Alumni and friends of the School of Science gathered on the clear and chilly morning of Oct. 16, 2024 to join Dean Nergis Mavalvala for a lecture by professor of physics Jesse Thaler. Professor Thaler is director of the Institute for Artificial Intelligence and Fundamental Interactions, or IAIFI, one of the inaugural NSF artificial intelligence research institutes. In his talk, “Deep Learning + Deep Thinking = Deeper Understanding,” Thaler discussed the significant advantages machine learning offers to solving problems faced across the field of physics, and he explained the efforts undertaken within IAIFI to foster interdisciplinary collaboration between researchers in AI and physics.

After thanking Dean Mavalvala, Professor Thaler began the morning by testifying, relating to the audience the story of his transformation “from an AI curmudgeon to an AI evangelist.” In 2017, he was confronted by Patrick Komiske and Eric Metodiev, two MIT graduate students who were using machine learning in their doctoral research to discriminate between quarks and gluons, subatomic particles that cannot be detected in isolation. At the time, Thaler was also working on the problem of distinguishing quarks from gluons and had found some success with his own human-derived techniques. He was skeptical that AI’s capacity for “deep learning” could surpass the “deep thinking” of a human being, and he challenged Komiske and Metodiev to explain “what they even mean[t] by quarks and gluons?!” The students’ PhD work convinced Thaler that the combination of AI computation and rigorous human decision making could chart a principled path toward distinguishing between quarks and gluons. He learned the lesson that research into the quark–gluon problem was improved by the use of machine learning, which both contributed to his AI conversion and demonstrated that other questions might benefit from similar attention.

“Deep learning is all the more powerful when combined with deep thinking.”

Thaler described IAIFI, which formed in 2020, as a group “dedicated to combining the deep learning revolution in artificial intelligence with deep thinking in physics, in order to try to gain a deeper understanding of physical systems and machine intelligence.” A collaboration between MIT, Harvard, Northeastern, and Tufts, IAIFI is dedicated to enabling breakthrough discoveries in physics and to the development of approaches to AI that incorporate first principles from fundamental physics. The institute seeks to cultivate pioneering research, to empower the next generation of talent working in AI physics, and to build a dynamic, interdisciplinary community. Its faculty, postdocs, and students bring together first principles from fundamental physics, rich datasets, and exciting discovery opportunities at the intersection of AI and theoretical physics, experimental physics, astrophysics, and other subfields. Additionally, researchers are innovating AI technology by infusing machine learning systems with physics principles.

Because AI-physics is a novel and collaborative endeavor, Thaler explained, IAIFI makes a priority of developing new talent and cultivating connections between the two fields. The institute’s postdoctoral fellowship program, which just welcomed its fourth cohort, has brought researchers from across the globe to MIT. Through coursework and an interdisciplinary PhD in physics, statistics, and data science, faculty within IAIFI are exposing more scientists to their new techniques and enticing others to follow their new approach to fundamental scientific research. Through summer school courses, workshops, colloquia, and informal gatherings, IAIFI provides opportunities for its participants to share their work and ideas in order to advance physics research and galvanize AI innovation by fusing deep thinking with deep learning.

According to Thaler, the future of AI and science is the future of science. Machine learning technology has only recently advanced to the point where it is of benefit to fundamental science. As projects in IAIFI progress, generative AI takes a larger role and impacts how physicists conceptualize and carry out computations. He ended with the notion that “deep learning is all the more powerful when combined with deep thinking.” Professor Thaler took questions from an enthusiastic crowd and lingered with guests at the conclusion of the breakfast.

Gayatri Pradhan and her two daughters
Gayatri Pradhan, MBA ’06, and her two daughters listen to the School of
Science breakfast talk given by Professor Jesse Thaler on Oct. 16, 2024.

Jesse Feiman | School of Science