Artificial intelligence (AI) is becoming a routine partner in medicine — scanning X-rays for early signs of cancer, flagging subtle changes in blood tests and highlighting polyps during colonoscopies that the human eye might miss. But a new study in Lancet Gastroenterology and Hepatology has raised an uncomfortable question: what happens to doctors’ skills when they start relying on AI every day?
The paper, “Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy”, suggests that while AI can sharpen diagnostic accuracy in the moment, it may also subtly change how clinicians work — and not always for the better.
The Polish-led research looked at more than 1,400 colonoscopies before and after four clinics introduced an AI tool designed to spot precancerous growths. The technology had already been shown to boost detection rates, but the research revealed that when experienced endoscopists later performed colonoscopies without AI, their ability to find those growths dropped sharply.
While focused on AI-assisted colonoscopies, the research findings could have implications for other specialties, from radiology to dermatology, where AI adoption is accelerating. Paired with fresh survey data on what doctors really want from AI, the picture that emerges is one of cautious optimism for AI in healthcare — and a warning that design and deployment matter as much as the technology itself.
From clinical support to clinical crutch?
The researchers found that several months after the routine introduction of AI technology, the adenoma detection rate (ADR) in non-AI procedures fell from 28.4% to 22.4% — a 20% relative decrease. While AI-assisted procedures still performed well (25.3% ADR), the authors say this is the first real-world evidence of “deskilling” in a core medical task.
Lead author Dr Marcin Romańczyk from the Academy of Silesia called the findings “concerning given the adoption of AI in medicine is rapidly spreading”.
“We need to find out which factors may cause or contribute to problems when healthcare professionals and AI systems don’t work well together, and to develop ways to fix or improve these interactions.”
The team says more research is needed into how human–AI teamwork can avoid undermining core skills across different medical fields.
Why it matters beyond bowel screening
Colonoscopy is just one test — but the potential implications are likely far wider, in terms of both how AI is used in healthcare and whether the current enthusiasm for rapid uptake will persist.
- Skill erosion could hit any specialty where AI takes on tasks previously done by humans: radiology, dermatology, cardiology imaging, pathology.
- Previous lab studies hinted that doctors change their behaviour when AI is involved — looking less closely at scans or images because they expect the algorithm to catch errors.
- If those habits persist when the AI is turned off, it could mean slower or less accurate diagnoses in other contexts.
Doctors’ attitudes: cautious optimism
The concerns come against a backdrop of measured curiosity from clinicians, with a separate survey released last week by So What? Research showing they rely on AI mainly for real-world support, particularly with things that would otherwise take away from patient-focused time, but that “they don’t expect it to work miracles”.
In a survey of more than 400 Australian doctors on their feelings about AI in healthcare:
- 54% saw both positives and negatives;
- 35% were mostly positive; and
- only 11% were strongly negative or dismissive.
Specialty also matters: gastroenterologists — the very group in the Lancet study — were among the most enthusiastic (67% seeing more positives than negatives), followed by rheumatologists (62%). Oncologists and endocrinologists were far more cautious, with just 27% and 23%, respectively, having mostly positive views of AI.
What doctors actually want from AI
The survey also asked doctors where AI could make the biggest difference. Their top priorities were more quotidian than futuristic:
- 65% want AI to reduce admin and paperwork
- 55% want support for clinical decision-making
- 52% are interested in improved diagnostics and personalised care
- 46% see value in enhanced patient monitoring
- 39% want help with drug discovery
- 34% are interested in predictive outbreak modelling
The message: AI will gain trust by solving everyday problems and supporting — not replacing — clinical judgment.
The focus on real-world utility is already reflected in some of the AI-driven health innovation emerging from local companies — from clinical decision support platforms and patient-monitoring tech to smarter medical transcription and translation services that save clinicians time.
ASX-listed Straker Ltd (ASX:STG), for example, has been developing advanced language AI that can streamline clinical communications, while other home-grown players are trialling AI tools for pathology triage and imaging analysis. The common thread for leaders in this space, at least for now, is technology that frees up doctors to spend more time with patients and less on paperwork.
Designing AI that helps without harming
The Lancet findings are a reminder that introducing AI is not just a technology upgrade — it is a behavioural shift. If clinicians change how they work when AI is present, those patterns can stick.
That means developers, healthcare providers and policymakers need to:
- Build systems that train as well as assist — prompting clinicians to stay engaged, not just confirm AI suggestions;
- Rotate AI use strategically to maintain skills; and
- Measure outcomes beyond AI-assisted performance, looking at how clinicians perform when the tech is absent.
AI has the potential to make doctors faster, sharper and more consistent. But as the Lancet colonoscopy study shows, it could also change how they approach their work — sometimes in ways that undermine hard-won skills.
With most doctors open to AI but wary of hype, the real opportunity lies in tools that work with clinicians, not instead of them — preserving the instincts and expertise that no algorithm can replace.