On Explaining Unfairness: An Overview

Fairness and explainability must work together: The paper establishes that algorithmic fairness and explainability are complementary foundational elements for responsible AI, and their interplay is critical—suggesting that understanding why an AI system is unfair is just as important as identifying that unfairness exists.

Explanations serve three distinct purposes in addressing unfairness: The paper categorizes explanation approaches into three types: (1) enhancing fairness metrics themselves, (2) helping researchers understand the root causes of unfairness in algorithms, and (3) supporting the development of methods to reduce unfairness—indicating that explanations aren't one-size-fits-all but must be tailored to different stages of the fairness problem-solving process.

Significant research gaps remain unexplored: The paper identifies "undiscovered literature paths" and gaps in the intersection of fairness and explainability research, suggesting that many practical questions about how to explain and address algorithmic bias remain unanswered and represent opportunities for future work.

Comprehensive frameworks are needed: By presenting taxonomies for both fairness and explanations separately, the paper implies that practitioners and researchers need structured, organized approaches to navigate these complex topics rather than ad-hoc solutions.