Microsoft has unveiled a new AI system that outperforms human doctors in diagnosing complex medical cases, describing the breakthrough as a step toward “medical superintelligence.”
Led by Mustafa Suleyman, CEO of Microsoft AI, the system mimics a panel of expert physicians and works alongside OpenAI’s latest o3 model. In testing, the AI accurately solved more than 80% of challenging case studies from the New England Journal of Medicine – compared to a 20% success rate by practicing doctors working alone.
Microsoft said the AI is not only more accurate but also more efficient at ordering tests, making it a cost-effective tool. Despite this, the company downplayed job disruption concerns, emphasizing that AI would support rather than replace clinicians. “Doctors build trust and navigate ambiguity in ways AI can’t,” it said.
The research focused on real-world diagnosis rather than memorized answers, which Microsoft said can misrepresent an AI model’s abilities when measured by multiple-choice exams like the U.S. Medical Licensing Examination.
The system uses a “diagnostic orchestrator” – an agent-like AI that selects appropriate tests and diagnoses, emulating multidisciplinary consultation. Microsoft tested it using over 300 NEJM case studies converted into interactive challenges and paired it with models from OpenAI, Meta, Google, Anthropic, and others.
Suleyman told The Guardian he believes near-perfect diagnostic AI could be available within a decade. “It will be a massive weight off the shoulders of health systems around the world,” he said.
Microsoft acknowledged that the technology is not yet ready for clinical deployment and requires further testing on more common symptoms and real-world cases. Still, the company sees potential for AI to enhance care delivery and help patients manage routine health needs independently.
