Human-in-the-Loop, or Rubber Stamp? The “Ceremonial Human” Problem
“There’s always a human in the loop” is meant to be reassuring. Too often, the human only confirms what the system already decided.
“Don’t worry — there’s always a human in the loop.” It’s the standard reassurance whenever an AI system touches an important decision. And it can be true in the way that matters. It can also be theatre.
The uncomfortable version is what we call the ceremonial human: a person who formally approves, signs, and carries the legal responsibility — but who, in practice, only confirms a choice the system already made. The options were narrowed, ranked, and interpreted before they arrived. The human’s job has quietly shrunk to pressing “accept.”
When that happens, “human oversight” stops being a safeguard and becomes a liability shield. If the decision is wrong or unfair, the human is accountable — but had no real power to change it. The machine decided; the person absorbs the blame.
So the honest test of “human-in-the-loop” is not whether a human is present. It’s whether the human can meaningfully do otherwise. Can they see why the system produced its output? Can they access the underlying facts, not just the score? Do they have the authority, time, and information to overrule it — and is overruling a real, used option, not a theoretical one?
Where the answer is no, the loop is decorative. Meaningful oversight has to sit at the real boundary of the decision — before the options are compressed, not after.
→ Test any “human in the loop” claim with The Ten Questions.
