Chelsea Finn

There is a question Chelsea Finn asked herself in college that still drives her work: can an AI learn something new without needing millions of examples? Humans do it all the time: you see three pictures of an animal you have never seen before and you already recognize it the next time. Classic neural networks, by contrast, needed massive data to learn anything at all. Finn wanted to change that.

In 2017, still a doctoral student at UC Berkeley under Pieter Abbeel and Sergey Levine, she published MAML (Model-Agnostic Meta-Learning) at the ICML conference. The central idea was elegant: instead of training a model to do a specific task, train it to learn new tasks quickly, from just a few examples. A model that knows "how to learn" rather than just "what to learn." The paper became one of the most-cited works of the decade in machine learning.

For that work she received the ACM Doctoral Dissertation Award, was named to the MIT Technology Review 35 Under 35 list, and earned a Sloan Research Fellowship and the NSF CAREER Award. Since 2019 she has been a professor at Stanford (Computer Science and Electrical Engineering), and in 2024 she co-founded Physical Intelligence (Pi), a robotics startup building a general-purpose physical AI that applies her ideas about efficient learning directly to robots that move in the real world. She also received the Presidential Early Career Award for Scientists and Engineers (PECASE) in 2025.

To me, Chelsea Finn represents something that is easy to forget in this field: the best breakthrough does not always come from having more data or more computing power, but from asking the right question. She asked how humans learn and built an algorithm inspired by that intuition. Every time you see an AI model adapt to something new with minimal information, there is a little of that Berkeley question working underneath. And that is what makes her story worth telling.