Physicians' Performance Drops When AI Colonoscopy Tool Becomes Unavailable

A recent observation in the medical field highlights a critical aspect of human-AI collaboration. Physicians who began utilizing an AI-powered tool for colonoscopies demonstrated a notable decline in their diagnostic performance when the system was not accessible. This suggests a potential over-reliance or skill degradation when AI assistance becomes a standard part of their workflow. Specifically, prior to the introduction of the AI tool, specialists consistently identified adenomas in approximately 28.4% of colonoscopy procedures. However, in the period following the AI tool's adoption, the adenoma detection rate for procedures conducted without AI support dropped to 22.4%. This indicates a significant reduction in diagnostic accuracy when the AI aid was absent, compared to their baseline performance before its integration. This phenomenon raises important questions about the long-term impact of AI integration on human expertise and the necessity of maintaining core skills even with advanced technological support.