Human coaches leverage their deep understanding of a team’s goals and the coach’s own rich past experience to figure out when and how to intervene to improve the team’s performance. Today’s AI cannot compete with human coaches. Because AI cannot recognize detailed goals, mental states, and behaviors, it often fails to accurately predict human actions, and as a result, cannot effectively intervene—besides being ineffective, AI suggestions based on surface analyses can disrupt and annoy the team.

PSI-Coach allows AI systems to unobtrusively monitor each team member, then, like an experienced coach, help the team achieve high-level performance. PSI-Coach blends automated recognition of human mental states, deep understanding of individual and group goals and behaviors, with a capacity to accurately predict the probability of a positive outcome for multiple interventions. It also includes human-factors engineering and interactive narrative technologies so that its own reasoning and resulting recommendations are clear and easy to understand. These features let an AI recommend actions at the perfect time and in the perfect way to influence a team member or the team as a whole to make the changes necessary to meet their team goals most effectively.

PSI-Coach also uses Sherlock™, our prototyping platform for building applications that collect, analyze, and visualize physiological data, then reason about physiological, neurological, and behavioral states. The platform includes functional near-infrared spectroscopy (fNIRS) sensors, eye trackers, and standoff cardiac monitors embedded in webcams, as well as mouse/keyboard analytics. Sherlock lets PSI-Coach inform and validate research by linking physiological measures of human cognitive states to AI models.