In Harvard study, AI offered more accurate diagnoses than emergency room doctors

A Harvard Medical School study published in Science compared OpenAI's o1 and GPT-4o models against human emergency room physicians in 76 real diagnostic cases. The o1 model matched or outperformed human doctors at each diagnostic touchpoint, with strongest advantage during initial triage (67% exact/close diagnosis vs. 55–50% for physicians). Researchers emphasized models received identical information from electronic medical records with no preprocessing.