Microsoft researchers have developed an AI tool that mimics the step-by-step diagnostic process of doctors, achieving over twice the accuracy of general physicians in solving complex medical cases. The tool, detailed in a study published in the journal Science on October 22, 2026, uses an “orchestrator” to coordinate questions, exams, and tests, simulating a panel of clinical experts collaborating to solve diagnostic cases.
The AI model differs from existing medical AI systems, which typically analyze all patient symptoms and lab results at once. Instead, it decides what to ask about a patient’s history, orders tests based on initial findings, and updates its hypotheses as new information becomes available. When tested on over 300 unusual medical dilemmas and emergency room scenarios, the tool outperformed human doctors using textbooks and the internet.
Dominic King, corporate vice-president of the health unit at Microsoft AI, noted that while AI excels at multiple-choice questions, real-world diagnostics require a different approach. The tool also proved more cost-effective in designing diagnostic work-ups compared to human doctors. Harsha Nori, head of AI at Microsoft AI Health, emphasized the importance of measuring both accuracy and the efficiency of the diagnostic process.
Next, Microsoft plans to collaborate with Beth Israel Deaconess Medical Center in Boston to test the tool’s ability to triage patients in emergency department waiting rooms.