Screening social anxiety with the Social Artificial Intelligence Picture System.

Abstract

Social anxiety disorder (SAD) is a prevalent anxiety disorder marked by strong fear and avoidance of social scenarios. Early detection of SAD lays the foundation for the introduction of early interventions. However, due to the nature of social avoidance in social anxiety, the screening is challenging in the clinical setting. Classic questionnaires also bear the limitations of subjectivity, memory biases under repeated measures, and cultural influence. Thus, there exists an urgent need to develop a reliable and easily accessible tool to be widely used for social anxiety screening. Here, we developed the Social Artificial Intelligence Picture System (SAIPS) based on generative multi-modal foundation artificial intelligence (AI) models, containing a total of 279 social pictures and 118 control pictures. Social scenarios were constructed to represent core SAD triggers such as fear of negative evaluation, social interactions, and performance anxiety, mapping to specific dimensions of social anxiety to capture its multifaceted nature. Pictures devoid of social interactions were included as a control, aiming to reveal response patterns specific to social scenarios and to improve the system's precision in predicting social anxiety traits. Through laboratory and online experiments, we collected ratings on SAIPS from five dimensions. Machine learning results showed that ratings on SAIPS robustly reflected and predicted an individual's trait of social anxiety, especially social anxiety and arousal ratings. The prediction was reliable, even based on a short version with less than 30 pictures. Together, SAIPS may serve as a promising tool to support social anxiety screening and longitudinal predictions.