Microsoft’s medical AI system makes diagnoses 4 times more accurately than doctors

Microsoft has introduced the MAI Diagnostic Orchestrator (MAI-DxO) artificial intelligence system, which, according to the company, is capable of making medical diagnoses four times more accurately than real doctors. At the same time, its work costs less.

The system was tested on 304 complex clinical cases published in the authoritative New England Journal of Medicine. MAI-DxO did not just give a result, but went through the diagnostic process step by step – just like a doctor does. Especially for such tests, Microsoft has developed a new test protocol called Sequential Diagnosis Benchmark.

The peculiarity of MAI-DxO is that it works as a “collective mind” – it combines the answers and analytical capabilities of several advanced AI models at once. The “virtual consilium” includes systems from OpenAI, Google, Anthropic, Meta and even Grok from Elon Musk. Together, they analyze symptoms, suggest hypotheses, refine data, and ultimately issue a final diagnosis.

The results were impressive. The AI ​​achieved an accuracy of 80 percent, while the average among doctors under similar conditions was only 20. In addition, MAI-DxO turned out to be more economical: it chooses more affordable but more effective diagnostic methods and reduces overall costs by about 20 percent.

The head of Microsoft’s AI direction, Mustafa Suleiman, who previously worked at Google, called this development a step towards creating a super-intelligent medical system. According to him, such solutions can radically change the approach to diagnosis, and in the future – to treatment.

So far, Microsoft has not disclosed how exactly it intends to use the technology. Among the possible options are its implementation in the Bing search engine as a self-diagnosis tool or the creation of assistive solutions for doctors. It is already known that MAI-DxO will soon begin testing in real clinics.

However, the technology is controversial. First, doctors were forbidden to use any auxiliary tools during the test, which casts doubt on the correctness of the comparison. Second, AI is not yet able to take into account the emotional state of the patient or the characteristics of a particular medical institution – for example, the availability of equipment. Despite this, most experts agree: if the system shows comparable effectiveness in real practice, it will be a real breakthrough – albeit not the most joyful for doctors.