Artificial intelligence can help professionals work faster and make difficult tasks easier. But a new philosophy paper raises a concern: when people rely on AI too often, they may gradually lose skills without realizing it.

The paper, led by University of Helsinki philosopher Kaisa Kärki, reviews earlier research on “deskilling”—the decline of abilities when technology takes over parts of a job. One study found that experienced doctors who had become accustomed to AI assistance showed a drop in their ability to detect cancer during colonoscopies performed without AI. The doctors had an average of 27 years of experience and had completed at least 2,000 procedures.

Research involving breast cancer screening and accounting also showed signs of skill loss after professionals began using AI tools. The effects were not the same for everyone: some skills seemed more vulnerable than others. Skills practiced until they become automatic may be less likely to fade than skills that are newer or used less often.

The concern is not limited to experienced workers. Kärki also describes “never-skilling,” which can happen when newer professionals never get the chance to learn a task because AI handles it for them. Programming research cited in the paper, for example, found that learners gained more when they solved coding problems without AI, working through mistakes themselves.

Workers may not be able to judge their own skill levels accurately. And because AI can assist with a task while the person is still involved, the work may feel like practice—even when the professional is doing less of the thinking that builds expertise.

Kärki argues that professional organizations should help protect essential skills. Possible measures include regular work or training periods without AI, routine skill checks, and asking professionals to form their own judgment before viewing an AI recommendation. Aviation offers one example: pilots continue practicing manual flying and use simulators that recreate autopilot failures.

The paper does not test which safeguards work best, and its findings do not mean AI should be abandoned. Instead, it suggests that professions need to think carefully about which skills must remain strong—and build in ways to practice and measure them. If AI is becoming part of the job, keeping human expertise sharp may need to become part of the job, too.