Princeton researchers used machine learning techniques to uncover a new understanding of how water turns to ice, key to controlling myriad industrial processes, refining climate models and enabling basic research. This work was cited in support of the 2024 Nobel Prize for Physics as one of a few remarkable findings that illustrate AI's transformative power as a research tool. Image courtesy the researchers
When this year’s Nobel Prize in physics went to Geoffrey Hinton and Princeton’s John Hopfield for their work on the foundations of modern artificial intelligence, the Nobel committee detailed clear examples of how the new technology has advanced science and technology across many fields.
One of the cases cited in the committee’s 12-page document was Princeton University research from 2022 that revealed deep new insights into the formation of ice.
“With its roots in the 1940s, machine learning based on artificial neural networks has developed over the past three decades into a versatile and powerful tool, with both everyday and advanced scientific applications,” the committee writes.
After outlining the history, illuminating how physical models came to encompass biological phenomena and enable artificial neural networks to become the powerful tools they are today, the committee turns to the ways in which these tools have become essential “for modelling and analysis in almost all of physics.”
They give around a dozen examples, from finding the elusive Higgs boson to predicting protein structures from amino acid sequences. One key example is probing quantum-mechanical many-body problems, one of the most important areas in physics. To support this claim, the committee chose a 2022 Princeton paper from postdoc Pablo Piaggi, graduate student Jack Weis, and professors Athannasios Panagiotopoulos, Pablo Debenedetti and Roberto Car.
The Princeton paper details a machine learning model of water that simulates ice formation using fundamental theories to describe each interaction between hundreds of thousands of atoms — previously considered impossible due to the problem’s vast complexity. The model derived its insights from the atoms’ electronic structures, a first-principles approach that would have been unthinkable without the computational efficiencies the researchers leveraged with artificial neural networks.
“Until recently, simulating ice nucleation with quantum accuracy was deemed impossible due to the prohibitive computational cost of quantum-mechanical calculations,” the authors write. “Recent progress enabled by machine learning has made these calculations tractable and thus greatly extended the field,” a development that is as important to climate models and climate mitigation technologies as it is to food systems and industrial processes.
As the Nobel committee notes, machine learning is now “revolutionizing science, engineering and daily life.” As true at Princeton — birthplace of computing, incubator of quantum mechanics, cradle of artificial intelligence — as anywhere.
