Social Gaze Elicits Social Learning

Social gaze is a fundamental social cue that regulates approach and avoidance in social interactions. Everyday social interactions are characterized by repeated eye gaze exchanges. However, it remains unknown whether and how individuals learn from experiencing direct or averted eye gaze in others. Here, we used reinforcement learning modeling to investigate how micro-level gaze behaviors contribute to dynamic changes in social preferences and social closeness. Participants repeatedly chose between two faces that predominantly (70%) responded with either direct or averted gaze when selected, and rated their perceived closeness towards these individuals. Results showed a significant increase in choice frequency and closeness ratings for the direct gaze partner compared to the averted gaze partner. Uncovering the underlying mechanisms, computational modeling revealed that the direction of others’ eye gaze elicits a learning signal (prediction error) that in turn changes the latent value of the interaction partner. The learning signal was calibrated by individual differences in perceiving direct gaze as rewarding. Initial attractiveness and valence of the interaction partner predicted this reward calibration, suggesting that pre-existing impressions shape how social feedback is valued. In turn, calibration was reflected in participants’ social evaluations of the direct gaze partner, indicating how learning experiences contributed to impression formation. Together, these findings demonstrate that even minimal gaze interactions can shape social preference and closeness through reinforcement learning mechanisms. They further suggest that eye contact is a powerful, yet individually calibrated, social signal with implications for understanding and improving social engagement in both everyday and clinical contexts.