When Can AI Reduce Individuals’ Anchoring Bias and Enhance Decision Accuracy? Evidence from Multiple Longitudinal Experiments

This study aimed to identify and explain the mechanism underlying decision-making behaviors adaptive to AI advice. We develop a new theoretical framework by drawing on the anchoring effect and the literature on experiential learning. We focus on two factors: (1) the difference between individuals’ initial estimates and AI advice and (2) the existence of a second anchor (i.e., previous-year credit scores). We conducted two longitudinal experiments in the corporate credit rating context, where correct answers exist stochastically. We found that individuals exhibit some paradoxical behaviors. With greater differences and no second anchor, individuals are more likely to make adjustment efforts, but their initial estimates remain strong anchors. Yet, in multiple-anchor contexts individuals tend to diminish dependence on their initial estimates. We also found that the accuracy of individuals was dependent on their debiasing efforts.