Taking things up from where I left them at the end of the last episode I go into more detail about the various ways you might try to ensure that a conscious AI is moral, and examine the problems each of them might experience. In particular I examine how an AI might twist instructions in a way that is ultimately harmful.