Fei-Fei Li

Li studied physics at Princeton, receiving her bachelor’s degree in 1999, then completed graduate work in electrical engineering at Caltech. Stanford records a master’s degree in 2001 and a doctorate in 2005. Her research interests now include deep learning, robotic learning, spatial intelligence, and ambient intelligence for healthcare. These connect visual perception with systems that must interpret or act within their surroundings.[1]

Li directed the Stanford Artificial Intelligence Laboratory from 2013 to 2018 and has co-directed Stanford’s Human-Centered AI Institute since 2019. Her Stanford biography records a Google vice-presidency and the Google Cloud chief scientist role for AI and machine learning in 2017–2018. She now leads World Labs, whose other founders are Justin Johnson, Ben Mildenhall, and Christoph Lassner. AMD announced an agreement to acquire World Labs on September 28, 2026. The announcement places Li’s proposed AMD executive vice-president and chief scientist appointment after closing, expected by year-end subject to approvals. It does not establish that she has already started that role.[1][2][3]

ImageNet made recognition measurable

ImageNet gave computer-vision researchers a shared collection of labeled images and a common way to compare recognition systems. Li co-authored the ImageNet Large Scale Visual Recognition Challenge paper, which describes tests for classifying object categories and detecting objects within images. The paper also examines annotation challenges and compares computer performance with human accuracy. Its value is the combination of data and evaluation: a researcher can ask whether a method improves on the same task, rather than compare results from unrelated image collections.[4]

A concrete route into spatial intelligence

World Labs describes Marble as its first product, generating persistent three-dimensional worlds from inputs including images, video, text, and layouts. For an introduction to the broader idea, begin with her TED2024 talk. It connects visual perception, prediction, and action without requiring familiarity with model architectures. The ImageNet paper is the next step for understanding her earlier contribution to evaluating machine vision.[2][7][4]

In her essay on world models, Li separates three functions: renderers produce observations such as images, simulators represent the state of an environment, and planners choose actions. The distinction helps explain why a plausible-looking video is insufficient evidence that a system understands geometry or physical dynamics. She argues that simulation matters because people and computer programs need environments they can inspect and interact with, not only watch.[5]