CBD 2026: What AI Can't Do (Yet) — And Why: The Everyday Role of Creative Leaps in Human Learning with Douglas Guilbeault, PhD
Today's AI systems learn through processes of brute-force statistical optimization that require orders of magnitude more data and compute than any person, and yet the concepts that AI systems develop remain surprisingly brittle and inflexible. By comparison, humans routinely achieve deep and robust understanding from remarkably little, given the pervasive constraints they face on memory, attention, and information access. Drawing on new research showing that a simple threshold rule governs how people learn everything from grammar to social conventions, this keynote argues that human learning is distinguished not by gradual optimization but by creative, categorical transitions: sudden leaps across semantic space from disordered, random exploration to stable conceptual insights. I argue that the capacity to inhabit uncertainty, chaos, and ineffability — and to construct shared poetic meanings from within these spaces — is a distinctly human ability that may reflect a fundamentally different paradigm than what underlies current large language models. What enables this human capacity for creative leaps remains a challenging open question, one that invites us to reconsider how qualitative dimensions of mind — from bodily experience and emotions to subjective narratives, metaphors, and artistic culture — can function as engines of insight that shape discovery even in the most rigorous domains of mathematics and science. The talk closes by considering what is lost when we treat human minds as mere prediction machines, and why cultivating exploratory intuition, poetic reasoning, artistic experimentation, and contemplative practice matters more than ever in the age of AI.
Douglas R. Guilbeault is an Associate Professor of Organizational Behavior at the Stanford Graduate School of Business and the Spence Faculty Scholar for 2025–2026. His research sits at the intersection of computational social science, analytic sociology, cognitive science, and machine learning, examining how communication networks and technologies shape the emergence and spread of culture — from linguistic categories and behavioral conventions to collective beliefs — and how human learning relates to and coevolves with today's AI systems. He is co-director of the Computational Culture Lab, and his research has appeared in top scientific venues including Nature, the Proceedings of the National Academy of Sciences, and Management Science, as well as popular journalistic venues including The Atlantic, Wired, and Harvard Business Review. He has received research awards from the International Conference on Computational Social Science, the Cognitive Science Society, and the International Communication Association. He holds a PhD in Communication from the University of Pennsylvania's Annenberg School, along with degrees in philosophy, rhetoric, and cognitive linguistics.
This session is part of the FREE Contemplation by Design Summit, Oct. 14-25, 2026.
The full summit schedule is posted at: https://med.stanford.edu/contemplation/summit.html.