“4 times more precise than doctors”: Microsoft unveils an AI that calls into question the very future of medical diagnosis
Mai – DXO: Microsoft AI which surpasses doctors in diagnostic precision – and upsets the balance of modern medicine. © Shutershock
“It is a real step towards medical superintendent.” With these words, Mustafa Suleyman, boss of Microsoft AI, may sign the turn of an era where the most complex medical diagnoses would no longer be reserved for the best specialists … but for artificial intelligence. And it’s not science fiction.
Advertisement
May-DXO: Microsoft AI which diagnoses better than experienced doctors
May-Dxo, for Microsoft AI Diagnostic Orchestratorattacks that medicine has more delicate: clinical reasoning. It is not simply a question of associating symptoms with a disease, as would a search engine. The system reproduces the sequential process of a diagnosis, by analyzing real clinical cases step by step.
We show that the Microsoft AI Diagnostic Orchestrator (Mai – DXO) diagnoses properly up to 85 % of the cases of the NEJM, a rate more than four times higher than that of a group of experienced doctors.
To test its capacities, Microsoft used 304 cases extracted from the New England Journal of Medicine, a world reference. Each case has been broken down in several times – interrogation, hypotheses, exams, final diagnosis – via a tool called sequential diagnosis benchmark (sd bench). The idea? Compare several AI (GPT-4, Gemini, Claude 3, Grok, Llama, Deepseek, etc.) to a panel of experienced doctors.
The result is clear: May-Dxo, combined with Openai O3, properly resolved 85.5 % of cases, against barely 20 % for doctors, on average. And without multiplying unnecessary examinations: the system itself chooses the most relevant and less expensive tests, reducing the overall invoice by 20 %.
The model achieves maximum performance at 85.5 % of diagnostic success, with increased cost effectiveness.
Microsoft’s official graphic showing that Mai – Dxo outpasses other AI and human doctors in diagnostic precision, while maintaining reduced costs. © Microsoft
Its principle is based on a multi-agent orchestration. Each AI plays a different role in the reasoning chain, like discussion specialists. Some models emit hypotheses, others require exams, a last slice. Mustafa Suleyman, Ex-Deepmind and now boss of Microsoft AI, describes this as a “chaining of debates between agents”-a mechanism which, according to him, could bring us closer to a medical superintendent.
This orchestration mechanism – several agents who debate in chain – is what brings us closer to medical superintendent.
That said, Mai-dxo is not yet ready for the hospital. The test remains theoretical, without real interaction with patients. Microsoft is considering progressive uses: integration into Bing for the general public, or as an assistant for doctors via Copilot. But above all deployment, rigorous clinical validation will be necessary.
Because in fact, medicine, it is not only to resolve school cases. It is to manage uncertainty, emotions, blurred cases. This is where AI will have to prove that it also knows how to show nuance.
Advertisement
Want to save even more? Discover Our promo codes Selected for you.
