AI nailed emergency diagnoses better than doctors in Harvard trials

A Harvard study found OpenAI’s o1 reasoning model diagnosed emergency-room cases more accurately than doctors in text-based tests.

The advantage was strongest when information was limited, but it was not statistically significant in the fuller-data trial.

Researchers said AI could support faster second opinions in emergency care, but it cannot replace bedside clinical judgment yet.

Safety, accountability, and non-text clinical signals remain major barriers to real-world deployment.