For years, AI has been promoted by healthcare professionals and companies for its potential in medicine, including enhancing diagnostics and surpassing doctors in tasks like cancer detection. Some enthusiasts even envision AI one day discovering a “cure for cancer.”
However, a recent study indicates that doctors who consistently used AI experienced a decline in their skills within a few months.
The study, published in The Lancet on Wednesday, revealed that clinicians who regularly used AI became overly dependent on its recommendations over six months. This dependence led them to become “less motivated, less focused, and less responsible when making cognitive decisions without AI assistance.”
This research adds to a growing body of evidence highlighting the possible negative effects of AI on its users. A previous study from MIT suggested that ChatGPT could produce inaccurate or fabricated scientific information.
How the study was conducted
Researchers from various European institutions conducted an observational study across four endoscopy centers in Poland participating in the Artificial Intelligence in Colonoscopy for Cancer Prevention (AICOLON) trial. Funding was provided by the European Commission and the Japan Society for the Promotion of Science.
As part of the trial, these centers implemented AI tools for detecting polyps—growths that can be either benign or cancerous—in late 2021. The study examined 1,443 colonoscopies performed without AI assistance out of a total of 2,177 colonoscopies conducted between September 2021 and March 2022. Nineteen experienced endoscopists carried out these colonoscopies.
The researchers compared the quality of colonoscopies performed three months before and three months after the introduction of AI. Colonoscopies were randomly conducted either with or without AI assistance. Among those performed without AI, 795 occurred before regular AI use, and 648 took place after the AI tools were implemented.
What the study found
Three months before AI implementation, the adenoma detection rate (ADR) was approximately 28%. After AI was introduced, the ADR dropped to 22% when clinicians performed unassisted colonoscopies. ADR, a standard metric for colonoscopy quality, represents “the proportion of screening colonoscopies performed by a physician that detect at least one histologically confirmed colorectal adenoma or adenocarcinoma.” Higher ADR is linked to a reduced risk of colorectal cancer, as adenomas are precancerous growths.
The study indicated that AI improved detection rates when used, but clinicians’ detection skills declined when AI assistance was removed.
Researchers attributed this decline to “the natural human tendency to over-rely” on decision support systems like AI.
“Imagine you want to travel somewhere and can’t use Google Maps,” said Marcin Romańczyk, study co-author and assistant professor at the Medical University of Silesia. “We call it the Google Maps effect. We try to get somewhere, and it’s impossible to use a regular map. It works very similarly.”
Implications of the study
Omer Ahmad, a consultant gastroenterologist at University College Hospital London, who wrote an editorial accompanying the study (but wasn’t involved in the research), told TIME that AI exposure likely weakened doctors’ visual search habits and gaze patterns essential for polyp detection.
“In essence, dependence on AI detection could dull human pattern recognition,” Ahmad stated. He added that regular AI use might also “reduce diagnostic confidence” when AI assistance is unavailable or impair the endoscopists’ colonoscope maneuvering skills.
Catherine Menon, principal lecturer in the University of Hertfordshire’s Department of Computer Science, commented that, “Although de-skilling resulting from AI use has been raised as a theoretical risk in previous studies, this study is the first to present real-world data that might potentially indicate de-skilling arising from the use of AI in diagnostic colonoscopies.” Menon expressed concern that over-reliance on AI could make healthcare practitioners vulnerable to technological failures.
Other experts advise caution when interpreting the results of a single study.
Venet Osmani, a professor of clinical AI and machine learning at Queen Mary University of London, noted to SMC that the overall number of colonoscopies (both AI-assisted and non-AI-assisted) increased during the study. Osmani suggested that the increased workload may have caused clinician fatigue and lower detection rates.
Allan Tucker, a professor of artificial intelligence at Brunel University of London, also pointed out that clinician performance improved overall with AI assistance. He added to SMC that concerns about deskilling due to automation bias “is not unique to AI systems and is a risk with the introduction of any new technology.”
“The ethical question then is whether we trust AI over humans,” Tucker stated. “Often, we expect there to be a human overseeing all AI decision-making, but if the human experts are putting less effort into their own decisions as a result of introducing AI systems this could be problematic.”
“This is not simply about monitoring technology,” Ahmad says. “It’s about navigating the complexities of a new human-AI clinical ecosystem.” He emphasizes the importance of establishing safeguards, suggesting that, beyond this study, efforts should focus on “preserving essential skills in a world where AI becomes ubiquitous.”
“`
