May 3, 2026
Paging Dr. Chatbot
OpenAI's o1 correctly diagnosed 67% of ER patients vs. 50-55% by triage doctors
ER docs just got shown up by AI — and the comments are in full panic mode
TLDR: A Harvard study found OpenAI’s model beat ER triage doctors at spotting the right diagnosis from basic patient notes, fueling talk that medicine is changing fast. But the comments stole the show: some want AI second opinions immediately, while others say the real test is doctors with AI, not doctors versus it.
The big headline is wild enough: a Harvard study says OpenAI’s o1 system correctly identified emergency room cases 67% of the time, while human triage doctors landed around 50–55%. Translation for non-hospital people: when doctors had only the quick intake notes and basic patient info, the AI was better at guessing what was actually wrong. Unsurprisingly, the community did not respond with calm reflection. It responded with a full-blown "wait, are we replacing doctors now?" spiral.
The hottest reactions weren’t just about the score — they were about trust. One commenter said human doctors nearly killed them and immediately wanted a place to upload their ER data into an AI for a rematch. Another argued doctors sometimes play it safe to avoid lawsuits rather than chase the less obvious diagnosis, which turned the thread into a mini-drama about whether medicine is really about healing or defensive paperwork. And then came the classic internet move: if AI can beat ER doctors, why stop there? One user casually announced they already use language models to diagnose their dogs and started pitching an AI-powered vet marketplace.
There was skepticism too. Some commenters basically squinted at the study and asked, "Hold on — isn’t o1 old already?" Others wanted the more realistic showdown: not AI versus doctors, but doctors using AI versus doctors without it. That’s where the mood landed — less "goodbye, doctors" and more "give the humans a robot sidekick already".
Key Points
- •A Harvard study published in Science found that OpenAI’s o1 outperformed doctors on some text-based emergency triage diagnostic tasks.
- •In one experiment using records from 76 Boston ER patients, the AI achieved 67% exact or near-exact diagnostic accuracy versus 50%–55% for human doctors.
- •When more detailed patient information was available, the AI’s diagnostic accuracy rose to 82%, compared with 70%–79% for expert humans, though the difference was not statistically significant.
- •In five clinical case studies on longer-term treatment planning, the AI scored 89% versus 34% for 46 doctors using conventional resources.
- •Researchers said the study does not show AI can replace physicians because it excluded non-text clinical signals such as visual appearance and distress, and they highlighted ongoing concerns about AI error and liability.