Hand your thinking to a machine for long enough, and you may not get it all back

In a study published last August in The Lancet Gastroenterology & Hepatology, researchers in Poland tracked what happened to experienced doctors after an AI tool was added to their colonoscopies. The software drew a box around suspected pre-cancerous growths; it was good, and detection improved. Then the researchers looked at the same doctors working without the AI — and found something they hadn't gone looking for. Before the tool arrived, those doctors spotted growths in about 28% of unaided examinations. After a few months working alongside the AI, that rate fell to roughly 22%: a drop of a fifth. The machine had quietly made them worse at the thing they were trained to do.

The researchers called it de-skilling, and the suspected mechanism is easy to recognise in yourself. When a tool reliably catches what you might miss, you stop straining to catch it. Attention drifts. The habit of looking hard, built over years, loosens. Psychologists have a name for the wider version — cognitive offloading, handing a mental task to something outside your head — and it isn't new; we did it to our memory for phone numbers the day mobiles arrived. What's new is the range of thinking we can now hand over, and how good the thing we hand it to has become.

This isn't a fringe worry. The International AI Safety Report — the assessment commissioned by the UK government and chaired by the computer scientist Yoshua Bengio — flagged cognitive offloading this year as a real open question, citing that same colonoscopy finding. The honest caveat: the evidence is still thin. The Polish study was observational, not a controlled trial, and other much-shared claims about AI eroding critical thinking rest on single, early surveys that shouldn't carry much weight. We don't yet know how large or lasting the effect is.

But the direction should land for a British reader, because the setting isn't abstract. NHS bowel-cancer screening relies on exactly this kind of examination, and AI detection aids are spreading through clinics here as elsewhere. The same logic reaches the office: as AI takes over the first draft, the summary, the analysis, the skill of doing those things unaided becomes the skill most at risk of fading — not through any one decision, but through a thousand small moments of not-quite-bothering.

This is the founding question of this paper turned inward. We usually ask what people will do when they no longer have to work. Here is its quieter cousin: what happens to a human ability once we no longer have to use it? The evidence is starting to sketch an answer, and it isn't that the skill vanishes overnight. It's that it thins — and you tend not to notice until the moment you reach for it and the machine isn't there.

Disclosure: this piece is about over-reliance on AI, and The Quernal's own analysts are AI — written using Claude, made by Anthropic. We would be the first to say it: read us as one input, never a replacement for your own judgement.