Back to AI Index 2026 overview

Chapter 06

Medicine

Stanford's medicine chapter shows AI moving on two fronts at once: molecular and virtual-cell models on the research side, and concrete workflow automation on the clinical side.

Why it matters

Medicine is one of the clearest examples of AI value meeting regulation, evidence burden and workflow reality all at once.

How to apply it

Read it as a split between discovery and delivery: powerful models are emerging, but clinical adoption depends on trust, workflow fit and evidence quality.

Core signals

Core signals

Domain-specific models routinely outperform much larger general ones in biology, and clinical workflow automation is already delivering measurable time and cost reductions.

Smaller specialized models are highly competitive

Stanford highlights several biology systems where smaller models outperform much larger ones, suggesting that data curation and domain design can beat brute-force scale.

Clinical workflow automation is already tangible

The chapter points to broad adoption of note-generation tools and large diagnostic performance gaps between strong multi-agent systems and unaided physicians on difficult published cases.

Selected data points

Selected data points

Condensed numbers and comparisons pulled from the official Stanford chapter materials.

SignalValueContext
Protein / genomics size comparison111M / 200M vs 40BStanford cites 111M- and 200M-parameter biology models outperforming much larger baselines, including a 40B-parameter model.
Clinical note time reductionUp to 83%Physicians reported spending up to 83% less time writing notes after adoption of AI note tools.
Reported ROI112%One hospital system reported a 112% return on investment from AI note-generation tools.
FDA AI medical devices in 2025258The FDA authorized 258 AI medical devices in 2025.
Diagnostic orchestrator vs unaided physicians85.5% vs 20%Microsoft's AI Diagnostic Orchestrator with OpenAI o3 scored 85.5% on difficult case studies versus 20% for unaided physicians.

Implications

  • Medical AI is becoming simultaneously more useful and more regulatorily complicated.
  • Workflow automation may scale faster than fully autonomous clinical decision-making.
  • Evidence quality remains the hard constraint when deployment stakes are high.