Algorithmic Exclusion

Abstract

Is it “bias” if a system designer, purposefully or not, excludes a group from an algorithm’s development? No. Should the goal of “AI ethics” be to ensure that everyone is included in algorithmic systems? Also no.

In this chapter, we argue that the term “algorithmic exclusion” more accurately represents the way algorithmic systems engage with marginalized populations, but that even so, the goal should not necessarily be to include everyone. We define two subtypes of algorithmic exclusion (direct and indirect) and discuss how algorithmic exclusion relates to “category-based erasure.” We then discuss how answering questions of inclusion/exclusion must be done in relationship to who benefits from technologies.

Keywords: algorithmic exclusion, AI ethics, disparate impact, predatory inclusion, erasure, data violence