World model vs game engine: what each is actually for

Press a key in a game and something very specific happens. The engine reads your input, updates a list of objects, checks which ones touch, applies the rules and draws the result. Every step is code someone wrote, and the same input gives the same output.

Now press a key in an AI world model and a different thing happens. A neural network looks at the last few frames and your action, and predicts what the next frames should look like. Nobody wrote the rules. The network learned them, imperfectly, from watching a great deal of footage.

Both can produce a playable world. Treating them as rivals, though, hides what each is good for. This guide sets out the world model vs game engine question plainly: how each works, where each fails, and what a sensible combination looks like.

What a game engine does

An engine is a machine for keeping a world consistent. It holds an explicit state: positions, health, inventory, who owns the key. Each frame it applies rules to that state (physics, collision, scripts, AI behaviour) and then renders the result as pixels.

Three properties matter here.

• The state is inspectable. A designer can open the game and read that the door is locked and the guard is at position X.
• The rules are exact. If gravity pulls at a set rate, it always does.
• Rendering is separate from truth. The picture is a view of the state. Change the art style and the world still behaves the same.

The cost is authorship. Every object, rule and texture has to be made or bought by a person. The world can only contain what someone put there.