.
You press <<1>> and get 7.8730616431 points.
You press <<1>> and get 7.9118344028 points.
You press <<1>> and get 8.134691905 points.
You press <<1>> and get 7.3146307967 points.
You press <<1>> and get 7.5832954876 points.

101
You are playing round 3:
You press <<1>> and get 45.5086514616 points.
You press <<1>> and get 45.4708060494 points.
You press <<1>> and get 45.9992623342 points.
You press <<1>> and get 45.2562607277 points.
You press <<1>> and get 45.4858045741 points.
You press <<1>> and get 45.5714527483 points.
You press <<1>> and get 45.6341629546 points.
You press <<1>> and get 45.0394158823 points.
You press <<1>> and get 45.4305272183 points.
You press <<1>> and get 45.5876540303 points.

You are playing round 4:

You press <<1>> and get 11.3846751956 points.
You press <<1>> and get 11.5363886375 points.
You press <<1>> and get 11.7231131775 points.
You press <<1>> and get 11.6522950289 points.
You press <<1>> and get 11.526071202 points.
You press <<1>> and get 11.8137491734 points.
You press <<1>> and get 11.3114379632 points.
You press <<1>> and get 11.9434174388 points.
You press <<1>> and get 11.2375196571 points.
You press <<1>> and get 11.7253908566 points.

You are playing round 5:

You press <<1>> and get 6.3035553498 points.
You press <<1>> and get 6.1091934289 points.
You press <<1>> and get 5.8593245302 points.
You press <<1>> and get 6.7472106803 points.
You press <<1>> and get 5.8960908181 points.
You press <<1>> and get 6.4056392625 points.
You press <<1>> and get 6.2178725578 points.
You press <<1>> and get 6.0009585684 points.
You press <<1>> and get 6.5478514159 points.
You press <<1>> and get 6.1982281156 points.

You are playing round 6:

You press <<1>> and get 7.6893330342 points.
You press <<1>> and get 7.3522351643 points.
You press <<1>> and get 7.5121866021 points.
You press <<1>> and get 7.7384186564 points.
You press <<1>> and get 6.8107475324 points.
You press <<4>> and get 24.8575884026 points.
You press <<2>> and get 13.8675683188 points.
You press <<1>> and get 7.8646759927 points.

102
You press <<4>> and get 24.9148267139 points.
You press <<2>> and get 13.7239940234 points.

You are playing round 7:

You press <<1>> and get 4.8637779497 points.
You press <<4>> and get 21.4211685718 points.
You press <<2>> and get 9.9283331067 points.
You press <<1>> and get 4.3283739798 points.
You press <<1>> and get 4.2486815414 points.
You press <<2>> and get 9.7358045234 points.
You press <<3>> and get 15.5522984698 points.
You press <<4>> and get 21.1157043351 points.
You press <<4>> and get 21.2293801786 points.
You press <<4>> and get 21.2346665196 points.

You are playing round 8:

You press <<5>> and get 21.2600225368 points.
You press <<6>> and get 14.9281922416 points.
You press <<6>> and get 15.1378822594 points.
You press <<6>> and get 15.0143768307 points.
You press <<8>> and get 2.1704472833 points.
You press <<7>> and get 8.4601420045 points.
You press <<5>> and get

Horizon task
Data source: [114]

Number of experiments: 1
Number of participants: 78
Number of choices: 43680

Example prompt:
You are participating in multiple games involving two slot machines, labeled F and N.
The two slot machines are different across different games.
Each time you choose a slot machine, you get some points.
You choose a slot machine by pressing the corresponding key.
Each slot machine tends to pay out about the same amount of points on average.
Your goal is to choose the slot machines that will give you the most points across the
experiment.
The first 4 trials in each game are instructed trials where you will be told which slot
machine to choose.
After these instructed trials, you will have the freedom to choose for either 1 or 6
trials.

Game 1. There are 5 trials in this game.

103
You are instructed to press F and get 63 points.
You are instructed to press N and get 60 points.
You are instructed to press N and get 64 points.
You are instructed to press N and get 74 points.
You press <<N>> and get 65 points.

Game 2. There are 10 trials in this game.

You are instructed to press N and get 36 points.
You are instructed to press N and get 41 points.
You are instructed to press F and get 40 points.
You are instructed to press F and get 33 points.
You press <<N>> and get 44 points.
You press <<N>> and get 34 points.
You press <<F>> and get 48 points.
You press <<F>> and get 50 points.
You press <<N>> and get 38 points.
You press <<F>> and get 35 points.

Game 3. There are 5 trials in this game.

You are instructed to press F and get 18 points.
You are instructed to press N and get 51 points.
You are instructed to press N and get 41 points.
You are instructed to press F and get 23 points.
You press <<N>> and get 42 points.

Game 4. There are 10 trials in this game.

You are instructed to press F and get 65 points.
You are instructed to press N and get 55 points.
You are instructed to press N and get 68 points.
You are instructed to press N and get 61 points.
You press <<F>> and get 54 points.
You press <<N>> and get 74 points.
You press <<N>> and get 71 points.
You press <<N>> and get 42 points.
You press <<F>> and get 63 points.
You press <<N>> and get 53 points.

Game 5. There are 10 trials in this game.

You are instructed to press N and get 18 points.
You are instructed to press F and get 53 points.
You are instructed to press F and get 50 points.
You are instructed to press N and get 19 points.
You press <<F>> and get 46 points.
You press <<F>> and get 44 points.
You press <<N>> and get 10 points.

104
You press <<F>> and get 42 points.
You press <<F>> and get 39 points.
You press <<F>> and get 44 points.

Game 6. There are 10 trials in this game.

You are instructed to press F and get 13 points.
You are instructed to press N and get 51 points.
You are instructed to press F and get 1 points.
You are instructed to press N and get 36 points.
You press <<N>> and get 38 points.
You press <<N>> and get 36 points.
You press <<N>> and get 52 points.
You press <<N>> and get 36 points.
You press <<N>> and get 27 points.
You press <<F>> and get 12 points.

Game 7. There are 5 trials in this game.

You are instructed to press F and get 65 points.
You are instructed to press N and get 50 points.
You are instructed to press F and get 74 points.
You are instructed to press F and get 55 points.
You press <<N>> and get 72 points.

Game 8. There are 10 trials in this game.

You are instructed to press N and get 29 points.
You are instructed to press F and get 42 points.
You are instructed to press F and get 40 points.
You are instructed to press F and get 42 points.
You press <<F>> and get 53 points.
You press <<F>> and get 47 points.
You press <<N>> and get 40 points.
You press <<N>> and get 41 points.
You press <<F>> and get 38 points.
You press <<F>> and get 51 points.

Game 9. There are 5 trials in this game.

You are instructed to press N and get 60 points.
You are instructed to press N and get 66 points.
You are instructed to press F and get 61 points.
You are instructed to press F and get 65 points.
You press <<N>> and get 54 points.

Game 10. There are 10 trials in this game.

You are instructed to press F and get 44 points.
You are instructed to pr

105
Weather prediction task
Data source: [115]

Number of experiments: 1
Number of participants: 23
Number of choices: 4600

Example prompt:
You will be playing a game in which you pretend to be a weather forecaster.
In each trial, you will see between one and three tarot cards.
Your task is to decide if the combination of cards presented predicts rainy weather
(by pressing E) or fine weather (by pressing J).

You are seeing the following: card 2, card 4. You press <<E>>. You are correct, the
weather is indeed rainy.
You are seeing the following: card 1, card 2. You press <<J>>. You are correct, the
weather is indeed fine.
You are seeing the following: card 3, card 4. You press <<J>>. You are wrong, the
weather is rainy.
You are seeing the following: card 2, card 3. You press <<J>>. You are correct, the
weather is indeed fine.
You are seeing the following: card 1. You press <<E>>. You are wrong, the weather
is fine.
You are seeing the following: card 1, card 2, card 3. You press <<E>>. You are wrong,
the weather is fine.
You are seeing the following: card 1, card 2, card 4. You press <<J>>. You are wrong,
the weather is rainy.
You are seeing the following: card 1, card 2, card 4. You press <<E>>. You are
correct, the weather is indeed rainy.
You are seeing the following: card 3, card 4. You press <<E>>. You are correct, the
weather is indeed rainy.
You are seeing the following: card 3, card 4. You press <<E>>. You are correct, the
weather is indeed rainy.
You are seeing the following: card 1. You press <<J>>. You are correct, the weather
is indeed fine.
You are seeing the following: card 1, card 4. You press <<J>>. You are correct, the
weather is indeed fine.
You are seeing the following: card 4. You press <<E>>. You are wrong, the weather
is fine.
You are seeing the following: card 3, card 4. You press <<E>>. You are correct, the
weather is indeed rainy.
You are seeing the following: card 1, card 2. You press <<J>>. You are correct, the
weather is indeed fine.
You are seeing the following: card 3, card 4. You press <<E>>. You are correct, the
weather is indeed rainy.

106
You are seeing the following: card 2. You press <<J>>. You are correct, the weather
is indeed fine.
You are seeing the following: card 1, card 4. You press <<J>>. You are wrong, the
weather is rainy.
You are seeing the following: card 1, card 4. You press <<J>>. You are correct, the
weather is indeed fine.
You are seeing the following: card 1. You press <<J>>. You are correct, the weather
is indeed fine.
You are seeing the following: card 1, card 3, card 4. You press <<J>>. You are
correct, the weather is indeed fine.
You are seeing the following: card 2. You press <<J>>. You are correct, the weather
is indeed fine.
You are seeing the following: card 1, card 2. You press <<J>>. You are correct, the
weather is indeed fine.
You are seeing the following: card 2, card 3, card 4. You press <<E>>. You are
correct, the weather is indeed rainy.
You are seeing the following: card 3. You press <<J>>. You are correct, the weather
is indeed fine.
You are seeing the following: card 1, card 2, card 3. You press <<J>>. You are wrong,
the weather is rainy.
You are seeing the following: card 3, card 4. You press <<E>>. You are correct, the
weather is indeed rainy.
You are seeing the following: card 1, card 3. You press <<J>>. You are correct, the
weather is indeed fine.
You are seeing the following: card 1, card 4. You press <<J>>. You are wrong, the
weather is rainy.
You are seeing the following: card 2, card 3, card 4. You press <<E>>. You are wrong,
the weather is fine.
You are seeing the following: card 2, card 4. You press <<E>>. You are correct, the
weather is indeed rainy.
You are seeing the following: card 4. You press <<J>>. You are wrong, the weather
is rainy.
You are seeing the following: card 1, card 2. You press <<J>>. You are correct, the
weather is indeed fine.
You are seeing the following: card 1, card 3, card 4. You press <<E>>. You are
correct, the weather is indeed rainy.
You are seeing the following: card 4. You press <<J>>. You are wrong, the weather
is rainy.
You are seeing the following: card 1, card 2. You press <<J>>. You are correct, the
weather is indeed fine.
You are seeing the follow

Iowa gambling task

Data source: [116]

107
Number of experiments: 9
Number of participants: 511
Number of choices: 55435

Example prompt:
You see in front of you four decks of cards labeled H, V, J, and D.
You get a loan of 2000$ of play money.
You have to select one card at a time, from any of the four decks, for 100 trials.
You select a card from a deck by pressing the corresponding key.
After turning a card, you win some money, the amount varies with the deck.
You sometimes also have to pay a penalty, which also varies with the deck.
Your goal is to maximize profit on the loan of the play money.

You press <<H>>. You win 100.0$ and lose 200.0$.

You press <<H>>. You win 100.0$ and lose 150.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 250.0$.
You press <<J>>. You win 50.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<H>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<J>>. You win 50.0$ and lose 0.0$.
You press <<J>>. You win 50.0$ and lose 0.0$.
You press <<H>>. You win 100.0$ and lose 300.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 1250.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<J>>. You win 50.0$ and lose 50.0$.
You press <<H>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<J>>. You win 50.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<H>>. You win 100.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<H>>. You win 100.0$ and lose 0.0$.
You press <<J>>. You win 50.0$ and lose 50.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<H>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.

108
You press <<H>>. You win 100.0$ and lose 250.0$.
You press <<J>>. You win 50.0$ and lose 50.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<H>>. You win 100.0$ and lose 350.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<J>>. You win 50.0$ and lose 0.0$.
You press <<H>>. You win 100.0$ and lose 300.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 1250.0$.
You press <<H>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<J>>. You win 50.0$ and lose 50.0$.
You press <<H>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<J>>. You win 50.0$ and lose 50.0$.
You press <<H>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 250.0$.
You press <<J>>. You win 50.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<H>>. You win 100.0$ and lose 150.0$.
You press <<H>>. You win 100.0$ and lose 200.0$.
You press <<H>>. You win 100.0$ and lose 350.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 1250.0$.
You press <<J>>. You win 50.0$ and lose 0.0$.
You press <<J>>. You win 50.0$ and lose 50.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<V>>. You win 100.0$ and lose 0.0$.
You press <<D>>. You win 50.0$ and lose 0.0$.
Yo

109
Virtual subway network
Data source: [117]

Number of experiments: 4
Number of participants: 789
Number of choices: 227923

Example prompt:
Imagine that you are a tourist and you have to navigate the subway network in an
unfamiliar town.
In each round, you will have to navigate from a starting station to a goal station.
Please try to plan your trip as quickly as possible.
During your trip, you will see the name of the current station and its neighboring
stations in all four directions.
If there is no neighboring station in a particular direction, there will be a circle instead
of a station name.
You can go north by pressing G, west by pressing B, south by pressing V, and east by
pressing C.
When you reach the goal station, press Z to end the round and start the next round.

The new starting station is 1 and the goal station is 3.

Your station: 1. Neighboring stations: circle on the north, 2 on the east, circle on the
south, and 9 on the west. You press <<B>>.
Your station: 9. Neighboring stations: circle on the north, 1 on the east, 8 on the
south, and circle on the west. You press <<V>>.
Your station: 8. Neighboring stations: 9 on the north, circle on the east, 7 on the
south, and circle on the west. You press <<V>>.
Your station: 7. Neighboring stations: 8 on the north, 6 on the east, circle on the
south, and circle on the west. You press <<C>>.
Your station: 6. Neighboring stations: circle on the north, 5 on the east, circle on the
south, and 7 on the west. You press <<C>>.
Your station: 5. Neighboring stations: 4 on the north, circle on the east, circle on the
south, and 6 on the west. You press <<G>>.
Your station: 4. Neighboring stations: 3 on the north, circle on the east, 5 on the
south, and circle on the west. You press <<G>>.
Your station: 3. Neighboring stations: 2 on the north, circle on the east, 4 on the
south, and circle on the west. You press <<Z>>.
You are successful.

The new starting station is 1 and the goal station is 3.

Your station: 1. Neighboring stations: circle on the north, 2 on the east, circle on the
south, and 9 on the west. You press <<B>>.
Your station: 9. Neighboring stations: circle on the north, 1 on the east, 8 on the
south, and circle on the west. You press <<V>>.
Your station: 8. Neighboring stations: 9 on the north, circle on the east, 7 on the

110
south, and circle on the west. You press <<V>>.
Your station: 7. Neighboring stations: 8 on the north, 6 on the east, circle on the
south, and circle on the west. You press <<C>>.
Your station: 6. Neighboring stations: circle on the north, 5 on the east, circle on the
south, and 7 on the west. You press <<C>>.
Your station: 5. Neighboring stations: 4 on the north, circle on the east, circle on the
south, and 6 on the west. You press <<G>>.
Your station: 4. Neighboring stations: 3 on the north, circle on the east, 5 on the
south, and circle on the west. You press <<G>>.
Your station: 3. Neighboring stations: 2 on the north, circle on the east, 4 on the
south, and circle on the west. You press <<Z>>.
You are successful.

The new starting station is 4 and the goal station is 6.

Your station: 4. Neighboring stations: 3 on the north, circle on the east, 5 on the
south, and circle on the west. You press <<V>>.
Your station: 5. Neighboring stations: 4 on the north, circle on the east, circle on the
south, and 6 on the west. You press <<B>>.
Your station: 6. Neighboring stations: circle on the north, 5 on the east, circle on the
south, and 7 on the west. You press <<Z>>.
You are successful.

The new starting station is 1 and the goal station is 7.

Your station: 1. Neighboring stations: circle on the north, 2 on the east, circle on the
south, and 9 on the west. You press <<B>>.
Your station: 9. Neighboring stations: circle on the north, 1 on the east, 8 on the
south, and circle on the west. You press <<V>>.
Your station: 8. Neighboring stations: 9 on the north, circle on the east, 7 on the
south, and circle on the west. You press <<V>>.
Your station: 7. Neighboring stations: 8 on the north, 6 on the east, circle on the
south, and circle on the west. You press <<Z>>.
You are successful.

The new starting station is 9 and the goal station is 7.

Your station: 9. Neighboring stations: circle

Multi-task reinforcement learning

Data source: [93]

Number of experiments: 2
Number of participants: 380
Number of choices: 76760

Example prompt:
You will explore a castle, walking from room to room.

111
In each room, you will find different amounts of resources: wood, stone, and iron.
In each room, there are three doors that lead to different rooms.
The doors are labeled I, P, and G.
You have to choose the right doors to find the most valuable resources.
You choose a door by pressing the corresponding key.
At the beginning of each round, you will be shown how valuable the resources are.
These values are given as market prices for wood, stone, and iron.
Multiplying the prices with the amounts of resources and adding them up yields a
reward.
You want to maximize the cumulative reward.
After every round, you will start in room 0 again and see the new market prices.

The current market prices are 1 for wood, -1 for stone, and 0 for iron.
You are in room 0. You press <<P>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.
You are in room 2. You press <<P>> and you find 100 wood, 100 stone, and 0 iron.
You get 0 points.
The current market prices are -1 for wood, 1 for stone, and 0 for iron.
You are in room 0. You press <<I>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.
You are in room 1. You press <<G>> and you find 70 wood, 70 stone, and 70 iron.
You get 0 points.
The current market prices are -1 for wood, 1 for stone, and 0 for iron.
You are in room 0. You press <<P>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.
You are in room 2. You press <<G>> and you find 0 wood, 90 stone, and 0 iron. You
get 90 points.
The current market prices are -2 for wood, 1 for stone, and 0 for iron.
You are in room 0. You press <<P>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.
You are in room 2. You press <<G>> and you find 0 wood, 90 stone, and 0 iron. You
get 90 points.
The current market prices are 1 for wood, -2 for stone, and 0 for iron.
You are in room 0. You press <<I>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.
You are in room 1. You press <<G>> and you find 70 wood, 70 stone, and 70 iron.
You get -70 points.
The current market prices are -1 for wood, 1 for stone, and 0 for iron.
You are in room 0. You press <<P>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.
You are in room 2. You press <<G>> and you find 0 wood, 90 stone, and 0 iron. You
get 90 points.
The current market prices are 1 for wood, -2 for stone, and 0 for iron.
You are in room 0. You press <<P>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.

112
You are in room 2. You press <<G>> and you find 0 wood, 90 stone, and 0 iron. You
get -180 points.
The current market prices are 1 for wood, -1 for stone, and 0 for iron.
You are in room 0. You press <<G>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.
You are in room 3. You press <<P>> and you find 0 wood, 100 stone, and 60 iron.
You get -100 points.
The current market prices are 1 for wood, -2 for stone, and 0 for iron.
You are in room 0. You press <<I>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.
You are in room 1. You press <<G>> and you find 70 wood, 70 stone, and 70 iron.
You get -70 points.
The current market prices are -1 for wood, 1 for stone, and 0 for iron.
You are in room 0. You press <<P>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.
You are in room 2. You press <<G>> and you find 0 wood, 90 stone, and 0 iron. You
get 90 points.
The current market prices are 1 for wood, -2 for stone, and 0 for iron.
You are in room 0. You press <<I>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.
You are in room 1. You press <<G>> and you find 70 wood, 70 stone, and 70 iron.
You get -70 points.
The current market prices are -2 for wood, 1 for stone, and 0 for iron.
You are in room 0. You press <<P>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.
You are in room 2. You press <<G>> and you find 0 wood, 90 stone, and 0 iron. You
get 90 points.
The current market prices are 1 for wood, -2 for stone, and 0 for iron.
You are in room 0. You press <<I>> and you find 0 wood, 0 stone, and 0 iron. You
get 0 points.
You are in room

Horizon task
Data source: [118]

Number of experiments: 1
Number of participants: 36
Number of choices: 15290

Example prompt:
You are participating in multiple games involving two slot machines, labeled M and
U.
The two slot machines are different across different games.
Each time you choose a slot machine, you get some points.
You choose a slot machine by pressing the corresponding key.

113
Each slot machine tends to pay out about the same amount of points on average.
Your goal is to choose the slot machines that will give you the most points across the
experiment.
The first 4 trials in each game are instructed trials where you will be told which slot
machine to choose.
After these instructed trials, you will have the freedom to choose for either 1 or 6
trials.

Game 1. There are 10 trials in this game.

You are instructed to press U and get 45 points.
You are instructed to press M and get 25 points.
You are instructed to press U and get 38 points.
You are instructed to press M and get 12 points.
You press <<U>> and get 38 points.
You press <<U>> and get 42 points.
You press <<U>> and get 44 points.
You press <<U>> and get 35 points.
You press <<U>> and get 42 points.
You press <<U>> and get 45 points.

Game 2. There are 10 trials in this game.

You are instructed to press U and get 71 points.
You are instructed to press M and get 35 points.
You are instructed to press U and get 71 points.
You are instructed to press M and get 26 points.
You press <<U>> and get 61 points.
You press <<U>> and get 55 points.
You press <<U>> and get 61 points.
You press <<U>> and get 61 points.
You press <<U>> and get 67 points.
You press <<U>> and get 69 points.

Game 3. There are 5 trials in this game.

You are instructed to press U and get 50 points.
You are instructed to press M and get 43 points.
You are instructed to press M and get 46 points.
You are instructed to press U and get 45 points.
You press <<U>> and get 61 points.

Game 4. There are 10 trials in this game.

You are instructed to press U and get 52 points.
You are instructed to press M and get 51 points.
You are instructed to press U and get 57 points.
You are instructed to press M and get 54 points.
You press <<U>> and get 67 points.

114
You press <<U>> and get 58 points.
You press <<U>> and get 68 points.
You press <<U>> and get 57 points.
You press <<U>> and get 68 points.
You press <<U>> and get 65 points.

Game 5. There are 10 trials in this game.

You are instructed to press U and get 32 points.
You are instructed to press M and get 32 points.
You are instructed to press M and get 38 points.
You are instructed to press U and get 33 points.
You press <<M>> and get 37 points.
You press <<M>> and get 48 points.
You press <<M>> and get 38 points.
You press <<M>> and get 49 points.
You press <<M>> and get 36 points.
You press <<M>> and get 48 points.

Game 6. There are 5 trials in this game.

You are instructed to press U and get 74 points.
You are instructed to press M and get 51 points.
You are instructed to press M and get 72 points.
You are instructed to press M and get 76 points.
You press <<U>> and get 57 points.

Game 7. There are 10 trials in this game.

You are instructed to press U and get 40 points.
You are instructed to press U and get 34 points.
You are instructed to press M and get 31 points.
You are instructed to press M and get 37 points.
You press <<U>> and get 41 points.
You press <<U>> and get 43 points.
You press <<U>> and get 47 points.
You press <<U>> and get 41 points.
You press <<U>> and get 44 points.
You press <<U>> and get 45 points.

Game 8. There are 5 trials in this game.

You are instructed to press U and get 56 points.
You are instructed to press M and get 59 points.
You are instructed to press M and get 48 points.
You are instructed to press M and get 48 points.
You press <<U>> and get 71 points.

Game 9. There are 5 trials in this game.

115
You are instructed to press U and get 46 points.
You are instructed to press M and get 46 points.
You are instructed to press U and get 47 points.
You are instructed to press U and get 56 points.
You press <<U>> and get 47 points.

Game 10. There are 10 trials in this game.

You are instructed to press M and get 49 points.
You are instructed to p

Horizon task
Data source: [23] and unpublished data from the authors

Number of experiments: 5
Number of participants: 221
Number of choices: 138875

Example prompt:
You are participating in multiple games involving two slot machines, labeled C and
A.
The two slot machines are different across different games.
Each time you choose a slot machine, you get some points.
You choose a slot machine by pressing the corresponding key.
Each slot machine tends to pay out about the same amount of points on average.
Your goal is to choose the slot machines that will give you the most points across the
experiment.
The first 4 trials in each game are instructed trials where you will be told which slot
machine to choose.
After these instructed trials, you will have the freedom to choose for either 1 or 6
trials.

Game 1. There are 5 trials in this game.

You are instructed to press A and get 66 points.
You are instructed to press A and get 80 points.
You are instructed to press C and get 29 points.
You are instructed to press A and get 75 points.
You press <<A>> and get 81 points.

Game 2. There are 10 trials in this game.

You are instructed to press A and get 69 points.
You are instructed to press A and get 50 points.
You are instructed to press C and get 51 points.
You are instructed to press A and get 64 points.
You press <<C>> and get 42 points.
You press <<A>> and get 54 points.

116
You press <<A>> and get 64 points.
You press <<A>> and get 64 points.
You press <<A>> and get 57 points.
You press <<C>> and get 55 points.

Game 3. There are 10 trials in this game.

You are instructed to press A and get 31 points.
You are instructed to press C and get 43 points.
You are instructed to press A and get 26 points.
You are instructed to press C and get 36 points.
You press <<C>> and get 26 points.
You press <<C>> and get 41 points.
You press <<C>> and get 44 points.
You press <<C>> and get 44 points.
You press <<C>> and get 43 points.
You press <<C>> and get 53 points.

Game 4. There are 10 trials in this game.

You are instructed to press C and get 65 points.
You are instructed to press A and get 77 points.
You are instructed to press A and get 52 points.
You are instructed to press C and get 73 points.
You press <<C>> and get 61 points.
You press <<C>> and get 81 points.
You press <<C>> and get 70 points.
You press <<C>> and get 67 points.
You press <<A>> and get 62 points.
You press <<C>> and get 68 points.

Game 5. There are 10 trials in this game.

You are instructed to press A and get 70 points.
You are instructed to press C and get 19 points.
You are instructed to press A and get 43 points.
You are instructed to press C and get 41 points.
You press <<A>> and get 53 points.
You press <<C>> and get 19 points.
You press <<A>> and get 61 points.
You press <<A>> and get 68 points.
You press <<A>> and get 62 points.
You press <<C>> and get 46 points.

Game 6. There are 10 trials in this game.

You are instructed to press C and get 63 points.
You are instructed to press A and get 44 points.
You are instructed to press C and get 49 points.

117
You are instructed to press C and get 47 points.
You press <<A>> and get 52 points.
You press <<A>> and get 52 points.
You press <<C>> and get 55 points.
You press <<A>> and get 51 points.
You press <<A>> and get 34 points.
You press <<C>> and get 56 points.

Game 7. There are 5 trials in this game.

You are instructed to press C and get 61 points.
You are instructed to press A and get 44 points.
You are instructed to press A and get 41 points.
You are instructed to press A and get 47 points.
You press <<A>> and get 29 points.

Game 8. There are 5 trials in this game.

You are instructed to press C and get 51 points.
You are instructed to press A and get 76 points.
You are instructed to press C and get 54 points.
You are instructed to press A and get 84 points.
You press <<C>> and get 58 points.

Game 9. There are 10 trials in this game.

You are instructed to press C and get 54 points.
You are instructed to press A and get 14 points.
You are instructed to press A and get 15 points.
You are instructed to press C and get 46 points.
You press <<A>> and get 20 points.
You press <<C>> and get 44 points.
You press <<A>> and get 16 points.
You press <<C>> and get 46 points.
You press

Aversive learning
Data source: [119]

Number of experiments: 1
Number of participants: 57
Number of choices: 18240

Example prompt:
You are going to predict the probability of electric shocks associated with two visual
stimuli.
First, you will have to indicate the probability that a stimulus predicts a shock at the
current moment in time on a rating bar (between 0 and 100 percent).

118
After that, the outcome for each stimulus will be presented visually.
An upcoming shock will be indicated by a square over the stimulus, while a no-shock
outcome will be indicated by a circle.
Finally, shocks will be delivered after you have learned about the outcome visually.
If both stimuli indicate a shock, they will be presented one after the other in random
order.
The shock probability fluctuates over time such that one stimulus has a stable prob-
ability while the other varies.

Stimulus J and K are shown on the screen. You predict that the shock probability for
stimulus J is <<50.0>>% and the shock probability for stimulus K is <<50.0>>%.
After that, a circle is shown over stimulus J, and a square is shown over stimulus K.
Finally, a shock is delivered for stimulus K.
Stimulus J and K are shown on the screen. You predict that the shock probability for
stimulus J is <<50.0>>% and the shock probability for stimulus K is <<59.33>>%.
After that, a square is shown over stimulus J, and a circle is shown over stimulus K.
Finally, a shock is delivered for stimulus J.
Stimulus J and K are shown on the screen. You predict that the shock probability for
stimulus J is <<55.72>>% and the shock probability for stimulus K is <<59.33>>%.
After that, a circle is shown over stimulus J, and a circle is shown over stimulus K.
Finally, no shocks are delivered.
Stimulus J and K are shown on the screen. You predict that the shock probability for
stimulus J is <<55.72>>% and the shock probability for stimulus K is <<59.33>>%.
After that, a circle is shown over stimulus J, and a circle is shown over stimulus K.
Finally, no shocks are delivered.
Stimulus J and K are shown on the screen. You predict that the shock probability for
stimulus J is <<55.72>>% and the shock probability for stimulus K is <<59.33>>%.
After that, a circle is shown over stimulus J, and a circle is shown over stimulus K.
Finally, no shocks are delivered.
Stimulus J and K are shown on the screen. You predict that the shock probability for
stimulus J is <<55.72>>% and the shock probability for stimulus K is <<59.33>>%.
After that, a circle is shown over stimulus J, and a square is shown over stimulus K.
Finally, a shock is delivered for stimulus K.
Stimulus J and K are shown on the screen. You predict that the shock probability for
stimulus J is <<55.72>>% and the shock probability for stimulus K is <<62.42>>%.
After that, a circle is shown over stimulus J, and a circle is shown over stimulus K.
Finally, no shocks are delivered.
Stimulus J and K are shown on the screen. You predict that the shock probability for
stimulus J is <<55.72>>% and the shock probability for stimulus K is <<62.42>>%.
After that, a circle is shown over stimulus J, and a circle is shown over stimulus K.
Finally, no shocks are delivered.
Stimulus J and K are shown on the screen. You predict that the shock probability for
stimulus J is <<55.72>>% and the shock probability for stimulus K is <<62.42>>%.
After that, a square is shown over stimulus J, and a square is shown over stimulus K.
Finally, shocks are delivered for both stimulus J and stimulus K.

119
Stimulus J and K are shown on the screen. You predict that the shock probability for
stimulus J is <<58.2>>% and the shock probability for stimulus K is <<65.0>>%.
After that, a circle is shown over stimulus J, and a square is shown over stimulus K.
Finally, a shock is delivered for stimulus K.
Stimulus J and K are shown on the screen. You predict that the shock probability for
stimulus J is <<58.2>>% and the shock probability for stimulus K is <<67.48>>%.
After that, a circle is shown over stimulus J, and a circle is shown over stimulus K.
Finally, no shocks are delivered.
Stimulus J and K are shown on the screen. You predict that the shock probability for
stimulus J is <<58.2>>% and the shock probability

Spatially correlated multi-armed bandit

Data source: [94]

Number of experiments: 1
Number of participants: 78
Number of choices: 9360

Example prompt:
You will be presented with a series of 16 different environments to explore.
In each trial, you can select an option between numbers 1 and 30 by pressing the cor-
responding key.
By selecting any of these options, you will earn points associated with each unique
option.
Imagine these options 1 through 30 as lying next to each other in an ordered line;
options closer to each other tend to have similar rewards as rewards tend to cluster
together.
For each environment, you will be able to make either 5 or 10 choices.
When you made all your choices in a given environment, you will start making choices
in the next unexplored environment.
The rewards underlying the different options are different in each environment so you
will learn them anew for each environment.
Each environment starts with the value of a single option revealed.
When you choose the number corresponding to a different option, you will be told the
value of that option and receive those points.
Previously revealed options, including the starting option, can also be reselected,
although there may be small changes in the point value.
It is your task to gain as many points as possible across all 16 environments.

Environment number 1:
The value of option 4 is 76. You have 5 choices to make in this environment.
You press <<5>> and receive 61 points.
You press <<3>> and receive 64 points.
You press <<6>> and receive 42 points.
You press <<2>> and receive 49 points.

120
You press <<11>> and receive 22 points.

Environment number 2:
The value of option 2 is 65. You have 10 choices to make in this environment.
You press <<3>> and receive 68 points.
You press <<1>> and receive 40 points.
You press <<9>> and receive 10 points.
You press <<17>> and receive 51 points.
You press <<18>> and receive 35 points.
You press <<16>> and receive 54 points.
You press <<14>> and receive 48 points.
You press <<23>> and receive 64 points.
You press <<22>> and receive 53 points.
You press <<23>> and receive 65 points.

Environment number 3:
The value of option 22 is 37. You have 5 choices to make in this environment.
You press <<13>> and receive 30 points.
You press <<7>> and receive 67 points.
You press <<8>> and receive 77 points.
You press <<6>> and receive 32 points.
You press <<9>> and receive 57 points.

Environment number 4:
The value of option 16 is 34. You have 10 choices to make in this environment.
You press <<5>> and receive 77 points.
You press <<6>> and receive 53 points.
You press <<4>> and receive 79 points.
You press <<3>> and receive 41 points.
You press <<22>> and receive 25 points.
You press <<29>> and receive 41 points.
You press <<26>> and receive 46 points.
You press <<12>> and receive 64 points.
You press <<11>> and receive 41 points.
You press <<13>> and receive 55 points.

Environment number 5:
The value of option 19 is 26. You have 5 choices to make in this environment.
You press <<8>> and receive 42 points.
You press <<3>> and receive 44 points.
You press <<24>> and receive 37 points.
You press <<29>> and receive 35 points.
You press <<15>> and receive 48 points.

Environment number 6:

121
The value of option 16 is 44. You have 10 choices to make in this environment.
You press <<9>> and receive 32 points.
You press <<4>> and receive 43 points.
You press <<21>> and receive 26 points.
You press <<26>> and receive 5 points.
You press <<1>> and receive 13 points.
You press <<6>> and receive 9 points.
You press <<12>> and receive 27 points.
You press <<18>> and receive 36 points.
You press <<23>> and receive 23 points.
You press <<17>> and receive 21 points.

Environment number 7:
The value of option 2 is 36. You have 5 choices to make in this environment.
You press <<9>> and receive 49 points.
You press <<16>> and receive 72 points.
You press <<17>> and receive 59 points.
You press <<15>> and receive 49 points.
You press <<17>> and receive 58 points.

Environment number 8:
The value of option 9 is 66. You have 10 choices to make in this environment.
You press <<10>> and receive 66 points.
You press <<8>> and receive 37 points.
You press <<11>> and receive 38 points.
You pr

Serial reaction time task

Data source: [120]

Number of experiments: 2
Number of participants: 238
Number of choices: 238000

Example prompt:
Press the instructed key.

Act as fast and accurately as possible.

The instruction is to press D, you press <<D>> in 1375 ms. That is correct.
The instruction is to press K, you press <<K>> in 1960 ms. That is correct.
The instruction is to press J, you press <<J>> in 1043 ms. That is correct.
The instruction is to press F, you press <<F>> in 718 ms. That is correct.
The instruction is to press D, you press <<D>> in 568 ms. That is correct.
The instruction is to press K, you press <<K>> in 527 ms. That is correct.

122
The instruction is to press J, you press <<J>> in 587 ms. That is correct.
The instruction is to press K, you press <<K>> in 565 ms. That is correct.
The instruction is to press J, you press <<J>> in 541 ms. That is correct.
The instruction is to press F, you press <<F>> in 631 ms. That is correct.
The instruction is to press D, you press <<D>> in 456 ms. That is correct.
The instruction is to press K, you press <<K>> in 480 ms. That is correct.
The instruction is to press J, you press <<J>> in 627 ms. That is correct.
The instruction is to press J, you press <<J>> in 1816 ms. That is correct.
The instruction is to press F, you press <<F>> in 731 ms. That is correct.
The instruction is to press D, you press <<D>> in 563 ms. That is correct.
The instruction is to press K, you press <<K>> in 401 ms. That is correct.
The instruction is to press J, you press <<J>> in 472 ms. That is correct.
The instruction is to press D, you press <<D>> in 414 ms. That is correct.
The instruction is to press K, you press <<K>> in 412 ms. That is correct.
The instruction is to press J, you press <<J>> in 439 ms. That is correct.
The instruction is to press F, you press <<F>> in 612 ms. That is correct.
The instruction is to press K, you press <<K>> in 468 ms. That is correct.
The instruction is to press J, you press <<J>> in 497 ms. That is correct.
The instruction is to press F, you press <<F>> in 397 ms. That is correct.
The instruction is to press D, you press <<D>> in 318 ms. That is correct.
The instruction is to press K, you press <<K>> in 374 ms. That is correct.
The instruction is to press J, you press <<J>> in 327 ms. That is correct.
The instruction is to press F, you press <<F>> in 604 ms. That is correct.
The instruction is to press D, you press <<D>> in 300 ms. That is correct.
The instruction is to press K, you press <<K>> in 555 ms. That is correct.
The instruction is to press J, you press <<J>> in 314 ms. That is correct.
The instruction is to press K, you press <<K>> in 441 ms. That is correct.
The instruction is to press J, you press <<J>> in 474 ms. That is correct.
The instruction is to press F, you press <<D>> in 394 ms. That is incorrect.
The instruction is to press K, you press <<K>> in 586 ms. That is correct.
The instruction is to press K, you press <<K>> in 528 ms. That is correct.
The instruction is to press J, you press <<J>> in 466 ms. That is correct.
The instruction is to press F, you press <<F>> in 434 ms. That is correct.
The instruction is to press D, you press <<D>> in 355 ms. That is correct.
The instruction is to press J, you press <<K>> in 390 ms. That is incorrect.
The instruction is to press F, you press <<F>> in 769 ms. That is correct.
The instruction is to press D, you press <<D>> in 416 ms. That is correct.
The instruction is to press K, you press <<K>> in 395 ms. That is correct.
The instruction is to press K, you press <<K>> in 601 ms. That is correct.
The instruction is to press J, you press <<J>> in 475 ms. That is correct.
The instruction is to press J, you press <<J>> in 549 ms. That is correct.
The instruction is to press F, you press <<F>> in 438 ms. That is correct.
The instruction is to press D, you press <<D>> in 327 ms. That is correct.
The instruction is to press K, you press <<K>> in 438 ms. That is correct.
The instruction is to press J, you press <<J>> in 342 ms. That is correct.

123
The instruction is to press F, you press <<F>> in 456 ms. That is correct.
The instruction is to press D, you press <<D>> in 380 ms. That is correct.
The instruction is to press K, you press <<K¿

Decisions from description

Data source: [60]

Number of experiments: 1
Number of participants: 1981
Number of choices: 28153

Example prompt:
You will choose from two monetary lotteries by pressing W or H.
The lotteries offer different points with different probabilities.
Your choice will trigger a random draw from the chosen lottery that will be added to
your bonus.
Your goal is to maximize your bonus.
You will be presented with multiple choice problems consisting of different lotteries
varying in outcomes and probabilities.

Lottery W offers 4.0 points with 80.0% probability or 0.0 points with 20.0% probabil-
ity.
Lottery H offers 3.0 points with 100.0% probability.
You press <<H>>.

Lottery W offers 4.0 points with 20.0% probability or 0.0 points with 80.0% probabil-
ity.
Lottery H offers 3.0 points with 25.0% probability, or 0.0 points with 75.0% probabil-
ity.
You press <<H>>.

Lottery W offers -3.0 points with 100.0% probability.

Lottery H offers -32.0 points with 10.0% probability, or 0.0 points with 90.0% proba-
bility.
You press <<W>>.

Decisions from experience

Data source: [60]

Number of experiments: 79
Number of participants: 3942
Number of choices: 1015249

Example prompt:
You can sample from two monetary lotteries by pressing K or D.

124
The lotteries offer different points with different probabilities.
Initially, you will not know the outcomes and probabilities of the lotteries, but you
can learn about them through sampling.
Whenever you sample, a random draw from the selected lottery will be generated,
which does not affect your bonus.
You can sample from the lotteries in whatever order and for as long as you like.
Whenever you feel ready, you can stop sampling by pressing X and then choose one
lottery for real by pressing the corresponding key.
This choice will then trigger a random draw from the chosen lottery that will be added
to your bonus.
Your goal is to maximize your bonus.
You will be presented with multiple choice problems consisting of different lotteries
varying in outcomes and probabilities.

You encounter a new choice problem:

You press <<K>> and observe 4.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 4.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 3.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 4.0 points.

125
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 4.0 points.
You press <<X>> to stop sampling and then press <<K>>.

You encounter a new choice problem:

You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 3.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 4.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<K>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.

126
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 3.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.
You press <<D>> and observe 0.0 points.

Changing bandit
Data source: [121]

Number of experiments: 1
Number of participants: 30
Number of choices: 141000

Example prompt:
You are participating in multiple games involving two slot machines, labeled M and
V.
The two slot machines are different in different games.
Each time you choose a slot machine, you get points (choosing the same slot machine
will not always give you the same points).
You select a slot machine by pressing the corresponding key.
The expected points change randomly, abruptly, and independently with a hazard rate
(which you will be told).
When the points change, the new expected point value assigned to that slot machine
is sampled from a uniform distribution (from 1 to 99 points).
For example, if the hazard rate is 0.1, the expected points of the machines change
with 10%.
Your goal is to choose the slot machine that will give you the most points.

Game 1. The hazard rate is 0.1. There are 100 trials in this game.
You press <<M>> and get 65 points.
You press <<V>> and get 58 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<V>> and get 5 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<V>> and get 60 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<V>> and get 39 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.

127
You press <<M>> and get 65 points.
You press <<V>> and get 87 points.
You press <<V>> and get 82 points.
You press <<V>> and get 82 points.
You press <<V>> and get 56 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<V>> and get 28 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 65 points.
You press <<M>> and get 10 points.
You press <<V>> and get 28 points.
You press <<V>> and get 28 points.
You press <<V>> and get 28 points.
You press <<V>> and get 28 points.
You press <<V>> and get 28 points.
You press <<V>> and get 28 points.
You press <<V>> and get 28 points.
You press <<M>> and get 13 points.
You press <<V>> and get 28 points.
You press <<V>> and get 28 points.
You press <<V>> and get 28 points.
You press <<V>> and get 28 points.
You press <<V>> and get 28 points.
You press <<M>> and get 13 points.
You press <<V>> and get 29 points.
You press <<M>> and get 88 points.
You press <<M>> and get 88 points.
You press <<M>> and get 88 points.
You press <<M>> and get 88 points.
You press <<M>> and get 88 points.
You press <<M>> and get 88 points.

128
You press <<V>> and get 47 points.
You press <<M>> and get 88 points.
You press <<M>> and get 53 points.
You press <<V>> and get 18 points.
You press <<M>> and get 3 points.
You press <<V>> and get 18 points.
You press <<V>> and get 18 points.
You press <<V>> and get 18 points.
You press <<V>> and get 18 points.
You press <<V>> and get 18 points.
You press <<V>> and get 18 points.
You press <<V>> and get 18 points.
You press <<M>> and get 49 points.
You press <<M>> and get 49 points.
You press <<M>> and get 49 points.
You press <<M>> and get 49 points.
You press <<M>> and get 49 points.
You press <<M>> and get 49 points.
You press <<M>> and get 49 points.
You press <<M>> and get 49 points.
You press <<V>> and get 71 points.
You press <<V>> and get 71 points.
You press <<M>> and get 68 points.
You press <<V>> and get 59 points.
You press <<M>> and get 68 points.
You press <<M>> and get 68 points.
You press <<M>> and get 68 points.
You press <<M>> and get 68 points.
You press <<M>> and get 68 points.
You press <<M>> and get 68 points.
You press <<M>> and get 68 points.
You press <<M>> and get 68 points.
You press <<M>> and get 68 points.
You press <<M>> and g

Probabilistic reasoning
Data source: [122]

Number of experiments: 2
Number of participants: 128
Number of choices: 19740

Example prompt:
You have to estimate the probability of a series of weather-related queries.
Please respond by typing your estimated probabilities.

129
What is the probability that the weather will be icy and not frosty on a random day
in England? You estimate <<20>>%.
What is the probability that the weather will be not normal or not typical on a random
day in England? You estimate <<50>>%.
What is the probability that the weather will be not frosty on a random day in
England? You estimate <<65>>%.
If the weather in England is not frosty on a random day, what is the probability that
the weather will also be not icy on the same day? You estimate <<40>>%.
If the weather in England is icy on a random day, what is the probability that the
weather will also be not frosty on the same day? You estimate <<20>>%.
If the weather in England is typical on a random day, what is the probability that the
weather will also be normal on the same day? You estimate <<50>>%.
What is the probability that the weather will be icy and frosty on a random day in
England? You estimate <<35>>%.
What is the probability that the weather will be normal on a random day in England?
You estimate <<50>>%.
What is the probability that the weather will be icy or frosty on a random day in
England? You estimate <<35>>%.
What is the probability that the weather will be not icy or not frosty on a random
day in England? You estimate <<60>>%.
If the weather in England is not typical on a random day, what is the probability that
the weather will also be not normal on the same day? You estimate <<40>>%.
What is the probability that the weather will be normal and typical on a random day
in England? You estimate <<30>>%.
What is the probability that the weather will be not normal on a random day in
England? You estimate <<20>>%.
If the weather in England is not frosty on a random day, what is the probability that
the weather will also be icy on the same day? You estimate <<20>>%.
If the weather in England is not normal on a random day, what is the probability that
the weather will also be typical on the same day? You estimate <<40>>%.
If the weather in England is icy on a random day, what is the probability that the
weather will also be frosty on the same day? You estimate <<60>>%.
If the weather in England is not icy on a random day, what is the probability that the
weather will also be not frosty on the same day? You estimate <<60>>%.
What is the probability that the weather will be typical on a random day in England?
You estimate <<70>>%.
What is the probability that the weather will be typical or not normal on a random
day in England? You estimate <<50>>%.
What is the probability that the weather will be typical and not normal on a random
day in England? You estimate <<30>>%.
What is the probability that the weather will be not normal and not typical on a
random day in England? You estimate <<40>>%.
If the weather in England is frosty on a random day, what is the probability that the
weather will also be not icy on the same day? You estimate <<60>>%.

130
What is the probability that the weather will be frosty or not icy on a random day in
England? You estimate <<65>>%.
If the weather in England is not typical on a random day, what is the probability that
the weather will also be normal on the same day? You estimate <<34>>%.
If the weather in England is normal on a random day, what is the probability that the
weather will also be not typical on the same day? You estimate <<45>>%.
If the weather in England is normal on a random day, what is the probability that the
weather will also be typical on the same day? You estimate <<10>>%.
What is the probability that the weather will be icy or not frosty on a random day in
England? You estimate <<40>>%.
What is the probability that the weather will be frosty on a random day in England?
You estimate <<40>>%.
If the weather in England is frosty on a random day, what is the probability that the
weather will also be icy on the same day? You estimate <<40>>%.
If the weather in England is not icy on a random day, what is the probability that the
wea

Two-step task
Data source: [123]

Number of experiments: 1
Number of participants: 139
Number of choices: 55878

Example prompt:
You are participating in a space treasure game.
In this game, you will be visiting two alien planets in search of treasure.
Each planet has two aliens on it.
The blue aliens live on the blue planet.
The red aliens live on the red planet.
When you visit a planet, you can choose an alien to trade with by pressing the corre-
sponding button.
When you trade with an alien, it will either give you treasure or junk.
Your goal is to figure out, and trade with, the aliens that are most likely to give you
treasure.
To visit a planet, you will choose one rocket ship from two by pressing the correspond-
ing button.
They have different designations.
Each rocket ship has a planet it will fly to most of the time.
But sometimes they will take you to the other planet!
Remember the following hints:
1. How likely an alien is to give you treasure will change over time, but this change
will be slow.
2. Whether you get treasure depends only on the alien you choose to trade with.
3. If there is an alien you want to trade with, remember to pick the rocket ship that

131
is most likely to take you to that alien’s planet.

You are presented with two spaceships called S and C. You press <<S>>. You end
up on the blue planet. You see a blue alien named D and a blue alien named R. You
press <<R>>. You find junk.
You are presented with two spaceships called S and C. You press <<S>>. You end
up on the blue planet. You see a blue alien named D and a blue alien named R. You
press <<D>>. You find treasure.
You are presented with two spaceships called S and C. You press <<S>>. You end
up on the blue planet. You see a blue alien named D and a blue alien named R. You
press <<D>>. You find junk.
You are presented with two spaceships called S and C. You press <<C>>. You end
up on the red planet. You see a red alien named G and a red alien named V. You
press <<V>>. You find junk.
You are presented with two spaceships called S and C. You press <<S>>. You end
up on the red planet. You see a red alien named G and a red alien named V. You
press <<G>>. You find treasure.
You are presented with two spaceships called S and C. You press <<S>>. You end
up on the blue planet. You see a blue alien named D and a blue alien named R. You
press <<D>>. You find junk.
You are presented with two spaceships called S and C. You press <<S>>. You end
up on the blue planet. You see a blue alien named D and a blue alien named R. You
press <<R>>. You find junk.
You are presented with two spaceships called S and C. You press <<C>>. You end
up on the red planet. You see a red alien named G and a red alien named V. You
press <<G>>. You find junk.
You are presented with two spaceships called S and C. You press <<C>>. You end
up on the blue planet. You see a blue alien named D and a blue alien named R. You
press <<D>>. You find junk.
You are presented with two spaceships called S and C. You press <<S>>. You end
up on the blue planet. You see a blue alien named D and a blue alien named R. You
press <<R>>. You find treasure.
You are presented with two spaceships called S and C. You press <<S>>. You end
up on the red planet. You see a red alien named G and a red alien named V. You
press <<V>>. You find treasure.
You are presented with two spaceships called S and C. You press <<S>>. You end
up on the red planet. You see a red alien named G and a red alien named V. You
press <<V>>. You find treasure.
You are presented with two spaceships called S and C. You press <<S>>. You end
up on the blue planet. You see a blue alien named D and a blue alien named R. You
press <<D>>. You find treasure.
You are presented with two spaceships called S and C. You press <<S>>. You end
up on the blue planet. You see a blue alien named D and a blue alien named R. You
press <<R>>. You find junk.
You are presented with two spaceships called S and C. You press <<S>>. You end

132
up on the red planet. You see a red alien named G and a red alien named V. You
press <<V>>. You find junk.
You are presented with two spaceships called S and C. You press <<S>>. You end
up on the blue planet. You see a blue alien named D and a blue alien

Evaluation data
Two-step task (modified cover story)
Data source: [26]

Number of experiments: 1
Number of participants: 24
Number of choices: 9702

Example prompt:

You are playing the role of a musician living in a fantasy land.

You play the flute for gold coins to an audience of genies, who live inside magic lamps
on Pink Mountain and Blue Mountain.
Pink Mountain has genies H and J, and Blue Mountain has genies A and E.
Each genie lives in a lamp with the corresponding letter on it.
When you arrive on a mountain, you can pick up a lamp and rub it.
If the genie is in the mood for music, he will come out of his lamp, listen to a song,
and give you a gold coin.
Each genie’s interest in music changes with time.
To go to the mountains, you chose one of two magic carpets, which you purchase from
a magician, who enchants them to fly.
Magic carpet K generally flies to Pink Mountain, and magic carpet O generally flies
to Blue Mountain.
However, on rare occasions a strong wind blowing from that mountain makes flying
there too dangerous because the wind might blow you off the carpet.
In this case, the carpet is forced to land instead on the other mountain.
You can take a magic carpet or pick up a lamp and rub it by pressing the corresponding
key.
Your goal is to get as many coins as possible over the next 201 days.

You are presented with magic carpets K and O. You press <<K>>. You end up on
Pink Mountain. You see lamp H and lamp J. You rub lamp <<H>>. You receive 0
coins.
You are presented with magic carpets O and K. You press <<K>>. You end up on
Pink Mountain. You see lamp H and lamp J. You rub lamp <<J>>. You receive 1
coins.
You are presented with magic carpets K and O. You press <<K>>. You end up on
Pink Mountain. You see lamp H and lamp J. You rub lamp <<J>>. You receive 1
coins.

133
You are presented with magic carpets O and K. You press <<K>>. You end up on
Pink Mountain. You see lamp H and lamp J. You rub lamp <<J>>. You receive 1
coins.
You are presented with magic carpets K and O. You press <<K>>. You end up on
Pink Mountain. You see lamp H and lamp J. You rub lamp <<J>>. You receive 0
coins.
You are presented with magic carpets O and K. You press <<O>>. You end up on
Blue Mountain. You see lamp A and lamp E. You rub lamp <<A>>. You receive 1
coins.
You are presented with magic carpets O and K. You press <<O>>. You end up on
Blue Mountain. You see lamp A and lamp E. You rub lamp <<A>>. You receive 0
coins.
You are presented with magic carpets K and O. You press <<O>>. You end up on
Blue Mountain. You see lamp A and lamp E. You rub lamp <<E>>. You receive 1
coins.
You are presented with magic carpets K and O. You press <<O>>. You end up on
Blue Mountain. You see lamp A and lamp E. You rub lamp <<E>>. You receive 0
coins.
You are presented with magic carpets O and K. You press <<K>>. You end up on
Pink Mountain. You see lamp H and lamp J. You rub lamp <<J>>. You receive 0
coins.
You are presented with magic carpets O and K. You press <<K>>. You end up on
Pink Mountain. You see lamp H and lamp J. You rub lamp <<H>>. You receive 0
coins.
You are presented with magic carpets O and K. You press <<O>>. You end up on
Blue Mountain. You see lamp A and lamp E. You rub lamp <<A>>. You receive 1
coins.
You are presented with magic carpets K and O. You press <<O>>. You end up on
Pink Mountain. You see lamp H and lamp J. You rub lamp <<J>>. You receive 1
coins.
You are presented with magic carpets K and O. You press <<O>>. You end up on
Blue Mountain. You see lamp A and lamp E. You rub lamp <<A>>. You receive 0
coins.
You are presented with magic carpets O and K. You press <<K>>. You end up on
Blue Mountain. You see lamp A and lamp E. You rub lamp <<A>>. You receive 1
coins.
You are presented with magic carpets K and O. You press <<O>>. You end up on
Pink Mountain. You see lamp H and lamp J. You rub lamp <<J>>. You receive 0
coins.
You are presented with magic carpets K and O. You press <<K>>. You end up on
Pink Mountain. You see lamp H and lamp J. You rub lamp <<J>>. You receive 0
coins.
You are presented with magic carpets O and K. You press <<O>>. You end up on
Blue Mountain. You see lamp A and lamp E. You rub lamp <<A>>. You receive 0

134
coins.
You are presented with magic carpets O and K. You pr

Maggie’s farm (modified problem structure)

Data source: [27]

Number of experiments: 1
Number of participants: 658
Number of choices: 921200

Example prompt:

You are participating in multiple games involving three apple trees, labeled S, F, and
N.
The three apple trees are different across different games.
Each time you choose an apple tree, you get an apple of a given size.
You choose an apple tree by pressing the corresponding key.
Each apple tree tends to provide apples of about the same size on average.
Your goal is to choose the apple trees that will give you the largest apples across the
experiment.
The first few trials in each game are instructed trials where you will be told which
apple tree to choose.
After these instructed trials, you will have the freedom to choose for either 1 or 6
trials.

Game 1. There are 8 trials in this game.

You are instructed to press F and get an apple with size 3.0 centimeters.
You are instructed to press N and get an apple with size 2.0 centimeters.
You press <<S>> and get an apple with size 4.0 centimeters.
You press <<S>> and get an apple with size 4.0 centimeters.
You press <<S>> and get an apple with size 4.0 centimeters.
You press <<F>> and get an apple with size 5.0 centimeters.
You press <<N>> and get an apple with size 5.0 centimeters.
You press <<N>> and get an apple with size 4.0 centimeters.

Game 2. There are 10 trials in this game.

You are instructed to press F and get an apple with size 9.0 centimeters.
You are instructed to press N and get an apple with size 2.0 centimeters.
You are instructed to press F and get an apple with size 10.0 centimeters.
You are instructed to press F and get an apple with size 10.0 centimeters.
You press <<F>> and get an apple with size 10.0 centimeters.
You press <<F>> and get an apple with size 8.0 centimeters.
You press <<F>> and get an apple with size 9.0 centimeters.
You press <<F>> and get an apple with size 7.0 centimeters.
You press <<F>> and get an apple with size 9.0 centimeters.

135
You press <<F>> and get an apple with size 10.0 centimeters.

Game 3. There are 8 trials in this game.

You are instructed to press S and get an apple with size 2.0 centimeters.
You are instructed to press F and get an apple with size 6.0 centimeters.
You press <<F>> and get an apple with size 5.0 centimeters.
You press <<F>> and get an apple with size 6.0 centimeters.
You press <<F>> and get an apple with size 6.0 centimeters.
You press <<F>> and get an apple with size 5.0 centimeters.
You press <<F>> and get an apple with size 6.0 centimeters.
You press <<F>> and get an apple with size 7.0 centimeters.

Game 4. There are 11 trials in this game.

You are instructed to press F and get an apple with size 5.0 centimeters.
You are instructed to press S and get an apple with size 6.0 centimeters.
You are instructed to press F and get an apple with size 4.0 centimeters.
You are instructed to press F and get an apple with size 4.0 centimeters.
You are instructed to press N and get an apple with size 2.0 centimeters.
You press <<S>> and get an apple with size 6.0 centimeters.
You press <<S>> and get an apple with size 5.0 centimeters.
You press <<S>> and get an apple with size 3.0 centimeters.
You press <<S>> and get an apple with size 4.0 centimeters.
You press <<N>> and get an apple with size 2.0 centimeters.
You press <<S>> and get an apple with size 4.0 centimeters.

Game 5. There are 5 trials in this game.

You are instructed to press S and get an apple with size 6.0 centimeters.
You are instructed to press F and get an apple with size 7.0 centimeters.
You are instructed to press S and get an apple with size 6.0 centimeters.
You are instructed to press S and get an apple with size 5.0 centimeters.
You press <<S>> and get an apple with size 6.0 centimeters.

Game 6. There are 11 trials in this game.

You are instructed to press S and get an apple with size 6.0 centimeters.
You are instructed to press S and get an apple with size 4.0 centimeters.
You are instructed to press S and get an apple with size 5.0 centimeters.
You are instructed to press F and get an apple with size 3.0 centimeters.
You are instructed to press N and get an apple with size 4.0 centimeters.
You press <<S>> and get an apple with size 6.0 centime

Logical reasoning (entirely novel domain)

Data source: [28]

Number of experiments: 1
Number of participants: 3543

136
Number of choices: 99204

Example prompt:

You’re about to answer a set of 20 questions about logical reasoning. How many of
the 20 questions do you think you will answer correctly?
You say <<12>>.
Compared to other participants in this study, how well do you think you will do?
Marking 90% means you will do better than 90% of participants, marking 10% means
you will do better than only 10%, and marking 50% means that you will perform bet-
ter than half of the participants.
You say <<70>>%.
On a scale of 0 to 10, how difficult is solving logical reasoning problems for the aver-
age participant?
You say <<6>>.
On a scale of 0 to 10, how difficult is solving logical reasoning problems for you?
You say <<4>>.

You will be presented with brief passages or statements and will be required to eval-
uate their reasoning or determine what inferences you can logically draw from the
passage.
Your task is to use the buttons D, Z, F, O, and X to select the best answer choice,
even though more than one choice may present a possible answer.

Q1. Life imitates art. Which of the following, if true, most strongly supports the pre-
vious statement?
The choices are:
Z: When Warren Beatty filmed Reds, he tried to suggest not only the chaos of the
Russian Revolution but also its relationship to the present.
X: The number of professional ballet companies has increased over the last five years,
but the number of dance majors has decreased.
D: On Tuesday, the business section of the newspaper had predicted the drop in inter-
est rates that occurred on Friday.
O: Truman Capote wrote In Cold Blood as a result of a series of brutal slayings by
two crazed killers.
F: Soon after the advent of color television, white shirts became less popular as dressy
attire for men, and pastel-colored shirts began to sell well.
You press <<F>>.

Q2. On average, federal workers receive salaries 35.5 percent higher than private-sector
salaries. For instance, federal workers in California average $19,206 a year, 25 percent
higher than the average pay in the private sector, which is $15,365. This information
would best support which of the following opinions?
The choices are:
Z: Private-sector salaries in California are above average.

137
X: The private sector is being paid fairly.
O: Federal jobs are more secure than private-sector jobs.
D: Public-sector work is more difficult than private-sector work.
F: Federal pay is out of line.
You press <<Z>>.

Q3. No high jumper entered the track meet unless he or she was a track club member.
No track club member both entered the meet and was a high jumper. Which of the
following conclusions can be correctly drawn from the two previous sentences?
The choices are:
F: No one but high jumpers entered the meet.
D: Only track club members entered the meet.
X: No track club members entered the meet.
Z: No high jumper entered the meet.
O: Some track club members entered the meet.
You press <<Z>>.

Q4. About 33% of American men between 25 and 50 are overweight. Research has
shown that in most cases men between 25 and 50 who are overweight are more sub-
ject to heart disease than men who are not overweight. Which of the following is the
most logical conclusion to this argument?
The choices are:
D: Therefore, 33% of the American men between 25 and 50 should lose weight.
O: Therefore, if 33% of the American men between 25 and 50 were to lose weight,
they would reduce their risk of heart disease.
X: Therefore, if the men between 25 and 50 who are overweight were to lose weight,
they would reduce their risk of heart disease by 33%.
Z: Therefore, if 33% of American men were to lose weight, they would reduce their
risk of heart disease.
F: Therefore, if the overweight men between 25 and 50 were to lose weight, their risk
of heart disease would be reduced.
You press <<F>>.

Q5. All computer geniuses are also brilliant mathematicians. Therefore, some com-
puter geniuses don’t require calculators for simple multiplication facts. Which of the
following is the least necessary assumption for the previous conclusion to be logically
correct?
The choices are:
F: Some brilliant mathematicians don’t require calcul

Two-step task (neural alignment)

Data source: [41]

Number of experiments: 1
Number of participants: 94

138
Number of choices: 28153

Example prompt:

You are playing multiple rounds of a game.

Your goal is to collect as many gold coins as possible as you visit different states.
If you are in state C, you have the choice between options B and H.
If you are in state S, you have the choice between options G and X.
Picking one of these options may result in a gold coin.
How likely an option leads to a gold coin slowly changes during the game.
Picking option O generally leads to state C, and picking option N generally leads to
state S.
However, on rare occasions you will end up in the other state.
You can select an option by pressing the corresponding key.

You are presented with options O and N. You press <<N>>. You end up in state
S. You are presented with option G and option X. You press <<X>>. You receive 1
coins.
You are presented with options O and N. You press <<N>>. You end up in state
S. You are presented with option G and option X. You press <<X>>. You receive 1
coins.
You are presented with options N and O. You press <<N>>. You end up in state
S. You are presented with option G and option X. You press <<X>>. You receive 1
coins.
You are presented with options N and O. You press <<O>>. You end up in state
C. You are presented with option B and option H. You press <<B>>. You receive 1
coins.
You are presented with options N and O. You press <<O>>. You end up in state
S. You are presented with option G and option X. You press <<X>>. You receive 1
coins.
You are presented with options N and O. You press <<O>>. You end up in state
C. You are presented with option B and option H. You press <<B>>. You receive 1
coins.
You are presented with options O and N. You press <<N>>. You end up in state
S. You are presented with option G and option X. You press <<X>>. You receive 0
coins.
You are presented with options O and N. You press <<N>>. You end up in state
C. You are presented with option B and option H. You press <<H>>. You receive 1
coins.
You are presented with options N and O. You press <<O>>. You end up in state
C. You are presented with option B and option H. You press <<H>>. You receive 1
coins.
You are presented with options N and O. You press <<O>>. You end up in state
S. You are presented with option G and option X. You press <<G>>. You receive 0
coins.

139
You are presented with options O and N. You press <<N>>. You end up in state
S. You are presented with option G and option X. You press <<X>>. You receive 1
coins.
You are presented with options N and O. You press <<O>>. You end up in state
C. You are presented with option B and option H. You press <<H>>. You receive 1
coins.
You are presented with options O and N. You press <<N>>. You end up in state
S. You are presented with option G and option X. You press <<X>>. You receive 0
coins.
You are presented with options O and N. You press <<N>>. You end up in state
S. You are presented with option G and option X. You press <<G>>. You receive 0
coins.
You are presented with options N and O. You press <<O>>. You end up in state
C. You are presented with option B and option H. You press <<B>>. You receive 1
coins.
You are presented with options O and N. You press <<N>>. You end up in state
S. You are presented with option G and option X. You press <<X>>. You receive 1
coins.
You are presented with options N and O. You press <<O>>. You end up in state
S. You are presented with option G and option X. You press <<X>>. You receive 1
coins.
You are presented with options O and N. You press <<N>>. You end up in state
S. You are presented with option G and option X. You press <<X>>. You receive 1
coins.
You are presented with options O and N. You press <<N>>. You end up in state
S. You are presented with option G and option X. You press <<X>>. You receive 0
coins.
You are presented with options O and N. You press <<N>>. You end up in state
C. You are presented with option B and option H. You press <<H>>. You receive 1
coins.
You are presented with options O and N. You press <<N>>. You end up in state
C. You are presented with option B and option H. You press <<H>>. You receive 1
coins.
You are presented with options N and O.

Sentence reading (neural alignment)

Data source: [42]

Number of experiments: 1
Number of participants: 5
Number of choices: 0

Example prompt:

We were sitting on the couch.

140
References
[1] Anderson, J. The Architecture of Cognition (1983).

[2] Newell, A. Unified Theories of Cognition (Harvard University Press, Cambridge,

1990).

[3] Lake, B. M., Ullman, T. D., Tenenbaum, J. B. & Gershman, S. J. Building

machines that learn and think like people. Behavioral and brain sciences 40,
e253 (2017).

[4] Lake, B. M. & Baroni, M. Human-like systematic generalization through a

meta-learning neural network. Nature 623, 115–121 (2023).

[5] Wu, C. M., Meder, B. & Schulz, E. Unifying principles of generalization: past,
present, and future. Annu. Rev. Psych 76, 1–33 (2024).

[6] Lake, B. M., Salakhutdinov, R. & Tenenbaum, J. B. Human-level concept

learning through probabilistic program induction. Science 350, 1332–1338
(2015).

[7] Goddu, M. K. & Gopnik, A. The development of human causal learning and
reasoning. Nature Reviews Psychology 1–21 (2024).

[8] Chu, J. & Schulz, L. E. Play, curiosity, and cognition. Annual Review of
Developmental Psychology 2, 317–343 (2020).

[9] Silver, D. et al. Mastering the game of go without human knowledge. nature
550, 354–359 (2017).

[10] Kahneman, D. & Tversky, A. Prospect theory: An analysis of decision under

risk (2013).

[11] Riveland, R. & Pouget, A. Natural language instructions induce compositional

generalization in networks of neurons. Nature Neuroscience 27, 988–999 (2024).

[12] Bommasani, R. et al. On the opportunities and risks of foundation models.

arXiv preprint arXiv:2108.07258 (2021).

[13] Dubey, A. et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783
(2024).

[14] Binz, M. & Schulz, E. Using cognitive psychology to understand gpt-3.

Proceedings of the National Academy of Sciences 120, e2218523120 (2023).

[15] Binz, M. & Schulz, E. Turning large language models into cognitive models
(2024).

141
[16] Hofman, J. M. et al. Integrating explanation and prediction in computational
social science. Nature 595, 181–188 (2021).

[17] Rocca, R. & Yarkoni, T. Putting psychology to the test: Rethinking model
evaluation through benchmarking and prediction. Advances in Methods and
Practices in Psychological Science 4, 25152459211026864 (2021).

[18] Petroni, F. et al. Inui, K., Jiang, J., Ng, V. & Wan, X. (eds) Language models as
knowledge bases? (eds Inui, K., Jiang, J., Ng, V. & Wan, X.) Proceedings of the
2019 Conference on Empirical Methods in Natural Language Processing and the
9th International Joint Conference on Natural Language Processing (EMNLP-
IJCNLP), 2463–2473 (Association for Computational Linguistics, Hong Kong,
China, 2019). URL https://aclanthology.org/D19-1250.

[19] Dettmers, T., Pagnoni, A., Holtzman, A. & Zettlemoyer, L. Qlora: Efficient fine-
tuning of quantized llms. Advances in Neural Information Processing Systems
36 (2024).

[20] Nosofsky, R. M. The generalized context model: An exemplar model of

classification. Formal approaches in categorization 18–39 (2011).

[21] Peterson, J. C., Bourgin, D. D., Agrawal, M., Reichman, D. & Griffiths, T. L.
Using large-scale experiments and machine learning to discover theories of
human decision-making. Science 372, 1209–1214 (2021).

[22] Daw, N. D., Gershman, S. J., Seymour, B., Dayan, P. & Dolan, R. J. Model-
based influences on humans’ choices and striatal prediction errors. Neuron 69,
1204–1215 (2011).

[23] Wilson, R. C., Geana, A., White, J. M., Ludvig, E. A. & Cohen, J. D. Humans
use directed and random exploration to solve the explore–exploit dilemma.
Journal of experimental psychology: General 143, 2074 (2014).

[24] Palminteri, S., Wyart, V. & Koechlin, E. The importance of falsification in

computational cognitive modeling. Trends in cognitive sciences 21, 425–433
(2017).

[25] van Baar, J. M., Nassar, M. R., Deng, W. & FeldmanHall, O. Latent motives
guide structure learning during adaptive social choice. Nature Human Behaviour
6, 404–414 (2022).

[26] Feher da Silva, C. & Hare, T. A. Humans primarily use model-based inference
in the two-stage task. Nature Human Behaviour 4, 1053–1066 (2020).

[27] Dubois, M. & Hauser, T. U. Value-free random exploration is linked to

impulsivity. Nature Communications 13, 4542 (2022).

142
[28] Jansen, R. A., Rafferty, A. N. & Griffiths, T. L. A rational model of the dunning–
kruger effect supports insensitivity to evidence in low performers. Nature Human
Behaviour 5, 756–763 (2021).

[29] Kool, W., Cushman, F. A. & Gershman, S. J. When does model-based control
pay off? PLoS computational biology 12, e1005090 (2016).

[30] Kool, W., Gershman, S. J. & Cushman, F. A. Cost-benefit arbitration between

multiple reinforcement-learning systems. Psychological science 28, 1321–1333
(2017).

[31] Awad, E. et al. The moral machine experiment. Nature 563, 59–64 (2018).

[32] Akata, E. et al. Playing repeated games with large language models. arXiv
preprint arXiv:2305.16867 (2023).

[33] Demircan, C. et al. Evaluating alignment between humans and neural net-
work representations in image-based learning tasks. The Thirty-eighth Annual
Conference on Neural Information Processing Systems (2024).

[34] Singh, M., Richie, R. & Bhatia, S. Representing and predicting everyday
behavior. Computational Brain & Behavior 5, 1–21 (2022).

[35] Xu, H. A., Modirshanechi, A., Lehmann, M. P., Gerstner, W. & Herzog, M. H.
Novelty is not surprise: Human exploratory and adaptive behavior in sequential
decision-making. PLOS Computational Biology 17, e1009070 (2021).

[36] Hick, W. E. On the rate of gain of information. Quarterly Journal of

experimental psychology 4, 11–26 (1952).

[37] Coda-Forno, J., Binz, M., Wang, J. X. & Schulz, E. Salakhutdinov, R. et al. (eds)
CogBench: a large language model walks into a psychology lab. (eds Salakhut-
dinov, R. et al.) Proceedings of the 41st International Conference on Machine
Learning, Vol. 235 of Proceedings of Machine Learning Research, 9076–9108
(PMLR, 2024). URL https://proceedings.mlr.press/v235/coda-forno24a.html.

[38] Kipnis, A., Voudouris, K., Buschoff, L. M. S. & Schulz, E. metabench–a sparse
benchmark to measure general ability in large language models. arXiv preprint
arXiv:2407.12844 (2024).

[39] Yamins, D. L. et al. Performance-optimized hierarchical models predict neural

responses in higher visual cortex. Proceedings of the national academy of sciences
111, 8619–8624 (2014).

[40] Schrimpf, M. et al. The neural architecture of language: Integrative modeling

converges on predictive processing. Proceedings of the National Academy of
Sciences 118, e2105646118 (2021).

143
[41] Feher da Silva, C., Lombardi, G., Edelson, M. & Hare, T. A. Rethinking model-
based and model-free influences on mental effort and striatal prediction errors.
Nature Human Behaviour 7, 956–969 (2023).

[42] Tuckute, G. et al. Driving and suppressing the human language network using
large language models. Nature Human Behaviour 8, 544–561 (2024).

[43] Lad, V., Gurnee, W. & Tegmark, M. The remarkable robustness of llms: Stages
of inference? arXiv preprint arXiv:2406.19384 (2024).

[44] Cochran, W. G. The combination of estimates from different experiments.

Biometrics 10, 101–129 (1954).

[45] DeepSeek-AI et al. Deepseek-r1: Incentivizing reasoning capability in llms via

reinforcement learning (2025). URL https://arxiv.org/abs/2501.12948. 2501.
12948.

[46] Hilbig, B. E. & Moshagen, M. Generalized outcome-based strategy classification:

Comparing deterministic and probabilistic choice models. Psychonomic bulletin
& review 21, 1431–1443 (2014).

[47] Agrawal, M., Peterson, J. C. & Griffiths, T. L. Scaling up psychology via scien-
tific regret minimization. Proceedings of the National Academy of Sciences 117,
8825–8835 (2020).

[48] Binz, M., Gershman, S. J., Schulz, E. & Endres, D. Heuristics from bounded
meta-learned inference. Psychological review 129, 1042 (2022).

[49] Musslick, S. et al. Automating the practice of science – opportunities, challenges,

and implications (2024). URL https://arxiv.org/abs/2409.05890. 2409.05890.

[50] Musslick, S., Strittmatter, Y. & Dubova, M. Closed-loop scientific discovery in

the behavioral sciences (2024).

[51] Rmus, M., Jagadish, A. K., Mathony, M., Ludwig, T. & Schulz, E. Towards
automation of cognitive modeling using large language models. arXiv preprint
arXiv:2502.00879 (2025).

[52] Dillion, D., Tandon, N., Gu, Y. & Gray, K. Can ai language models replace
human participants? Trends in Cognitive Sciences 27, 597–600 (2023).

[53] Huben, R., Cunningham, H., Smith, L. R., Ewart, A. & Sharkey, L. Sparse
autoencoders find highly interpretable features in language models (2024). URL
https://openreview.net/forum?id=F76bwRSLeK.

[54] Chefer, H., Gur, S. & Wolf, L. Transformer interpretability beyond attention
visualization (2021).

144
[55] Vaswani, A. Attention is all you need. Advances in Neural Information
Processing Systems (2017).

[56] Smith, J. T., Warrington, A. & Linderman, S. Simplified state space layers for
sequence modeling.

[57] Zador, A. et al. Catalyzing next-generation artificial intelligence through

neuroai. Nature communications 14, 1597 (2023).

[58] Doerig, A. et al. The neuroconnectionist research programme. Nature Reviews

Neuroscience 24, 431–450 (2023).

[59] Ruggeri, K. et al. The globalizability of temporal discounting. Nature Human

Behaviour 6, 1386–1397 (2022).

[60] Wulff, D. U., Mergenthaler-Canseco, M. & Hertwig, R. A meta-analytic review of

two modes of learning and the description-experience gap. Psychological bulletin
144, 140 (2018).

[61] Frey, R., Pedroni, A., Mata, R., Rieskamp, J. & Hertwig, R. Risk preference
shares the psychometric structure of major psychological traits. Science advances
3, e1701381 (2017).

[62] Enkavi, A. Z. et al. Large-scale analysis of test–retest reliabilities of self-

regulation measures. Proceedings of the National Academy of Sciences 116,
5472–5477 (2019).

[63] Henrich, J., Heine, S. J. & Norenzayan, A. Most people are not weird. Nature
466, 29–29 (2010).

[64] Schrimpf, M. et al. Integrative benchmarking to advance neurally mechanistic

models of human intelligence. Neuron 108, 413–423 (2020).

[65] Schrimpf, M. et al. Brain-score: Which artificial neural network for object
recognition is most brain-like? BioRxiv 407007 (2018).

[66] Poldrack, R. A. et al. The past, present, and future of the brain imaging data
structure (bids). Imaging Neuroscience 2, 1–19 (2024).

[67] Markiewicz, C. J. et al. The openneuro resource for sharing of neuroscience

data. Elife 10, e71774 (2021).

[68] Schulze Buschoff, L. M., Akata, E., Bethge, M. & Schulz, E. Visual cognition in
multimodal large language models. Nature Machine Intelligence 1–11 (2025).

[69] Vere, S. A. A cognitive process shell. Behavioral and Brain Sciences 15, 460–461
(1992).

145
[70] Loshchilov, I. & Hutter, F. Decoupled weight decay regularization (2019). URL
https://arxiv.org/abs/1711.05101. 1711.05101.

[71] Schaefer, A. et al. Local-global parcellation of the human cerebral cortex from
intrinsic functional connectivity mri. Cerebral cortex 28, 3095–3114 (2018).

[72] Esteban, O. et al. fmriprep: a robust preprocessing pipeline for functional mri.
Nature methods 16, 111–116 (2019).

[73] Fonov, V. S., Evans, A. C., McKinstry, R. C., Almli, C. R. & Collins, D. Unbi-
ased nonlinear average age-appropriate brain templates from birth to adulthood.
NeuroImage 47, S102 (2009).

[74] Penny, W. D., Friston, K. J., Ashburner, J. T., Kiebel, S. J. & Nichols, T. E.
Statistical parametric mapping: the analysis of functional brain images (Elsevier,
2011).

[75] contributors, N. nilearn. URL https://github.com/nilearn/nilearn.

[76] Yax, N., Oudeyer, P.-Y. & Palminteri, S. Assessing contamination in

large language models: Introducing the logprober method. arXiv preprint
arXiv:2408.14352 (2024).

[77] Warner, B. et al. Smarter, better, faster, longer: A modern bidirectional encoder
for fast, memory efficient, and long context finetuning and inference (2024). URL
https://arxiv.org/abs/2412.13663. 2412.13663.

[78] Wang, Z. et al. Helpsteer2-preference: Complementing ratings with preferences.

arXiv preprint arXiv:2410.01257 (2024).

[79] Teknium, R., Quesnelle, J. & Guang, C. Hermes 3 technical report. arXiv
preprint arXiv:2408.11857 (2024).

[80] Lin, S., Hilton, J. & Evans, O. Truthfulqa: Measuring how models mimic human
falsehoods. CoRR abs/2109.07958 (2021). URL https://arxiv.org/abs/2109.
07958.

[81] Anderson, J. R. & Lebiere, C. The newell test for a theory of cognition.
Behavioral and brain Sciences 26, 587–601 (2003).

[82] Anderson, J. R. The adaptive character of thought (Erlbaum, Hillsdale, NJ,

1990).

[83] Binz, M. et al. Meta-learned models of cognition. Behavioral and Brain Sciences
1–38 (2023).

[84] Andreas, J. Goldberg, Y., Kozareva, Z. & Zhang, Y. (eds) Language models as
agent models. (eds Goldberg, Y., Kozareva, Z. & Zhang, Y.) Findings of the

146
 Association for Computational Linguistics: EMNLP 2022, 5769–5779 (Associa-
tion for Computational Linguistics, Abu Dhabi, United Arab Emirates, 2022).
URL https://aclanthology.org/2022.findings-emnlp.423.

[85] Paszke, A. et al. PyTorch: an imperative style, high-performance deep learning

library (Curran Associates Inc., Red Hook, NY, USA, 2019).

[86] Defazio, A. et al. The road less scheduled (2024). URL https://arxiv.org/abs/
2405.15682. 2405.15682.

[87] Van den Bos, W. & McClure, S. M. Towards a general model of temporal
discounting. Journal of the experimental analysis of behavior 99, 58–73 (2013).

[88] Sutton, R. S. & Barto, A. G. Reinforcement learning: An introduction (MIT

press, 2018).

[89] Gershman, S. J. A unifying probabilistic view of associative learning. PLoS

computational biology 11, e1004567 (2015).

[90] Czerlinski, J., Gigerenzer, G. & Goldstein, D. G. How good are simple heuristics?
(1999).

[91] Pedroni, A. et al. A prospect theory account of the “hot” columbia card task
(2018).

[92] Hebart, M. N., Zheng, C. Y., Pereira, F. & Baker, C. I. Revealing the multidi-
mensional mental representations of natural objects underlying human similarity
judgements. Nature human behaviour 4, 1173–1185 (2020).

[93] Tomov, M. S., Schulz, E. & Gershman, S. J. Multi-task reinforcement learning

in humans. Nature Human Behaviour 5, 764–773 (2021).

[94] Wu, C. M., Schulz, E., Speekenbrink, M., Nelson, J. D. & Meder, B. Generaliza-
tion guides human exploration in vast decision spaces. Nature human behaviour
2, 915–924 (2018).

[95] Badham, S. P., Sanborn, A. N. & Maylor, E. A. Deficits in category learning in

older adults: Rule-based versus clustering accounts. Psychology and Aging 32,
473 (2017).

[96] Bahrami, B. & Navajas, J. 4 arm bandit task dataset (2022). URL osf.io/f3t2a.

[97] Collsiöö, A., Juslin, P. & Winman, A. Is numerical information always bene-
ficial? verbal and numerical cue-integration in additive and non-additive tasks.
Cognition 240, 105584 (2023).

[98] Cox, G. E., Hemmer, P., Aue, W. R. & Criss, A. H. Information and processes
underlying semantic and episodic memory across tasks, items, and individuals.

147
 Journal of Experimental Psychology: General 147, 545 (2018).

[99] Feng, S. F., Wang, S., Zarnescu, S. & Wilson, R. C. The dynamics of explore–
exploit decisions reveal a signal-to-noise mechanism for random exploration.
Scientific reports 11, 3077 (2021).

[100] Flesch, T., Balaguer, J., Dekker, R., Nili, H. & Summerfield, C. Comparing
continual task learning in minds and machines. Proceedings of the National
Academy of Sciences 115, E10313–E10322 (2018).

[101] Garcia, B., Lebreton, M., Bourgeois-Gironde, S. & Palminteri, S. Experiential

values are underweighted in decisions involving symbolic options. Nature human
behaviour 7, 611–626 (2023).

[102] Gershman, S. J. Deconstructing the human algorithms for exploration. Cognition

173, 34–42 (2018).

[103] Collins, A. G., Brown, J. K., Gold, J. M., Waltz, J. A. & Frank, M. J. Working
memory contributions to reinforcement learning impairments in schizophrenia.
Journal of Neuroscience 34, 13747–13756 (2014).

[104] Hebart, M. N. et al. Things-data, a multimodal collection of large-scale datasets

for investigating object representations in human brain and behavior. Elife 12,
e82580 (2023).

[105] Krueger, P. M., Callaway, F., Gul, S., Griffiths, T. L. & Lieder, F. Identifying
resource-rational heuristics for risky choice. Psychological Review (2024).

[106] Kumar, S., Dasgupta, I., Daw, N. D., Cohen, J. D. & Griffiths, T. L. Disen-
tangling abstraction from statistical pattern matching in human and machine
learning. PLoS computational biology 19, e1011316 (2023).

[107] Lefebvre, G., Lebreton, M., Meyniel, F., Bourgeois-Gironde, S. & Palminteri,
S. Behavioural and neural characterization of optimistic reinforcement learning.
Nature Human Behaviour 1, 0067 (2017).

[108] Levering, K. R., Conaway, N. & Kurtz, K. J. Revisiting the linear separability
constraint: New implications for theories of human category learning. Memory
& cognition 48, 335–347 (2020).

[109] Ludwig, T., Siegel, M. & Schulz, E. Human multi-task learning: the why and
what (2023). URL http://dx.doi.org/10.32470/CCN.2023.1528-0.

[110] Plonsky, O., Apel, R., Erev, I., Ert, E. & Tennenholtz, M. When and how can
social scientists add value to data scientists? a choice prediction competition for
human decision making. Unpublished Manuscript (2018).

148
[111] Popov, V. & Dames, H. Intent matters: Resolving the intentional versus inci-
dental learning paradox in episodic long-term memory. Journal of Experimental
Psychology: General 152, 268 (2023).

[112] Sadeghiyeh, H. et al. Temporal discounting correlates with directed exploration

but not with random exploration. Scientific reports 10, 4020 (2020).

[113] Schulz, E., Franklin, N. T. & Gershman, S. J. Finding structure in multi-armed

bandits. Cognitive psychology 119, 101261 (2020).

[114] Somerville, L. H. et al. Charting the expansion of strategic exploratory behavior

during adolescence. Journal of experimental psychology: general 146, 155 (2017).

[115] Speekenbrink, M., Channon, S. & Shanks, D. R. Learning strategies in amnesia.

Neuroscience & Biobehavioral Reviews 32, 292–310 (2008).

[116] Steingroever, H. et al. Data from 617 healthy participants performing the iowa
gambling task: A “many labs” collaboration. Journal of Open Psychology Data
3, e5–e5 (2015).

[117] Tomov, M. S., Yagati, S., Kumar, A., Yang, W. & Gershman, S. J. Discovery of
hierarchical representations for efficient planning. PLoS computational biology
16, e1007594 (2020).

[118] Waltz, J. A., Wilson, R. C., Albrecht, M. A., Frank, M. J. & Gold, J. M.
Differential effects of psychotic illness on directed and random exploration.
Computational Psychiatry (Cambridge, Mass.) 4, 18 (2020).

[119] Wise, T., Michely, J., Dayan, P. & Dolan, R. J. A computational account of
threat-related attentional bias. PLoS computational biology 15, e1007341 (2019).

[120] Wu, S., Éltető, N., Dasgupta, I. & Schulz, E. Chunking as a rational solution
to the speed–accuracy trade-off in a serial reaction time task. Scientific reports
13, 7680 (2023).

[121] Xiong, H.-D., Ji-An, L., Mattar, M. & Wilson, R. Neural network modeling
reveals diverse human exploration behaviors via state space analysis (2023).

[122] Zhu, J.-Q., Sanborn, A. N. & Chater, N. The bayesian sampler: Generic bayesian
inference causes incoherence in human probability judgments. Psychological
review 127, 719 (2020).

[123] Zorowitz, S. & Niv, Y. Data from two-step task pilots (2023). URL osf.io/2tgjd.

149