This year’s Nobel Prizes in Physics and Chemistry both centered around AI; the Nobel Prize in Physics was awarded for research that formed the basis of modern machine learning, while the prized research in Chemistry uses AI models to solve problems concerning proteins.
We asked three of our AI researchers to explain these breakthroughs and what they mean for the AI field.
How Nobel Prize-winning Physics research revolutionized AI
The Nobel Prize in Physics was awarded to John Hopfield and Geoffrey Hinton for having laid the foundation for modern machine learning. The two researchers used theories about human memory to develop artificial neural networks, which make it possible for AI to learn from experience, much like people do. Artificial neural networks are models based on the biological neural networks in animals and humans.
Assistant Professor Stefano Sarao Mannelli at the Department of Computer Science and Engineering explains why this research is considered groundbreaking for AI:
“Hinton's and Hopfield's research gave new momentum to the field of neural networks after some backlash in the 1960s, and paved the way for fundamental discoveries in machine learning, especially deep learning, which is at the basis of modern AI. AI has now become part of the toolbox that scientists use for discovery, and this affects all areas including physics, chemistry and biology.”
Symbiotic relationship between Science and AI
The Nobel Prize award acknowledges the incredible importance that neural networks have in our lives today, Stefano Sarao Mannelli continues, as well as on the significant impact statistical physics has in the development and understanding of neural networks. This recognition also highlights the important connection between AI and other scientific fields.
“Among those working in AI, we know that there is a clear symbiotic relationship with science. On one hand, AI is frequently used to accelerate scientific discovery; on the other hand, insights from fields like physics, psychology, and neuroscience have profoundly influenced the development of algorithms and our understanding of their functioning.”, says Stefano Sarao Mannelli.
AI helped solve 50-year-old problem with proteins’ structures
The Nobel Prize in Chemistry was awarded to a trio who have all applied AI to their research. David Baker has, as the Nobel Committee phrases it “succeeded with the almost impossible feat of building entirely new kinds of proteins” while Demis Hassabis and John Jumper have “developed an AI model to solve a 50-year-old problem: predicting proteins’ complex structures”.
Proteins are the nano-sized machines in our body that regulate the different biological processes. David Baker used computerised methods to create proteins with new functions. According to Simon Olsson, Associate Professor in Data Science and AI at the Department of Computer Science and Engineering, this involves a complicated search problem where each step could involve experimental scientists spending months in the lab. AI made it possible to accelerate the rate of checking results. Demis Hassabi and John Jumper on the other hand, managed to create an AI model, Alphafold2, for predicting 3D structures of a protein.
“Determining the three-dimensional structures of proteins is a scientific field of its own, and again involves searching over an enormous space in a highly efficient manner typically guided by time-consuming and expensive experiments.”, says Simon Olsson.
Open source data important factor
Rocío Mercado, Assistant Professor in Data Science and AI at the Department of Computer Science and Engineering, attributes the success of Hassabi and Jumper’s model AlphaFold2 to its advanced data-driven approaches, the design of clever AI architectures for improved data efficiency, and also the use of large, high-quality datasets from open-source data.
“It's important to highlight that AlphaFold2 builds on decades of work by structural biologists, who have painstakingly collected and shared protein data openly through initiatives like the Protein Data Bank. Although there's more work to be done in the space of protein structure prediction and protein engineering, Alphafold2 has already had a profound impact, particularly in drug discovery, where it has enabled the exploration of previously unknown protein structures. This Nobel Prize underscores the benefit of AI-driven tools in addressing complex scientific challenges with significant societal benefits.”, says Rocio Mercado.
Research from the department of Computer Science and Engineering follows in the same footsteps
Simon Olsson’s research group is working on expanding the findings that Hassabis and Jumper’s AI model, Alphafold2, are based on. Instead of ‘just’ finding the 3D structures of molecules, such as proteins, the groups focus on the movements of molecules. Today the simulations of these movements occupy a lot of computing resources globally.
“Consequently, one goal in our team is to develop AI methods to accelerate the simulation of these techniques by designing systems that allow us to predict how a protein might move on much longer time scales without explicit simulation. Such technology could enable understanding of fundamental processes on time-and-length scales that are more aligned with biology”
In another lab within the department, Assistant Professor Rocío Mercado and her team are working on generative AI methods to engineer molecules and materials to specification, much like Baker's team has worked on engineering proteins to specification.
“Different molecular modalities present unique sets of challenges; in the materials science and chemical synthesis domains, for instance, data sharing initiatives haven't really taken off as in the structural biology domain, so we must devise clever ways that bridge molecular simulation and multi-modal deep learning to maximize the data efficiency of our methods. The goal for me and my team is that in the next 5 to 10 years, chemists are regularly leveraging the tools we develop in our team to help them design and prioritize molecules, such that our methods enable scientific discovery in the chemical and related sciences.”