AI OUTDIAGNOSES EMERGENCY ROOM DOCTORS IN HARVARD STUDY PUBLISHED IN SCIENCE AND THE MEDICAL ESTABLISHMENT IS NOT HAPPY ABOUT IT A study published in the journal Science by researchers at Harvard Medical School and Beth Israel Deaconess Medical Center has landed like a grenade in the medical profession. OpenAI’s o1 reasoning model correctly identified the exact or near diagnosis in 67 percent of real-world triage cases. The two attending physicians assessed in the same experiment scored 55 percent and 50 percent respectively. The AI received the same electronic health record data the doctors worked from, with no additional information and no preprocessing by researchers. The study matters because it used real patients from a real Boston emergency room, not constructed hypotheticals designed to make the technology look good. The model was tested at multiple diagnostic touchpoints from the initial triage moment forward, and its advantage over the physicians was most pronounced at the earliest stage when the fewest data points are available to anyone. The medical establishment pushed back hard. Emergency physicians pointed out that the study compared the AI against non-specialist attending physicians rather than emergency medicine specialists, and that real clinical care involves physical examination, imaging interpretation, and the nonverbal cues a patient gives that no text-based system can detect. The researchers stopped well short of advocating clinical deployment and called for formal prospective trials before anyone draws policy conclusions. But the study is peer-reviewed and published in one of the most prestigious scientific journals in the world. The argument that AI cannot match human clinical judgment at the diagnostic stage just became significantly harder to sustain. Keywords: AI medical diagnosis, OpenAI o1 doctors, Harvard AI study, AI emergency room