New Framework Addresses Algorithmic Fairness Limitations in ML Systems

Modern machine learning systems are increasingly integrated into society, influencing access to opportunities and reflecting existing societal inequalities. While algorithmic fairness has emerged to counter systematic disadvantages for marginalized groups, early approaches often relied on oversimplified models. This research introduces a novel framework designed to overcome two key limitations: the reliance on single-point estimates for auditing and the tendency to view individuals in isolation rather than within their broader structural context. The proposed framework aims to provide a more robust and effective approach to ensuring fairness in AI systems. By considering the structural embeddedness of individuals and moving beyond simplistic metrics, it seeks to develop mitigation strategies that are more effective in real-world, complex socio-technical environments where algorithms mediate human opportunities. This work contributes to a deeper understanding of how to build more equitable AI.