Position: Beyond Sensitive Attributes, ML Fairness Should Quantify Structural Injustice via Social Determinants

A research position paper argues that algorithmic fairness frameworks should move beyond focusing on sensitive attributes like race and gender to examine structural injustice through social determinants—contextual variables that shape outcomes systemically. The authors demonstrate through college admissions models, census data analysis, and healthcare screening applications that fairness interventions centered solely on sensitive attributes can paradoxically create new forms of structural injustice.

This position paper addresses a fundamental limitation in how the machine learning and AI communities conceptualize fairness. Traditional algorithmic fairness research treats discrimination as a problem of disparate outcomes across demographic groups, leading to technical solutions focused on de-biasing sensitive attributes. However, the authors contend this approach misses structural injustice—systematic disadvantage embedded in social and economic contexts that precede and shape individual attributes. The distinction matters because contextual factors like neighborhood quality, educational access, and healthcare infrastructure are often treated as noise in algorithmic systems rather than audited as sources of systematic bias.

The practical implications are substantial. When mitigation strategies target only sensitive attributes without examining underlying social determinants, they can mask or even reinforce structural inequalities. The healthcare screening case study illustrates this concretely: algorithmic fairness interventions may achieve demographic parity while leaving systemic disparities in health outcomes untouched. This creates regulatory and ethical risks for organizations deploying such systems, as apparent fairness metrics conceal deeper injustices.

For AI developers and organizations deploying high-stakes systems, this research signals a need for more sophisticated fairness auditing that incorporates social determinants analysis. The work establishes a cross-disciplinary framework pulling from public health, sociology, and social policy perspectives. Future technical developments will likely need to incorporate contextual data auditing and social determinant measurement into fairness assessments, fundamentally expanding what fairness infrastructure must track and validate.

• →Current algorithmic fairness approaches focusing on sensitive attributes miss structural injustice embedded in social determinants and contextual variables
• →Fairness interventions targeting only demographic parity can create new forms of structural injustice while appearing compliant with fairness metrics
• →High-stakes domains like healthcare and education require auditing social determinants alongside sensitive attributes to identify systemic disparities
• →Contextual factors shaping outcomes should be treated as auditible signals rather than noise to be normalized in algorithmic systems
• →Future fairness frameworks must precede mitigation strategies with structural injustice assessment using cross-disciplinary insights from public health and sociology