Microsoft AI Beats Doctors in Complex Diagnosis Tests, Study Finds

Microsoft’s artificial intelligence technology has delivered a striking result in medical research, outperforming general physicians on a series of challenging diagnostic tests. The system, known as MAI-DxO, achieved substantially higher diagnostic accuracy in simulated cases, raising new questions about how AI could support doctors in the future.

A study published in the scientific journal Science on October 22, 2026, examined whether an AI system could approach medical diagnosis in a more realistic way—by gathering information, selecting tests and revising its conclusions step by step rather than simply producing an answer from a list of symptoms.

The findings are promising, but they come with an important qualification: the research evaluated historical and simulated cases, not routine treatment of patients in a live hospital setting.

Microsoft AI Achieves 83% Accuracy in Complex Cases

According to Microsoft's published research, MAI-DxO achieved approximately 83% diagnostic accuracy on 304 challenging cases drawn from clinical case discussions published by the New England Journal of Medicine.

In the same evaluation, general physicians using textbooks and online reference resources achieved approximately 36% accuracy. Physicians working without those external resources scored approximately 22%.

The researchers also tested the system on 325 emergency department cases from the MIMIC-IV-ED dataset. MAI-DxO achieved around 80% accuracy, compared with approximately 47% for the physicians in that evaluation.

These figures refer to the specific study benchmarks. They should not be interpreted as proof that AI is more effective than every doctor or specialist in everyday clinical practice.

How Does Microsoft's MAI-DxO System Work?

Traditional medical AI systems often analyse the available symptoms, laboratory results and other information together before suggesting a diagnosis. MAI-DxO takes a more structured approach.

The system coordinates several AI agents that simulate different clinical perspectives. It works through a case in stages, using new information to refine its assessment and determine what should happen next.

• Reviews the patient's information: It identifies relevant details from the clinical presentation.
• Chooses questions: It determines which additional questions could help clarify the case.
• Selects diagnostic tests: It considers which investigations may provide useful evidence.
• Reassesses possible diagnoses: It updates its hypotheses as results become available.
• Considers costs: It evaluates the potential value of investigations rather than focusing only on the final answer.

This step-by-step method is designed to resemble the reasoning process used in clinical diagnosis. The goal is not just to name a disease, but to develop a more organised diagnostic plan.

AI Could Help Reduce Diagnostic Testing Costs

Accuracy was not the only outcome measured in the research. Microsoft also assessed the estimated costs associated with diagnostic work-ups—the questions, tests and procedures used to investigate a medical problem.

The researchers reported that some configurations of MAI-DxO reached high diagnostic accuracy while using less costly work-ups. In one cost-optimised configuration, the system achieved about 81% accuracy on the complex cases at less than half the estimated cost of the comparison systems.

However, these were estimates based on the study's evaluation methods, not evidence that hospitals will automatically save the same amount of money by adopting the technology. Actual costs and outcomes would depend on clinical workflows, local prices and patient needs.

Can AI Replace Doctors in the Future?

The findings highlight the growing potential of AI to assist with complicated diagnostic problems. Such tools could eventually help clinicians consider possible explanations, identify useful tests or review cases that require additional attention.

But there is a major difference between performing well on a benchmark and treating a real patient. Doctors must interpret physical examinations, account for incomplete or changing information, communicate with patients and make decisions that involve risks, ethics and individual circumstances.

The findings also need to be interpreted in the context of the research design. Microsoft developed the system, and the reported results come from an evaluation of its performance under defined test conditions. Independent validation and real-world evidence will be important before drawing broad conclusions.

Microsoft Plans Further Hospital-Based Evaluation

Microsoft has announced plans to work with Beth Israel Deaconess Medical Center in Boston to evaluate the system in clinical settings. One planned area of investigation is whether AI can help with triage in emergency department waiting rooms.

Testing in actual healthcare environments could help researchers determine whether the system performs reliably when information is incomplete, cases vary widely and clinical decisions have real consequences.

Until that evidence is available, MAI-DxO should be viewed as a research-stage diagnostic technology rather than a replacement for professional medical assessment.

What This Breakthrough Means for the Future of Healthcare

Microsoft's findings demonstrate how AI systems are moving beyond simple question-and-answer tools towards more structured problem-solving. By coordinating multiple AI agents and choosing the next investigative step, MAI-DxO attempts to make diagnostic reasoning more systematic.

If future clinical studies confirm its usefulness, technology of this kind could become an additional resource for medical teams, potentially supporting complex case reviews and more efficient testing decisions.

For now, the strongest conclusion is that MAI-DxO performed exceptionally well in specific diagnostic benchmarks. Whether that advantage translates into safer, more affordable and more effective care for real patients remains an important question for further research.

Sources and Further Reading

• Microsoft AI — Building Medical Superintelligence
• Reuters — Microsoft develops AI tool that can diagnose sequentially like a doctor
• TechCrunch — Microsoft's AI outperforms doctors on hard cases
• Science — Scientific journal

Medical disclaimer: This article reports on research findings and is for informational purposes only. It is not medical advice. Do not use AI-generated diagnoses to replace a qualified healthcare professional.

Published by Trend Tracker Live | AI News | Health Technology | Medical Research