Nobel Prize in Physics: How Hopfield and Hinton’s AI Transformed Our World

If you’ve enjoyed an AI-generated video, fraud protection, or voice-to-text, you can thank scientists like physicist John Hopfield and computer scientist Geoffrey Hinton. On Oct. 8, 2024, they received the Nobel Prize in Physics for their pioneering work on artificial neural networks, which, though inspired by biology, relied heavily on statistical physics.

How neurons compute

Artificial neural networks are inspired by how biological neurons work. In 1943, Warren McCulloch and Walter Pitts introduced a simple model showing how neurons connect, receive signals, and send out their own signals.

Neurons weigh incoming signals differently, similar to valuing some friends’ opinions more than others when deciding to buy a phone. For example, if Alice and Bob say yes, but Charlie, who you trust more for tech advice, says no, you might weigh his opinion more heavily. This could shift your decision from buying to not buying the phone.

Hopfield network

Artificial neural networks were initially inspired by biology, but their development soon incorporated ideas from logic, mathematics, and physics. Physicist John Hopfield applied physics concepts to study a specific type of recurrent neural network, now known as the Hopfield network, focusing on how it evolves over time.

These dynamics are also relevant in social networks, where information spreads quickly, leading to viral memes and echo chambers from simple interactions.

Hopfield was a pioneer in using physics models, especially those studying magnetism, to understand neural networks. He showed that these dynamics could give such networks a form of memory.

Boltzmann machines and backpropagation

In the 1980s, Geoffrey Hinton, Terrence Sejnowski, and others expanded on Hopfield’s work to develop a new class of models called Boltzmann machines, named after the 19th-century physicist Ludwig Boltzmann. These models, as the name suggests, were based on Boltzmann’s work in statistical physics.

While Hopfield networks could store and correct patterns, similar to a spellchecker, Boltzmann machines went further. They could generate new patterns, laying the groundwork for today’s generative AI.

The Role of Back propagation in Neural Networks

In the 1980s, Hinton was also involved in a major breakthrough: backpropagation. To make neural networks perform useful tasks, the right connection weights between neurons must be chosen. Backpropagation is an essential algorithm that helps adjust these weights based on how the network performs on a training dataset. However, training networks with many layers remained difficult.

In the 2000s, Hinton and his team found a clever solution. They used Boltzmann machines to pretrain each layer of a multilayer network, followed by a fine-tuning algorithm to further refine the weights. These multilayer networks, now called deep networks, sparked the deep learning revolution.

AI gives back to physics.

The Nobel Prize in physics highlights how ideas from the field have fueled the growth of deep learning. Now, deep learning is giving back to physics by enabling fast, accurate simulations of systems ranging from molecules to the Earth’s climate.

By honoring Hopfield and Hinton with the Nobel Prize, the committee signals optimism in our ability to use these breakthroughs to enhance human well-being and support a sustainable future.

Read original article on: Science Alert

Read more: Scitke