World Model
An internal representation that allows an AI system to simulate and predict how the environment behaves.
Overview
A world model is an AI system's learned internal representation of how its environment works, enabling it to predict the consequences of actions, simulate future states, and plan accordingly. World models capture the dynamics, physics, and rules of the environment without requiring explicit programming of these rules.
Key Details
In model-based reinforcement learning, world models allow agents to 'imagine' outcomes without real-world interaction, dramatically improving sample efficiency. For language models, world models describe the implicit understanding of real-world concepts that emerges from training on text. Video generation models like Sora demonstrate world modeling by generating physically plausible scenes. Building accurate world models is considered a key step toward more general AI systems.