Professor: Human-Level AI Needs Physical World, Not Just Language

Carnegie Mellon professor and former Meta AI researcher Deepak Pathak argues that large language models cannot achieve human-level intelligence because they lack physical-world experience. He advocates embedding 'sensory motor common sense' into AI models using adaptive reinforcement learning. His robots first train in simulation, then learn to navigate real environments, climb stairs, and imitate human actions through a method called WHIRL (In-the-Wild Human Imitating Robot Learning). Pathak co-founded Skild AI with Abhinav Gupta to build a general-purpose 'brain' for robots. Last month, Skild AI announced a $300 million funding round from SoftBank, Jeff Bezos and others. Pathak believes physical reasoning, not language, is the core of intelligence.