The evidence that AI tools quietly erode professional skill

Researchers are beginning to measure something many workers already suspected: leaning on AI can make you worse at your own job. A new piece in Nature gathers the early evidence for what scientists now call deskilling, and the sharpest example comes from medicine. In a study of endoscopists in Poland, doctors found adenomas, the growths that can develop into colon cancer, in 28.4% of colonoscopies before they started using an AI detection tool. After several months with the tool, their detection rate when working without it dropped to 22.4%. The explanation is plain. Once software highlights suspicious areas, clinicians look less carefully on their own, and lead author Yuichi Mori says they become less focused and less motivated.

The concern reaches well beyond one specialty. In a survey of US healthcare workers, 70% of nurses and 77% of physicians said they worry about losing skills through over-reliance on AI. To test the same effect in software, Anthropic ran a randomized controlled trial with 52 engineers on coding tasks, part of a wider effort to put real numbers on a problem that has so far been mostly anecdotal. Kevin Crowston of Syracuse University frames the practical question well: the point is to decide in advance which skills you want to keep sharp and which you are willing to hand over.

Why it matters

If you use AI at work, the real risk is not that the tool fails. Your own judgment can quietly fade while the tool keeps performing well, so it is worth choosing now which skills you practice unaided, so they are still there the day you need them.