In a carefully constructed benchmark study published in Nature, researchers placed artificial intelligence systems alongside a human specialist to see who could better identify oral mucosal lesions — including cancers — from biopsy-confirmed cases. The specialist prevailed, achieving 70% accuracy against a best AI result of 66%, while general-purpose language models fell far short, some barely surpassing chance. The study is less a verdict on AI than a map of where the technology stands: promising in specialized forms, unreliable in general ones, and not yet ready to stand alone where the stak