Human-Centered Artificial Intelligence for Mental Health

Abstract: Artificial intelligence (AI) offers promising solutions for advancing prevention, diagnosis, and intervention in mental health. However, many approaches face challenges in terms of real-world usability and interpretability. In this talk, I will outline our human-centered approach to AI for mental health, with a focus on developing algorithms and systems that work in practical settings. Specifically, I will present our work on multimodal automated detection of psychosocial stress, integrating computer vision, remote photoplethysmography, and behavioral signal processing to capture this key risk factor for mental health. Furthermore, I will discuss two computational approaches to support psychiatric diagnosis: (1) a multimodal video-based analysis of social interactions, and (2) a network-based analysis of messaging patterns. Finally, I will address the crucial role of explainability in AI for mental health, detailing current challenges in interpreting affective models and proposing methods to improve model trustworthiness for clinical use.