AI Is Quietly Eroding Expert Skills, Early Studies Warn

Professionals across medicine and software are starting to show measurable skill loss as they lean on AI tools, and the people most affected are worried about it. A recent survey of US health-care workers found 70% of nurses and 77% of physicians fear that over-reliance on AI is degrading their own abilities. Emerging research suggests that concern is well-founded.

The clearest evidence so far comes from a Polish study of veteran endoscopists, each with at least 2,000 colonoscopies under their belt. After they began using a real-time AI system that flags precancerous adenomas, their unaided performance fell: detection rates on AI-free procedures dropped from 28.4% before the tool’s introduction to 22.4% afterward. The study authors argue that constant AI assistance leaves clinicians less focused and less rigorous when they have to decide on their own — a phenomenon researchers are calling AI-driven ‘deskilling.’

There is no established defense against the effect yet, and experts say it should become a major research priority over the coming decade. The same question is being probed in software engineering, where Anthropic ran a randomized controlled trial putting 52 engineers through a basic coding task, half of them prompted to use an AI assistant. The broader takeaway: outsourcing cognitive work to AI carries a hidden cost, and individuals and institutions need to decide deliberately which skills are worth keeping sharp.