Fei-Fei Li: Spatial Intelligence is the Next Frontier in AI

TL;DR

Fei-Fei Li argues that spatial intelligence – understanding, reasoning about, and generating 3D worlds – is the next frontier for AI and essential for AGI. She traces the arc from ImageNet (object recognition) to image captioning to generative AI, and announces her new company World Labs, co-founded with Justin Johnson, Ben Mildenhall, and Christoph Lassner, which aims to build world models that g

Key insights

• Visual intelligence in evolution took 540 million years to develop, whereas language took less than 500,000 years; vision was the driver of the evolutionary arms race and animal intelligence.
• Language is fundamentally 1D (sequential tokens), while vision is high-dimensional (3D space, geometry, physics, causality, action), making it a harder problem for AI.
• The data for spatial intelligence is not readily available on the internet – it exists only in human heads, requiring hybrid approaches including real-world data and synthetic data.