Seminar Human Oversight of AI Redistributive Decisions

Abstract

Artificial intelligence is increasingly integrated into high-stakes decisions—public benefits allocation, hiring, and beyond—making human oversight both practically valuable and normatively required. This paper investigates whether individuals exhibit AI aversion when overseeing redistributive decisions made by an artificial agent, and disentangles two potential mechanisms: the black-box effect, arising from uncertainty about how the AI reaches its decisions, and intrinsic AI aversion, a reluctance to rely on algorithmic judgment. I develop a theoretical framework for the revision of redistributive choices under incomplete information and agent heterogeneity, and test its predictions in a two-session online experiment. The main finding is that subjects do not exhibit AI aversion: participants do not intervene more when they oversee an AI rather than a human, and their decisions are driven by the expected fairness cost of non-intervention rather than by who made the redistribution. The results suggest that human oversight of AI other-regarding decisions is unlikely to generate excessive scrutiny, though this may also limit its effectiveness when AI decisions are biased or mistaken.