The Second Law of Intelligence: Controlling Ethical Entropy in Autonomous Systems