Surbhi Goel

Surbhi Goel is the Magerman Term Assistant Professor of Computer and Information Science at the University of Pennsylvania, affiliated with the Theory group, the ASSET Center, and the Warren Center for Network and Data Sciences. Her research lies at the intersection of theoretical computer science and machine learning, with a focus on making modern AI systems trustworthy. Her work has been recognized by an Alfred P. Sloan Research Fellowship, a Schmidt Sciences AI2050 Early Career Fellowship, an Amazon Research Award, a JP Morgan AI Fellowship, a Simons‐Berkeley Research Fellowship, and Rising Star awards in ML and EECS. Her research is additionally supported by grants from NSF, UK’s AI Security Institute, OpenAI, and Microsoft Research. Before joining Penn, she was a postdoc at Microsoft Research NYC. She received her Ph.D. from the University of Texas at Austin, advised by Adam Klivans, where her thesis received the Bert Kay Dissertation Award. She is also the co-founder of Learning Theory Alliance (LeT‐All), a community building and mentorship initiative and recently organized the Special Year on Transformers and LLMs at Simons Institute.

AI2050 Project

AI designed to converse and collaborate with people promises immense societal benefits, from medicine to education. Yet, the black-box nature of these systems leads to unpredictable and often harmful errors, undermining the trust essential for their widespread and safe adoption. Goel’s project addresses this trust deficit by using theoretically grounded approaches to understand why these systems fail during conversations, find ways to predict these failures, and empower the system to use these signs to verifiably avoid risky decisions. The goal is to build a future where AI is safe by design.