Human–AI Collaboration: Performance, Uncertainty, and Human Preferences
Theme: Joint Human–AI Systems (TDAI Speaker Series)
Speaker: Mark Steyvers
Professor, Cognitive Sciences, University of California, Irvine
Date & Time: Tuesday, April 14, 2026 · 4:00–5:00 PM (ET)
Location: Pomerene Hall Room 350 (Project Zone)
Audience/Format: Research Seminar · Open to all
Host: Peter Kvam
This seminar explores how humans and artificial intelligence systems collaborate on decision-making tasks, and why strong AI performance alone does not guarantee effective teamwork. Drawing on controlled experiments with known ground truth, Dr. Steyvers examines when human–AI collaboration improves outcomes and when it fails, highlighting the role of complementary strengths, uncertainty, and human judgment.
Abstract
Recent advances in AI systems have produced models that match or exceed human performance on several benchmarks, yet strong AI capability does not automatically translate into effective human–AI teamwork. This talk examines when collaboration between humans and AI improves performance and when it falls short, focusing on controlled tasks with known ground truth and forms of collaboration such as prediction aggregation, AI-assisted decision making, and joint action.
Human–AI complementarity can emerge when people and models make different types of errors, allowing combined predictions to outperform either alone. However, these gains are often limited when humans rely directly on AI advice, as people tend to overestimate AI accuracy and struggle to assess when outputs are correct. The talk explores how AI systems can better communicate uncertainty in ways that align with human interpretation, and how individuals can learn to calibrate their trust in AI through experience and feedback.
It also highlights findings from real-time human–AI collaboration, showing that effective teamwork depends not only on performance but also on factors such as cooperativeness, fairness, and meaningful participation. Together, these insights illustrate how uncertainty, trust, and complementary strengths shape the success of human–AI collaboration.
Speaker
Mark Steyvers is a Professor of Cognitive Sciences at the University of California, Irvine. His research focuses on cognitive modeling, human–AI interaction, and decision-making, with an emphasis on understanding how humans and intelligent systems can effectively collaborate.