Comment by flender

And if it knew every possible board configuration and optimal move, it could potentially do as well as it could, but instead if it were to just recognize “this looks like a chess game” and use an optimized tool to determine the next move, that would be a better use of training, it would seem.