NANCY Seminar – Yves Oytana (Université de Franche-Comté)

Event details :

Title: Advisory algorithms, automation bias and liability sharing, co-written with Marie OBIDZINSKI

Abstract: We examine the optimal design of a liability-sharing rule when the use of an AI-generated prediction by a human user can generate an external benefit or harm. The AI provides a prediction about the state of the world, while the user exerts a judgment effort to assess the payoff from a risky decision. Our main assumption is that the user may exhibit automation bias, leading her to overestimate the algorithm’s accuracy. We show that this bias can make it optimal to increase the AI manufacturer’s liability share.