The recent Nobel Prize awards in the fields of Physics and Chemistry highlight how artificial intelligence has penetrated science and how closely its methods are intertwined with the fundamental principles of natural sciences. In 2024, pioneers were recognized in both Physics and Chemistry, whose discoveries and developments have elevated AI to new heights. What is remarkable is not only the significance of these works but also the close connection between physical and chemical principles and the algorithms that enable AI. But what exactly is causing AI to play such a dominant role in scientific research today?
In Physics, John Hopfield and Geoffrey Hinton were honored for their work on neural networks. What might initially appear as an algorithmic achievement is deeply rooted in the principles of physics. In the 1980s, Hopfield developed the first model of a neural network that seeks out a low-energy state. This method is based on a physical concept derived from statistical mechanics and enables the network to learn associative connections. Similar to particles moving toward a low-energy state in a system, the neural network learns to strengthen or weaken node connections by adjusting their weights. This is not only a method inspired by physics but also the foundation of machine learning.
Hinton took it a step further by applying the Boltzmann law, which describes how systems prefer certain states based on energy. Based on this, he developed the so-called Boltzmann machine, a neural network that operates with probabilities. The introduction of this statistical method enabled generative models that are now used in many areas, from language processing to image generation. What Hinton and Hopfield created was not just a tool for computer science, but a model based on physical principles that revolutionized AI research.
Equally profound is the contribution of this year's Nobel laureates in the field of Chemistry. Demis Hassabis and John Jumper from Google DeepMind developed the AI system AlphaFold, which can predict protein folding with unprecedented accuracy. This prediction was long considered one of the greatest challenges in protein chemistry, as the three-dimensional structure of a protein is crucial to its function. AlphaFold uses neural networks to predict folding solely from the amino acid sequence – a task that previously took years with conventional methods. With AI, this process has been reduced to minutes. The fact that this is now possible using neural networks shows how closely AI and biochemical research are intertwined.
Here too, it is again physical principles that make AI's success possible. Neural networks like AlphaFold are based on the interactions between amino acids that interact with each other in the protein chain. The AI system learns from hundreds of thousands of known protein structures how these interactions affect the folding of the chain. It is then able to recognize patterns in the data and predict the shape a protein will take. This process is similar to the statistical methods Hinton used for the Boltzmann machine and is another example of how physical models can be integrated into chemistry and biology.
But why is AI receiving so much attention now? The answer lies in a series of technological advancements that have come together in recent years. Only the increasing computing power and the availability of large datasets have made it possible for AI models like AlphaFold or neural networks in physics to function at all. In both physics and chemistry, immense amounts of data are generated today, whether in the analysis of particles in high-energy physics or in the study of biomolecules in biochemistry. AI provides the ability to process this data and identify patterns that human researchers could hardly detect.
