Microsoft’s New AI Tool Diagnoses Diseases Better Than Human Doctors, Claims Tech Giant

Microsoft has introduced a powerful new AI tool that can diagnose diseases with four times more accuracy than human doctors. Mustafa Suleyman, head of Microsoft's AI division, says this new system could also make medical testing cheaper.

How It Works

To test the AI, Microsoft used 304 real medical cases from the New England Journal of Medicine. These were turned into step-by-step problems, just like how a doctor would approach a diagnosis. The tool, called MAI Diagnostic Orchestrator (MAI-DxO), asks several top AI models to work together-like a team of expert doctors-to figure out the right diagnosis.

The Results

AI accuracy: 80%

Doctor accuracy: 20%

Cost reduction: 20% (choosing more affordable tests)

Suleyman explained, "These AI agents debate and discuss like a team, which brings us closer to what we call medical superintelligence."

Real-World Impact

Microsoft believes this tool could cut medical costs, especially in the U.S., where healthcare is very expensive. Dominic King, a Microsoft VP, said the tool is not only good at finding the right diagnosis but also does it cheaply.

However, the team is still working on improving the system. One concern is that the training data might not represent all types of people equally, which could lead to biased results.

What's Next?

Microsoft hasn't decided how to release this tool yet. It might be added to Bing to help users check symptoms or made into a tool for doctors. "We'll keep testing it in real-world situations," said Suleyman.

AI is already being used in some hospitals, like helping radiologists read scans. But using it for general diagnosis (like what your family doctor does) brings challenges, especially around fairness and trust.

Experts React

David Sontag from MIT called the study strong because the AI worked like a real doctor. But he also said the human doctors in the study didn't have access to tools they normally use, so the comparison may not be fully fair.

Eric Topol from Scripps Research praised the AI's ability to handle tough cases but said it should be tested in clinical trials before being used widely.