Human-Centered AI

Why this matters

AI systems are increasingly deployed in settings where a human is nominally “in the loop” — a clinician reviewing a diagnosis, a loan officer reviewing a risk score — but the interface design largely determines whether that oversight is meaningful or just a rubber stamp.

What we work on

• Confidence-communication strategies that improve appropriate reliance on model output (not just more or less reliance).
• Interaction techniques that surface underlying evidence rather than opaque scores.
• Field methods for studying how time pressure and workload change reliance behavior outside the lab.

Representative outcomes

Our CHI paper on human-in-the-loop diagnostic interfaces (Honorable Mention, ACM CHI) grew directly out of this focus area, and its findings are now informing a longitudinal field deployment with the university health system.