Microsoft's New AI System Outperforms Human Doctors in Diagnosing Patients

Microsoft has taken a step towards medical superintelligence, according to Mustafa Suleyman, CEO of the company's AI arm. The new AI tool can diagnose disease more accurately and at a lower cost than human physicians.

The experiment tested the tool's ability to diagnose patients, simulating the work of a human doctor. The Microsoft team used 304 case studies from the New England Journal of Medicine to create a test called the Sequential Diagnosis Benchmark (SDBench). A language model analyzed each case step-by-step to mimic a doctor's diagnostic process.

Microsoft's researchers developed the MAI Diagnostic Orchestrator (MAI-DxO) system, which queries multiple AI models to mimic human experts working together. In the experiment, MAI-DxO outperformed human doctors with 80% accuracy and reduced costs by 20%.

According to Suleyman, this orchestration mechanism is a step towards medical superintelligence. The company recruited Google AI researchers for this project, reflecting the competition for AI expertise in the tech industry.

AI is already used in US healthcare, including aiding radiologists. Microsoft's technology could be integrated into Bing for user diagnosis or to assist medical professionals in patient care.

The project showcases how AI models can diagnose diseases accurately by replicating human diagnostic processes. Microsoft sees this as a path to medical superintelligence.

The research also highlights the potential of AI in reducing healthcare costs. Dominic King, a Microsoft vice president involved in the project, emphasizes the model's accuracy and cost-effectiveness in diagnosis.

According to the source: WIRED.