Metaphors for AI, and why I don’t like them — LessWrong