NEJM AI Grand Rounds From Clinician to Chief Health AI Officer: A Conversation with Dr. Karandeep Singh
38 snips
Sep 17, 2025 Dr. Karandeep Singh, a physician-scientist and Chief Health AI Officer at UC San Diego Health, merges programming with medicine. He shares insights from his journey in biomedical informatics, discussing the crucial gap between theoretical AI capabilities and real-world clinical applications. Singh highlights his experience with sepsis prediction tools, the unexpected challenges in AI governance, and the importance of ethical considerations. His story emphasizes that successful AI must prioritize patient care and remain aware of its limitations.
AI Snips
Chapters
Transcript
Episode notes
From Kid Programmer To Clinician-Inventor
- Karandeep Singh taught himself many programming languages while training as a physician and built tools to solve clinical problems.
- He discovered biomedical informatics during nephrology fellowship and chose to pursue AI at scale while continuing clinical practice.
Prediction Versus Recognition Gap
- Singh found a vendor sepsis model identified sepsis well after clinicians already treated patients, not earlier as advertised.
- The model often tracked the digital footprint of care rather than prospectively predicting sepsis.
Beware Feedback Loops In Deployed Models
- Feedback loops can make deployed models look better or worse depending on how clinician actions affect the outcome.
- Evaluations must account for interactions between model predictions and clinical workflows.