Nature: AI Tools Are Degrading Doctors' and Coders' Skills

There is a specific kind of finding that lands harder than the leaderboard headlines, and Nature just published a good example. Its news feature pulls together the first wave of controlled studies asking whether professionals who use AI every day are quietly getting worse at their jobs. The early answer, hedged but consistent across the studies cited, is yes.

The clearest case comes from colonoscopy. A multicentre observational study of Polish endoscopists, published in The Lancet Gastroenterology and Hepatology, found that after routine AI assistance was introduced at their centres, the doctors' adenoma detection rate on unaided procedures fell from 28.4% to 22.4%, a relative decline of about 20%. The specific number matters because adenoma detection is a hard, measurable outcome that maps onto real patient risk, not a lab exercise.

On the software side, Nature cites an Anthropic randomized trial of 52 engineers in which those given AI coding assistants scored 50% on a post-task quiz against 67% for the control group, with the gap particularly stark on questions requiring error diagnosis in code. Kevin Crowston, quoted in the piece, puts the mechanism plainly: people are "basically borrowing skills from the AI, but are not developing those skills themselves." Robert Wachter's phrasing is the one worth sitting with, that "even highly skilled professionals might get worse at tasks that their job requires as they become more dependent on AI tools."

The honest caveat is that this is a first wave of small, mostly short-term studies, and the piece is careful to frame the evidence as early results rather than settled science. What the reporting does not tell you is whether the deskilling reverses with mandatory AI-off practice, whether seniors are more resilient than juniors, or how much of the coder quiz gap is real skill loss versus effort withdrawal when the AI is doing the thinking. Surveys cited by Nature suggest the workforce already senses the problem: 70% of nurses and 77% of physicians told US researchers they are worried about losing their skills because of over-reliance on AI systems.

The interesting move for anyone building or buying these tools now is to stop treating augmented output as the whole measure. If Nature's summary of the field holds up, the leaders will be the hospitals, firms, and vendors that bake unaided-practice reps and audit into the workflow before the deskilling becomes the story regulators care about.