Microsoft develops AI tool that can diagnose ’sequentially’ like a doctor
Microsoft says its MAI-DxO system outperformed physicians on simulated diagnostic cases; prospective clinical evaluation remains ahead.
TL;DR
- Microsoft reported 83% diagnostic accuracy on complex New England Journal of Medicine cases, compared with 36% for physicians with reference materials.
- On emergency-department cases, Microsoft reported 80% for MAI-DxO versus 47% for physicians.
- The results are simulated evaluations; a prospective hospital study is planned, according to the company.
Microsoft said its MAI-DxO system reached 83% accuracy on complex cases drawn from the New England Journal of Medicine. It compared that result with 36% for physicians using reference materials. [1] [2]
The system coordinates several AI agents through questions, examinations and tests instead of giving one immediate answer. Microsoft also reported 80% accuracy on emergency-department cases from MIMIC-IV-ED, versus 47% for physicians. [1] [2]
Reuters described the system as reasoning through cases sequentially, while TechCrunch reported that Microsoft presented it as part of its pursuit of “medical superintelligence.” The company said it plans a prospective evaluation with Beth Israel Deaconess Medical Center. [2] [3]
The figures come from simulated benchmark cases and do not show improved patient outcomes in routine care. Hospital performance, safety and clinical benefit remain untested in the reported prospective study. [1] [2] [3]
Why it matters
The result puts multi-agent coordination and sequential reasoning at the center of a prominent healthcare AI claim. The gap between benchmark performance and evidence from real clinical settings remains decisive.
Editor's note
Benchmark figures are company-reported; Reuters and TechCrunch provide secondary coverage. Simulation results do not establish patient benefit.