Bridging reinforcement-learning and drift-diffusion modeling to uncover the cognitive processes underlying collective foraging

Inferences drawn from cognitive-computational models necessarily depend on their assumptions, yet models from distinct frameworks are rarely tested against each other. Here, we directly compare different variants of social reinforcement-learning and drift-diffusion models within the same collective foraging paradigm, investigating the cognitive processesgoverning the integration of social information. We designed an interactive immersive-reality experiment, where participants—either alone or in groups of five—foraged for rewards in environments with different resource distributions. Our behavioral results revealed that groups outperformed solitary foragers and even optimal Bayesian agents in environments with sparse rewards but large differences in patch quality—precisely as predicted by our numerical simulations from both frameworks. Computational analyses showed that reinforcement-learning and drift-diffusion models generated similar behav-ioral predictions, with both frameworks accurately reproducing participants’ behavior across conditions. However, at the cognitive level, both frameworks partly diverged in explaining the process of social information integration, overall suggesting that participants integrated the choices and payoffs of others into their personal evaluation of the environment. This study deepens our understanding of social decision-making processes in naturalistic collectives and emphasizes the importance of cross-framework comparisons.