Doctors who got used to AI in colonoscopies spotted fewer growths without it, in one study
Detection without AI fell from 28.4% to 22.4% at four Polish clinics after the tool arrived. A bigger later trial found no such drop.
A colonoscopy looks for polyps, small growths that can turn into bowel cancer. Doctors are scored on how often they find at least one in a patient: the higher that share, the fewer cancers turn up later. AI tools now help by marking suspicious spots on the screen. The worry is that doctors who get used to the help look less carefully when it is switched off.
Four Polish centres started using such a tool at the end of 2021. Researchers compared the colonoscopies those doctors did without AI in the three months before and the three months after, 1,443 patients in all. The share of patients with a growth found fell from 28.4% to 22.4%, a drop of 6 percentage points (p=0.0089). The authors suggest doctors leaned on the tool, but their study design cannot show that directly.
A larger trial, planned in advance and published in Endoscopy in June 2026, found no such drop. Thirteen doctors did 5,013 colonoscopies in three phases: before AI, with AI, then after it was taken away. The seven less experienced doctors found more growths while the AI was running, from 31.9% to 39.5%. When it was removed, their rate went back to where it started rather than below it. Experienced doctors did not change at any point. The authors report neither upskilling nor deskilling.
What it means
The question is real and the evidence is split: one study found a drop, a larger and better-designed one did not. Neither found a problem with the tool while it is switched on, and in the planned trial it raised detection for less experienced doctors. What is unresolved is whether relying on it dulls a doctor’s own eye over time, which is a question for how clinics train and check their staff.
Caveats
- The Polish study only looked back at records, so anything else that changed in those units over those six months could explain the fall. Three months either side is a short window, and 1,443 patients is small for a 6-point difference.
- The two studies measure different things: any pre-cancerous growth in the Polish one, growths of 5 mm or more in the planned trial.
- The planned trial had only 13 doctors, so it could miss a modest effect, especially among the experienced ones.
- Neither study measures what patients care about most, cancers and deaths prevented or missed.
- Other researchers published a challenge to the Polish paper’s interpretation, with a reply from the authors, in the same journal in late 2025.