Microsoft's AI Panel Beats Doctors in Medical Diagnosis Tests
A new AI system from Microsoft is showing some impressive results when it comes to diagnosing tough medical cases. The system, called MAI-DxO, was tested against real-world cases that doctors often struggle with. In a study published in Science, the AI got the right answer 83% of the time. That's more than double what human doctors managed when they worked alone with their reference materials.
What makes this AI different is how it works. Instead of one model doing all the thinking, MAI-DxO uses multiple AI agents that act like a team of doctors. They talk to a virtual patient, order tests, and discuss what might be wrong. It's like having a whole panel of experts in one room, debating the best diagnosis.
The cases used were pretty challenging ones, taken from the New England Journal of Medicine. These aren't your everyday sick visits. They're the kind of complex problems that usually require specialists or teams of doctors. Still, the AI handled them better than generalist doctors working on their own.
Microsoft isn't the only big tech company claiming AI wins in healthcare. Google and OpenAI have made similar announcements. But the study shows that much of the improvement comes from using more powerful AI models. The MAI-DxO system adds extra benefits on top of that.
When it comes to cost, the AI is also making progress. Using OpenAI's GPT-5, the system cut down the cost per case by about $3,000 while improving accuracy slightly. With the newer GPT-5.5 model, accuracy jumped from 72% to nearly 87% on complex cases. But on emergency room cases, the AI didn't show much improvement either way.
The system helps most when dealing with harder cases and when paired with less advanced models. For example, Google's Gemini 2.5 Flash went from getting 46% right to nearly 69% with MAI-DxO's help. Even so, the gap between the best and worst models was still bigger than what the AI system added.
Microsoft's team says that as AI models get better, the room for improvement shrinks naturally. They're working with Mayo Clinic on a new healthcare-focused AI model set to launch next year. The goal is to make real progress in how medical records are handled and how doctors can predict patient outcomes.
The people behind this project have deep roots in AI and healthcare. One of them previously co-founded DeepMind, which Google bought back in 2014. They're now focused on building what they call medical superintelligence, hoping to close the gap between the demand for medical expertise and the supply of doctors.