Rethinking Fairness: An Interdisciplinary Survey of Critiques of Hegemonic ML Fairness Approaches

"Fairness" itself is contested: The paper reveals that computer scientists and scholars from other disciplines define fairness very differently. Technical definitions often miss broader social justice concerns, suggesting that simply making an algorithm "fair" by computational standards may not produce just outcomes for marginalized communities.

Technical fixes alone cannot solve social problems: The survey identifies "technological solutionism" as a major limitation—the tendency to treat complex social inequities as problems that algorithms can solve. Abstraction in AI systems can obscure real-world harms and create a false sense that fairness has been achieved when structural issues remain unaddressed.

Current fairness work often serves as "ethics washing": Rather than preventing harm, fairness measures can be used to avoid regulation and create the appearance of responsibility without substantive change. This allows organizations to claim ethical commitment while maintaining problematic systems.

Marginalized communities are excluded from decision-making: The paper emphasizes an absence of participatory design and democratic deliberation in AI fairness research. Those most affected by AI systems are typically not involved in defining what fairness means or how systems should be built, undermining legitimacy and effectiveness.

Data practices entrench existing inequities: The survey highlights how data collection is often non-consensual, lacks transparency, and perpetuates historical biases. These foundational problems mean that fairness interventions applied later cannot fully remedy systems built on compromised data.