Yes, I agree. My point is that as soon as you are giving your AI "real world" problems, where the AI itself is a stone on its internal go board, you have to start worrying about these issues.