AI-Driven Deskilling in Medicine: Evidence Mounts That Clinicians Lose Diagnostic Abilities When Relying on Artificial Intelligence

As artificial intelligence tools become increasingly embedded in clinical workflows, a growing body of evidence suggests that healthcare professionals may be losing hard-won diagnostic skills when they rely on these systems—a phenomenon researchers are calling AI-driven "deskilling."

A pivotal study published last October in The Lancet Gastroenterology and Hepatology demonstrated just how quickly AI assistance can erode human expertise. Physicians in Poland who specialized in endoscopy—each having performed at least 2,000 colonoscopies during their careers—were given access to an AI system that analyzes colonoscopy images in real time and flags adenomas, a type of precancerous intestinal lesion. The tool was available on some days but not on others.

The results were striking. During the three-month period before the AI tool was introduced, the specialists detected at least one adenoma (搜索)搜索该疾病 in 28.4% of colonoscopies. After the tool was introduced, however, the adenoma detection rate for colonoscopies performed without AI assistance fell to 22.4%—a decline of six percentage points. The study authors concluded that continuous exposure to such tools can cause clinicians to become "less motivated, less focused, and less responsible when making cognitive decisions without AI assistance."

"It kind of forgot what to look for," said Trent Cash, a behavioral scientist at the University of Waterloo in Canada, describing the pattern observed in the colonoscopy study. Cash and his colleagues recently argued in Trends in Cognitive Sciences that learned skills may indeed atrophy when taken over by AI. "The evidence is extremely clear that if we offload a specific skill to AI, we're probably not going to retain that skill particularly well."

A Pattern Across Domains

The deskilling effect extends beyond medicine. Among high school students learning a new math concept, those who used an AI tool like ChatGPT to solve practice problems performed worse than students who never used the tool once it was taken away, according to research by economist Alp Sungu of the Wharton School at the University of Pennsylvania and colleagues, published in the Proceedings of the National Academy of Sciences.

Similarly, a preprint from Carnegie Mellon University researchers found that people using an AI assistant to solve practice SAT reading comprehension questions struggled to produce correct answers when the AI was removed—and were more likely to simply give up and skip questions. "These findings are particularly concerning because persistence is foundational to skill acquisition," the team wrote.

The Cognitive Loop: Use It or Lose It

Cash frames the issue as a "use it or lose it" principle, which he calls keeping oneself in the "cognitive loop." The key distinction, emerging research suggests, lies in whether AI replaces mental work or supports it.

"When AI replaces the work, skills fade," Cash and colleagues argue. But AI tools designed with learning guardrails—offering hints and encouragement rather than fully formed answers—can preserve and potentially enhance skills. In Sungu's study, an AI tutor that provided problem-solving guidance without giving away solutions helped students learn new math concepts as effectively as traditional book-based study.

"The struggle of working through a new concept is where real learning happens," Sungu said. "When you're solving a problem, you need to get your hands dirty."

Clinical Implications and Unanswered Questions

The implications for healthcare are significant. A survey of US healthcare workers published earlier this year found that 70% of nurses and 77% of physicians are worried about losing their skills because of over-reliance on AI systems. Robert Wachter, a physician at the University of California, San Francisco and author of a book on AI in healthcare, noted that the colonoscopy study suggests even highly skilled professionals may get worse at core tasks as they become more dependent on AI tools.

Yuichi Mori, a physician-researcher at the University of Oslo and co-author of the Lancet study, cautioned that more studies are needed to confirm the deskilling phenomenon. "There is no established solution against deskilling right now," Mori said. "It should be a very hot research topic in the next decade."

Preserving Human Expertise

Experts emphasize that individuals and institutions must make deliberate choices about which skills to protect. "Just being aware that this phenomenon exists hopefully provokes some self-reflection about which skills people want to maintain and which they're willing to outsource," said Kevin Crowston, an information scientist at Syracuse University.

Benjamin Lira Luttges, a behavioral scientist at Wharton, offered a guiding principle: "Outsource the crap, not the craft." His research found that motivated learners could benefit from AI-generated examples—such as rewritten cover letters—by actively engaging with and refining the AI's output rather than passively accepting it.

Cash remains skeptical of doomsday scenarios. "Humans are resilient," he said. "I don't think that AI is going to suddenly turn our brains to mush." But the evidence is accumulating that without conscious effort to stay in the cognitive loop, the skills that define clinical expertise may quietly slip away.