What do magnets and memory have in common? Ask a physicist who reinvented neural networks in 1982.

The answer

In 1982, Caltech physicist John Hopfield published a landmark paper in the Proceedings of the National Academy of Sciences introducing what are now called Hopfield networks. Inspired by the physics of spin glasses, these recurrent neural networks store patterns as stable energy minima, allowing them to retrieve a complete memory from partial or noisy input - a property called content-addressable or associative memory. Each neuron connects to every other neuron with symmetric weights, and the network provably converges to a stored pattern by minimizing an energy function analogous to the Ising model in statistical mechanics. Hopfield's work bridged physics and neuroscience, reinvigorated neural network research during the expert-systems era, and earned him a share of the 2024 Nobel Prize in Physics.