Normative Reasoning for AI