Microsoft’s AI diagnostic system beats generalist doctors in simulated hard cases

Microsoft has developed MAI-DxO, an AI system designed to handle diagnosis as a sequence of decisions rather than a one-shot guess. In a study published in Science, it correctly identified 83% of 304 difficult published cases, versus 36% for participating generalist doctors who could use reference tools; it also chose testing paths with lower estimated costs. The system works less like a single chatbot and more like a virtual clinical team: it asks for history, requests tests, revises possible explanations, and weighs whether another test is worth ordering. That is a more realistic test than medical multiple-choice questions, but it does not show the software is ready to replace doctors, since the study used simulations and case records rather than routine care for live patients.