<p>This study compares cooperative behavior between patients with schizophrenia and a non-clinical student group using an incentivized experimental paradigm and controlling for beliefs, emotions, and demographics. Participants in both groups played one-shot and finitely repeated Prisoner’s Dilemma (RPD) games against an artificial agent trained to mimic human decision-making. The experimental design encouraged continuous belief updating and behavioral adaptation. Our results show that patients with schizophrenia cooperate significantly more than controls in one-shot Prisoner’s Dilemma (PD) games and display lower adaptability to the environment in which they interact. Specifically, unlike controls, their likelihood of cooperation is unaffected by the type of PD game (one-shot or repeated) and does not significantly decline over rounds in the RPD games. Our regression analysis shows that beliefs about a partner’s cooperation, positive emotions, and the type of game are all significant predictors of cooperative behavior in the non-clinical sample. In contrast, among clinical participants, only beliefs significantly predict cooperation. Patients also show a reduced ability to predict others’ behavior, and their adopted strategies are largely unidentifiable. Together, our experimental results reveal systematic differences between the two samples in their interactions with artificial agents.</p>