Episode 64: The Promise and Pitfalls of Medical AI

The speaker reports that integration friction — measured in hundreds of clicks to incorporate results into reports — is a greater barrier than the choice between rule-based and AI systems.

The speaker reports that AI product development creates a paradigm where it is easy to take advantage of certain populations because the goal is to have something very generalizable, resulting in a difficult problem of how to ensure solutions trained on certain data benefit those regions, particularly in countries without high-end medicine.

The speaker reports that a home-brewed AI-driven hanging protocol tool was trivial to build a decade ago, yet mature PACS-integrated versions that remember a radiologist's preferred sequence layout for neurodegenerative studies remain unavailable.