Measure · A framing I propose

The Automation Bias Index

Automation bias is not a new discovery. The tendency to lean on reliable automation and stop checking it has been documented for decades. What we lack is a way to watch it grow inside a specific system, before it has hollowed the oversight out.

Automation bias is the well-studied human tendency to over-trust automated systems and under-use one's own judgment. I do not claim to have discovered it; Parasuraman and Riley described it carefully in the nineties, and the literature since is large. What I propose, at v0.1 and as a framing rather than a finished metric, is an Automation Bias Index: a way to track how fast deference rises as a model gets more reliable, so a team can see the oversight emptying before it is empty.

What automation bias is

The short version: when a machine is usually right, people stop double-checking it, and their own skill at catching its errors fades. It is rational in the moment and dangerous over time, because the rare error arrives into a reviewer who is no longer watching. This is the mechanism underneath the rubber stamp.

Why measure its slope

Because the danger is not the bias itself but how quickly it sets in. The steeper the climb from healthy skepticism to reflexive approval, the faster meaningful human oversight decays into theater. An index that tracks that slope turns a known psychological effect into something a specific deployment can monitor and design against.