More is Different in AI - More is the Same in Physics

A new paper by Prof. Ido Kanter explores the relevance of Philip W. Anderson’s influential concept “More is Different” in the context of modern artificial intelligence. The study has now been officially published in Physica A.

The work compares macroscopic physical systems with machine-learning architectures through the phenomenon of spontaneous symmetry breaking, a central concept in statistical physics.

The analysis reveals an intriguing contrast: from an information perspective, physical systems often follow a principle that can be described as “More is the Same,” where the macroscopic state cannot necessarily be inferred from a limited subset of microscopic components. In contrast, AI models demonstrate the “More is Different” principle. Even a single node in a trained neural network can encode information about the global task, while interactions among multiple nodes give rise to computational capabilities that exceed the sum of their individual contributions.

These findings highlight a fundamental distinction between physical systems with fixed interactions and adaptive AI architectures that evolve through learning, offering new insight into how complexity and information emerge in both domains.

📄 Published paper (Physica A):
https://www.sciencedirect.com/science/article/pii/S0378437126002700

📰 Media coverage:
https://scienmag.com/when-more-means-different-exploring-the-divide-between-physics-and-ai/