Group-dependent learning and decision neural signals underlie the persistence of intergroup bias

Intergroup bias persists despite repeated interactions that should enable belief updating. In this study, we combined a behavioral reinforcement-learning framework with fMRI to examine how group membership shapes trial-by-trial learning and value-based decisions about others. Using a multi-player game with ingroup and outgroup actors, we modeled participants’ evolving beliefs about others’ likelihoods of inflicting or avoiding harm. We assessed how these belief-updating processes are implemented at the neural level. Behaviorally, participants were more likely to inflict harm on out-group players, regardless of the out-group players' actions. The imaging data focused on regions within the social-learning and decision-making networks. Learning-related activity in the right supramarginal gyrus (rSMG) and the precuneus tracked participants’ trial-by-trial beliefs about others’ intentions. The rSMG encoded group-dependent prediction errors, showing heightened responses to unexpected prosocial acts by outgroup members and unexpected antisocial acts by ingroup members. The precuneus tracked the predictability of harmful outcomes independent of group membership. Ventral-medial prefrontal cortex (vmPFC) activity was modulated by decision factors (distance from the star and players, perceived aggressiveness), and this modulation depended on group association. Together, these findings show group identity biases neural learning signals and value computations, sustaining intergroup bias despite opportunities to update beliefs.