From a young age, Miles Cranmer was captivated by physics. His grandfather, a physics professor at the University of Toronto, provided him with books on the topic, while his parents took him to university open houses in southern Ontario, Canada, with the Perimeter Institute for Theoretical Physics being a standout. “I recall someone discussing infinity when I was very young, and it fascinated me, ” Cranmer shared. During high school, he interned at the Institute for Quantum Computing at the University of Waterloo, describing it as “the best summer of my life at that point. ” This experience led him to pursue an undergraduate degree in physics at McGill University. One evening in his second year, 19-year-old Cranmer came across an interview with renowned theoretical physicist Lee Smolin in Scientific American. Smolin stated that reconciling quantum theory and relativity would “take generations. ” “That just triggered something in my mind, ” Cranmer remarked.

“I can’t accept that — it needs to happen faster. ” For him, the key to accelerating scientific advancement lay in artificial intelligence. “That night, I concluded, ‘We need to implement AI for science. ’” He began exploring machine learning, ultimately integrating it into his doctoral research in astrophysics at Princeton University. Nearly a decade later, Cranmer, now at the University of Cambridge, has witnessed AI starting to revolutionize science, albeit not to the extent he envisions. While single-purpose systems like AlphaFold can produce scientific predictions with remarkable accuracy, researchers still don’t have “foundation models” tailored for general scientific discovery. These models would function akin to a scientifically accurate version of ChatGPT, adept at generating simulations and predictions across various research fields. In 2023, Cranmer and over two dozen scientists launched the Polymathic AI initiative aimed at developing these foundation models.