AI in medicine: Doctors are likely to quickly lose skills in cancer screening (www.heise.de)

A Polish before-and-after study published in The Lancet Gastroenterology & Hepatology found that experienced endoscopists’ adenoma detection rate (ADR) fell by more than 20% within months after they began using AI assistance during colonoscopy. Researchers measured detection success at four endoscopy centers once before and once after AI support was introduced and concluded AI-assisted workflows produced a “negative influence” on clinicians’ tumor-search behavior. The authors label this rapid loss of competence “deskilling,” and note the effect appeared even in highly experienced physicians — implying it may be larger for trainees and could grow as AI systems become more accurate. For the AI/ML community this is a major practical and ethical signal: real-world deployment can change human attention and perception in ways that undermine clinical outcomes, and prior controlled studies may have overestimated net benefit if they didn’t capture downstream deskilling. Key implications include the need to test AI tools in longitudinal, real-world settings; to monitor metrics like ADR after deployment; to design interfaces and training that preserve operator skills; and to study whether deskilling is reversible. Regulators, hospitals and developers should therefore treat high offline model performance as necessary but not sufficient — human-AI interaction effects must be measured and mitigated before wide clinical roll-out.