The DOI matches its Crossref record; 75 of the 75 references it cites were found in a registry or an index; 37 of the 59 sources its claims cite were checked for support.
Interpretable abstractions of artificial neural networks predict behavior and neural activity during human information gathering
Open accessPublished version of a preprint
Abstract
Humans and other animals are driven to acquire information about opportunities in their environments, yet how they evaluate what is worth learning remains unclear. Here we combine artificial neural networks with symbolic regression to extract an expressive yet interpretable model that specifies how human participants evaluate decision-relevant information during choice. The recovered function depends primarily on the relative evidence accumulated across options rather than absolute uncertainty about each, revealing that participants seek information symmetry across alternatives rather than minimizing uncertainty option by option. This account outperforms standard models of uncertainty-based exploration and generalizes to an independent dataset. Using ultrahigh-field (7T) functional magnetic resonance imaging optimized for midbrain and brainstem, we simultaneously measured activity across five neuromodulatory nuclei and two cortical regions. Ventral tegmental area activity showed opposed coding of information and selection values, a pattern suited to arbitrating between sampling and choosing, and anterior cingulate cortex and anterior insula tracked value-of-information computations.
Overview
CoChat Summary
TL;DR
The paper combines a hybrid artificial neural network with symbolic regression to model how humans value information during sampling, recovering a four-parameter function based on relative evidence across options rather than absolute uncertainty. The symbolic model predicts behavior and 7T fMRI activity, with VTA showing opposed coding of information and selection values and ACC/AI tracking value-of-information computations.
Key findings
- In 20 participants, sampling increased with higher initial uncertainty (β = −0.069, s.e. = 0.013, z = −5.18, P = 2.27 × 10−7) and with smaller final red-dot discrepancy (β = −0.131, s.e. = 0.019, z = −6.87, P = 6.48 × 10−12); uncertainty in the blocked option did not significantly influence sampling (β = −0.006, s.e. = 0.006, z = −1.01, P = 0.310).
- The hybrid-ANN model fit behavior better than linear and UCB models for all participants, and the symbolic model performed comparably to the ANN. The symbolic functions generalized to an independent two-armed bandit dataset, outperforming UCB (Wilcoxon W = 15.0, n = 89, median difference = −0.07, P = 4.31 × 10−16) and an offset UCB variant (W = 286.0, n = 89, median difference = −0.03, P = 2.21 × 10−12).
- Symbolic regression recovered a four-parameter function depending on the ratio Na/Nu, interpreted as attentional inertia (β1), information satiation (β2), undirected exploration (β3) and directed exploration (β4); separate first-visit and subsequent-visit parameters yielded eight total parameters. This supports an information-symmetry principle: participants balance knowledge across options rather than minimizing uncertainty option by option.
- ANN-derived VoI better fit neural data than linear (Cohen’s d = −1.025, Wilcoxon P = 2.12 × 10−12) and UCB (d = −1.105, P = 5.64 × 10−13); the symbolic function performed almost as well as the ANN (d = −0.129, P = 0.006). VTA showed positive coding of summed VoS (β = 11.3, P = 0.020) and negative coding of summed VoI (β = −46.1, P = 1.93 × 10−5); SN was negatively associated with VoI sum (β = −24.9, P = 6.38 × 10−5) and difference (β = −22.7, P = 0.008); AI and ACC tracked VoI-related signals; VSN, DRN and LC showed no significant associations after Bonferroni correction.
Methods
- Twenty participants (14 females), aged 19–32, completed four 25-min sessions of an information-sampling task inside a 7T MRI scanner. Trials presented three patches of 100 moving dots; green covers revealed initial information simultaneously, gray covers revealed dots sequentially every 150 ms, and on 80% of trials one patch was blocked, leaving two options for subsequent analyses. Sampling was capped at 24 s for two-option trials and 33 s for three-option trials.
- The hybrid model computed values for four actions (stay, switch, select attended, select unattended). It compared linear, UCB and ANN-based VoI functions; the ANN was a Lipschitz-bounded deep network with four hidden layers of 32 neurons each and tanh activations, trained with Adam and L-BFGS. Symbolic regression (SymbolicRegression.jl) was applied to the trained ANN’s predictions to recover closed-form equations.
- fMRI data were acquired at 7T with 1-mm isotropic voxels and TR = 1.378 s. A GLM used VoI as a parametric modulator while controlling for outcome and VoS; model comparisons used per-voxel MSE, Cohen’s d with bootstrap confidence intervals and Wilcoxon signed-rank tests. ROI analyses used mixed-effects models weighted by inverse variance, with Bonferroni correction across five subcortical ROIs.
Limitations
- The paper states that the specific symbolic equations are tailored to the two-option structure and encode relative comparisons between attended and unattended options; unlike UCB, they do not naturally extend to n-armed bandits. Possible extensions include replacing Nunattended with the mean across unattended options or weighting unattended options by salience or recency.
- The paper states that the task does not permit dissociation of decision uncertainty, evidence uncertainty and reward uncertainty, which are highly correlated in the design; orthogonal manipulation of these uncertainty types is an important direction for future research.
- The paper reports that VSN, DRN and LC showed no significant association with the sum or difference of VoI or VoS after Bonferroni correction, and notes that within the sensitivity limits of the measurement and analysis, these nuclei do not strongly encode these decision variables in this task. Evident from the text: the neural analyses are restricted to two-option trials, and the sample size is 20 participants.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures












![Extended Data Fig. 7 Neural model comparison across whole brain. Ridge plots showing the bootstrap distributions of Cohen’s d effect sizes for pairwise model comparisons in predicting neural activity (n = 80 sessions, two-sided Wilcoxon signed-rank tests). The ANN vs. Symbolic comparison (bottom, black) shows a small effect size (d = − 0.129, 95% CI [ − 0.494, 0.078], Wilcoxon P = 0.006), with a confidence interval overlapping zero, indicating equivalent performance. In contrast, ANN vs. Linear (dark green, d = − 1.025, 95% CI [ − 1.264, − 0.849], Wilcoxon P = 2.12 × 10−12) and ANN vs. UCB (dark green, d = − 1.105, 95% CI [ − 1.366, − 0.918], Wilcoxon P = 5.64 × 10−13) show large negative effects, indicating substantially better ANN performance. Symbolic vs. Linear (dark red, d = − 0.882) and Symbolic vs. UCB (dark red, d = − 0.953) also show large effects, demonstrating that the symbolic function outperforms standard heuristics while matching ANN performance. Darker shaded regions within each distribution represent 95% confidence intervals. Vertical reference lines indicate conventional effect-size thresholds: no effect (0), small (0.2), medium (0.5), and large (0.8). Negative Cohen’s d values indicate the first model has lower MSE (better fit). Source data](https://figures.cochat.ai/paper-figures/9a9bffbd821b413597d71f9262257492/interpretable-abstractions-of-artificial-neural-networks-predict-behavior-and-figure-13.jpg)

Verification
Claims the paper makes
10 Valid · 2 Not valid · 5 Not checked · 134 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Here we exploit high-resolution (1-mm isotropic), rapid-repetition-time (1.378 s), accelerated ultrahigh-field (7T) functional magnetic resonance imaging (fMRI)27,28 to measure activity simultaneously across all five neuromodulatory nuclei and two interconnected cortical regions that project to them: the anterior insula (AI) and anterior cingulate cortex (ACC)3,22,29–33.”
§ Interpretable abstractions of artificial neural networks predict behavior and neural activity during human information gathering > Main
What the sources say
- Could not verifycites [27] Bromberg-Martin, E. S. & Hikosaka, O. Midbrain dopamine neurons signal preference for advance information about upcoming rewards. Neuron 63, 119–126 (2009). sourceExperimental — accuracy not yet measured
The citation appears attached to the description of the anterior insula and anterior cingulate cortex as regions that project to the neuromodulatory nuclei, not to the citing authors’ fMRI specifications. Our copy’s Experimental Procedures (Neural Recording; page unavailable) describes recordings around the substantia nigra and ventral tegmental area, but the passages found for anterior insula, anterior cingulate cortex, and their projections do not address that connection. (LLM, conf=0.74)
- Partly supportedcites [28] Khalighinejad, N., Garrett, N., Priestley, L., Lockwood, P. & Rushworth, M. F. S. A habenula-insular circuit encodes the willingness to act. Nat. Commun. 12, 6329 (2021). sourceExperimental — accuracy not yet measured
The Discussion (page unavailable) says the anterior insula sends direct projections to the habenula, and the Results identify activity in the anterior insula and a cluster extending into the anterior cingulate. The paper does not establish that both cortical regions project to all five neuromodulatory nuclei. The imaging specifications are attached to different citations in the citing sentence; this paper’s Methods report 1.5-mm functional resolution and a 1.962-s repetition time, not the quoted values. (LLM, conf=0.86)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
Every source the paper cites for this claim was checked and supports it.
In the paper
“The model has some generalizability: it predicted behavior in a different explore–exploit task20 (Fig. 5b).”
§ Interpretable abstractions of artificial neural networks predict behavior and neural activity during human information gathering > Discussion
What the sources say
- Supportedcites [20] Gershman, S. J. Deconstructing the human algorithms for exploration. Cognition 173, 34–42 (2018). source
Section 4.4 'Re-analysis of Wilson et al. (2014)' states that, to obtain further validation of the hybrid model, Gershman fit it to data collected in a different explore-exploit task (Gaussian rewards truncated 1-100, fixed block parameters, 10 forced-choice trials, 320 blocks), and that these data 'were decisively better explained by the hybrid model, with a protected exceedance probability of 0.999'. The Discussion (5) repeats that quantitative model comparison including a re-analysis of Wilson et al. (2014) supported the hybrid model. This matches the citing sentence's claim of generalizability to a different explore-exploit task; the only difference of specificity is that the model's reward variance was re-fit (grid search) on the new dataset rather than applied with frozen parameters. (LLM, conf=0.85)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Remarkably, our symbolic functions also outperformed the UCB model in predicting participants’ choices in this different context (Wilcoxon signed-rank test: W = 15.0, n = 89, median difference = −0.07, P = 4.31 × 10−16).”
§ Interpretable abstractions of artificial neural networks predict behavior and neural activity during human information gathering > Results > The ANN can be transformed into an interpretable symbolic function
What the sources say
- Partly supportedcites [20] Gershman, S. J. Deconstructing the human algorithms for exploration. Cognition 173, 34–42 (2018). sourceExperimental — accuracy not yet measured
Our copy of Gershman (2018) does supply the two-armed bandit dataset the citing sentence points at: Methods 3.1 reports 44 participants in Experiment 1 and 45 in Experiment 2 (89 in total), both running two-arm Gaussian-reward bandit tasks, and Results 4.1/4.4 report that a model with an extra uncertainty term beyond UCB (the 'hybrid') better explained choices, decisively so in the re-analysis of Wilson et al. (2014) (protected exceedance probability 0.999). However, the paper nowhere uses 'symbolic functions' and never reports a Wilcoxon signed-rank comparison of predictive accuracy: model comparison is Bayesian random-effects on BIC-approximated model evidence (Methods 3.3), so the specific statistics attributed here (W = 15.0, median difference = −0.07, P = 4.31 × 10⁻¹⁶) are not in this source and appear to be the citing authors' own analysis. The part of the claim our copy does not support is the symbolic-function-versus-UCB Wilcoxon test and its numbers. (LLM, conf=0.68)
boostedstates an associationn = 89- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Crucially, however, the number of green dots in the blocked option did not significantly influence sampling behavior (β = −0.006, s.e. = 0.006, z = −1.01, P = 0.310)”
§ Interpretable abstractions of artificial neural networks predict behavior and neural activity during human information gathering > Results > Sampling behavior adaptively scales with task difficulty and uncertainty
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“and when the final discrepancy in red dots was smaller (β = −0.131, s.e. = 0.019, z = −6.87, P = 6.48 × 10−12; Fig. 2b).”
§ Interpretable abstractions of artificial neural networks predict behavior and neural activity during human information gathering > Results > Sampling behavior adaptively scales with task difficulty and uncertainty
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
73 more claims
Create a free account to see all 151 claims and to verify the 22 references still open.
- Probably not a substantial claim. 37
- It may state two separate claims as one. 46
- The quote only partly states this claim. 1
Statistics recomputed
Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.
- Consistentz = -1.01reported p = .31 · recomputed p = .312
“z = −1.01, p = .310”
- Consistentz = 1.41reported p = .158 · recomputed p = .159
“z = 1.41, p = .158”
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ Interpretable abstractions of artificial neural networks predict behavior and neural activity during human information gathering > Results > The ANN integrates evidence from both patches to compute VoI
“the value of staying increases with attended evidence while simultaneously decreasing with unattended evidence, producing a diagonal gradient (Fig. 4b, right).”
§ Interpretable abstractions of artificial neural networks predict behavior and neural activity during human information gathering > Results > The ANN integrates evidence from both patches to compute VoI
“The ANN produces diagonal arrows instead, indicating that additional attended evidence simultaneously decreases the value of staying and increases the value of switching.”
Citations
Cites
Advances in functional and structural MR image analysis and implementation as FSL
Smith, S. M., Jenkinson, M., Woolrich, M. W., Beckmann, C. F., Behrens, T. E. J., Johansen-Berg, H., Bannister, P. R., De Luca, M., Drobnjak, I., Flitney, D. E., et al. (2004) · NeuroImage · doi:10.1016/j.neuroimage.2004.07.051 · Cited by 13,630 · cited 2×
“Preprocessing of fMRI data was performed with the FMRIB Software Library 65,66 .”
Jenkinson, M., Bannister, P., Brady, M., & Smith, S. (2002) · NeuroImage · doi:10.1006/nimg.2002.1132 · Cited by 11,888 · cited 3×
“Further non-task confound regressors were added to reduce noise in BOLD signal, including: (1) head motion parameters estimated using MCFLIRT during pre-processing 68 ; (2) regressors for voxel-wise estimates of physiological noise arising from cardiac and respiratory activity, estimated using FSL’s Physiological Nois…”
InfluentialJenkinson, M., Beckmann, C. F., Behrens, T. E. J., Woolrich, M. W., & Smith, S. M. (2012) · NeuroImage · doi:10.1016/j.neuroimage.2011.09.015 · Cited by 11,314
Fast robust automated brain extraction
Smith, S. M. (2002) · Human Brain Mapping · doi:10.1002/hbm.10062 · Cited by 10,496 · cited 1×
“The Brain Extraction Tool 67 was used to separate brain from non-brain matter in structural and functional images.”
Accurate and robust brain image alignment using boundary-based registration
Greve, D. N., & Fischl, B. (2009) · NeuroImage · doi:10.1016/j.neuroimage.2009.06.060 · Cited by 3,596 · cited 1×
“The whole-brain EPI was registered to the subject-specific structural images using Boundary-Based Registration (BBR) incorporating fieldmap correction 69 .”
Temporal Autocorrelation in Univariate Linear Modeling of FMRI Data
Woolrich, M. W., Ripley, B. D., Brady, M., & Smith, S. M. (2001) · NeuroImage · doi:10.1006/nimg.2001.0931 · Cited by 2,852 · cited 1×
“In the first level, a univariate general linear model was used to compute parameter estimates for each regressor in each session 70 .”
Discrete coding of reward probability and uncertainty by dopamine neurons
Fiorillo, C. D., Tobler, P. N., & Schultz, W. (2003) · Science · doi:10.1126/science.1077349 · Cited by 2,085
Uncertainty, Neuromodulation, and Attention
Yu, A. J., & Dayan, P. (2005) · Neuron · doi:10.1016/j.neuron.2005.04.026 · Cited by 1,774
Multilevel linear modelling for FMRI group analysis using Bayesian inference
Woolrich, M. W., Behrens, T. E. J., Beckmann, C. F., Jenkinson, M., & Smith, S. M. (2004) · NeuroImage · doi:10.1016/j.neuroimage.2003.12.023 · Cited by 1,728 · cited 1×
“Finally, a random-effects analysis was conducted at the third level, where subject-identity was a random effect 71 .”
Behavioral Theories and the Neurophysiology of Reward
Schultz, W. (2006) · Annual Review of Psychology · doi:10.1146/annurev.psych.56.091103.070229 · Cited by 1,637 · cited 1×
“Activity in individual neurons in both habenula and VTA reflects both the value of a stimulus in terms of the reward that it predicts 56–58 but also value of information 20,21,26,59 in line with the suggestion that VTA may compare the relative advantage to be gained from making a selection and…”
Research works printed but not linked: 5
- Green, D. M. & Swets, J. A. Signal Detection Theory and Psychophysics (Wiley, 1966).
- Zupko, J. John Buridan: Portrait of a Fourteenth-Century Arts Master (Univ. Notre Dame Press, 2003).
- Sutton, R. S. & Barto, A. G. Reinforcement Learning: An Introduction, Vol. 1 (MIT Press, 1998).
- Weinhardt, D., Eckstein, M. K. & Musslick, S. Computational discovery of human reinforcement learning dynamics from choice behavior. OpenReview.net https://openreview.net/forum?id=x2WDZrpgmB (2024).
- Wang, R. & Manchester, I. Direct parameterization of Lipschitz-bounded deep networks. In Proc. 40th International Conference on Machine Learning (eds Krause, A. et al.) 36093–36110 (ACM, 2023).
Cited by
AI-Discovered Cognitive Models Reveal Novel Insights into Human and Animal LearningPreprint
Kasenberg, D., Castro, P. S., Eckstein, M. K., Éltető, N., Dabney, W., Wang, C., Engelcke, M., Mohanta, R., Dev, A., Botvinick, M., et al. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.05.18.725921 · Cited by 4
Metadata
- Authors
- D’Ambrogio, S., Grohn, J., Khalighinejad, N., Mattar, M. G., Hunt, L., & Rushworth, M. F. S.
- Affiliations
- University of Oxford (GB)
- New York University (US)
- Corresponding author
- D’Ambrogio, Simone
- ORCID
- Venue
- Nature Neuroscience
- Published
- 2026-06-26
- Volume
- 29
- Issue
- 8
- Pages
- 2036-2047
- ISSN
- 1097-6256 (ISSN-L)
- 1546-1726
- PMC
- PMC13433322
- OpenAlex
- W4411641310
- Cited by
- 1
- References
- 69
- Citations by year
- 2026: 1
- FWCI
- 4.12
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- Nature Neuroscience (publishedVersion, cc-by)
- bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by)
- Research Square (acceptedVersion, cc-by)
- PubMed (publishedVersion)
- PubMed Central (publishedVersion, cc-by)
- Funders
- RCUK | Biotechnology and Biological Sciences Research Council (BB/W003392/1)
- Wellcome Trust (221794/Z/20/Z)
- Topics
- Life Sciences › Neuroscience › Cognitive Neuroscience › Neural and Behavioral Psychology Studies
- Social Sciences › Decision Sciences › General Decision Sciences › Decision-Making and Behavioral Economics
- Life Sciences › Neuroscience › Cellular and Molecular Neuroscience › Neurotransmitter Receptor Influence on Behavior
- Keywords
- value of information, information seeking, decision-making, symbolic regression, ventral tegmental area, anterior cingulate cortex, anterior insula, evidence accumulation
- MeSH terms
- Soft Computing, Adult, Brain, Brain Mapping, Choice Behavior, Decision Making, Female, Humans, Magnetic Resonance Imaging, Male, Neural Networks, Computer, Ventral Tegmental Area, Young Adult
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-09-28, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-09-28 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Interpretable abstractions of artificial neural networks predict behavior and neural activity during human information gathering. https://app.cochat.ai/papers/10.1038/s41593-026-02342-9
- The paper
- D’Ambrogio, S., Grohn, J., Khalighinejad, N., Mattar, M. G., Hunt, L., & Rushworth, M. F. S. (2026). Interpretable abstractions of artificial neural networks predict behavior and neural activity during human information gathering. Nature Neuroscience. https://doi.org/10.1038/s41593-026-02342-9
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.