I'm assuming point two about "no knowledge of the game is put into the program" is more that none is put into the neural network? Or am I misunderstanding what is meant there? Seems the scoring function is clearly how you score the game?
This statement refers to the fact that, there are no rules like "you should put a O to block X.X" and things like that. Inside the program there is just the rule to understand if the game ended and who won. In the NN there is nothing at all of course, it's just random weights at start.
Ah, that makes sense. Still having fun reading over the code. Thanks for sharing!