This year’s Nobel Prizes have been announced and it’s a big win for AI. Both the Physics and Chemistry prizes are distinctly lacking when it comes to Physicists and Chemists this year, and it’s thanks to the ground-breaking work of computer scientists in developing and implementing AI into research.
The Physics prize has come out as the more controversial of the two, with the statistical physicist John J. Hopfield of Princeton University and computer scientist Geoffrey E. Hinton of the University of Toronto splitting the prize between them, for their roles in building the foundations that enable machine learning and AI today. For the Chemistry prize, half went to David Baker (a biochemist from the University of Washington) for computational protein design, and the other half was split between two computer scientists at Google DeepMind, Demis Hassabis (DeepMind’s co-founder) and John M. Jumper (the current director of DeepMind) for protein structure prediction.
So, what exactly is the work behind the sudden influx of computer scientist Nobel Laureates.
The Physics prize seems to be a final realisation from the Nobel prize committee to the relevance of AI in today’s world and, although many would argue it isn’t strictly a Physics field, the committee evidently were anticipating push-back, highlighting the importance of neural networks in the Physics world in their announcement.
In 1982, Hopfield published his first paper, titled “Neural networks and physical systems with emergent collective computational abilities”, here he introduced one of the first artificial neural networks, which came to be known as a Hopfield Network. This could learn in a limited way, having the basic ideas of assigning weights to nodes and making connections between them to form a network. Hopfield Networks have the ability to store patterns from data and check how closely a new pattern fits with what was stored, allowing the network to take an incomplete or distorted image and find the saved image which most resembled it. This may seem very basic, especially now that AI is powerful enough to build its own images and imitate human literature, but at the time this was ground-breaking.
Geoffrey Hinton followed on from Hopfield’s work in 1985 and introduced a probabilistic element to the deterministic Hopfield Network, this greatly improved its learning ability and earned Hinton the title of the “Godfather of AI”. Hinton’s Boltzmann Machine, so called due to its utilisation of the Boltzmann distribution, was trained using many examples of patterns the machine would likely interact with, enabling it to classify images or even create new examples of the type of pattern from which it was trained.
As the forerunners of the current AI boom, these two individuals deserve huge praise for their work, but considering how long ago this work was, and how long science has been utilising neural networks in data analysis, modelling and data generation, perhaps it is more highlighting how behind the Nobel committee appears to be. Especially considering how the acknowledgement of their work coincides with the popularization of Chat GPT, which perhaps forced their hand to recognise AI’s importance.
Basic neural network showing how nodes are connected to take an input and deliver an output. Image by Cesar Jung-Harada on Flickr.
The chemistry prize was far less controversial thanks to the applications of protein structural prediction, as understanding these structures is key to understanding how proteins functions. Baker utilised deep learning algorithms to allow proteins of a desired shape to be predicted, whilst the Google DeepMind team worked from the other end, designing a program known as AlphaFold, which takes a given sequence of amino acids and predicts how it folds. In 2020, DeepMind’s AlphaFold could predict structures with 90% accuracy and since then some 2 million people have utilised their software, leading to the number of known protein structures increasing by a factor of 1000. AlphaFold works thanks to an enormous database of known protein structures and sequences being fed into it, hence experimental verification of its results only serves to improve its accuracy. This is hugely important work as the process of finding structures experimentally can take from months up to years to complete, with Haemoglobin (the first protein structure to be determined) taking 33 years alone. In four years AlphaFold has allowed for 200 million more structures to be identified. These protein structures will be crucial to the future of the biochemistry industry, allowing for more effective medicines and sensors to be discovered and implemented.
Haemoglobin subunit gamma protein (one of the proteins that makes up haemoglobin) as generated on the AlphaFold site. We were required to cite the following in order to use the image:
Jumper, J et al. Highly accurate protein structure prediction with AlphaFold. Nature (2021).
So why the controversy in Physics, but not Chemistry? It likely comes down to the Chemistry prize being awarded for application of AI in the field, whereas the Physics prize appears to be more a prize for AI in general. What is abundantly clear, however, is that the Nobel committee this year have decided to recognise AI and its importance in a big way, with this decision giving voice to many of those emphasizing the need for regulation and discussion of its use. In interviews Demis Hassabis stresses that the risks and unknowns surrounding AI need to be discussed and researched now, and the “Godfather of AI” himself, appearing far more pessimistic, has stated, “I worry that the overall consequences of this might be systems that are more intelligent than us, that might eventually take control.”
Whether pessimistic or optimistic about AI in the future, if Nobel Laureates are highlighting the importance of its regulation and research. Perhaps this is why, despite anticipating the controversy, the Nobel Committee chose 2024 (a year with more focus on AI than ever) to highlight the work of these individuals and their teams.
Featured image: Created by Callum Oozeerally, using images by 紅色死神 on flickr and Vivienne on flickr