Details
- Microsoft AI CEO Mustafa Suleyman announced that MAI-DxO was published in the journal Science.
- The system was reported to be 1.7 times more accurate than physicians on emergency-department diagnostic cases.
- MAI-DxO also used fewer medical tests and incurred lower costs while reaching its diagnoses.
- The system coordinates multiple AI agents that question patients, recommend tests, compare hypotheses, and converge on a diagnosis.
- In separate complex cases from the New England Journal of Medicine, MAI-DxO achieved roughly 83% accuracy versus about 36% for generalist physicians using reference materials.
- The comparison involved simulated or benchmarked cases, not autonomous treatment of patients in routine clinical care.
- Specialists and physician teams were not included in the reported human comparison, limiting how directly the results translate to hospital practice.
- Microsoft is evaluating the system with clinical partners, including Beth Israel Deaconess Medical Center, against real-world care workflows.
Impact
The findings strengthen Microsoft’s position in medical AI by pairing diagnostic accuracy claims with lower testing and cost, but they do not establish readiness for unsupervised clinical use. The benchmark advantage may pressure Google, OpenAI, and healthcare-AI developers to demonstrate comparable performance in prospective hospital studies. Regulatory approval, clinician oversight, patient safety, and validation across diverse populations will determine whether MAI-DxO changes care delivery rather than remaining a research benchmark.