In this episode of Real Talk, hosts Scot Morris, OD, and Rehan Ahmed, MD, dive into the limitations of current artificial intelligence and the critical role of data architecture in healthcare.
Some key topics of discussion:
- The Limits of LLMs: The hosts discuss a compelling position paper from Google DeepMind titled “LLMs Can’t Jump.” While AI has mastered pattern recognition (induction) and logical deduction, it fundamentally lacks “abduction”—the ability to invent an entirely new creative explanation or concept when no pre-existing data pattern exists.
- Understanding Data Models: Dr. Morris breaks down the four core data models—relational, document (NoSQL), graph and vector—explaining how patient care is shifting toward becoming a structured data-modeling process.
- The EHR Problem: The episode concludes with a critique of current electronic health record (EHR) systems. The hosts argue that today’s EHRs are merely digitized 25-year-old billing tables that act as closed data silos. They call for an evolution toward open architecture to reduce clinical friction and allow better tools to connect medical data seamlessly.
Listen to the full episode here:
Listen to all episodes of Real Talk here
Subscribe to Real Talk on Apple Podcasts and Spotify
Author
