The abstract is read at the publisher; the summary is JClub's.
Hierarchical digital surrogate models trained on sparse, non-Gaussian investment sequences in repeated trust games accurately predict future investments and outperform theory-driven reinforcement-learning and Bayesian cognitive models.
- Why it matters: Small laboratory datasets have forced social decision-making models to rely on hand-tailored mechanisms, limiting mechanistic inference; data-driven latent dynamical modeling can reduce these constraints.
- What they did: The authors build hierarchical generative surrogates in three stages: first predicting future investments, then analyzing latent state-space dynamics for how uncertainty and social versus non-social cues are represented, and finally running in-silico cue and strategy perturbations.
- The result: The inferred dynamics both mirror structure in empirical behavior and enable virtual experiments that probe cue-driven transitions into trusting or mistrusting states and test novel cue-outcome combinations and trustee strategies.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.