Research
Tuesday Nov. 15, 2022
Colloquium: Krzysztof Gajos (Harvard) - Human Cognitive (Dis)Engagement during AI-Assisted Decision-Making
12:30 to 2 p.m.
Hybrid: Zoom & UTA 5.522 (Large Conference Room)

Abstract: People supported by AI-powered decision support tools were expected to make better decisions than either people or AI systems on their own. In practice, this is almost never the case. Based on the evidence that people over-rely on the AI recommendations, we posit that in most settings people do not cognitively engage with the AI-provided information (recommendations or explanations) leading to the poorer-than-expected outcomes. This problem has gone largely undiagnosed because the prevalent methods for evaluating human-AI interaction innovations rely on proxy tasks which, we demonstrate, produce misleading results. We then demonstrate that cognitive forcing (an intervention intended to push people toward more analytical processing of information) can reduce human over-reliance on the AI but possibly without addressing the root cause. Finally, we introduce incidental learning as an indirect but objective indicator of cognitive engagement. Using this as our dependent measure, we provide some of the most direct evidence to date that people really do not engage cognitively with the AI-provided information in the standard explainable AI settings, even when cognitive forcing is applied. Instead, we observed both high quality decisions and strong evidence of cognitive engagement when people were provided by an AI with informative explanations but no decision recommendations. These results provide early evidence that in some settings a better human-AI interaction paradigm may be to use the power of AI to construct actionable syntheses of the information to aid people in making their own well-informed decisions rather than recommending the ready made decisions to them.

Bio: Krzysztof Gajos is a Gordon McKay professor of Computer Science at the Harvard Paulson School of Engineering and Applied Sciences. Krzysztof’s current interests include 1. Principles and applications of intelligent interactive systems; 2. Tools and methods for behavioral research at scale (e.g., LabintheWild.org); and 3. Design for equity and social justice. He has also made contributions in the areas of accessible computing, creativity support tools, social computing, and health informatics. http://www.eecs.harvard.edu/~kgajos/

Event Category