The Microsoft AI diagnostic tool is one of the most talked-about projects in digital health. Microsoft says the system can solve hard medical cases step by step, much like a real doctor. In its tests, it beat a group of experienced physicians by a wide margin. This article explains how it works, what the results mean, and why it is not ready for hospitals yet.
What Is the Microsoft AI Diagnostic Tool?
The tool is called the Microsoft AI Diagnostic Orchestrator, or MAI-DxO. Microsoft AI, the company’s consumer AI division led by Mustafa Suleyman, presented the research on 30 June 2025. The company described it as a step toward what it calls medical superintelligence.
MAI-DxO does not hold medical knowledge of its own. Instead, it coordinates existing AI models. It acts like a virtual panel of doctors who work together on one case.
How Does It Diagnose Like a Doctor?
Most medical AI tools are tested with multiple-choice exams. Real patients do not come with answer options. Therefore, Microsoft built a different test called the Sequential Diagnosis Benchmark, or SDBench.
The test turns 304 complex case records from the New England Journal of Medicine into interactive challenges. The process works like this:
- The AI or doctor reads a short summary of the patient.
- It asks follow-up questions or orders tests, one at a time.
- Each question and test adds a cost.
- Finally, it gives a diagnosis, and the answer is checked against the published case.
Before giving the final answer, MAI-DxO also checks whether its diagnosis explains all the findings. This extra step is meant to improve safety.
Microsoft AI Diagnostic Tool Results
The results look striking. MAI-DxO solved up to 85 percent of the cases. The best setup, which paired it with OpenAI’s o3 model, reached 85.5 percent. In contrast, 21 experienced physicians averaged about 20 percent.
| Group | Accuracy on 304 NEJM Cases |
|---|---|
| MAI-DxO with OpenAI o3 | 85.5% |
| Experienced physicians (average) | About 20% |
In addition, Microsoft says MAI-DxO reached correct answers with lower testing costs than the physicians or any single AI model it tested.
Why the Results Need Careful Reading
These numbers do not mean AI is four times better than your doctor. Several limits matter:
- Very hard cases: The test used rare and complex puzzles, not everyday illnesses.
- Unassisted doctors: The physicians worked as generalists. In real life, they would ask colleagues and specialists for help.
- No live patients: Microsoft has said the system has not been used on live patient data outside controlled tests.
- Research stage: Microsoft calls it a research demonstration, not a finished clinical product.
Experts have said the next step is a proper clinical trial. Until then, the tool should not replace medical advice.
What Has Changed in 2026?
Microsoft has kept building in health. In March 2026, it launched Copilot Health, a section inside Copilot that helps people understand their own health information. It brings together health records, wearable data, and medical history in one place.
The service started as a preview for Microsoft 365 Personal, Family, and Premium subscribers in the United States. Microsoft says it runs in a separate environment with extra privacy controls. It also says personal health data is not used to train AI models.
Microsoft has also said that new publications will show how MAI-DxO may work across a wider range of clinical conditions. Meanwhile, the company reports more than 50 million health-related sessions every day across Bing and Copilot.
Could It Help Doctors and Patients?
If future trials succeed, the tool could support doctors with second opinions. It could also reduce unnecessary tests and costs. Because it works through a dialogue, a doctor can follow its reasoning step by step.
However, this is not a substitute for a doctor visit. If you have symptoms, see a qualified healthcare professional.
Frequently Asked Questions
What is MAI-DxO?
It is Microsoft’s AI Diagnostic Orchestrator. It coordinates AI models to investigate a medical case step by step.
How accurate is the Microsoft AI diagnostic tool?
In Microsoft’s tests, it solved up to 85 percent of 304 complex cases. The physicians averaged about 20 percent.
Can I use MAI-DxO today?
No. It is a research demonstration and is not available for clinical use.
What is Copilot Health?
It is a Copilot feature that helps users understand their own health data. It is in preview for certain US subscribers.
Conclusion
The Microsoft AI diagnostic tool shows how AI may one day support medical reasoning, not just answer exam questions. The results are impressive, yet they come from a controlled test of very hard cases. Real-world trials will decide whether the tool can help patients safely. For more tech and health updates, keep following Pakistan Post. You can also read Microsoft’s research on the Microsoft AI website.
