A Unifying Human-Centered AI Fairness Framework

TL;DR

<3-5 sentence high-level summary>

Abstract

The increasing use of Artificial Intelligence (AI) in critical societal domains has amplified concerns about fairness, particularly regarding unequal treatment across sensitive attributes such as race, gender, and socioeconomic status. While there has been substantial work on ensuring AI fairness, navigating trade-offs between competing notions of fairness as well as predictive accuracy remains challenging, creating barriers to the practical deployment of fair AI systems. To address this, we introduce a unifying human-centered fairness framework that systematically covers eight distinct fairness metrics, formed by combining individual and group fairness, infra-marginal and intersectional assumptions, and outcome-based and equality-of-opportunity (EOO) perspectives. This structure allows stakeholders to align fairness interventions with their values and contextual considerations. The framework uses a consistent and easy-to-understand formulation for all metrics to reduce the learning curve for non-experts. Rather than privileging a single fairness notion, the framework enables stakeholders to assign weights across multiple fairness objectives, reflecting their priorities and facilitating multi-stakeholder compromises. We apply this approach to four real-world datasets: the UCI Adult census dataset for income prediction, the COMPAS dataset for criminal recidivism, the German Credit dataset for credit risk assessment, and the MEPS dataset for healthcare utilization. We show that adjusting weights reveals nuanced trade-offs between different fairness metrics. Finally, through case studies in judicial decision-making and healthcare, we demonstrate how the framework can inform practical and value-sensitive deployment of fair AI systems.

Figures (4)

• Figure 1: Class distribution across protected groups in the Adult, COMPAS, MEPS, and German Credit datasets. For UCI Adult, income $\leq\$50K$ is labeled as "Low" and $>\$50K$ as "High." For COMPAS, "1" is the positive label, indicating a prediction of recidivism. For MEPS, the positive class represents high medical utilization (10 or more visits per year). In German Credit, "1" denotes creditworthiness and "0" indicates credit denial, with noticeably higher denial rates among younger applicants. These disparities highlight the presence of label imbalance and potential bias across demographic subgroups, underscoring the need for fairness-aware modeling.
• Figure 2: Trade-off between accuracy and fairness across datasets and fairness definitions. Each data-point represents the trade-off for different $\lambda$.
• Figure 3: Fairness trade-offs between infra-marginal individual fairness and intersectional group fairness across four datasets. Left column: outcome-based trade-offs; right column: equality-of-opportunity-based trade-offs. Each point corresponds to a different fairness weight configuration ([$w_{\hbox{I-M,ind}}, w_{\hbox{int,ind}}, w_{\hbox{I-M,grp}}$, $w_{\hbox{int,grp}}$]) with $w_{\hbox{int,ind}}, w_{\hbox{I-M,grp}}$ kept fixed at 0.
• Figure 4: Fairness relationships between outcome-based and EOO-based intersectional group fairness across four datasets. Each point corresponds to a different fairness weight configuration ([$w_{\hbox{I-M,ind}}, w_{\hbox{int,ind}}, w_{\hbox{I-M,grp}}$, $w_{\hbox{int,grp}}$]) with $w_{\hbox{int,ind}}, w_{\hbox{I-M,grp}}$ kept fixed at 0.