The Learned Intermediary Meets the Large Language Model: Mapping the Liability Architecture of Clinical AI
I am writing this paper from three vantage points that rarely converge in one author. I am a board-certified pediatric hospitalist who admits, manages, and discharges children at an academic children's hospital — I see ambient scribes generate notes I have to sign, predictive deterioration models fire alerts I have to triage, and decision-support systems recommend pathways I have to accept or override, every shift. I am a medical expert witness who has reviewed cases in pediatric standard of care, diagnostic error, and wrongful death. And I am a builder of clinical AI tools whose engineering team has had to confront, in our own product, the same model-validation, alert-burden, and safety-monitoring problems that the lawsuits will eventually litigate.
Suggested citation: Martello RD. The Learned Intermediary Meets the Large Language Model: Mapping the Liability Architecture of Clinical AI in an Era of Algorithmic Medicine. docexpertwitness.com White Paper Series, v5.1, May 2026.
AI & Medicine · 45 min read · Ralph D. Martello, MD, pediatric hospitalist and medical expert witness.