Harvard study finds AI more accurate than ER doctors in diagnoses
Original: In Harvard study, AI offered more accurate diagnoses than emergency room doctors
Why This Matters
Study demonstrates AI potential to enhance emergency medical diagnosis accuracy
Harvard Medical School study published in Science shows OpenAI's o1 model achieved 67% accuracy in emergency room triage diagnoses versus 55% and 50% for human physicians. Study examined 76 Beth Israel emergency room patients.
Harvard Medical School researchers compared OpenAI's o1 and 4o models against two attending physicians in diagnosing 76 Beth Israel emergency room patients. The AI models used identical electronic medical record information available at diagnosis time. OpenAI's o1 model provided exact or close diagnoses in 67% of triage cases, while human physicians achieved 55% and 50% accuracy rates respectively. The performance gap was most pronounced during initial emergency room triage when information is limited and urgency highest. Lead author Arjun Manrai stated the AI 'eclipsed both prior models and our physician baselines.' Researchers emphasized the study demonstrates need for prospective trials in real-world patient care settings rather than claiming AI readiness for life-or-death decisions.