Description: Department Seminar

Location: 368A Heady Hall

Contact: Bertan Turhan

Title: Learning from an Unknown DGP: Experimental Evidence on Belief Updating with AI Recommendations

Abstract: Understanding belief updating when the data-generating process (DGP) is unknown is a central but understudied problem in economics. We design an online experiment where subjects report beliefs before and after receiving qualitative recommendations from an AI assistant. This setting captures “black box” information sources, where the DGP is opaque but stable and informative. We compare three models of updating: quasi-Bayesian learning, which generalizes Bayesian learning, weighted inertial updating, and the contraction rule. Our findings provide the first empirical evidence on how individuals revise beliefs in response to AI recommendations, offering theoretical and practical insights into decision-making with opaque information structures.