Bengaluru: Doctors who use AI to detect polyps during colonoscopies are now worse at detecting them without it. According to a 2025 study in The Lancet Gastroenterology & Hepatology, their adenoma detection rate—a measure of how many potentially cancerous growths they caught—fell by six percentage points after they started relying on AI assistance.
During the three months before the AI tool was introduced, doctors found at least one adenoma during 28.4 per cent of colonoscopies. In the three months after, the adenoma detection rate for colonoscopies performed without AI assistance decreased to 22.4 per cent.
All these physicians had performed at least 2,000 colonoscopies during their careers.
This is the paradox sitting at the centre of AI’s rapid march into healthcare. Clinicians are using AI tools more than ever. For doctors, daily AI use tripled in a single year—from 10 per cent in 2025 to 38 per cent in 2026. For nurses, that number went from 16 per cent to 32 per cent year-over-year.
But this growing reliance is eroding their ability to do their work themselves. Work that they were able to do before.
A survey of US healthcare workers published earlier this month shows that this is at the top of their minds—70 per cent of nurses and 77 per cent of physicians are worried about losing their skills because of over-reliance on AI systems.
Researchers call it deskilling. The more you outsource a task, the less capable you become of doing it yourself. A 2026 Anthropic Research study found that workers who fully delegated tasks to AI improved their speed—but at the cost of actually learning the skill. The finding was in software development, but the implications map cleanly onto medicine, where the stakes are considerably higher.
“On a positive note, 77% of clinicians reported that they double-check AI answers with original sources or trusted databases like PubMed or UpToDate,” the study added.
Also read: From cancer to diabetic retinopathy, how AIIMS is betting big on AI to aid in diagnostics
Fact-checking session
The patients are accelerating it too. More than half of patients now use AI to research their diagnoses or look up medication side effects before they arrive. Nearly 60 per cent of patients said their doctors openly engage with AI-generated information. And around 56 per cent of doctors say they “review AI information provided by patients, working to explain how that content aligns, or in some cases, does not align, with evidence-based clinical resources.”
The doctor’s office has become a fact-checking session.
Both patients and clinicians — 70 per cent in each group — agree that AI is enabling better health literacy and engagement. The problem is the absence of guardrails around it.
“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 [to AI],” said Kevin Crowston, an information scientist at Syracuse University in New York, to Nature.