RESEARCH · RESEARCH · #2441
Microsoft’s Diagnostic Orchestrator (GPT-5) paper in Science reports higher diagnostic accuracy than doctors
Microsoft published a peer‑reviewed paper in Science describing Diagnostic Orchestrator, a multi‑agent medical AI system powered by OpenAI’s GPT‑5. In tests on case reports and emergency‑department data compiled into a new benchmark, SDBench, the system achieved substantially higher diagnostic accuracy than the human physicians in the study and lower estimated costs, though Microsoft and authors say clinical validation in real-world settings is still required.
KEY POINTS
- Microsoft published a peer‑reviewed paper in Science describing Diagnostic Orchestrator, a multi‑agent medical AI system powered by OpenAI’s GPT‑5.
- In tests on case reports and emergency‑department data compiled into a new benchmark, SDBench, the system achieved substantially higher diagnostic accuracy than the human physicians in the study and lower estimated costs, though Microsoft and authors say clinical validation in real-world settings is still required.
- If validated in clinical settings, a system that meaningfully outperforms physicians on diagnosis would have major implications for care delivery, regulation, and healthcare costs.
WHY IT MATTERS
If validated in clinical settings, a system that meaningfully outperforms physicians on diagnosis would have major implications for care delivery, regulation, and healthcare costs.