Debate has sparked about the boundary between physics and computing after the 2024 Nobel Prize for Physics was awarded to research that developed early artificial intelligence. The prize was awarded jointly to John Hopfield and Geoffrey Hinton for “foundational discoveries and inventions that enable machine learning with artificial neural networks”. The work that Hopfield and Hinton, starting in the 80s, was instrumental in the development of artificial intelligence over the last 40 years.
Artificial neural networks are systems inspired by complex biological networks in the brain used to process information. Today, they have a wide range of applications in artificial intelligence, such as in social media, personalised marketing and medicine.
John Hopfield developed the ‘Hopfield network’ in 1982, which has the ability to store and recreate patterns in data. The network is made up of a series of nodes and connections, where each node has a specific value (originally zero or one) and the connections vary in strength. The network is modelled with inspiration from the relationships between atoms in magnetic materials, which have specific properties due to each atom having a spin. Hopfield modelled the network to have a property that describes the entire system, similar to the energy of a magnetic material. When the network analyses a pattern it checks each node individually to see how changing its value changes the 'energy' of the system, minimising the total energy until it reaches a point where it can’t make any more improvements. The network can ultimately be used to distinguish between different patterns and recreate damaged data.
Hinton continued this development, publishing the ‘Boltzmann machine’ in 1985. The Boltzmann machine has two kinds of layers of nodes, visible layers and hidden ones, where the hidden nodes control the function of the network as a whole while the visible nodes process the information. Hinton drew inspiration for his machine from the Boltzmann equation – a model in physics that describes the statistical behaviour of molecules in a gas. The machine is a generative model, meaning it will learn certain patterns based on the information it is trained on and accordingly produce outputs based on the probability of those patterns.
Computer science and modern physics generally go hand in hand, but generally it is computing that aids the development of physics. In Hopfield and Hinton’s work, however, physical systems inspire techniques for artificial intelligence.
Half of this year’s Chemistry prize also went to AI research, but in this case the research in question was using artificial intelligence to solve a specific problem in chemistry, and has therefore received significantly less backlash. Demis Hassabis and John Jumper received a quarter of the prize each for “protein structure prediction”. In 2020 Hassabis and Jumper produced an AI model, AlphaFold2, that has now been used across the world to predict the structure of 200 million proteins.
The interdisciplinary nature of modern science and the role of emerging technology in solving scientific problems makes it harder to distinguish clear boundaries between the sciences. Looking at Hopfield and Hinton’s work on its own, it seems counterintuitive to classify it as physics rather than computer science. However, for them to gain recognition at the Nobel prizes, the classification is necessary.
The Nobel prizes has awarded prizes in Physics, Chemistry, Medicine, Literature and Peace since 1901, and since then has only added one category, Economic Sciences, in 1969. In the century since computer science has exploded, leading to massive inventions and discoveries that have become increasingly ingrained in the modern world. While there are other ways for computer scientists to gain recognition, most notably the Turing Award, which Geoffrey Hinton won in 2018, few awards have the same prestige and celebrity as a Nobel Prize. While Hopfield and Hinton were able to gain a Nobel prize through a tenuous link to physics, other computer scientists aren’t able to get the same title.
There’s no doubt that Hopfield and Hinton were instrumental in pioneering new machine learning models, but finding a place for them to gain the recognition they deserve has thrown a spanner in the works of the Nobel prize machine. Perhaps the lack of room for computing as its own discipline and the changing, interdisciplinary nature of science is a sign that the Nobel prizes need to adapt for the changing times.
