How to design adaptive systems to improve stress management using artificial intelligence

Biofeedback is a technique that relies on measuring bodily functions and providing feedback to the individual so that they can train and control those functions. Artificial intelligence has empowered these systems by making them context-aware, adapting models to users' physiological variations, and providing personalized feedback. However, incorporating AI techniques has opened up new challenges in designing, developing, and evaluating biofeedback systems. In this work, we conducted 25 semi-structured interviews with various specialists in medicine, psychology, human-computer interaction, and computer science to investigate what an AI-based system that considers differences in personal health data should look like. The results helped us answer ‘How can we design AI-assisted systems that take into account differences in personal physiological and AI knowledge between individuals to avoid misinterpretations?’ by defining six design considerations for biofeedback systems that use AI techniques trained with users' physiological signals. Finally, we discuss how these considerations could help researchers design systems for well-being.