Title: On the Identification of Fair Auditors to Evaluate Recommender Systems based on a Novel Non-Comparative Fairness Notion

Abstract: Decision-support systems are information systems that offer support to people's decisions in various applications such as judiciary, real-estate and banking sectors. Lately, these support systems have been found to be discriminatory in the context of many practical deployments. In an attempt to evaluate and mitigate these biases, algorithmic fairness literature has been nurtured using notions of comparative justice, which relies primarily on comparing two/more individuals or groups within the society that is supported by such systems. However, such a fairness notion is not very useful in the identification of fair auditors who are hired to evaluate latent biases within decision-support systems. As a solution, we introduce a paradigm shift in algorithmic fairness via proposing a new fairness notion based on the principle of non-comparative justice. Assuming that the auditor makes fairness evaluations based on some (potentially unknown) desired properties of the decision-support system, the proposed fairness notion compares the system's outcome with that of the auditor's desired outcome. We show that the proposed fairness notion also provides guarantees in terms of comparative fairness notions by proving that any system can be deemed fair from the perspective of comparative fairness (e.g. individual fairness and statistical parity) if it is non-comparatively fair with respect to an auditor who has been deemed fair with respect to the same fairness notions. We also show that the converse holds true in the context of individual fairness. A brief discussion is also presented regarding how our fairness notion can be used to identify fair and reliable auditors, and how we can use them to quantify biases in decision-support systems.