Will there be a paradigm shift when it comes to awarding Nobel Prizes? 2024 will go down in history as a turning point in the course of science. The statement was addressed by a renowned expert in Artificial Intelligence (AI) – and one of the most-cited scientists in the world. João Gama, a researcher at INESC TEC, made a bold statement: “AI won the 2024’s Nobel Prizes in Physics and Chemistry.”
However, since we are addressing the scientific domain, there are several initial interpretations of these developments. Ricardo Bessa, a researcher at INESC TEC and also one of the most-cited scientists in the world, stated that “at first glance, it may seem unusual to award a Nobel Prize in Physics or Chemistry to acknowledge the individuals’ work in AI. However, if there were a specific Nobel Prize in AI, there would likely be winners from fields like neuroscience, philosophy (yes, I’m involved in an AI project with philosophers), or even physics – where purely ‘physical’ neural networks may emerge (as recently published in the journal Nature).” Physicist Nuno Silva, also a researcher at INESC TEC, believes that “although it is natural to question the relevance of the Nobel Prize in Physics considering the advances it represents in this area as a fundamental science, the importance of this award in a more interdisciplinary and comprehensive perspective of this domain is quite clear.” Concerning the Nobel Prize in Chemistry in particular, one of the opinions was that of Hugo Penedones, a computer engineer and former student at the University of Porto, who was involved in creating the algorithm that led to the award – having already predicted what would later become reality on October 9, 2024.
We all know that AI is on the agenda. There are several books, opinion pieces, and a plethora of commentators – strangely enough, discussing a topic where there is a significant shortage of experts. So, we resorted to said experts to understand how John Hopfield, Geoffrey Hinton, David Baker, Demis Hassabis and John Jumper – the winners of 2024’s Nobel Prizes – advanced science through fundamental progress, with a major impact on society. The goal of the latest INESC TEC feature, called INESC TECWatch, is to help explain and provide accurate information on current topics according to our scientific domains. Whenever the topic falls within our areas of expertise, we will be here to provide objective, science-based opinions – something that is crucial today, yet seems insufficiently addressed in current public debates.
Shall we move on to the data on the Nobel Prize in Physics?
October 8, 2024. The Royal Swedish Academy of Sciences announced the winners of the Nobel Prize in Physics. Two relevant names: John Hopfield (American) and Geoffrey Hinton (Canadian). The first is a theoretical physicist at Princeton University; the second works as a psychologist and computer scientist at the University of Toronto. Hopfield developed associative memory, which – according to the statement issued by the Nobel Committee for Physics – “can store and reconstruct images and other types of patterns in data”. Hinton built upon Hopfield’s work, establishing the foundations for what scientists commonly refer to as machine learning. The two received the Nobel Prize “for foundational discoveries and inventions that enable machine learning with artificial neural networks,” (also by the Nobel Committee).
According to João Gama, Hinton is a pioneering figure in the deep learning community. In 2018, together with Yoshua Bengio and Yann LeCun, he received the Turing Award (to honour Allan Turing) – which is, according to the INESC TEC researcher, “often referred to as the Nobel Prize in Computing”; the acknowledgment stemmed from Hilton’s deep learning work. “Together with their students, they contributed significantly to advance computer vision, with deep implications for natural language processing,” explained João Gama.
Before dedicating his work to neural networks, Hopfield was a theoretical physicist who studied the behaviour of complex physical systems, e.g., condensed matter. In 1980, he presented ‘the Hopfield model’, where he used concepts from statistical physics to explain how neural networks can store and retrieve memories. What Hopfield proposed was that neural activity could be understood as a system with many components – neurons – that interact with each other and evolve into a state of balance, a common approach in physics.
Understand the bridges between Physics and AI
Let’s go back to the term machine learning. A practical example of a tool that uses machine learning is the popular ChatGPT. But how do machine learning and physics relate? Why did two AI experts receive the Nobel Prize in Physics? Especially since one of them – Geoffrey Hinton – resigned from Google in 2023, after warning about the risks of AI.
The relationship between AI and Physics concerns the modelling of complex systems and the understanding of how the human brain processes information. As we already know, both Hopfield and Hinton built learning models based on physical principles. Hinton’s work on deep neural networks has applications in fields like materials physics, where computers support the discovery of new materials with specific properties. Hopfield’s work explores how one can model the brain – a biological system – using principles of physics. In other words, both Hopfield and Hinton have made contributions to science involving biological systems (brain) and complex physical systems (neurons).
In addition, Nuno Silva recalled that “Hopfield was a pioneer in the mathematical development of models inspired by the functioning of neurons, and in the observation of emerging phenomena in the collective behaviour of systems with many neurons – with emphasis on the ability to reconstruct memories from pieces of partial information, with noise.”
The research of both Hinton and Hopfield makes fundamental contributions to science, making it possible to better understand how networks of neurons process information, while laying the foundations for modern AI. The relationship between AI and physics lies in the use of physical concepts to understand the dynamics of neural networks and their application in solving complex scientific problems.
As already mentioned, Nuno Silva considers it normal for most people to question the relevance of AI in a Nobel Prize awarded to Physics. However, he believes that the impact of the work done by the two Nobel Prize winners is great, particularly considering physics, since it encourages a much-needed reflection on the path towards a more comprehensive and interdisciplinary notion of this scientific domain.
Ricardo Bessa believes that the Nobel Prize acknowledges, above all, the role played by AI – namely the ability to quickly infer and extract knowledge from data and accelerate new discoveries in other scientific areas (in this case, physics). “In addition, they promote the fusion between physical/chemical models and machine learning, as is the case of physics-informed machine learning,” said the researcher. Also this week – and revisiting a topic previously discussed here – there was another award-winning field where AI played a vital role: Chemistry.
Nobel Prize in Chemistry: the facts
On October 9, 2024 – the day after the Nobel Prize in Physics was announced -, the results for the Nobel Prize in Chemistry were released. The Royal Swedish Academy of Sciences presented three names: David Baker, Demis Hassabis and John Jumper.
Let’s start with David Baker, an American researcher at the University of Washington, who – according to the Swedish Academy – was recognised for an “almost impossible feat”: the “computational protein design”. The rationale behind Demis Hassabis (the English collaborator of Google DeepMind) and John Jumper (American, also part of Google DeepMind) selection was different: they received the Nobel Price for protein structure prediction – “in proteins, amino acids are linked together in long strings that fold up to make a three-dimensional structure, which is decisive for the protein’s function,” mentioned the Swedish Academy.
The Chair of the Nobel Committee for Chemistry, Heiner Linke, said: “One of the discoveries being recognised this year concerns the construction of spectacular proteins. The other is about fulfilling a 50-year-old dream: predicting protein structures from their amino acid sequences. Both of these discoveries open up vast possibilities.” Another key aspect mentioned by the Committee was the discoveries’ potential, namely in terms of novel nano-material manufacturing and the faster development of specific drugs and vaccines – without disregarding the possibility of making the chemical industry a greener sector than it is today.
In 2003, Baker was able to design a new protein that was unlike any other. How did the scientist achieve this milestone? To understand it, we must explain that proteins are generally composed of 20 different amino acids, which can be described as the “building blocks of life”. David Baker was able to use them and design this new protein, 21 years ago. Thanks to this discovery, the research group he works with has been able to design more proteins, which can be used in pharmaceuticals, vaccines, nanomaterials or even tiny sensors.
Why are Hassabis and Jumper the latest Nobel Prize winners, if researchers have been predicting protein structures from amino acid sequences since 1970? In this case, it’s worth mentioning that the two scientists presented an AI model called AlphaFold2 four years ago.
Shall we explore the connection between the Nobel Prize in Chemistry and AI?
To do so, we resort to João Gama yet again, and to his explanation of AlphaFold.
“AlphaFold is a tool developed by DeepMind, a subsidiary of Alphabet – Google’s parent company. This tool uses AI to predict the three-dimensional structure of proteins from their amino acid sequences. This ability is crucial, because the shape of a protein determines its biological role. Understanding this is vital to many areas, e.g., the development of new drugs and the treatment of diseases,” explained the researcher.
In 2020 – the beginning of the COVID-19 pandemic -, this tool “surpassed other traditional methods of predicting protein structures in the Critical Assessment of Protein Structure Prediction (CASP13), a biennial competition dedicated to this domain,” said João Gama.
“AlphaFold was able to predict more than 200 million protein structures – almost every catalogued protein known to science! The technology behind AlphaFold is a neural network architecture called ‘transformer’, which is effective at dealing with sequential data and capturing long-range dependencies between amino acids,” concluded the INESC TEC researcher.
Here’s a noteworthy information: there is a Portuguese researcher involved in the creation of this algorithm. His name is Hugo Penedones, and he is a former student at the University of Porto, one of INESC TEC associate institutions. Hugo Penedones worked with the two winners at DeepMind – between 2015 and 2019 -, and together with other authors, they published the article ‘Improved protein structure prediction using potentials from deep learning‘ in the journal Nature. For the first two years, he worked at the headquarters in London; then, he moved to Google’s office in Zurich. He told us that during his time in London, he became involved in the earlier stages of the AlphaFold project.
“It all started in February/March 2016, during an internal Hackathon, where we were free to carry out any project we found interesting for three days. Together with Rich Evans (who came up with the idea) and Marek Barwiński, we decided to play around with the Protein Folding problem, developing some minor deep reinforcement learning and optimisation algorithms that interacted with the FoldIt game (developed by David Baker’s group). Our work was selected as the best project and generated enough enthusiasm from DeepMind’s leadership (including the CEO, Demis Hassabis) to start exploring this problem full-time,” stated Hugo Penedones.
This is how, between March 2016 and September 2017, the alumnus of the University of Porto was involved in the development of AlphaFold1 – which, according to João Gama, won the bi-annual CASP13 competition in 2018, leading to several articles (as the one mentioned before, published in Nature).
After the initial success, I was no longer involved in the process, but the team grew and continued to improve the system, leading to the AlphaFold 2 solution that won CASP14 (2020). with even more impressive results; Nature published another article on the topic – in which I am no longer a co-author since I did not contribute to that phase of the project. That one article gained quite the acclaim and may have led to the Nobel Prize,” he mentioned.
The computer engineer’s (who currently works at Inductiva) opinion about the Nobel Prize in Chemistry is quite clear; in fact, he had already made a public statement to a well-know Portuguese platform in which he predicted it. “Maybe the most surprising fact is that it happened quite quickly, just four years after CASP13. Typically, decisions take a bit longer, for validation purposes. But it was fair enough, because AlphaFold’s contribution was not to try to cure a specific disease, or some other result; the goal was to develop a computational method that could accurately predict the structure of proteins from their amino acid sequence.”
In short:
We started this opinion piece by exposing the vision of researcher João Gama, to whom “these Nobel Prizes were a milestone for science, and a paradigm shift: because we have moved from model-based science to data-driven science.” According to the INESC TEC researcher, “they have a significant impact on the areas in which Portugal and Europe should invest.”
We also realised that other AI researchers highlight other aspects. This is the case of Ricardo Bessa, who believes that “a key lesson we must learn from the 2024 Nobel Prizes is that science (and innovation) is increasingly multidisciplinary, focusing on fundamentals (and less on specific applications).”
Regarding Physics, Nuno Silva – who also emphasised multidisciplinarity – did not deny the impact of the winners’ work on this domain. “In fact, the success of the approaches is mainly due to the application of models and techniques of statistical physics, showing how it is possible to start from physical models and first principles to establish solid foundations and advance new fields of knowledge.”
As for Hugo Penedones, he is confident about the Nobel Prize in Chemistry, but surprised at how soon it was awarded – though he believes it is a deserved recognition of the work done.
Hence, we can assume that the discussion among scientists regarding this topic will be quite exhaustive, and that these Nobel Prizes have further ignited the debate on AI.
In the meantime, the Royal Swedish Academy of Sciences granted more Nobel Prizes: Victor Ambros and Gary Ruvkun (Physiology or Medicine), Han Kang (Literature), Nihon Hidankyo organisation (Peace), Dan Acemoglu, Simon Johnson and James A. Robinson (Economics). We will also seek to learn more about them through the insightful and clarifying commentary of fellow researchers dedicated to these areas.
News, current topics, curiosities and so much more about INESC TEC and its community!
