Modeling Individual Fairness Beliefs and Its Applications

Abstract

One of the criticisms made about data and algorithm-driven intelligent systems is that their results are viewed as being unfair or inequitable by individuals who believe in fairness criteria other than those embedded in the system design. In fact, computer and data scientists admit potential unfairness residing in intelligent systems. Accordingly, various approaches have been proposed to make intelligent systems fair. However, the consideration of a fundamental issue is missing in current efforts to design fair intelligent systems: Fairness is in the eye of the beholder . That is, the concept of fairness is very often highly subjective in most domains. Based on the premise that fairness is subjective, we propose a framework to represent and quantify individuals’ subjective fairness beliefs and provide methodologies to aggregate them. The proposed approach provides insight into how a population will assess the fairness of a decision or policy, which in turn can provide guidance for policy as well as designing intelligent systems.