Invitation Only
Speaker: Arvind Murugan, Ph.D. (University of Chicago)
Title: How Matter Finds the Unlikely
How does matter arrive at configurations that seem to have purpose? Two strategies for finding such rare functional states are known, but both work outside the material: Darwinian evolution distributes the search across a population; backpropagation offloads the search to a computer model. Through theory and experiments, we explore a third, physical learning, in which a single piece of matter, through its own dynamics, drives itself toward functional configurations by experiencing examples of the desired behavior.
This program faces two challenges. We need tunable, expressive parameters: mediated interactions in disordered molecular and mechanical systems provide these, and collective processes like phase transitions yield higher expressivity than components engineered to mimic neurons. We also need a process that tunes them: many local molecular processes naturally implement Hebbian learning, encoding correlations into interactions through mechanisms as simple as selective survival of molecular aggregates. Together, this produces Pavlovian conditioning, supervised classification, and generative tuning of probabilistic behavior. Physical learning expands what matter can do, and opens paths toward model-free synthetic biology and non-genetic adaptation in cells.