AI In Diagnosis Boosts Detection, But May Erode Doctors' Skills, Study Warns

‘Google Maps’ effect can make medics over-reliant on technology.

AI is improving the accuracy of colonoscopies, but new research suggests it could be weakening doctors' own ability to detect cancer, a phenomenon researchers liken to forgetting how to read a map after years of using GPS.

A new study published in The Lancet Gastroenterology & Hepatology found that doctors who regularly used AI for colonoscopy screening became less skilled at spotting cancer indicators when the technology was removed, showing declines within just a few months.

Colonoscopy is a key method for preventing bowel cancer, enabling doctors to detect and remove adenomas (precancerous growths) before they become malignant.

In recent years, AI assistance has generated enthusiasm, with numerous trials demonstrating improved detection rates when the technology is applied. However, the new study's results highlight concerns about "deskilling" — the gradual loss of professional expertise when clinicians depend too heavily on automated systems.

The research, conducted across four endoscopy centers in Poland, observed 2,177 colonoscopies between September 2021 and March 2022. Of these, 1,443 were performed without AI assistance.

Before AI was introduced, the adenoma detection rate (ADR) – a key measure of colonoscopy quality – stood at about 28 percent for non-AI-assisted procedures. Three months after the technology's rollout, the ADR for non-assisted procedures dropped to 22 percent.

While AI-assisted colonoscopies still boosted detection rates, the results suggest that when clinicians became accustomed to AI guidance, they were less effective without it.

The study attributes this to a "natural human tendency to over-rely" on decision-support tools, a cognitive phenomenon also seen in navigation habits.

"We call it the Google Maps effect," said Marcin Romańczyk, co-author of the study. "We try to get somewhere, and it's impossible to use a regular map. It works very similarly."

Omer Ahmad, a consultant gastroenterologist at University College Hospital London said AI reliance could weaken visual search habits and gaze patterns critical for polyp detection, as well as reduce diagnostic confidence.

Catherine Menon, principal lecturer at the University of Hertfordshire, noted that the findings provide some of the first real-world data to suggest deskilling from AI use in diagnostic colonoscopies.

She warned that overreliance could leave health care providers vulnerable if AI systems fail or are compromised.

However, some researchers urge caution in interpreting the results.

Venet Osmani, professor of clinical AI at Queen Mary University of London, pointed out that the total number of colonoscopies – both AI-assisted and not – increased during the study, potentially leading to fatigue that could explain the drop in ADR.

Allan Tucker, professor of AI at Brunel University London, said overall detection improved when AI was in use, and the risk of "automation bias" is not unique to AI but to any new technology that shifts human decision-making.

Researchers suggest there is need for safeguards to ensure core diagnostic skills are preserved as AI becomes more integrated into health care.

Romańczyk called for further studies to examine how AI use affects the skills of health care professionals in various medical specialties.

This article originally appeared on our sister site Computing.