The comforting story about AI has a human in it. The machine does the work, and a person stays on top — checking, catching mistakes, overruling it when it’s wrong. Keep a human in the loop, everyone says, and relaxes.
There’s a hole in that story, and it’s the whole point of this one. The human can only catch the machine’s mistakes if the human still knows how to do the work. Supervision is not a job title; it’s a skill, and it’s made of exactly the competence the machine is busy relieving you of. Take that away and the human in the loop becomes a human in the chair — present, responsible, and unable to tell whether anything is wrong.
The trap closes behind you
This is what makes deskilling different from ordinary forgetting: it removes the very faculty you’d need to notice it happening.
Aviation learned this the hard way. Autopilot made flying safer for decades — and then a generation of pilots arrived who had logged thousands of hours without ever really flying the plane, and when the automation handed control back in a crisis, some of them could not do the thing they were, on paper, in charge of. The system worked beautifully right up until the moment it needed a human who’d kept the skill, and the skill was the one thing the system had quietly stopped requiring.
You can’t supervise what you never learned to do.
The same shape is now spreading to every desk where a person is told to “review the AI’s output.” Review it how? With what? If you never learned to write the contract, audit the figures, diagnose the patient, or build the argument yourself, your review is theatre. You can rubber-stamp. You cannot supervise. And the more capable the machine gets, the more tempting it is to stop maintaining the skill that would let you catch it — right up until the day it’s wrong and you’re the one who signed.
Keeping a hand on the wheel
The answer is not to refuse the automation. Planes should have autopilot. Desks should have AI. The answer is to keep flying the plane often enough that you still can.
That means doing the work by hand sometimes even when the machine would do it — not as ritual, but as training, so the supervising skill stays real. It means organisations that resist the temptation to let competence quietly drain out of their people in exchange for short-term speed. And it means being honest about the difference between I checked it and I could have done it, because only the second one is supervision.
Keep a human in the loop, by all means. Just make sure it’s a human who still knows how to take the wheel.