Mentalization, inferring others' emotions and intentions, is crucial for human social interactions and is impaired in various brain disorders. While previous neuroscience research has focused on static mentalization strategies, we know little about how the brain adaptively selects which strategies to use at any given moment. Here we investigate this core aspect of mentalization with computational modeling and functional magnetic resonance imaging (fMRI) during interactive strategic games. We find that most participants can adapt their strategies to the changing sophistication of their opponents, though there are considerable individual differences. Model-based fMRI analyses identify a distributed brain network in which activity and connectivity track this mentalization-belief adaptation. The extent to which people update their beliefs about others' sophistication can be predicted out of sample from neural activity, providing a neural signature of adaptive mentalization. Our model elucidates the neural basis of mentalization ability and provides a method for assessing these capabilities in healthy and clinical populations.
Buergi et al. (2026) studied this question.