Mirko Kremer, da Frankfurt School of Finance & Management, vai apresentar "When Delegating AI-Assisted Decisions Drives AI Over-Reliance"

Abstract:We empirically test a theory of AI over-reliance in a delegated and AI-assisted decision-making environment that endogenizes institutional incentives for decision makers (DMs) to rely on algorithmic recommendations. Based on a simple behavioral model, we predict that managers blame DMs for overriding algorithmic recommendations, despite knowing that such overrides most likely reflect valuable human judgment. This prediction highlights the ambivalent role of private information in human-machine interactions. While tacit domain knowledge justifies keeping humans in the decision-making loop, the non-codifiable and non-communicable nature of such knowledge poses a challenge in delegated settings where managers try to infer decision quality from what they can observe: algorithmic recommendations and decision outcomes. As a result of managers’ blame-like behavior, DMs over-rely on algorithmic recommendations and thus make worse decisions. We design controlled laboratory experiments that support these predictions, highlighting challenges that hinder the integration of