If it were actually reasoning generally, though, it wouldn't need to be trained on each game. It could be told the rules and figure things out from there.