Microsoft has developed an artificial intelligence tool that solved complex medical cases more than twice as accurately as generalist doctors, according to a peer-reviewed study published in Science. The system, called the MAI Diagnostic Orchestrator, or MAI-DxO, works through a case the way a clinician might, deciding which questions to ask and which tests to order before settling on a diagnosis.
Many medical AI systems assess symptoms, lab results and scans all at once. MAI-DxO takes a different approach. It simulates a panel of clinical experts, gathers information, reviews test results and revises its working diagnosis as new evidence comes in, Reuters reported.
In the study, MAI-DxO, running on OpenAI’s GPT-5, reached the correct diagnosis in 83.2% of 304 challenging cases published in the New England Journal of Medicine. Generalist doctors, who had access to textbooks and the internet, scored 36.1%. The system also handled 325 emergency department admissions better than the human group, scoring 80.3% against 46.8%.
Researchers said the diagnostic work-ups cost less, too. Those figures are estimates based on the cases, not actual medical bills.
Dominic King, a corporate vice president in Microsoft AI’s health unit, told Reuters that medical AI has largely been judged on multiple-choice tests. “AI has become very good at getting a 100% pass rate on those,” King said. “No patient ever came to see me with a nice summary.”
Harsha Nori, head of AI at Microsoft AI Health, said the team also measured how well the system chose its tests, not just whether it reached the right answer. “We’re not just measuring the accuracy of the systems,” Nori told Reuters.
The results come with clear limits. The doctors in the comparison were generalists rather than specialists, and the cases were presented as text. Neither the AI nor the doctors had access to physical examinations or medical imaging. Complex cases like these are usually handled by specialists or physician teams, who were not part of the comparison.
Microsoft also tested the orchestration approach with several large language models. According to the study, much of the gain comes from the strength of the underlying model, with the orchestration layer adding improvements on top. Testing included models from OpenAI, Google, Anthropic, Meta and DeepSeek, and the best results came with OpenAI’s GPT-5 and the newer GPT-5.5. On complex cases, the system raised GPT-5.5’s accuracy from 72% to nearly 87%, at a slightly lower cost. On emergency department cases, the tool made little difference for either model.
Mustafa Suleyman, Microsoft AI’s chief executive and a co-founder of DeepMind, called the work a step toward what he described as medical superintelligence. Suleyman, who joined Microsoft in 2024, said the company is working with the Mayo Clinic to train a foundation model for healthcare. He said he expects healthcare AI to improve quickly, predicting it will get “an order of magnitude better in capability within the next 24 to 36 months.”
Microsoft is not the only company making such claims. Google and OpenAI have also said their AI systems can match or outperform doctors in certain tasks.
The system is not yet in clinical use. Microsoft plans to work with Beth Israel Deaconess Medical Center in Boston to test whether MAI-DxO can help triage patients in emergency department waiting rooms. Under the planned study, patients would receive standard care, including triage, specialist time, treatment and decisions about whether it is safe to discharge them. Patients who consent would also engage with MAI-DxO, and researchers would compare its recommendations with those of the human doctors.
“This technology is ready for real-world evaluations,” King said, “and you can imagine the complexity and diversity of issues that come from a Boston emergency department on a Friday evening.”
Microsoft already sells dictation and radiology tools to clinicians, and it views MAI-DxO as an extension of that work. Suleyman said the company wants to build the strongest clinical decision support products on the market in the coming years.
Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.