I think the issue is that there is 1 (or possibly very few) perfect way to play the game, while there are endless ways to play it badly. When we want the comput

I think the issue is that there is 1 (or possibly very few) perfect way to play the game, while there are endless ways to play it badly. When we want the computer to play badly we want it to play badly like a human would, so we're playing chess and trying to pass the turing test. When we want the computer to play perfectly we don't care if it seems like a human playing.