Why do simple learning rules yield AI systems that generalize far beyond their training data?
In this lecture, Prof. Eddy Chen argues that this reflects an abundance of learnable structure in nature. This abundance motivates Nomic Liberalism, a conception of laws developed from Minimal Primitivism. On this view, laws can be simple, predictive, representation relative, and present across many scales and domains beyond fundamental physics.
The lecture connects this philosophical account to phenomena in machine learning, including double descent, scaling laws, and the emergence of advanced capabilities from simple objectives such as next token prediction. Prof. Eddy Chen suggests that AI systems can be understood as discovering liberal laws, often in domains traditionally thought to lack lawful structure and in coordinate systems unlike familiar human concepts.
The talk also addresses the limits of learning. Results in quantum foundations show that, in high dimensional quantum systems, many quantum states are observationally indistinguishable. A satisfactory theory of induction must therefore explain both why learning succeeds so widely and why it must sometimes fail.
Prof. Eddy Chen is Associate Professor of Philosophy at the University of California, San Diego. He works at the intersection of philosophy of physics, metaphysics, and philosophy of artificial intelligence, with research on the foundations of quantum mechanics, the metaphysics of time, the nature of laws, and conceptual and normative questions raised by AI systems.
The lecture is followed by a reception.