Can Do. Should Do? Rules for Responsible AI