AI Fairness and Variable Selection: A Mathematical Approach

A new arXiv paper (2608.11251) proposes a mathematical framework for evaluating fairness in AI systems, addressing the growing regulatory demands such as the EU AI Act. The authors argue that traditional approaches often neglect philosophical ethics and social awareness, and that variable selection processes can introduce implicit bias, affecting equity across different subgroups. Their approach aligns mathematical methodologies with ethical considerations and regulatory requirements, advocating for interdisciplinary collaboration. The key finding is that excluding sensitive or critical variables may compromise equity between subgroups, whereas retaining all relevant variables allows for more granular fairness assessments and reduces implicit bias. The paper emphasizes the importance of understanding broader ethical and societal contexts in AI fairness.