Pareek S, Velloso E, Goncalves J. Trust Development and Repair in AI-Assisted Decision-Making during Complementary Expertise. FAccT '24: Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency. 2024; Pages 546-561. https://doi.org/10.1145/3630106.3658924
Summary: This study examines how trust is developed, eroded, and restored in AI-assisted decision-making, especially in situations where human and AI expertise complement each other. Through two experimental tasks, participants classified familiar and unfamiliar stimuli with AI assistance, with trust being influenced by the AI's accuracy. When trust was damaged, various Trust Repair Strategies (TRSs) were tested. The most effective strategy for restoring trust was a model update, which surpassed initial trust levels, followed by apology and promise. Denial was the least effective. The study offers insights into how human decision-makers calibrate trust and the factors affecting the success of TRSs in Human-AI collaboration.
Comments