Medical AI education materials show errors despite automated verification

A study conducted by researchers from Asan Medical Center and Samsung Changwon Hospital has revealed that generative AI-produced medical education materials contain errors even after passing initial automated verification.

Using a six-stage production system, the research team created 6,000 flashcards and 833 infographics for radiology specialists. While the team set a strict safety threshold of an error rate below 0.3%, a manual review of 1,284 flashcards by 20 medical professionals found an error rate of approximately 1.0% in materials that had already passed AI-based automated checks.

The findings, published in the journal npj Digital Medicine, emphasize the necessity of a ‘human-in-the-loop’ system. The researchers concluded that while generative AI can mass-produce professional medical content, expert human oversight is essential to ensure accuracy and safety in medical education settings.