Microsoft introduced an AI doctor that diagnoses 4 times more accurately than regular doctors

Microsoft has developed an artificial intelligence system for medical diagnosis called MAI-DxO, Microsoft AI Diagnostic Orchestrator. It handles complex clinical cases four times more accurately than real doctors. The project was developed by the Microsoft AI division under the leadership of Mustafa Suleyman, co-founder of DeepMind.

MAI-DxO uses a "data orchestrator" architecture: decision-making involves five AI agents with different specializations, simulating the work of a medical board. The system analyzes medical cases, discusses treatment options, and reaches a collective conclusion. This technology was tested on 304 of the most difficult cases from practice, described in the New England Journal of Medicine. For the experiment, the chain-of-debate methodology was used, allowing one to trace the logical steps in the diagnostic process.

The AI agents were based on large language models from OpenAI, Meta, Anthropic, Google, xAI, and DeepSeek. The best result was shown by the o3 model from OpenAI, solving 85.5% of tasks. For comparison: doctors' accuracy was about 20%, however, during the experiment they did not have access to reference materials and could not consult with colleagues — under real conditions, this indicator could be higher. The system also takes into account the economics of diagnosis, eliminating unnecessary tests and reducing potential costs.

Microsoft plans to integrate MAI-DxO into Copilot and Bing — services that process tens of millions of medical queries per day. Next-generation AI systems should not just catch up with humans, but significantly surpass them in accuracy, speed, and cost.

MAI-DxO could become a "new entry point into healthcare." However, the technology is not yet ready for clinical application and will be refined based on real-world practice.

Source: Financial Times