UT Mathematicians and Computer Scientists Team Up to Outline Future of Math and AI

A new Math and AI Group at The University of Texas at Austin is working on several fronts to address the future of mathematics in the age of artificial intelligence—from striving to advance an open research infrastructure to evaluating frontier models’ mathematical results to exploring strategies for training the next generation of mathematicians and mathematical learners. 

Core to the ethos of the group is working to keep mathematical research open and rigorous and working towards building a shared infrastructure for the scientific and mathematical community, including in partnership with other institutions. Ensuring frontier AI systems become subjects themselves of scientific inquiry, the Math and AI group is teaming with other institutions, including the Courant Institute at New York University, to build a future where “the broad mathematical community can dissect, improve, reproduce—and ultimately—understand” the impact of AI on mathematics, including with a goal of developing an open foundation model for the mathematical sciences, as part of a larger academic consortium. The University of Texas at Austin-based National Science Foundation Institute for the Foundation of Machine Learning (IFML) is involved in the work.

“Large language models present an extraordinary opportunity for mathematics and science, but whether they bring a golden age of discovery depends on the choices we make today,” said Francesco Maggi, chair of the Department of Mathematics at UT Austin. “Mathematical progress is not simply about producing more theorems and proofs. It is about developing new ideas, deepening human understanding, and educating future generations of scientists.”

Maggi asserted that universities have a responsibility to ensure that these technologies strengthen, rather than undermine, the open and independent scientific enterprise, while helping graduate and undergraduate students develop the critical thinking, creativity and intellectual independence necessary to use them effectively.

“By bringing mathematicians and computer scientists together and building an open research infrastructure in collaboration with other academic institutions, we want to help shape a future in which LLMs expand the horizons of human knowledge while keeping that knowledge accessible, understandable and firmly rooted in the scientific community,” Maggi said.

Among the other efforts underway is work UT mathematicians are doing with the First Proof project, an initiative that supports independent, transparent and rigorous engagement with the evolving capabilities of AI in research mathematics. To help assess AI’s math capabilities, UT’s Math and AI group will host a First Proof event with mathematicians from leading institutions Jan. 7–8, examining AI-generated solutions for mathematical correctness, clarity and attribution. 

UT researchers also are participating in the Math Community, a collaborative effort to examine and clarify AI-generated results and help bring new insights into the peer-reviewed literature.

Rachel Ward and UT mathematician Tim Perutz participated in a Harvard-led summit co-organized by UT professor emeritus Dan Freed that examined doctoral training in the age of AI. Its recommendations emphasize rigorous oral assessment and communication, alongside careful use of AI tools that supports students’ intellectual development. These are recommendations for the profession, rather than announced changes to UT’s degree requirements. At the undergraduate level, UT mathematics educators Mike Starbird and Ward are beginning to work to improve AI tutoring so that systems better support how students learn mathematics.

Learn more about the Math and AI Group.