Creating AI-generated role-playing videos from causal network model simulations of social anxiety disorder for virtual therapeutic contexts

Social Anxiety Disorder (SAD) is characterized by an excessive fear of negative evaluation that influences avoidance behaviors and a constant negative view of self. In order to assist in remote exposure therapy through creation of personalized content, this paper develops a second-order adaptive network model of SAD. We built a second-order adaptive network with nineteen literature-related states cover not only possible causes, threat appraisal, but also physiological arousal, fear/action regulation, safety and avoidance behaviors, and post-event processing, all connected by weighted links. These weights can be made adaptive by the self-modeling principle for networks and reflect the neural influences on such behaviors (e.g. amygdala spike, vmPFC brake, insula activation). Besides weights, the learning speeds are also adaptive and regulated by certain factors (e.g. BNST sustaining anxiety, dopamine relief leading to habituation, dACC conflict monitoring). Through simulations, two SAD cases were observed: the brief success and failure of ventromedial prefrontal cortex (vmPFC) regulation in ameliorating fear and the result of conducting safety behaviors leading to anxiety reduction. A scenario was then translated from these simulations into scripts that provided the foundation for an AI-generated role-play video. The result illustrates the modeled emotions, behaviors and coping strategies. This work demonstrates an adaptable, research-driven framework for generating susceptible remote exposures.