The History of AI

Nobel Prize in Physics for Hopfield and Hinton

Era 5 · Generative AI, reasoning, and agents (2022–2026) › AI for science and recognition

Lineage
Neurons and networks
Kind
Institutional

Who

Royal Swedish Academy of Sciences

What happened

The Royal Swedish Academy of Sciences awards the Nobel Prize in Physics to John Hopfield and Geoffrey Hinton "for foundational discoveries and inventions that enable machine learning with artificial neural networks."

Why it mattered

It recognizes the physical foundations of neural networks and brings machine learning to the best-known prize in physics.

The full story

The Academy's announcement explains the prize in terms of physics. Hopfield, of Princeton University, created an associative memory in 1982: a network that stores patterns, such as images, and reconstructs them from damaged or incomplete versions. Hopfield drew on the physics of atomic spin, and training adjusts the connections so that the stored patterns have low energy. Given a damaged image, the network updates its nodes step by step, lowering the energy, until it reaches the most similar stored pattern.

Hinton, of the University of Toronto, built on the Hopfield network to create the Boltzmann machine, using tools from statistical physics, the branch that studies systems made of many identical parts. Hinton's machine learns to recognize the characteristic features of a type of data and can classify images or generate new examples similar to the ones it has seen.

The prize, 11 million Swedish kronor, was split equally. Ellen Moons, chair of the committee, said the laureates' work had already been of the greatest benefit, and noted that physicists use neural networks in many fields, such as designing materials with specific properties.

The next day the Nobel Prize in Chemistry also went to work done with AI: AlphaFold.

A curve with two valleys, pattern A and pattern B, separated by a hill. A ball, the damaged image, sits on the slope, and a dashed arrow carries it down to the bottom of pattern B's valley.

In the Hopfield network each stored pattern is an energy valley; a damaged image rolls down into the nearest valley.

Under the hood

In the Hopfield network each node takes the value +1 or -1, like the spin of an atom, and each pair of nodes is linked by a weight wᵢⱼ. The network has an energy E = -½ Σ wᵢⱼ·sᵢ·sⱼ, which drops when nodes linked by positive weights agree.

Storing patterns means choosing the weights so that those patterns sit in energy valleys. Recalling means letting the network evolve: each node takes the sign of the weighted sum of its neighbors, the energy never rises, and the network settles into the nearest valley, which is the most similar stored pattern.

Hinton's Boltzmann machine changes two things. Its nodes are stochastic, with the probability of each state following the Boltzmann distribution of statistical physics, and it adds hidden nodes, which do not see the data and learn to represent its features.

Common misconceptions

  • What people often say: The Nobel Prize in Physics rewarded computer science work that has nothing to do with physics.

    In fact: The Academy rewarded methods built with tools from physics, spin physics in Hopfield's case and statistical physics in Hinton's, and noted that physicists use these networks in fields such as materials design.

Videos

Builds on

Threads that pass through here

Sources

How to cite

“Nobel Prize in Physics for Hopfield and Hinton.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/nobel-physics-hopfield-hinton

This entry was last reviewed on November 13, 2026.

Open in the interactive timeline