Microsoft took “A real step towards medical superintelligence,” says Mustafa Suleyman, general director of the artificial intelligence company. The technological giant claims that its up-to-date, up-to-date AI tool can diagnose the disease four times more accurately and at much lower costs than a panel of human doctors.
The experiment checked whether the tool can correctly diagnose a patient with ailment, imitating work usually performed by a human doctor.
Microsoft has used 304 cases from the Up-to-date England Journal of Medicine to develop a test called a sequential diagnosis reference point (SDBENCH). The language model divided each case into a step by step process that the doctor would do to achieve a diagnosis.
Then Microsoft researchers built a system called Mai Diagnostic Orchestrator (Mai-DXO), which asks a few leading AI-W OPENAI GPT, Google’s Gemini, Claude, Claude, Meta Lama and Grok Xai-Grok Xai-Grok Xai-Grok Xai-Grok-Grok GPT. XAI-Grok XAI-XAI-XAI-XAI-
In its experiment, Mai-DXO exceeded human doctors, reaching an accuracy of 80 percent compared to 20 percent of doctors. This also reduced costs by 20 percent, choosing cheaper tests and procedures.
“This mechanism of orchestration-agents who cooperate in this style of the debate chain-it just brings us closer to medical superintelligence,” says Suleyman.
The company bought several Google AI researchers to assist this effort – another sign of the war intensifying for the best specialist knowledge of AI in the technology industry. Suleyman was previously a director at Google working on artificial intelligence.
AI is already widely used in some parts of the American healthcare industry, including helping radiologists in interpreting scans. The latest Multimodal AI models can act as more general diagnostic tools, although the operate of artificial intelligence in healthcare raises its own problems, especially related to the deviation of training data, which are skewed individual demographic data.
Microsoft has not yet decided whether he would try to commercialize technology, but the same director who spoke provided anonymity, said that the company could integrate it with Bing to assist users diagnose ailments. The company can also develop tools that will assist medical experts improve and even automate patient care. “In the next few years you will see that we are doing more and more work, proving these systems in the real world,” says Suleyman.
The project is the latest from the growing research group showing how AI models can diagnose the disease. Over the past few years, both Microsoft and Google have published articles showing that vast language models can accurately diagnose the ailment when it receives access to medical records.
Up-to-date Microsoft studies differ from previous work, because it more accurately reproduces the way human doctors diagnose the disease – analyzing symptoms, testing tests and conducting further analyzes until the diagnosis is achieved. Microsoft describes the way he combined several Frontier AI models as a “path to medical superintelligence” in a blog on the blog about today’s project.
The project also suggests that artificial intelligence can assist reduce health care costs, which is a critical issue, especially in the USA. “Our model works extremely well, both for diagnosis and a very profitable diagnosis,” says Dominic King, vice president of Microsoft, who is involved in the project.