"Better than 14 hours in a Canadian ER": Microsoft's diagnostic AI solves 83% of hard cases, doctors 36%

"At this point I'd take the AI doctor over 14 hours in a Canadian ER just to get told take Advil." That reply, with 217 likes, sat under a Polymarket post saying Microsoft built an AI that diagnoses complex cases more than twice as accurately as general physicians. The post passed 6,900 likes and 390,000 views.

According to Reuters on Oct 22, Microsoft published results for its MAI Diagnostic Orchestrator (MAI-DxO) in Science. It works like a simulated panel of specialists, deciding step by step what to ask and which tests to order, and revising its diagnosis as results come in.

Per The Next Web, MAI-DxO running on OpenAI's GPT-5 solved 83.2% of 304 difficult NEJM cases, versus 36.1% for generalist doctors with textbooks and internet access. On 325 emergency admissions it scored 80.3% to 46.8%. Microsoft AI CEO Mustafa Suleyman wrote on X that it is "1.7x more accurate than physicians in the ED" and orders fewer tests.

There are caveats. The comparison was with generalists, not specialists, and cases were presented as text with no physical exam or imaging. The tool is not used in clinical care yet; Microsoft plans to test ER triage with Beth Israel Deaconess Medical Center in Boston. "No patient ever came to see me with a nice summary," said Dominic King of Microsoft AI's health unit.