MuZero in Nature: mastering games without knowing the rules

AlphaGo and AlphaZero were already famous for beating Go and chess champions. But they had an advantage: they knew the game rules in advance, knew which moves were legal at any moment, and what happened after each one.

MuZero starts without that information. It doesn't know what "capturing a piece" or "making a Go eye" means: it has to figure everything out by playing, building its own mental model of the game. And yet it reaches the same superhuman levels — and also works on visual games like Atari, where "explicit rules" don't exist.

It's an important step because it gets closer to the kind of intelligence useful in the real world, where no one hands you a manual before starting.