AI Fairness at Subgroup Level - A Structured Literature Review

AI applications in practice often fail to gain the required acceptance by stakeholders due to unfairness issues. Research has primarily investigated AI fairness on individual or group levels. However, increasing research indicates shortcomings in this two-fold view. Particularly, the non-inclusion of the heterogeneity within different groups leads to increasing demand for specific fairness consideration at the subgroup level. Subgroups emerge from the conjunction of several protected attributes. An equal distribution of classified individuals between subgroups is the fundamental goal. This paper analyzes the fundamentals of subgroup fairness and its integration in group and individual fairness. Based on a literature review, we analyze the existing concepts of subgroup fairness in research. Our paper raises awareness for this primary neglected topic in IS research and contributes to the understanding of AI subgroup fairness by providing a deeper understanding of the underlying concepts and their implications on AI development and operation in practice.