IIRC it got a little guidance other than the rules. Maybe it was given a value for piece activity, so e.g. in your case of R+K vs K it would naturally try to restrict the enemy king.
But I'm not sure. Maybe it was simply given Q>R>B>N>P
Then even if a game ends as a draw in 50 moves, one side might have more material. This way we can imagine how it would develop on it's own the idea of mobility being good, and how that would eventually lead to decisive games.
Hi folks something has really puzzled me and I don't know who to ask. Firstly, I work in computer chip design house so I know about computers. However, when I hear people say that alpha zero teaches itself chess skills starting out simply knowing just the rules. I just cannot fathom how it can achieve checkmate for the first time. I mean if it winds up with a R+K vs K, random shuffling of pieces will not result in checkmate in less than 50 moves. So I just cannot imagine how a random shuffling of pieces can generate a win for either side. And if it cannot achieve checkmate, how can it progress. Is there something I am missing?