India Pharma Outlook Team | Friday, 23 October 2026
Microsoft’s Step-by-Step AI works through cases the way clinicians do, and it proved more than twice as accurate as general physicians on the toughest examples.
Picture a patient whose symptoms refuse to fit any neat pattern. Experienced doctors pore over textbooks and search the internet, yet the diagnosis stays elusive.
Published in the journal Science on 22 October, the study suggests a new way of approaching diagnosis that could reshape clinical practice.
It measures not just the final accuracy but the careful sequence of questions, exams, and tests. The research also shows how sequential reasoning can cut both diagnostic errors and unnecessary costs.
Most medical AI systems take an “all-at-once” approach. They receive every symptom, lab result, and imaging study at the same time and produce a diagnosis. Microsoft’s system works differently.
At its core is an “orchestrator.” This component coordinates a simulated panel of clinical experts that collaborate on each case. The process unfolds in sequence:
The result is a diagnostic work-up that feels sequential and deliberate rather than a single data dump.
Also Read: 10 AI-Powered Manufacturing Technologies Improving Pharma Production
Researchers tested the tool on more than 300 highly unusual and confusing medical dilemmas. The orchestrator-guided model was more than twice as accurate as general physicians who had access to textbooks and the internet at both working up the patient and arriving at the correct diagnosis.
It also outperformed human physicians on scenarios drawn from a major database of emergency-room admissions. Traditional comparisons often rely on multiple-choice questions that AI systems now ace at near-perfect rates.
As Dominic King, Corporate Vice-President of the Health Unit at Microsoft AI, put it: “AI has become very good at getting a 100% pass rate on those. But of course, that’s not real-world. No patient ever came to see me with a nice summary.”
The same study found another practical advantage that, the diagnostic processes designed by the orchestrator cost less than those ordered by human doctors. Harsha Nori, Head of AI at Microsoft AI Health, noted that the team measured not only accuracy but also “how judiciously and carefully it goes through a diagnostic work-up.”
Faster, more accurate, and more economical diagnoses could reduce errors and cut down on unnecessary tests. In high-pressure settings such as emergency departments, an AI that reasons sequentially could support clinicians rather than replace them, helping them stay focused on the patient while the system handles the methodical steps of gathering information.
Microsoft’s approach marks a shift from AI that simply answers questions to AI that reasons through a case the way good doctors already do. Through mimicking the iterative nature of clinical thinking, the tool aims to bring greater reliability to the moments when diagnoses are hardest to reach.
The researchers plan to test the system’s ability to triage patients in emergency-department waiting rooms through collaboration with Beth Israel Deaconess Medical Center in Boston. If successful, the work could move from controlled studies into the messy, real-time flow of hospital care. The larger promise is straightforward: technology that learns to think the way experienced clinicians already do, one careful step at a time.
.jpg)
