Microsoft claims to have unlocked AI capabilities that make its new “Microsoft AI Diagnostic Orchestrator” (MAI-DxO) four times better than experienced doctors at diagnosing illnesses – and a whole host cheaper.
“The path to Medical Superintelligence” triumphed Microsoft in a blog on its results – which saw it test AI models on 304 interactive case challenges drawn from the New England Journal of Medicine (NEJM) case series.
Its MAI-DxO correctly diagnosed 85% of cases. A panel of 21 American and British doctors diagnosed just 19.9% of them. (The physicians, all generalists, were forbidden from using search engines, referring to specialists, and worked without access to colleagues, textbooks, or AI.)
Technically, in brief, Redmond used five agents operating as a team, built on top of a range of different LLMs. Dr Hypothesis (tracks diagnoses) Dr Test-Chooser (selects optimal tests) Dr Challenger (plays devil's advocate) Dr Stewardship (manages costs) Dr Checklist (provides quality control).
So: Goodbye future meatspace doctors, hello AI in a white jacket?
Not so fast: Redmond’s blog trumpets that the paper shows “AI can sequentially investigate and solve medicine’s most complex diagnostic challenges—cases that expert physicians struggle to answer.”
But doctors have raised some tough questions.
This post is for paying subscribers only
Join peers managing over $100 billion in annual IT spend and subscribe to unlock full access to The Stack’s analysis and events.
