Continual Learning: World Models in Natural & Artificial Intelligence with Adam Safron

What does it actually mean for a machine to “understand” the world? Are today’s auto-regressive LLMs truly reasoning, or is statistical text prediction fundamentally distinct from genuine deliberative agency and causal inference? And how do we move away from brittle, post-hoc safety patches toward intrinsic, system-level alignment?

This week begins Continual Learning, a new mini-series co-hosted with my friend and colleague, cognitive scientist Dr. Adam Safron of the Allen Discovery Center at Tufts University and the Active Inference Institute. For the last year, we’ve been working together with the support of Survival and Flourishing Fund to advance scientific understanding and silo-crossing conversation around AI capabilities, alignment, and regulation—centered on a special issue of Philosophical Transactions of The Royal Society A on World models in natural and artificial intelligence co-edited by Adam and Michael Levin. The next several episodes are a meaningful detour into this work.

Over the coming season, we’ll dive deep into cognitive neuroscience, complex systems science, the study of narratives, and Buddhist epistemology to explore what true world modeling entails. This episodes launches that investigation by identifying major major themes from the special issue and connecting dots between its papers. (Strap in, because we move a million miles an hour.) Some of the questions we raise include:

* How do we rigorously define what a world model is—and isn’t?

* Do machines need goals, intrinsic motivation, and deliberation to truly think?

* What is the relationship between world-modeling and agency?

* Where can we look for evidence of emergent structure in scaling LLMs, and what does it mean if we don’t find it?

* How can we structure scientific collaboration to ask better questions about the future of human-machine co-evolution…and what might it take for machines to actively participate in that inquiry?

Explore the entire open-access special issue here.

And whether you read the papers or not, be sure to check out this illuminating interactive discourse map by Van Bettauer of Ideoscopic to help you navigate where these researchers agree, disagree, and point toward future study.

(Long-time fans will also want to check out his re-imagined interface for AskFutureFossils.com, complete with simulated debates between my guests!)

Subscribe for amazing conversations with Nadav Amir, John Krakauer, Fritz Breithaupt, Michael Levin, and many more!