Generative AI for Health: Promises and Pitfalls

Public discourse has been dominated by sweeping proclamations about AI's transformative benefits, yet has confused the many different things that the term "AI" refers to. While attention has been focused on flashy generated videos and chatbots, there has been quiet, steady progress on a different kind of AI: The AI that can detect tumors, flag cardiac arrhythmias, monitor vaccine spoilage, and optimize medicine storage. This type of AI -- predictive, rather than generative -- enables faster, predictable, more accurate care throughout the world.

In this talk, I will explore how different types of AI deliver value in healthcare, where AI hype misses critical nuances, and where misunderstanding AI's capabilities and effects has already caused harm. Drawing on examples from medical imaging, clinical decision support, and patient-facing tools, this talk will offer a clear-eyed framework for evaluating AI health technologies, moving from the narrative that we must use AI or be left behind to the question of "what kind of AI is this, and what can it actually do?"