Achieving Socio-Economic Parity through the Lens of EU AI Act

The EU AI Act addresses discrimination but has a blind spot regarding economic inequality: While the EU AI Act establishes protections against unfair treatment in AI systems, the paper argues that existing fairness frameworks overlook how AI can perpetuate biases against economically disadvantaged groups, potentially reinforcing rather than reducing wealth disparities.

Common AI fairness measures fail to protect lower-income populations: Using real-world data analysis, the researchers demonstrate that popular fairness notions currently used to evaluate AI systems cannot adequately address discrimination driven by socio-economic status, meaning AI systems may appear "fair" by current standards while still disadvantaging poorer communities.

AI regulation should explicitly address economic disparities as a fairness concern: The paper calls for the EU AI Act and similar regulations to broaden their anti-discrimination provisions to specifically target socio-economic disparities, treating economic disadvantage as a critical fairness issue alongside other protected characteristics.

A new fairness framework is needed that accounts for both circumstances and choices: The researchers propose "Socio-Economic Parity" as a solution that distinguishes between factors beyond individuals' control (like family background) and those within their control (like work hours), enabling AI systems to support disadvantaged groups without making unfair assumptions about personal responsibility.

Stakeholder dialogue on AI fairness must expand to include economic justice: The paper emphasizes that conversations between policymakers, technologists, and society about what constitutes fair AI should explicitly incorporate principles of economic equity and resource equalization for underprivileged populations.