The Podcast by KevinMD Why physicians must lead the design of artificial intelligence in health care
Mar 31, 2026
Tod Stillson, a family physician, medical device inventor, and health care entrepreneur, warns about AI designed without clinician oversight. He discusses who currently builds clinical AI, five non-negotiable characteristics for trustworthy tools, limits of raw LLMs and the role of retrieval-augmented generation, and how clinicians can learn to shape AI governance and product design.
AI Snips
Chapters
Transcript
Episode notes
Big Tech Controls Core LLM Infrastructure
- Big corporations currently control core LLM infrastructure used in healthcare.
- Tod Stillson compares LLM providers to an electric company supplying a utility that others build clinical products on top of.
Require AI To Augment Not Replace Clinical Judgment
- Demand AI that augments clinical judgment rather than impersonating it.
- Require explicit data boundaries, narrow scope, explainability, auditability, and leverage of clinicians' skills when evaluating products.
RAG And Prompt Engineering Fix LLM Triage Gaps
- Raw LLMs can achieve high diagnostic accuracy but perform poorly at clinical triage without safeguards.
- Retrieval augmented generation (RAG) and prompt engineering dramatically improve triage reliability and safety.