Fred Hutch AI Club - October 2026 Talk
Doug Fridsma, MD, PhD, Chief Medical and Science Officer, Health Universe
In this talk, I draw on sixty years of AI in medicine, from ELIZA's illusion of understanding, through expert systems' brittle promise, to neural networks' hidden vulnerabilities and add my own experience living through healthcare's decade-long, multibillion-dollar shift from paper to EHRs. I make the central argument that capability has always outpaced our ability to evaluate and govern it, and agentic AI is no exception.
I’ll describe why today's large language models still carry familiar risks, including hallucination, poor calibration, a lack of built-in accountability, and why a single LLM behaves like a brilliant but unsupervised intern rather than a trustworthy clinical partner. I explain why current oversight models built for slow-moving technology, like point-in-time regulatory approval, can't keep pace with systems that evolve weekly, and why agentic systems that plan, verify, and escalate offer a safer path forward. Drawing on Health Universe's approach to building a secure, governed environment for clinical AI, I describe what it looks like to embed evaluation and oversight into these systems from the start rather than bolt them on afterward.
Finally, I close with where I believe healthcare is headed: continuous evaluation replacing one-time approval, task interoperability replacing data interoperability, agent ecosystems replacing monolithic systems, and patients becoming true participants in their own care and a call for clinicians to actively shape the standards that will determine whether this technology earns the trust medicine requires.
Free
