Microsoft AI Diagnostic Accuracy Surpasses Doctors by Four Times

Microsoft has unveiled research published in Science demonstrating that its multi-agent medical diagnostic system, MAI-DxO, achieves an accuracy rate of 83% in complex case simulations, surpassing the 36% average of general practitioners. By coordinating multiple AI agents to simulate expert consultations—including history taking, testing, and diagnostic debate—the system quadruples human diagnostic performance in specific scenarios, marking a significant step toward "medical superintelligence."

Cross-verified reports highlight the role of Microsoft's proprietary SDBench benchmark in validating these results. AI CEO Mustafa Suleyman emphasizes that this breakthrough addresses global healthcare supply-demand imbalances rather than replacing doctors. The multi-agent architecture overcomes single-model reasoning limitations, positioning AI as a high-efficiency tool that frees clinicians to focus on patient care.

This development reshapes the medical AI landscape, proving the viability of autonomous decision support in high-stakes environments. With plans to achieve human-indistinguishable performance within years, Microsoft is accelerating the shift from assistive diagnostics to proactive clinical intelligence, offering new paradigms for managing healthcare resources amid aging populations.