A Critical Survey on Fairness Benefits of Explainable AI

Explainability alone doesn't guarantee fairness: The paper finds that many claims linking explainable AI (XAI) to fairness are "vague and simplistic" and often lack proper grounding in what XAI can actually accomplish. Being able to explain how an AI system makes decisions doesn't automatically make those decisions fair.

Different fairness goals require different approaches: The research identifies seven archetypal claims about XAI's fairness benefits, suggesting that "fairness" isn't one-size-fits-all. The paper emphasizes the need to specify which particular fairness goal is being addressed and how XAI specifically helps achieve it, rather than treating XAI as a universal solution.

XAI is a tool, not a cure-all: The authors recommend viewing XAI "not as an ethical panacea but as one of many tools" for addressing algorithmic fairness. This means organizations need multiple complementary strategies—not just explanations—to ensure fair AI systems.

Context and stakeholders matter critically: The paper stresses that claims about XAI and fairness must clarify "who is the stakeholder that benefits from XAI." Different groups (affected individuals, regulators, developers) may need different types of explanations, and what helps one group may not help another.

Current research lacks rigor on this connection: Based on analysis of 175 scientific articles, the survey found that many claims are poorly aligned with XAI's actual capabilities and often lack "normative grounding"—meaning they don't clearly explain the ethical principles underlying their fairness claims.