Doctors May Be Losing Skills To AI Without Realizing

Professionals May Struggle to Notice AI Eroding Their Skills

Doctors with an average of 27 years of experience and at least 2,000 colonoscopies behind them showed signs of losing some of their ability to spot cancer after getting used to AI assistance, according to research cited in a new paper. If skill can slip that way in veterans, workers in many other fields may be at risk too.

Kaisa Kärki, a philosopher at the University of Helsinki, ran no experiments of her own. Her paper, published in the journal Human Affairs, pulls together earlier studies on “deskilling,” the loss of professional abilities when tasks get handed to machines, across medicine, accounting, programming, and aviation.

Kärki’s main claim is that individual workers should not be left to tackle deskilling alone. Professionals may not realize how much their skills have slipped until something goes wrong, so she argues that professional organizations, the groups that speak for a field, should take charge instead.

AI Deskilling Sets In Because Hard Practice Gets Skipped

Kärki leans on a psychology principle called “use it or lose it”: abilities fade when they go unused. Research she cites found that adults who kept using their skills extensively showed no skill decay by age 65. Other work found that newly formed brain cells tend to die off unless a person takes on difficult learning tasks.

A driver who always follows GPS may struggle to find the way when the signal drops, and AI can work the same way on professional skills: it makes tasks easy, which removes the struggle that builds them. In a programming study she cites, learners who worked through coding problems without AI assistance learned far more, because they had to hunt down and fix their own errors.

A human tendency to avoid mental strain makes it worse. When AI is available, many professionals lean on it more than necessary, and the hardest tasks, the ones that build skill fastest, are the ones people are most likely to avoid.

Kärki also raises “never-skilling,” a term for newer workers who never get to build certain skills because AI handles the task before they can practice it.

Even Veteran Doctors Showed Signs of AI Deskilling

In the colonoscopy study, the dip in cancer detection appeared when doctors did exams without AI after growing used to it. The study was observational, meaning researchers tracked what happened rather than assigning doctors to different groups.

Breast cancer screening showed a similar pattern. Very experienced doctors who read mammograms, the X-rays used to screen for breast cancer, showed deskilling, though less experienced colleagues showed much more.

Outside medicine, accountants who had worked without AI tools for more than a decade showed a high degree of deskilling after becoming used to AI-assisted work, according to a 2018 study. Not every skill appears equally vulnerable, though. Skills practiced until they become automatic seem to fade less than skills learned for the first time.

Professionals Judge Their Own Skill Loss Poorly

People are not great at grading themselves. A review of earlier research cited in the paper found only a moderate link between how people rate their own abilities and how they perform when tested.

With AI, skill decay often comes from how a task gets done, not from stopping the task, so a worker using AI may still feel like they are practicing the skill.

Professionals are not entirely in the dark. A 2026 Wolters Kluwer Health survey found that 74% of physicians were worried about their own level of deskilling. Worry is not measurement, however. Individual doctors lack the tools to test their own skills, and by the time a loss shows up, errors may already be happening in real settings such as cancer diagnosis.

Professional Groups Should Fix AI Deskilling, Paper Argues

AI makes mistakes, and spotting them takes skill. In one cancer-care test the paper cites, the best AI programs answered 85% correctly, yet over 80% of their wrong answers had a high likelihood of moderate to severe harm to patients.

Professional groups can see across a whole field, Kärki says, so they are better placed to decide which core skills need protecting. Fixes the paper points to include stretches of work done without AI, training periods with no AI tools, regular skill checks, and requiring professionals to make their own decision before consulting an AI recommendation.

In aviation, after autopilot systems induced deskilling in pilots, the field added routine manual flying time and simulator sessions that mimic autopilot failures. Similar routines have been called for in medicine, though the paper does not test which fix works best.

Regulation has gaps too. The EU’s Artificial Intelligence Act tells designers of high-risk AI to keep human supervisors aware of the risk of leaning on it too heavily. Kärki says it is unclear how that awareness would be guaranteed, and the law does not address skill loss that builds up afterward.

A doctor with decades of experience may lose a skill without noticing, and blaming that doctor for the loss makes little sense. Fixing it, Kärki argues, falls to the professions themselves, with individual workers contributing through those same organizations.

---

Disclaimer: This article is a news summary of a published philosophy paper. It is not medical, legal, or professional advice. Readers with concerns about their own health or work should consult a qualified professional.

---