“Human in the loop”

Everyone agrees on this one. Whatever the AI does, a human should stay in the loop.

Regulators often ask for it. Vendors promise it. Org charts display it.

But watch what actually happens inside that loop.

The system produces a recommendation. The human clicks approve. Four hundred times a day.

Research has a name for this: automation bias. We defer to the machine even when we're formally in charge — especially when it's right 95% of the time.

Because the 5% is exactly where we've stopped looking.

So the human doesn't provide judgment. The human provides a signature.

And notice what the signature is for. When the system gets it wrong, the failure now has an owner: the person who clicked approve.

Not the vendor. Not the executive who set a quota that made real review impossible.

The loop doesn't preserve human judgment — it launders responsibility downward.

• A human in the loop is not the same as a human on the hook.
• Oversight you can't afford to exercise isn't oversight. It's theater.

Kant would say moral agency can't be delegated to a system that has none.

Aristotle would say judgment dies when there's no time to exercise it.

Both point to the same design question.

Not “Is there a human in the loop?”

But:

Does that human have the time, the authority, and the protection to say no?

If overriding the machine is slower, riskier, and harder to defend than agreeing with it, you haven't kept a human in the loop.

You've built a machine with a human-shaped alibi.