Can AI diagnose better than doctors? Microsoft’s new study explained
Microsoft's MAI-DxO system correctly diagnosed 83% of hard medical cases, compared with 36% for doctors, in a study published in Science. Here's how it works, what it really means, and its limits.
Can an AI system work out what’s wrong with a patient better than a doctor? A new study from Microsoft, published in the respected journal Science, suggests that in some tests, the answer may be yes.
Microsoft’s AI system, called MAI-DxO (Microsoft AI Diagnostic Orchestrator), correctly diagnosed about 83% of very difficult medical cases. Experienced general doctors working alone got about 36% right.
Microsoft AI chief Mustafa Suleyman called it a step toward “medical superintelligence“. But before you replace your doctor with a chatbot, it’s important to understand what the study did and didn’t show.
What did the study test?
The researchers used two sets of real, past medical cases:
- 304 complex cases from the New England Journal of Medicine (NEJM). These are famously hard “puzzle” cases that doctors use for training.
- 325 emergency department cases from Beth Israel Deaconess Medical Center in Boston.
The AI and the doctors received information step by step, just like in real life. They could “ask” questions and “order” tests, and each test had an estimated cost.
The results
- Complex NEJM cases: MAI-DxO about 83% correct, doctors about 36%
- Emergency cases: MAI-DxO about 80% correct, doctors about 47%
On the complex cases, the system also kept the estimated cost of tests lower than a basic AI model working alone.
How does MAI-DxO work?
MAI-DxO isn’t one single AI. It’s an “orchestrator”: it makes several AI “agents” work together like a virtual panel of doctors. Different agents suggest possible illnesses, pick the next tests, question each other’s ideas and check costs, before agreeing on a diagnosis.
It works on top of large AI models. Microsoft’s best results came from OpenAI’s GPT-5 and GPT-5.5 models.
The big idea: instead of one AI guessing, a team of AI agents argues the case out, a bit like doctors at a hospital meeting.
The important limits
The headline numbers are impressive, but experts and reporters point out several important limits:
- Not real patients: the tests were simulations using past cases, not live patients in a clinic.
- Doctors worked alone: in real life, hard cases are handled by specialists and teams, who weren’t part of the comparison.
- Rare puzzles, not everyday illness: the study didn’t test common problems, missing information or fast-changing situations.
- The AI model matters most: TechCrunch reported that much of the improvement comes from stronger base models. On emergency cases, the orchestrator added little over GPT-5.5 alone.
- Microsoft did the study: all 15 authors are Microsoft employees, though the paper was peer-reviewed by Science.
Suleyman himself said the system needs clinical testing before it can be used with patients.
What happens next?
Microsoft is partnering with Beth Israel Deaconess Medical Center to test MAI-DxO in a real hospital. Patients will get normal care, and those who agree will also interact with the AI so its suggestions can be compared with doctors’. Microsoft’s health VP, Dominic King, called it potentially the biggest study of its kind.
Microsoft is also working with the Mayo Clinic on a healthcare AI model planned for next year. There’s no release date for MAI-DxO as a product.
What does this mean for you?
- Don’t use a chatbot instead of a doctor. This system isn’t available to the public, and general chatbots can make serious mistakes.
- AI is more likely to become a doctor’s assistant, helping doctors catch rare conditions and avoid unnecessary tests.
- Expect more of this: Google and OpenAI have made similar claims, and Suleyman expects health AI to improve a lot within the next two to three years.
Frequently asked questions
Is AI better than doctors at diagnosis?
In this simulation, Microsoft’s AI beat generalist doctors working alone on difficult cases. But it hasn’t been proven with real patients, and doctors in real hospitals work in teams with specialists.
What is MAI-DxO?
It stands for Microsoft AI Diagnostic Orchestrator. It makes several AI agents work together like a panel of doctors to reach a diagnosis step by step.
Can I use MAI-DxO?
No. It’s a research system. Microsoft is starting hospital trials and hasn’t announced a public product.
Where was the study published?
In the peer-reviewed journal Science, in October 2026.
Will AI replace doctors?
Not soon. The more likely future is AI helping doctors make faster, more accurate decisions, with humans staying responsible for patient care.
