Beyond the Chatbot: The Rise of World Models and Physical AI
Researchers are shifting from text prediction to systems that simulate physical reality to unlock true autonomy and AGI.
The frontier of artificial intelligence is moving from the digital screen into the physical world. Researchers and tech giants are pivoting from Large Language Models (LLMs) toward "world models"—AI systems designed to learn through observation to simulate physical cause-and-effect.
Unlike LLMs, which operate on next-token prediction to generate text, world models build internal representations of environments, including spatial, physical, and causal dynamics. This shift is backed by massive capital; Physical AI scaleups raised approximately $16.1 billion in the first three quarters of 2025, a trend highlighted by Meta's significant investment in Scale AI.
The Gap in Machine Intuition
Current AI chatbots can describe physical events using language, but they lack an intuitive understanding of physics because they have no direct interaction with the material world. This absence of grounded experience creates a fundamental limitation for the development of truly autonomous robots that must navigate and manipulate complex environments reliably.
Melanie Mitchell of the Santa Fe Institute notes that the pursuit of understanding physical reality has become "a kind of shorthand for all the things that current AI systems can’t do well." While generative AI can create convincing content, it cannot yet reason through the physical consequences of an action in a three-dimensional space.
The Path to AGI
This transition from generative AI to "physical AI" is viewed as a critical requirement for achieving Artificial General Intelligence (AGI). The ability to model reality allows machines to perform complex real-world tasks with human-like intuition rather than relying on static datasets.
At Google I/O 2026, Demis Hassabis emphasized the stakes of this evolution, stating that AGI is gated by the ability to simulate reality with perfect fidelity. In this view, the race for intelligence is no longer just about who can process the most text, but who can most accurately simulate the laws of physics.
What to Watch
The industry is now monitoring whether these simulations can translate into reliable robotic hardware. The primary challenge remains the transition from a simulated world model to a physical agent that can operate in the unpredictable real world. As investment continues to pour into Physical AI, the focus will shift toward whether these models can move beyond simulation to achieve a level of dexterity and reasoning that matches human capability.