Prescient, Socially Intelligent Coach (PSI-Coach, ASIST)
Abstract
Improving team performance requires a deep understanding of what a team is doing--whether those actions are rational, irrational, or idiosyncratic-- and administering effective, non-annoying interventions so the team can reach high levels of achievement. Recognizing detailed user goals, mental states, and behaviors (in all of their human complexity) from low-level actions in dynamic open worlds is a challenging task requiring real time inference of complex, human cognitive processes. Predicting and planning well-timed, effective interventions takes a human coach years to learn and has repeatedly failed (in sometimes spectacular ways) in research and commercial AI systems. To meet these challenges, we created a Prescient, Socially Intelligent Coach (PSI-Coach). PSI-Coach is designed to unobtrusively monitor each team member to: (1) recognize their goals, mental states, and behaviors--without the assumption of rationality--by combining probabilistic programming inference with a cognitive architecture optimized to capture human variation (this architecture has modeled the minds of 4th-gen fighter pilots, missile defense operators, medical team members, and many others over the last 30 years); (2) infer detailed plans and goals (even when people do multiple things at the same time or change and adapt their behavior) using algorithms that reverse-engineer goals and mental states from dynamic streams of actions; (3) recognize shared mental models and whether they are in alignment or skewed using joint-behavior inference and analysis; (4) analyze these goals, mental states, behaviors, and shared mental models to compute practical, real-time team performance indicators based on empirical results of effective team performance collected over 10 years on more than 1,200 teams; and (5) use all of this information to predict team performance and plan effective, strategically timed interventions that maximize team performance.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 28, 2023
- Accession Number
- AD1227781
Entities
People
- Avi Pfeffer
- Bryan Loyall
- Curt Wu
- James Niehaus
- Kenny Lu
- Peter Weyhrauch
- Rob Hyland