Moving boundaries: An appreciation of John Hopfield

TL;DR

The paper examines how John Hopfield's work helped move the boundaries between physics, biology, and artificial intelligence, framing biological physics as an integral part of physics. It traces a throughline from his early studies of dielectric behavior and allostery to neural-network models and the later AI revolution, including Boltzmann machines and deep learning. A central thread is the use of energy landscapes and attractor dynamics to describe computation, memory, and learning, linking thermodynamics, information processing, and neuroscience. The discussion highlights the scientific and practical impact of this cross‑disciplinary trajectory, including advances like AlphaFold and emergent analyses of behavior, and it argues for a view of physics as a mode of inquiry that extends beyond traditional disciplinary boundaries.

Abstract

The 2024 Nobel Prize in Physics was awarded to John Hopfield and Geoffrey Hinton, "for foundational discoveries and inventions that enable machine learning with artificial neural networks." As noted by the Nobel committee, their work moved the boundaries of physics. This is a brief reflection on Hopfield's work, its implications for the emergence of biological physics as a part of physics, the path from his early papers to the modern revolution in artificial intelligence, and prospects for the future.

Figures (3)

• Figure 1: Simple models. (A) In a classical dielectric, localized charges oscillate in response to an applied field. To generate spatial dispersion, there must be a path for energy flow, schematized by (effective) lateral springs connecting the charges hopfield+thomas_1963. (B) In hemoglobin, the iron atom is bound to the heme group by spring of stiffness $K_{\rm Fe}$ that changes its equilibrium length ($\ell_{\rm Fe} \rightarrow \ell_{\rm Fe} + \delta_L$) when a ligand such as oxygen binds to the iron. In addition the iron atom is held by the protein as whole, schematized as a spring of stiffness $K_{\rm P}$ with a different equilibrium length $\ell_{\rm P}$hopfield_1973.
• Figure 2: Energy landscape and trajectories in a model of neural networks hopfield+tank_1986. (A) Solid contours are above a mean level and dashed contours below, with X marking fixed points at the bottoms of energy valleys. (B) Corresponding dynamics, shown as a flow field.
• Figure 3: Neural networks with a feed--forward architecture, or "perceptrons." (A) An early version from $\sim$1960 block_1962. (B) A modern version lecun+al_2015. The first steps in the modern AI revolution involved similar networks, with many hidden layers, that achieved human--level performance on image classification and other tasks.