Fei-Fei Li

Created ImageNet: 14 million labeled images across 22,000 categories, built with 49,000 crowd workers over years. The decisive insight: deep neural networks could be trained to understand visual categories if given enough labeled data. ImageNet + AlexNet (2012) is arguably the single experiment that launched the modern AI era.

Co-founded Stanford's Human-Centered AI Institute (HAI). The core thesis: AI should augment human capabilities, not replace them — and should be designed with human values, needs, and limitations at the center of the design process, not as an afterthought.

A persistent advocate for diversity in AI research (gender, race, geography, discipline). Her argument is not only moral but technical: a field dominated by one demographic builds systems that reflect those biases. AI4ALL, which she co-founded, has trained hundreds of underrepresented students in AI.

Founded World Labs in 2024 to build 'large world models' — AI systems that understand 3D space and physical dynamics, not just text or 2D images. Her argument: language models know facts, but they have no sense of how physical objects exist in space and time. Solving spatial intelligence is key to robotics, healthcare, and augmented reality.