AI reliance is measurably eroding skills
A colonoscopy study found doctors got worse at spotting growths after using AI, and Nature says the deskilling effect is showing up across professions.
Doctors got worse at finding precancerous growths after they got used to a machine finding them first. A multicentre study at four endoscopy centres in Poland, part of the ACCEPT trial and published in The Lancet Gastroenterology & Hepatology, watched what happened when experienced clinicians ran colonoscopies without AI after a stretch of using it. Their adenoma detection rate fell from 28.4% to 22.4%, a 6 percentage point drop, roughly 20% worse than before the tool arrived.
The colonoscopy work is observational, not a controlled trial, and some experts note the same centres ran far more procedures once AI arrived, so fatigue could be part of it. The researchers still called the pattern deskilling. On June 18, Nature pulled it together with other early findings under a blunt headline: routine AI use is starting to wear down hard-won professional skills. A survey it cites found 70% of nurses and 77% of physicians worried about losing their abilities, and the same alarm is being raised about software engineers who lean on coding assistants for work they used to do by hand.
Why this lands on operators
What changed is the direction. For years the pitch was that AI assistance only adds. This is measured evidence that handing a task over completely can subtract, by letting the underlying skill fade. That skill is also what lets you notice when the tool is wrong.
If you cannot do a job at all without the AI, you cannot check it either. That verification habit is the whole game, and it only works if you stay fluent enough to spot a bad answer. Keep your hand in on the work that matters, know where these tools reliably slip, and treat full automation of a critical task as a skills risk, not just a productivity win.