How can a robot use an AI model that predicts the results of actions to plan its next move?

Answer: It compares predicted outcomes before choosing an action

It repeats a recorded movement without checking the current scene — Replaying a recorded movement does not use predictions to adapt to the current scene. A world model can instead estimate what a candidate action would do from the present state, which matters when an object has moved.

It selects a movement from the most common words in a description — Word frequency alone does not predict the physical result of a movement. A world model connects states and actions to likely next states, allowing choices to depend on consequences rather than wording.

It compares predicted outcomes before choosing an action ✓ — Predicting action outcomes lets a robot compare possible moves before trying them. This kind of world model learns how situations change; a controller can use its predictions to pursue a goal. Predictions can be wrong, so the model does not guarantee reliable real-world behavior.