World model
월드 모델
An AI model that pictures inside itself how the world moves and looks ahead to what will happen next.
Overview
If you push a cup, it falls. If you throw a ball, it flies in a curve. People know this from experience and can picture it in their heads in advance. A world model is meant to let an AI learn this kind of order of the world from video and other data, and look ahead to what would happen if it took a certain action.
An AI that has learned only from text speaks well but does not know much about how objects move. So world models are meant to let robots and self-driving cars picture the result before they move, or to create virtual spaces that you can walk around in. Research is still very active, and people mean slightly different things by the term.
An everyday example
It is like playing chess and picturing a few moves ahead in your head: "if I move like this, my opponent will move like that."
Why it matters
As an "AI that understands the world" has been named as the next step after AI that talks well, big research labs are jumping in. It often appears together in news about robots, self-driving and video generation.