Microsoft's AI Tool Beats Doctors In Diagnosing Medical Conditions, Says New Study
Microsoft has made an AI tool that can find difficult illnesses more accurately than general doctors, according to a study published in the journal Science. In the tests, the AI correctly identified the illness in 83.2% of cases, while doctors got 36.1% right.
Note: The researchers tested the AI using medical cases that already existed. They did not test it on real patients in hospitals.
The tool is called MAI-DxO. It is designed to work like a group of doctors. It asks about a patient's symptoms, chooses which medical tests might help, looks at the results, and then suggests what illness the patient may have.
According to a report by Reuters, the AI also did better than doctors when tested on cases based on emergency room visits. Microsoft believes the tool could help doctors figure out difficult medical problems, especially when a patient's symptoms could be caused by several different illnesses.
How Does Microsoft's AI Tool Work?
Most AI tools try to answer a question using the information they are given. Microsoft's MAI-DxO works differently. It looks at a medical problem step by step, much like a doctor trying to find out what is wrong with a patient.
The tool uses several AI agents, which are AI programs that work together. A system called an orchestrator manages their work. First, the AI decides what questions to ask about the patient's medical history. Then it looks at the answers, suggests medical tests, and studies the results to work out which illnesses are most likely.
For example, a patient might have symptoms that could be caused by several different diseases. The AI can suggest more tests to help rule out some of those diseases before deciding which one is most likely.
Microsoft tested the tool using more than 300 difficult medical cases from the New England Journal of Medicine. The AI correctly identified the illness in 83.2% of cases. General doctors, who were allowed to use medical information to help them, got 36.1% right.
The researchers also tested the tool on 325 emergency room cases. MAI-DxO correctly identified the illness in 80.3% of cases, compared with 46.8% for the doctors.
These results look promising, but there is an important difference between studying an old medical case and treating a real patient. Also, the doctors in the study worked alone. They did not work in teams of specialist doctors, as they might when dealing with especially difficult illnesses.
The AI Tool Can Also Help Control Testing Costs
Finding the right illness is important, but medical tests can cost a lot of money. Microsoft's researchers looked at both how accurately the AI could identify illnesses and how much the tests needed to investigate each case might cost.
TechCrunch reported that the results changed depending on which AI model Microsoft used. When Microsoft used OpenAI's GPT-5.5 with its system, the percentage of difficult cases diagnosed correctly rose from 72% to almost 87%.
This was a different test from the main study, where the AI got 83.2% of cases right. The higher result of nearly 87% came specifically from the test that used GPT-5.5.
But the improvement was not the same for every kind of medical problem. In emergency room cases, GPT-5.5 got about 85% of diagnoses right whether it used Microsoft's system or worked without it. This suggests that having several AI agents work together may help more with some medical problems than with others.
Microsoft also tested AI models made by other companies, including Google, Anthropic, Meta, and DeepSeek. The results showed that the particular AI model used in the system makes a big difference to how well it works.
Microsoft said the AI might save money on medical tests in some situations. However, these possible savings were only estimates from simulated medical cases. The company has not yet shown that the tool can actually lower hospital costs when used in everyday patient care.
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 look at whether the tool can help medical teams check patients who are waiting in emergency rooms.
Patients who agree to join the research will still receive normal medical care. Researchers will then compare what the AI recommends with the decisions doctors make.
These tests are important because real patients can have many different symptoms, missing information in their medical records, or illnesses that are hard to recognize. Doctors also have to decide which tests are really needed and which treatments are safest for each patient.
The current study does not prove that MAI-DxO can take the place of doctors or help patients recover better. It only shows that the AI did well in carefully controlled tests. Researchers still need to find out how useful it will be in real hospitals.
Microsoft AI CEO Mustafa Suleyman has said the company's long-term goal is to build AI systems that can do complex medical tasks as well as, or better than, human medical experts. Microsoft is also working with the Mayo Clinic to create an AI model specially designed for healthcare. Suleyman said the company planned to release that model in 2027.
Microsoft is not the only company working on AI that can help identify illnesses. Google and OpenAI are also developing AI tools to help doctors make medical decisions.
For now, Microsoft's test results suggest that AI could help doctors understand difficult illnesses and choose which medical tests to use. But researchers need to do more studies before the tool can be trusted for regular patient care.
