The "godmother" of AI promotes spatial intelligence

In a recent essay, renowned AI expert Dr. Fei-Fei Li explains why spatial intelligence—that is, systems that can comprehend, reason about, and create 3D, physics-consistent worlds—will be the next major advancement in AI. The specifics: Li contends that although LLMs have mastered abstract knowledge, they are incapable of perceiving and acting in space, including motion and distance estimation. The cognitive foundation of human intelligence, according to her, is spatial comprehension, which is essential for advancing AI from language to perception and action. According to Li, world models will be essential to developing this intelligence, but they must be able to construct realistic three-dimensional worlds, comprehend inputs like pictures and activities, and forecast how those worlds will evolve over time. She continued by saying that by allowing AI to reason in the actual world, these models would eventually lead to new developments in robotics, research, healthcare, and design. One day, world models that comprehend how items move and interact might be used to test materials, simulate climate systems, or anticipate molecular interactions. Teaching AI practical physics is a problem, but momentum is growing quickly as Google, Tencent, and Li's World Labs compete to create spatially intelligent systems.