A study in Science details Microsoft's MAI-DxO system, which coordinates multiple AI agents to simulate a medical panel for complex diagnoses.
Tested across models from several providers, it achieved high accuracy on medical cases by orchestrating multi-agent reasoning workflows.
How the system works
The system, called MAI-DxO, coordinates multiple AI programs that work together like a team of doctors. Specifically, they can ask patients questions, order tests, and talk to each other to reach a diagnosis.
In a simulation using difficult cases from a well-known medical journal, the system gave the correct answer 83% of the time. In contrast, general practitioners and internists working individually with the help of literature had a success rate of 36% on the same test. However, cases of such difficulty are usually handled by specialist doctors or teams, who were not involved in the comparison.
The trials and next steps
Dominic King, vice president of health at Microsoft AI, said the biggest challenge in providing medical care is the growing gap between the demand and supply of medical expertise.
READ ALSO
Former OpenAI executive founded a company with robots for drug discovery
Microsoft tested the system using technologies from a variety of companies, including OpenAI, Google, Anthropic, Meta, and DeepSeek, with the best results coming from OpenAI's GPT-5 and GPT-5.5 models. Using GPT-5.5, the system increased its accuracy from 72% to nearly 87%, at a slightly lower cost.
The company is working with the Mayo Clinic to develop a basic health model that is planned to be released next year, while also starting a collaboration with a Boston hospital to evaluate the system's clinical scaling capabilities in different hospital settings. Google is also active in this area, having recently published a study on its own emergency chat system.
G
Hacks.gr on Google
Set it as a preferred source for cybersecurity updates.
Preferred source
Top comments (0)