On Explaining Proxy Discrimination and Unfairness in Individual Decisions Made by AI Systems