Fei-Fei Li’s Taxonomy of World Models: Why This Field Is Becoming Real

World models are becoming one of the most important ideas in artificial intelligence. In a recent essay, Fei-Fei Li and the World Labs team proposed a useful way to understand this fast-growing field: not all world models do the same thing.

The article explains that many systems are now called “world models,” but they can be separated into three main functions: renderers, simulators and planners.

A renderer generates observations, such as images or video frames. It can create visually impressive scenes, but visual realism does not always mean physical truth.

A simulator goes deeper. It tries to represent the real structure of the world: geometry, physics, dynamics, objects, motion and interactions. This is essential for robotics, autonomous vehicles, digital twins and physical AI.

A planner produces actions. It helps an agent decide what to do next based on what it observes and what it wants to achieve. This is where world models become directly connected to robotics and embodied intelligence.

The most important idea is that these three categories are starting to merge. A powerful future world model may be able to render a scene, simulate its physical behavior and plan actions inside it.

This is why the article matters. It confirms that world models are not just a trend or a buzzword. They are becoming a foundation for spatial intelligence, physical AI and machines that can understand and interact with the real world.

For Causal World Model, this reinforces our central thesis: the next generation of AI will not only speak or generate content. It will understand space, time, physics and action.

Source: Dr. Fei-Fei Li, “A Functional Taxonomy of World Models,” Substack, June 3, 2026.