Definition
The level of automation accuracy above which relying on the system helps more than it hurts, and below which reliance is a net cost. Put at roughly 70 percent accuracy under high workload.
What is at stake
Most of this glossary describes what reliance costs. This is the older literature saying reliance is not always a cost, that the question is empirical and that it depends on the system. It is the honest counterweight to the rest of this glossary, and worth carrying for that reason.
Where the word came from
- The ironies of automationLisanne Bainbridge, 1983
- Automation biasMosier and Skitka, 1990s
- The crossover pointParasuraman and Manzey, 2010, which puts a number on where reliance turns
How episcope came to it
Raja Parasuraman and Dietrich Manzey, 2010, in a peer-reviewed review article in Human Factors that brought complacency and bias together under one account of attention. Parasuraman was an Indian-American engineering psychologist at George Mason University, who died in 2015 and largely founded the study of automation complacency. The paper reached episcope through a 2024 review of AI and skill decay led by Brooke Macnamara, which cites it.
What to watch for
Two limits travel with the figure. It comes from process control and flight-deck automation rather than from generative AI. And it measures the task outcome, not the operator's skill six months later, which is the question Bainbridge's ironies and Macnamara's review are actually asking. The paper also names the interesting design problem: Dell'Acqua's recruiters did worse with the better AI, and Daron Acemoglu's garbling policy, from a 2026 economics working paper, argues for degrading precision on purpose. Both suggest maximum accuracy is not the optimum, and this is the older literature saying the same thing with a threshold attached.
Papers and articles behind it
Parasuraman, R., & Manzey, D. H. (2010). Complacency and Bias in Human Use of Automation: An Attentional Integration. Human Factors, 52(3), 381-410.