Microsoft has unveiled an experimental system called the Microsoft AI Diagnostic Orchestrator (MAI-DxO) that it says can diagnose complex medical cases with far greater accuracy and lower cost than experienced physicians. In internal tests released this week, the tool correctly identified ailments in up to 85.5% of 304 challenging patient cases drawn from the New England Journal of Medicine, versus 20% for a panel of 21 U.S. and U.K. doctors.
The study used a new Sequential Diagnosis Benchmark that simulates the step-by-step process clinicians follow—asking questions, ordering tests and refining hypotheses. MAI-DxO coordinates a virtual “panel” of large language models from OpenAI, Google, Anthropic, Meta, xAI and others, assigning roles such as hypothesis generation, cost control and error checking. Microsoft says the orchestrated approach not only improved accuracy but trimmed average test-ordering expenses by roughly one-fifth compared with human physicians and standalone AI models.
Mustafa Suleyman, who heads Microsoft’s AI unit, described the results as a “genuine step toward medical super-intelligence,” while stressing that the system is intended to support, not replace, clinicians. The research has been submitted for peer review, and the company said additional clinical trials and regulatory approvals would be needed before any real-world deployment.
Microsoft is exploring whether the technology could be integrated into products such as Bing or provided directly to healthcare organisations. The company said further validation on more common cases and broader patient populations will determine the pace at which the tool moves from laboratory benchmark to bedside use.
Can #AI diagnose medical problems? How does AI help physicians? How is AI impacting the practice of medicine?
Jason Wiesner, MD (@xrayviznMD) chair of the imaging service line at @SutterHealth, explores these questions and more. 📺
Microsoft is developing an artificial intelligence system that brought good results in initial tests in carrying out complex diagnoses.
The system was supplied with about 300 complex cases presented in a scientific journal.
#JornaldaCultura #JC
