Conference on Cognitive Computational Neuroscience

Mentalizing - the ability to infer beliefs, desires and emotions of others - is indispensable for sustaining complex social interactions. Previous work has highlighted the role of recursive reasoning (i.e. “I think that you think that…”), but existing modeling approaches are either limited in scope or fairly complex and computationally expensive. To bridge this gap, we introduce and empirically characterize CHASE – a Cognitive Hierarchy ASsEssment model. To validate the model, we ran a series of behavioral experiments (N = 186) and an fMRI study (N = 50) in which participants played Rock-Paper-Scissors against artificial and human opponents. We designed the artificial opponents to mimic human gameplay at different levels of sophistication, forcing subjects to adapt their strategy. We find that the model captures how most subjects are capable of dynamically adjusting their strategy to the different opponents, and that neural activity in the temporoparietal junction (TPJ) is linked to these model-predicted adaptive shifts. Furthermore, multivariate decoding points to separable neural patterns for the different levels of mentalization as characterized by the model. Together, our combined approach provides a simple yet powerful new way of systematically assessing mentalizing capabilities.