Definition
Disengaging because the AI looks reliable, and making worse decisions than a weaker AI would have led to. In Dell'Acqua's recruiter experiment, the group given the higher-quality AI spent less time on each application, clicked less and followed the recommendation more. It ended up less accurate than the group given the mediocre one.
What is at stake
The better the tool, the less the person checks. Most AI rollouts buy quality on the assumption that quality is the whole gain. Past a certain point, quality is also what removes the checking, and the experienced people in the room are the ones who defer most.
Where the word came from
- The ironies of automationLisanne Bainbridge, 1983, the operator kept for the emergency and deprived of the practice
- Automation biasMosier and Skitka, 1990s
- Falling asleep at the wheelDell'Acqua, 2022, the same effect measured on a hiring task with a better and a worse AI
How episcope came to it
Fabrizio Dell'Acqua, 2022, a Harvard Business School working paper, circulated without peer review. A pre-registered field experiment with 181 professional recruiters reviewing 44 applications each, randomly given a perfect AI, a good one, a weak one or none, and told which they had.
What to watch for
A working paper rather than a journal article, and its finding is easy to overstate. It does not recommend a worse AI. It shows that the checking a person does shrinks as the tool improves, so the checking has to be designed for rather than assumed.
Papers and articles behind it
Dell'Acqua, F. (2022). Falling Asleep at the Wheel: Human/AI Collaboration in a Field Experiment on HR Recruiters. Harvard Business School working paper.