When Should Your Team Override AI? A Trust Calibration Routine
A five-stage routine (co-sensing, co-framing, co-deciding, action and feedback, reconfiguration) helps teams calibrate reliance on AI, using override rate and accuracy.
Summary
The paper addresses two opposite failures in working with AI: automation bias, where people accept machine output without question, and algorithm aversion, where they reject it even when it is right. Drawing on evidence from human-AI teams, the authors propose a trust calibration routine in five stages: co-sensing, co-framing, co-deciding, action and feedback, and reconfiguration. Teams track two indicators, how often people override the AI and how accurate the AI is, so that reliance on it rises or falls with its demonstrated performance.
Editor's opinion
The question is not whether to trust AI but when, and the answer should come from data rather than instinct. Teams should track how often they override the AI and how accurate it is, and adjust their reliance on that evidence, so they neither follow it blindly nor dismiss it out of habit.
Key concepts
human-AI teams · trust calibration · override rate · decision quality