![\"\"](\"https:\/\/www.europesays.com\/ie\/wp-content\/uploads\/2025\/12\/EmailGraphic.png\")
<\/p>\nThere\u2019s a reason that the boom of artificial intelligence is being referred to as a \u201cbrave new world\u201d of technology. There is a lot about AI that is still unknown and we are discovering not only new ways to improve it but also new ways to apply it.\u00a0<\/p>\n
The same could be said for many areas of physics as well. After all, some fields \u2014 like quantum physics \u2014 have formally only existed for about a hundred years, which is relatively new compared to humans\u2019 long history with other sciences.\u00a0<\/p>\n
It might make sense, then, for researchers to band together to study the two fields in tandem.\u00a0<\/p>\n
It\u2019s the idea behind a National Science Foundation-funded institute, whose goal is to see how the two fields can synergize. \u201cThe basic premise is that AI can help us do better physics, and something that is less expected is that physics can also help us understand AI better,\u201d said James Halverson, a Northeastern professor of physics.\u00a0\u00a0<\/p>\n
Launched in 2020, the NSF AI Institute for Artificial Intelligence and Fundamental Interactions, or IAIFI, brings together researchers from Northeastern University, Harvard, the Massachusetts Institute of Technology, Boston University and Tufts University to better understand the intersection of physics and AI.\u00a0<\/p>\n
In the past six years, IAIFI has produced more than 300 peer-reviewed articles, which have been cited more than 27,000 times, according to Google Scholar.\u00a0<\/p>\n
\tNortheastern Global News, in your inbox.<\/p>\n
Sign up for NGN\u2019s daily newsletter for news, discovery and analysis from around the world.<\/p>\n
<\/p>\n
Now, the institute is building on that work with nearly $25 million over the next five years in new funding from the National Science Foundation<\/p>\n
\u201cThe renewal of IAIFI by the NSF allows us to scale the intersection of physics and AI at a pivotal moment,\u201d said Halverson.\u00a0<\/p>\n
\u201cI\u2019m excited to see how these tools will help us unlock deeper AI and physics research in the years ahead.\u201d\u00a0<\/p>\n
Halverson gave Northeastern Global News (NGN) a taste of some of the institute, where much of his work focuses on understanding the similarities between the type of AI known as neural networks and the fundamental particle interactions that govern a lot of modern physics.\u00a0<\/p>\n
While on the surface this work may seem largely abstract, it has translated to direct advances in the development of new AI systems and our ways of studying the universe\u2019s fundamental building blocks, Halverson said.<\/p>\n
One of the institute\u2019s most widely cited studies is a research paper Halverson worked on that poses a new architecture for neural networks based on the Kolmogorov-Arnold representation theorem, a well-known mathematical statement used to help break down complex equations.\u00a0\u00a0\u00a0<\/p>\n
Compared to many neural network architectures, the institute\u2019s proposed option is more interpretable, is backed by the mathematical theorem, and has been useful in many quantum physics applications, including knot theory, an important subject of quantum physics that explores mathematical loops and knots in 3D dimensional structures, Halverson said\u00a0<\/p>\n
Halverson worked on the paper with Fabian Ruehle, a Northeastern professor of physics and mathematics and a fellow IAIFI member, who added another important area of focus for the researchers, which is to get machine learning models to give accurate results, make less factual mistakes, and explain their reasoning.\u00a0\u00a0\u00a0\u00a0<\/p>\n
Robin Walters, a Northeastern professor in the Khoury College of Computer Sciences and the university\u2019s newest member of IAIFI, studies AI for physical systems.\u00a0<\/p>\n
One of his areas of research at the institute is reducing the complexity involved with what\u2019s known as loss landscapes. In simple terms, a loss landscape is a visual representation of a neural network\u2019s loss value, and is depicted in a multi-dimensional structure similar to a mountainside.\u00a0<\/p>\n