The Principles and Limits of Algorithm-in-the-Loop Decision Making

Abstract

Investigates what happens when humans use AI as a tool rather than a replacement. The study finds that algorithm-in-the-loop decision making does not automatically improve outcomes — humans often defer to the algorithm even when they should override it.

Key Findings

• Human oversight of AI decisions does not reliably improve accuracy
• People anchor heavily on algorithmic recommendations even when incorrect
• The "human in the loop" paradigm may provide false assurance of human control

Source