AI Beats Doctors in Harvard Emergency Triage Trial

A study published in the journal Science shows that OpenAI's o1 reasoning model outperformed human doctors in emergency room triage ^[1][2]. During a trial at Harvard, the AI achieved 82% accuracy in clinical reasoning and diagnosis ^[1]. This score exceeds the 70% to 79% range recorded by expert human doctors in the same high-pressure settings ^[1].

The research highlights how reasoning-based AI models manage complex cases where rapid decisions are necessary ^[2]. Analysts found that the AI provided more consistent diagnostic support compared to standard tools ^[1]. While the results are promising, researchers emphasized that AI should support rather than replace human medical staff ^[2].