Microsoft has unveiled a new medical tool, fueled by artificial intelligence, that it claims far outperforms human doctors in diagnosing complex ailments.
Urgent: These 3 AI Stocks Could be the Catalyst to Your Early Retirement... Free Picks Here
Mustafa Suleyman, who formed an AI health unit for the tech giant last year, told Financial Times for an interview published Monday that the company is nearing "AI models that are not just a little bit better, but dramatically better, than human performance: faster, cheaper, and four times more accurate."
The Microsoft AI Diagnostic Orchestrator, or "MAI-DxO," he added, will be "truly transformative."
The new system is the first initiative emerging from the AI unit, which Suleyman formed by poaching staff from DeepMind, a research lab he co-founded. Google now owns DeepMind.
Suleyman, the chief executive of Microsoft AI, said that the system's trial is a step on the way to "medical superintelligence," which would both help solve staffing issues and long waiting times for health systems.
The AI model uses an "orchestrator" creating panels of five AI agents that act as "doctors." Each of the agents has a role, such as choosing testing, and they interact to determine how to move forward.
Researchers fed MAI-DxO with 34 studies from the New England Journal of Medicine that describe how doctors have solved some of the most complicated cases, allowing researchers to test if the program could determine a correct diagnosis.
The decision-making process uses a "chain of debate" technique, making AI reasoning models give an accounting of how they solve problems.
"This is a landmark study," Eric Topol, a cardiologist and founder and director of the Scripps Research Translational Institute, commented.
"While this work was not done in the setting of real-world medical practice, it is the first to provide evidence for the efficiency potential of generative AI in medicine — accuracy and cost savings," he said.
The Microsoft model used leading large language models from several key sources, including OpenAI, Meta, Anthropic, Google, xAI, and DeepSeek.
The orchestrator made the language models perform better, but paired with OpenAI's o3 reasoning model, it solved 85.5% of the medical journal's cases, compared to 20% by human doctors.
However, the doctors' success rate could have been improved through the use of textbooks or advice from other colleagues, which they were not permitted.
The technology could be deployed soon through Microsoft's Bing search engine and the company's Copilot AI chatbot, both of which handle 50 million health questions every day.
Special: Buffett Bets $71 Billion in New Trump Era AI Advancement... Free Picks Here
DeepMind, however, has led the way on AI healthcare breakthroughs, with its lab chief, Sir Demis Hassabis, winning a joint chemistry Nobel prize last year after using AI to unlock biological secrets of proteins.
Meanwhile, Suleyman said the OpenAI model performed the best, but Microsoft believes that its aggregate orchestrator is what made the difference when it used the four world-class models.
Dominic King, who headed DeepMind's health unit before joining Microsoft in late 2024, said the MAI-DxO program "performed better than anything we've ever seen before."
"There is an opportunity here today to act almost as a new front door to healthcare," King said.
He stressed that the AI technology is still in its early stages, as it is not yet ready for a clinical environment and has not been peer-reviewed.
Sandy Fitzgerald ✉
Sandy Fitzgerald has more than three decades in journalism and serves as a general assignment writer for Newsmax covering news, media, and politics.
© 2025 Newsmax. All rights reserved.