AI Diagnosis: Microsoft claims superiority over human physicians

For patients facing difficult medical diagnoses, the landscape of diagnostic support may shift soon. Microsoft has released a new AI diagnostic tool that it says outperforms human generalists in handling complex cases. The system, designed to act as a collaborative panel of virtual specialists, aims to close the growing gap between medical demand and available expertise.

The results, published in the journal Science, show significant numerical differences. In simulations using complex cases from the New England Journal of Medicine, the system named MAI-DxO achieved an 83 percent accuracy rate. By comparison, primary care doctors and general internists working individually with reference materials were correct only 36 percent of the time. It is important to note that these cases are typically handled by specialists or teams, who were not included in this specific comparison.

Behind the technology is a coordinated effort by several AI agents that mimic a medical board. They interview the patient, order tests, and debate potential diagnoses. Mustafa Suleyman, the CEO of Microsoft AI, described this achievement as a step toward medical superintelligence. He is not alone in this ambition, as Google and OpenAI have also made similar claims regarding the capabilities of their respective AI models in recent times.

The data suggests that much of the performance boost comes from the underlying power of the AI models themselves rather than just the orchestration. Microsoft tested the system with models from various companies, including OpenAI, Google, Anthropic, Meta, and DeepSeek. The highest accuracy was achieved using OpenAI’s GPT-5 and GPT-5.5 models. However, the added complexity of the DxO system did not always improve cost or accuracy, particularly in emergency department scenarios.

For less powerful models, the improvement was more pronounced. A model from Google scored 46.4 percent on its own but rose to nearly 69 percent with the DxO assistance. Despite these gains, the gap between weaker and stronger base models remained larger than the specific boost provided by the orchestration layer. Microsoft researchers note that accuracy improvements naturally diminish as models approach their theoretical maximum performance.

Looking ahead, Suleyman stated that Microsoft is collaborating with the Mayo Clinic to develop a healthcare foundation model for release next year. He predicts a major breakthrough in medical record coding by 2027, which would allow for highly accurate predictions of patient outcomes. He believes that within 24 to 36 months, healthcare models will see an order of magnitude improvement in capability.

To validate these claims, Microsoft will partner with Beth Israel Deaconess Medical Center in Boston. This study will compare DxO recommendations against human doctors in real-world settings, including emergency rooms and urgent care. The goal is to provide compelling evidence that these systems can be deployed at scale, marking a significant move beyond simple benchmarking tasks.

Microsoft views this as a differentiator against other AI labs. With existing products in dictation and radiology, the company aims to provide the top clinical decision support tools in the coming years. Suleyman expressed strong confidence in the rapid progress of AI in healthcare, comparing the anticipated trajectory to previous advances in coding and chatbot technology.