Human-in-the-Loop

Human-in-the-loop means a person is intentionally involved at some point in an AI-driven process — reviewing outputs, approving decisions, correcting errors, or guiding system behavior. The concept spans a wide range: from a clinician authorizing an AI treatment recommendation before it takes effect, to an analyst reviewing a model-generated report before it goes to leadership. What the human does, when they do it, and with what authority varies enormously between implementations.

"A human reviews it" is one of the most overloaded phrases in AI governance. It can mean genuine independent judgment or reflexive approval under time pressure — and from the outside, both look the same. Automation bias is real: when AI systems produce confident-looking outputs consistently, reviewers adapt and begin approving without scrutinizing. A review process that works at low volume often breaks under production load when capacity doesn't scale with output volume. Calling something human-in-the-loop without specifying who reviews, with what authority, on what timeline, and with what training is governance language that provides cover without providing oversight.