Microsoft study says its AI diagnosed complex cases better than general doctors

In short: Microsoft says its MAI-DxO AI system diagnosed difficult simulated medical cases more accurately than general doctors in a published study.

What happened

Microsoft researchers reported results for an AI diagnostic system called MAI-DxO in the journal Science. The system was tested on complex cases drawn from the New England Journal of Medicine, using a simulation rather than real patients.

MAI-DxO is built to act like a small team. It uses multiple AI “agents” (separate AI helpers that each take a role, like a panel of doctors) to ask questions, suggest tests, and debate the most likely diagnosis.

In the simulation, MAI-DxO reached the correct diagnosis 83% of the time. In the same setup, primary care doctors and general internists working individually, with reference materials, were correct 36% of the time. The study notes that cases this complex are usually handled by specialists or a physician team, and those were not included in the comparison.

The study also suggests that much of the performance comes from the underlying AI model, not only from Microsoft’s coordinating layer. Microsoft tested models from several companies, and the best results came when MAI-DxO used OpenAI’s GPT-5 and GPT-5.5.

Why it matters

Tools like this could someday help doctors handle heavy workloads, especially when a case is confusing or time is short. But these results come from controlled simulations, and the system can be expensive to run per case, so real-world testing and careful oversight will matter before it is used widely.

Source: TechCrunch AI