Microsoft's AI Tool Beats Doctors In Diagnosing Medical Conditions, Says New Study

Microsoft has developed an AI tool that can diagnose complex medical conditions more accurately than general physicians, according to a study published in the journal Science. The tool correctly identified diseases in 83.2% of the cases tested, while doctors got 36.1% of them right.

Note: The study used existing medical cases rather than testing the AI on patients in hospitals.

Called MAI-DxO, the tool is designed to work like a team of doctors. It asks questions about a patient's symptoms, decides which medical tests are needed and studies the results before suggesting a diagnosis.

According to a Reuters report, the tool also performed better than doctors when tested on cases based on emergency department visits. Microsoft believes the technology could help doctors solve difficult medical cases, especially when the symptoms do not clearly point to one illness.

How Does Microsoft's AI Tool Work?

Most AI tools try to give an answer after looking at the information provided to them. Microsoft's MAI-DxO takes a different approach. It works through a medical case one step at a time, much like a doctor would.

The tool uses several AI agents that work together under a system called an orchestrator. It starts by deciding what to ask about the patient's medical history. The system then considers the answers, suggests tests and uses the results to narrow down possible causes.

For example, if a patient has symptoms that could point to several different illnesses, the tool can suggest further tests to help rule out some possibilities before reaching a conclusion.

Microsoft tested the tool on more than 300 difficult medical cases from the New England Journal of Medicine. It correctly diagnosed 83.2% of these cases, while general physicians who could refer to medical resources achieved 36.1% accuracy.

The researchers also tested it on 325 emergency department cases. MAI-DxO achieved 80.3% accuracy, compared with 46.8% for the doctors.

These results are promising, but there is an important difference between solving a case from an existing medical record and treating a real patient. The doctors in the study also worked individually, rather than as teams of specialists who might handle particularly difficult cases.

The AI Tool Can Also Help Control Testing Costs

Getting the diagnosis right is important, but medical tests can also be expensive. Microsoft's research examined how much it might cost to investigate a case, along with how accurately the AI could identify the illness.

TechCrunch reported that the results depended on which AI model was used. When Microsoft paired its system with OpenAI's GPT-5.5, diagnostic accuracy on complex cases increased from 72% to nearly 87%.

This is a separate test from the main study result of 83.2%. The higher figure applies specifically to the evaluation using GPT-5.5.

The improvement was not the same for every type of case. In emergency department scenarios, GPT-5.5 achieved around 85% accuracy both with and without Microsoft's system. This suggests that the extra layer of AI coordination may be more helpful for some medical problems than others.

Microsoft also tested AI models from companies including Google, Anthropic, Meta and DeepSeek. The results showed that the model powering the system plays an important role in how well it performs.

The company reported potential cost savings in some tests, but these were estimates based on simulated cases. It has not yet proved that the tool will reduce hospital costs in everyday use.

Microsoft Plans To Test The Tool In Hospitals

Microsoft is working with Beth Israel Deaconess Medical Center in Boston to test the AI in real healthcare settings. According to Reuters, the planned research will examine whether it can help medical teams assess patients waiting in emergency departments.

Patients who agree to take part will continue to receive standard medical care. Researchers will compare the AI's recommendations with the decisions made by doctors.

This testing matters because real patients may have several symptoms, incomplete medical histories or conditions that are difficult to identify. Doctors must also decide which tests are necessary and what treatment is safest.

The current study does not prove that MAI-DxO can replace doctors or improve patient recovery. It shows that the tool performed well in controlled tests, but its usefulness in hospitals still needs to be established.

Microsoft AI CEO Mustafa Suleyman has described the company's long-term goal as building AI systems that can match or surpass human medical expertise in complex healthcare tasks. The company is also working with the Mayo Clinic on a healthcare-focused AI model, which Suleyman said was planned for release in 2027.

Microsoft is not alone in exploring AI for medical diagnosis. Google and OpenAI are also developing tools designed to support clinical decisions.

For now, Microsoft's results suggest that AI could help doctors investigate complicated illnesses and decide which tests to use. However, more research is needed before it can be considered a reliable tool for routine patient care.