Alex Townsend, Cornell University – Scientifically-Minded Artificial Intelligence
On Cornell University Week: A.I. is also coming to science labs.
Alex Townsend, associate professor in the mathematics department, examines the benefits for scientific discovery.
Alex Townsend is an Associate Professor at Cornell University in the Mathematics Department. His research is in Applied Mathematics and most recently focuses on developing machines to learn partial differential equations from data. Before Cornell, he was an Applied Math instructor at MIT (2014-2016) and a DPhil student at the University of Oxford (2010-2014). He was awarded a SIAM Computational Science and Engineering best paper prize in 2023, a Simons Fellowship in 2022, an NSF CAREER in 2021, a SIGEST paper award in 2019, the SIAG/LA Early Career Prize in applicable linear algebra in 2018, and the Leslie Fox Prize in 2015.
Scientifically-Minded Artificial Intelligence
Artificial intelligence has been in the news a lot recently. You’ve probably heard about AI’s ability to do impressive things, like generating targeted ads, making chatbots, and even defeating humans in chess. But you haven’t yet heard about AI doing science. We’re talking about a lab assistant who whispers scientific discoveries in your ear and never takes a coffee break. Researchers, including myself, are aiming to create this kind of scientifically-minded AI.
Right now, we can get AI to look at a patient’s mammogram and estimate the likelihood of breast cancer. So, AI is already doing remarkable things and saving lives. But AI offers no scientific insight, and AI’s only explanation for why a patient has cancer is “because it said so.” We’re trying to make a different kind of AI — a science teacher. We want to develop AI that can predict patterns in ways that give scientists insights.
In the 1660s, Newton realized that it is easier to find relationships between rates of change, like slopes, than to model the quantities themselves, for example, height above sea level. Newton discovered calculus, and the relationships between rates of change are now called differential equations. They’ve become the language of science, a way to describe the world and understand things like cloud formation and tsunami waves. . So, we’re working on getting AI to speak in Newton’s language of calculus. Instead of predicting the future, we want AI to look at data and tell us the underlying governing differential equation. This will make AI a bit like Sherlock Holmes’s trusty sidekick Dr. Watson, who helps us connect the dots, but a human scientist still solves the case.