Even before artificial intelligence tools like large language models (LLMs) captured the public’s imagination, scientist and engineers were using AI to advance groundbreaking discoveries.
AI/machine learning tools like neural networks, generative models, and reinforcement learning have helped scientists conduct simulations, generate pattern recognition, and make predictions to find new insights for more than a decade.
But how can institutions like the National Science Foundation (NSF), which funds much of the scientific research in the United States, most effectively continue to support AI research efforts in the future?
In March, Prof. Andrew Ferguson of the Pritzker School of Molecular Engineering and Department of Chemistry at University of Chicago co-organized an NSF-supported workshop of nearly 200 scientists from across the country to envision the future of AI research within the NSF Directorate for Mathematical and Physical Sciences. NSF MPS supports fundamental research in astronomy, chemistry, materials, mathematics and physics to advance scientific knowledge and enhance national economic growth, security, and quality of life.
With colleagues from astronomy, chemistry, materials science, mathematics and statistics, and physics, Ferguson and the team spent two days together at the Massachusetts Institute of Technology soliciting community input, discussing ideas, and developing actionable suggestions to guide support of AI-based research in mathematics and physical sciences. Over the past several months, they turned that discussion into a white paper that is meant to provide a roadmap for the community, from funding entities like the NSF to individual researchers, in contributing to these fields in the age of AI.
“AI is here to stay,” Ferguson said. “We’re really just at the tip of the iceberg on what’s becoming possible with it in mathematics and the physical sciences. The recommendations we made could change quickly as the field evolves, but hopefully this can be useful in helping identify important priorities and opportunities.”
Encouraging cross-disciplinary research and educating an AI-aware workforce
To bolster research related to AI and MPS, the team recommended supporting cross-disciplinary research and educational efforts by funding grants where AI will have the biggest impact.
Building an interdisciplinary community could include funding workshops and conferences, but it could also include sharing knowledge in new ways.
“Ideas from one field can cross-pollinate another field, and AI spans a lot of different domains,” Ferguson said. “For example, investigators could be embedded into other research groups that do something different. That would help mid-career investigators continue to educate themselves in AI while also building a new community.”
For education, the workshop team concluded that AI literacy should be integrated into all levels of education to help develop the AI-educated workforce of the future.
“Scientists coming up in the field need to understand what it means to be a scientist in the age of AI,” Ferguson said. “The skills you need to be successful are different now than they were even five years ago. There’s an adage that goes around the physical sciences community that AI will not replace scientists, but scientists who use AI will replace scientists who don’t.”
Part of the success of the workshop came through understanding how different fields use AI, Ferguson said. For example, Ferguson and other researchers in materials science refer to “big data” when talking about large data sets of information about materials at the atomic scale. But physicists working with Large Hadron Collider at CERN also refer to their datasets as “big data,” and those datasets are at a much larger scale—so large that they cannot even store all the data they capture from particle collisions.
“What's interesting and important in the development and deployment of AI techniques for a materials scientist can be quite different from that for a physicist,” Ferguson said. “There’s also valuable two-way feedback with AI whereby the ideas and concepts from physical sciences can feed into frontier AI developments. Understanding how the thermodynamics of complex systems work can actually help design new AI algorithms, for example. The idea that there can be this feedback and bi-directional information transfer was an important concept that spanned the discussions at the workshop.”
How academia can contribute to AI
The ultimate goal is to accelerate the pace of scientific discovery, leading to new breakthroughs in difficult problems, like understanding the nature of dark energy and dark matter, developing new drugs, designing new materials, and identifying new fundamental particles.
“Industry is leading academia in AI areas that require enormous resources, like training large, complicated models,” Ferguson said. “But academia can excel in fundamental science or in pursuing blue-sky ideas.”
Ferguson said the NSF was pleased to receive the community white paper and will hopefully find the recommendations and identified opportunities valuable in designing new calls for funding. “It was a real privilege to be a part of this group,” he said. “It was a big endeavor, but hopefully it is useful for the community and for federal funding bodies.”
This work was supported by the U.S. National Science Foundation (NSF) under Award Number 2512945. Any opinions, findings, and conclusions or recommendations expressed are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.