Microsoft has unveiled a new artificial intelligence tool capable of outperforming experienced doctors when diagnosing complex medical conditions. The system, developed by the company’s AI division under the leadership of Mustafa Suleyman, was tested using over 300 real-world case studies from the New England Journal of Medicine. When paired with OpenAI’s advanced model “o3,” the AI tool successfully diagnosed roughly 85% of cases—compared to just 20% by a panel of 21 human physicians.
The AI system works by replicating the thought process of a medical team: gathering patient information, ordering tests, and interpreting results. A key component, known as the “diagnostic orchestrator,” collaborates with major AI models including those from OpenAI, Meta, Anthropic, Google, and others to guide the diagnostic process. The orchestrator functions as a kind of digital coordinator, determining which model is best suited for each part of the task and synthesizing their outputs.
In contrast to the human doctors in the trial—who had no access to reference materials or peer consultation—the AI tool leveraged a wide base of medical knowledge and performed complex reasoning to reach its conclusions. Microsoft noted that this setup was designed to test each side independently and fairly.
Beyond diagnostic accuracy, the AI system was also found to be more efficient in ordering fewer medical tests while still arriving at correct conclusions. This efficiency could reduce healthcare costs and ease some pressure on overstretched systems worldwide. However, the company stressed that it does not intend for AI to replace physicians. Instead, it envisions the tool as a support system—particularly valuable for second opinions, diagnostic assistance in under-resourced areas, and streamlining routine evaluations.
Despite strong early results, Microsoft acknowledged limitations in its research. The current testing focused on complex or rare conditions, not the day-to-day ailments seen in most clinics. The physicians involved in the trial were also placed in a constrained environment, lacking tools they might usually rely on. Future versions of the AI tool will be tested in real-world settings, where such variables can be reintroduced.
Suleyman believes that near-perfect AI diagnostic tools could be available within a decade. Speaking with The Guardian and Yahoo Finance, he emphasized that transparency in how the AI reaches decisions will be crucial to gaining public trust. Because the orchestrator logs each interaction and rationale behind a diagnostic suggestion, doctors and patients alike will be able to trace the logic behind each outcome.
The company described its work as a step toward “medical superintelligence,” where AI systems could eventually assist with every aspect of clinical reasoning. However, Microsoft continues to position the technology as an assistant rather than a substitute—supporting doctors in both time-consuming and intellectually demanding aspects of care.
As the medical field continues to explore how AI can aid decision-making, tools like Microsoft’s diagnostic orchestrator may help reshape how health care is delivered, especially in areas where access to specialists is limited or resources are stretched thin.
This image is the property of The New Dispatch LLC and is not licenseable for external use without explicit written permission.
