Pupil dilation reflects prediction errors in learning about others' personality traits

Learning about others' personality traits is crucial for building personal relationships. This learning process can be conceptualized as reductions in prediction errors (PEs), defined as discrepancies between expectations about others and their subsequent feedback. While pupil dilation is a well-established marker of various learning processes, it remains less clear how it tracks the updating of estimations about others' personalities and reflects the dynamics of social learning. To test this, we conducted a pilot experiment and a preregistered main experiment where participants learned about either other people's trait profiles or a self-matched profile based on participant's own trait ratings. We hypothesized that pupil size would scale with PE magnitude during personality trait learning. Using linear mixed-effects models (LMMs), we confirmed that absolute PEs decreased over time in the main experiment, indicating successful learning. We then modeled pupillometry data to examine trial-by-trial relationships between absolute PEs and pupil dilation. In both experiments, larger absolute PEs were associated with larger pupil dilation, even after controlling for signed PEs and feedback. Furthermore, learning about self-matched profile yielded smaller PEs and correspondingly reduced pupil dilation. These findings suggest that pupil dilation tracks error magnitude and represents a critical physiological index of personality-trait learning about others.