If we train AI on human works and humans only do what the AI says, we're at a maximum.