Researchers have developed data-driven hierarchical dynamical surrogate models capable of accurately predicting social decision-making from sparse behavioral data. Historically, laboratory investigations of social choices, such as trust building and adaptive strategies, have faced constraints from small datasets encoding rich cognitive processes. These data limitations traditionally confined modeling efforts to hand-tailored frameworks embedding strong assumptions about cognitive mechanisms. To address these constraints, the research team extracted generative models directly from behavioral investment sequences without imposing restrictive priors.
The modeling framework was derived from sparse, non-Gaussian investment sequences gathered during repeated trust games. In the first phase of their evaluation, the researchers established predictive validity by demonstrating that the inferred digital surrogates accurately forecast future investments despite the scarce training data. Furthermore, these data-driven models outperformed conventional theory-driven reinforcement-learning algorithms and Bayesian cognitive frameworks, establishing an empirical baseline for subsequent mechanistic investigations into social interaction dynamics.
In the second stage, the investigators examined the latent dynamics governing the surrogate models. This analysis revealed that the models successfully captured mechanistic structure in state space, specifically organizing how investments and choice uncertainty operate. Crucially, the internal representation of social versus non-social cues within the models paralleled patterns observed directly in empirical human behavior, confirming that the inferred dynamical architecture reflects genuine behavioral processes.
In the third stage, the team leveraged the generative capacity of the hierarchical digital surrogates to conduct in-silico experiments extending beyond the original laboratory observations. These virtual simulations enabled the researchers to probe how specific cues steer participants into trusting or mistrusting states. Additionally, the surrogates simulated human behavioral responses to previously unobserved cue-outcome combinations as well as novel trustee strategies.
Together, the findings demonstrate how generative digital surrogates provide a mechanistic, data-driven pathway for analyzing social interaction dynamics. By both accurately reproducing observed decision-making and facilitating hypothesis generation within virtual environments, these dynamical models overcome historical data scarcity barriers. The investigation was conducted by Manuel Brenner, Ismail Guennouni, Daniel Durstewitz, Stephanie N. L. Schmidt, Stefanie Lis, Peter Kirsch, and Georgia Koppe.
Subject of Research: Data-driven dynamical surrogate modeling of social decision-making in trust games
Article Title: Hierarchical data-driven dynamical surrogate models of social decision-making
Article References: Brenner, M., Guennouni, I., Durstewitz, D., Schmidt, S. N. L., Lis, S., Kirsch, P., & Koppe, G. (2026). Hierarchical data-driven dynamical surrogate models of social decision-making. PLOS Computational Biology, 22(11), e1014940. https://doi.org/10.1371/journal.pcbi.1014940
Image Credits: AI Generated
DOI: 10.1371/journal.pcbi.1014940
Keywords: social decision-making, trust games, dynamical surrogate models, latent dynamics, generative models, computational biology, reinforcement learning, Bayesian cognitive models, trust building, choice uncertainty, in-silico experiments, state space, behavioral data, social cues, non-Gaussian investment sequences, trustee strategies, adaptive strategies
Cite Scienmag News
Blake Davidson. (November 18, 2026). Data-Driven Dynamical Models Accurately Predict Social Decision-Making in Trust Games. Scienmag. https://scienmag.com/data-driven-dynamical-models-accurately-predict-social-decision-making-in-trust-games/
Blake Davidson. "Data-Driven Dynamical Models Accurately Predict Social Decision-Making in Trust Games." Scienmag, 18 November 2026, https://scienmag.com/data-driven-dynamical-models-accurately-predict-social-decision-making-in-trust-games/. Accessed 18 November 2026.
Blake Davidson. "Data-Driven Dynamical Models Accurately Predict Social Decision-Making in Trust Games." Scienmag. November 18, 2026. https://scienmag.com/data-driven-dynamical-models-accurately-predict-social-decision-making-in-trust-games/








