Microsoft's Sequential-Diagnosis AI Beats Physicians in Controlled Study

Microsoft researchers reported that a diagnostic orchestrator outperformed generalist physicians on interactive medical cases by asking questions and ordering tests step by step. The Science study found higher accuracy and lower estimated work-up costs in a controlled benchmark, but the system has not yet been shown to replace clinical care.

jajoy39

Oct 23, 2026 - 12:10

Oct 23, 2026 - 13:13

0 3

Microsoft company building and branding (file photo). Image source: Getty Images via BBC. https://www.bbc.co.uk/news/technology-64317078

Microsoft's Sequential-Diagnosis AI Beats Physicians in Controlled Study

Microsoft researchers say an AI system that works through a diagnosis step by step was more accurate than generalist physicians on a controlled set of complex medical cases. The system, called the Microsoft AI Diagnostic Orchestrator or MAI-DxO, decides what history to ask about, which examinations or tests to request and when it has enough evidence to commit to a diagnosis.

What the study tested

The paper in Science introduced the Sequential Diagnosis Benchmark, which turns 304 New England Journal of Medicine case-conference cases and 325 emergency-department admissions from MIMIC-IV-ED into interactive encounters. Participants begin with limited information rather than receiving a complete case summary.

On the complex cases, the study reported 83.2% accuracy for the orchestrator versus 36.1% for physicians with reference tools. On emergency-department cases, it reported 80.3% versus 46.8%. Microsoft said the system also maintained strong performance at lower estimated testing cost.

Why the result needs context

The benchmark measures an important part of diagnosis that conventional multiple-choice tests miss: deciding what to ask next and which tests are worth their cost. But the researchers caution that the results came from controlled, text-based cases. The comparison does not establish superiority over specialists, multidisciplinary teams or real-world clinical practice.

Microsoft said it plans a prospective evaluation with Beth Israel Deaconess Medical Center in Boston, where patients will interact with the system and its performance will be compared with physicians' independent assessments. That work will test workflow, safety and generalization beyond curated cases.

A research result, not a medical product

The findings show a potential way to make medical AI more systematic and cost-aware. They do not mean patients should use an unapproved model for diagnosis or that clinicians can be removed from care decisions. Real-world validation and appropriate clinical oversight remain necessary.

Illustrative file photo of a laptop and stethoscope; not a demonstration of the Microsoft diagnostic system or the Quest breach. Photo: National Cancer Institute / Daniel Sone / Unsplash. Unsplash License; cropped to 16:9. Image source; license.