According to a study published in the journal Science, Microsoft has developed an artificial intelligence tool that diagnoses complex diseases with a higher degree of accuracy than the average physician. In test cases, the tool correctly identified 83.2% of the diseases, compared to a 36.1% success rate for physicians.
The tool, known as MAI-DxO, is designed to function like a medical team. It inquires about a patient’s symptoms, determines which medical tests are required, analyzes the results, and ultimately provides diagnostic recommendations. According to Reuters, the tool outperformed physicians in tests based on emergency room cases. Microsoft believes this technology could assist doctors in addressing complex or challenging cases—particularly when symptoms do not clearly point to a specific condition. How does Microsoft’s AI tool work? While most artificial intelligence tools attempt to provide an answer after reviewing the information provided, Microsoft’s MAI-DxO takes a different approach: it processes medical cases step-by-step, much like a physician would.
This tool employs multiple artificial intelligence agents that work in concert within a system known as a “coordinator.” It begins by determining what medical history information to solicit from the patient. The system then evaluates the patient’s responses, recommends specific medical tests, and uses the results to narrow down the potential causes. For instance, if a patient presents with symptoms that could point to several different conditions, the tool can suggest further testing to help rule out certain possibilities before arriving at a conclusion. Microsoft tested the tool using more than 300 complex clinical cases from The New England Journal of Medicine. The results showed that the tool achieved an 83.2% diagnostic accuracy rate for these cases, compared to a 36.1% accuracy rate for general practitioners who refer patients to other medical facilities.
Researchers also conducted tests on 325 emergency department cases. MAI-DxO achieved an accuracy rate of 80.3%, compared to 46.8% for the physicians. While these results are encouraging, there is a significant distinction between resolving cases based on existing medical records and treating actual patients. Furthermore, the physicians in the study worked independently, rather than as part of a specialized team—a common approach for handling particularly complex cases. AI tools could also help manage the costs of medical testing. Accurate diagnosis is critical, yet medical tests can be very expensive. Microsoft’s research examined both the costs associated with case investigation and the accuracy of AI in identifying diseases. According to TechCrunch, the results varied depending on the AI model employed. When Microsoft integrated its system with OpenAI’s GPT-5.5, diagnostic accuracy for complex cases rose from 72% to nearly 87%. This specific test is separate from the 83.2% accuracy figure reported as the primary study result; the higher figure pertains specifically to the evaluation using GPT-5.5. The degree of improvement varied across cases. In emergency department scenarios, GPT-5.5 achieved an accuracy rate of approximately 85%, regardless of whether it was used in conjunction with Microsoft’s system. This suggests that an additional AI-driven layer of support may be more beneficial for certain types of medical issues than for others. Microsoft also tested AI models from companies such as Google, Anthropic, Meta, and DeepSeek, demonstrating that the underlying model plays a crucial role in the system’s overall performance. While the company reported potential cost savings in certain tests, these are estimates based on simulated cases; it has not yet been proven that the tool can reduce hospital costs in real-world, day-to-day operations. Microsoft plans to test the tool in a hospital setting. The company is collaborating with Beth Israel Deaconess Medical Center in Boston to evaluate its AI technology within an authentic clinical environment. According to Reuters, this planned study will assess whether the technology can assist medical teams in evaluating patients waiting in the emergency department. Patients who consent to participate will continue to receive standard medical care, while researchers will compare the AI’s recommendations against the decisions made by the physicians. This type of diagnostic assessment is critical, as patients may present with a wide range of symptoms, incomplete medical histories, or conditions that are difficult to diagnose. Physicians must also determine which tests are necessary and which treatment plans are the safest. Current research has not yet demonstrated that MAI-DxO can replace physicians or improve patient recovery outcomes. While the tool has performed well in controlled tests, its practical utility in a hospital setting requires further validation. Mustafa Suleyman, CEO of Microsoft AI, stated that the company’s long-term goal is to build AI systems capable of matching or even surpassing human medical experts in complex healthcare tasks. The company is also collaborating with the Mayo Clinic to develop an AI model tailored for the healthcare sector, which Suleyman noted is slated for release in 2027. Microsoft is not the only company exploring the application of artificial intelligence in medical diagnostics; Google and OpenAI are also developing tools designed to assist in clinical decision-making. At present, Microsoft’s research suggests that AI can assist physicians in diagnosing complex conditions and determining appropriate diagnostic tests. However, further research is required before it can be considered a reliable tool for routine patient care.