Fei-Fei Li on Spatial Intelligence: Data is Harder Than Models

FinanceYF5 · x · 2026-08-09

Fei-Fei Li shared her multidimensional thoughts on AI development at AASF 2026, advocating for a "middle ground" amidst extreme AI rhetoric:

Spatial and Physical Intelligence: She views this as the next frontier for AI. Most intelligence in evolution developed before language; animals lack language but possess amazing navigation and physical manipulation abilities. Unlocking this will bring order-of-magnitude changes to scientific discovery, disaster relief, manufacturing, agriculture, and healthcare.
Three Layers of World Models: Divided into rendering (showing what the world looks like), simulation (capturing physical and dynamic reality, like flowing water), and planning. World Labs focuses primarily on the simulation layer.
Data is Underestimated: She emphasizes that data is currently a harder challenge than models, a belief she has held since creating ImageNet over a decade ago.
Robotics and Healthcare: The primary use case for robots should be healthcare, as there is a global shortage of exhausted nurses, and much of this high-demand labor isn't even counted in GDP.
Technology and Humanity: She calls for not letting technology strip away human agency. Tech should be a tool to augment humanity, creativity, and productivity, rather than making decisions for us.

More from AGI Musings