Artificial intelligence (AI) is already part of our daily lives: chatbots that answer questions, generators of incredible images, tools that improve photos of black holes, and even programs that predict protein structures
In October 2024, scientists John Hopfield and Geoffrey Hinton received the Nobel Prize in Physics precisely for their fundamental contributions to these systems.
But why a Physics prize for something that seems so closely linked to computing? The answer lies in deep and surprising connections between AI and concepts in physics, such as magnetism and quantum fields.
It all began in the 1920s, with the German physicist Wilhelm Lenz, who proposed to his student Ernst Ising a model to understand magnetism in materials like iron.
Imagine a block of iron formed by atoms, each with a “spin”-a kind of small magnetic arrow that can point up or down.
Neighboring spins tend to align because this reduces the energy of the system, making it more stable.
Temperature causes random fluctuations, but over time, magnetic domains emerge where the spins become organized.
If we apply an external magnetic field, the entire material can align and form a permanent magnet.
This is the famous Ising model, which explains phenomena like ferromagnetism in a simple and elegant way.
To better visualize this, think of a ball rolling through a mountainous landscape.
The ball’s position represents the spin configuration, and the height indicates the energy.
The ball naturally rolls towards the lower valleys, which are the most stable states where the spins are aligned.
This concept of an “energy landscape” would become essential decades later.
Moving forward to 1982, scientist John Hopfield was inspired by this model to create something revolutionary: a neural network capable of memorizing patterns.
Instead of atoms with spins, he used artificial neurons that can be “active” or “inactive.” Each connection between neurons (synapse) has a weight that can be positive (favorable to alignment), zero, or negative (favorable to the opposite).
The network evolves by seeking states of lower energy, exactly as in the Ising model.
The great power of this Hopfield network lies in the possibility of adjusting the weights of the connections.
By doing so, it’s like sculpting the energy landscape: deep “wells” are created around specific configurations.
When the network receives an incomplete or noisy pattern, it naturally “rolls” to the nearest well, retrieving the memorized pattern.
This is the principle of associative memory that underlies many modern AIs.
Geoffrey Hinton and other researchers further developed these ideas, leading to the deep neural networks we know today.
In training, the machine receives thousands of examples and automatically adjusts the weights to recognize recurring patterns – the so-called machine learning.
But the link between AI and physics goes much further.
A neural network can be seen as a mathematical function that transforms inputs into outputs.
If we represent the inputs as points in a space, the network assigns a value to each point, functioning as a “field.” When the weights are randomly adjusted and the network is very wide (with many neurons), the distribution of the values “”it produces tends towards a bell-shaped curve – the famous Gaussian distribution.
This surprising behavior was discovered decades ago and shows that infinitely wide neural networks behave like Gaussian processes.
This is where quantum physics comes in.
In quantum mechanics, quantum fields describe particles and their interactions.
A free quantum field (without interactions) has fluctuations that exactly follow a Gaussian distribution.
Wide neural networks behave analogously to these fields without interactions.
In real networks, which are not infinite, small corrections arise, similar to the interactions between particles in physics.
This equivalence allows tools from quantum field theory – such as Feynman diagrams – to be used to study neural networks, and vice versa.
Scientists at the IAIFI institute, for example, are already using neural networks to simulate quantum fields with interactions, opening new avenues for solving complex problems in particle physics that would be impossible even for supercomputers.
On the other hand, AI helps experimental physics: it cleans noise in gravitational wave signals, analyzes data in search of new particles, reconstructs images distorted by gravitational lensing, and simulates materials or phase transitions.
In short, physics not only inspired the birth of neural networks, but also offers ways to understand them better, transforming what often seems like a “black box” into something more transparent.
And AI, in turn, returns the favor, helping to unravel mysteries of the universe.
This exchange of ideas between two such different areas shows how scientific knowledge advances in an integrated and surprising way, revealing that even the most modern machines have deep roots in the fundamental laws of nature.
Published in 04/26/2026 01h04
Text adapted by AI (Grok) and translated via Google API in the English version. Images from public image libraries or credits in the caption. Information about DOI, author and institution can be found in the body of the article.
Reference article: