Microsoft reports its AI diagnoses conditions far better than doctors
In MAI-DxO tests, the system achieved 83% accuracy, but it has not yet been tested in real clinical settings
The tool functions as a virtual medical board: multiple AI agents analyze symptoms and medical history, select necessary examinations, and formulate a diagnosis. For testing, Microsoft used its own benchmark, SDBench, compiled from medical reports and emergency department data. The company plans to make it publicly accessible.
In scenarios based on hundreds of publications from The New England Journal of Medicine, the accuracy of general practitioners was 36% when using reference materials and 22% without them. In materials from the Beth Israel Deaconess Medical Center, the system achieved 80% accuracy, while the average accuracy of doctors reached 47%. Additionally, its diagnoses were typically associated with lower estimated costs.
Microsoft plans to test the technology alongside Beth Israel Deaconess in the coming months. Mustafa Suleyman, head of Microsoft’s AI division, believes that doctors will retain a key role, while AI can take on some routine tasks and free up more time for specialists to spend with patients.
