Microsoft AI outscores generalists in simulated diagnosis

Microsoft’s Diagnostic Orchestrator, known as DxO, scored higher than generalist physicians on simulated medical cases in a peer-reviewed study published in Science, including complex cases and emergency department scenarios. The results measure diagnostic accuracy in controlled tests, not performance in patient care. On 304 complex cases drawn from New England Journal of Medicine clinicopathological conferences, DxO identified the correct diagnosis 83.2% of the time. Generalist physicians averaged 36.1% when they could use reference materials and 22.3% without them, according to the study. The highest-scoring individual physician on those cases reached 51.8%, the researchers reported. For a separate set of 325 emergency department-to-admission encounters, DxO scored 80.3%, compared with 46.8% for physicians. The study’s authors said the system’s strongest results came when it ran on OpenAI’s GPT-5 model. Unlike systems that receive a complete case description at once, DxO works through a case step by step. It begins with a brief presentation, then chooses questions and tests, updates possible diagnoses as information arrives and decides when it has enough evidence to reach a...