Microsoft's AI Tool Outperforms Doctors in Medical Diagnosis
Microsoft's developed MAI-DxO artificial intelligence diagnostic system achieved a higher accuracy rate than physicians in complex cases during a simulation study published in the journal Science.
Microsoft has announced a new diagnostic system called MAI-DxO (Diagnostic Orchestrator) that reasons over medical symptoms and histories by coordinating multiple AI agents. According to the study published in the journal Science, the system exhibited higher diagnostic accuracy compared to general practitioner physicians in simulations.
The MAI-DxO system, announced by Microsoft AI CEO Mustafa Suleyman, outperformed human physicians in benchmarks derived from medical case reports and emergency department data. Powered by OpenAI's GPT-5 and GPT-5.5 models, the system achieved an accuracy rate of approximately 83 percent in complex cases sourced from The New England Journal of Medicine. The accuracy of general practitioners using references in the same cases was recorded at 36 percent, while those not using references stood at 22 percent. Additionally, in emergency department cases from Beth Israel Deaconess Medical Center, the artificial intelligence reached an accuracy rate of 80 percent, whereas human doctors remained at 47 percent.
It was noted that the complex cases used in the study are normally handled by specialists, but specialists were not included in the comparison. Furthermore, Microsoft researchers emphasized that while the system holds promise, additional clinical evaluations are needed to measure and verify its performance in a real hospital environment. Within this scope, the company plans to begin real-world clinical evaluations at Beth Israel Deaconess Medical Center and is also partnering with Mayo Clinic for health-focused foundational model training.