Toward a Taxonomy of Algorithmic Harms for Disability: A Systematic Review

Abstract

Understanding how algorithmic systems reinforce ableist norms, structures, and beliefs can help reduce their harmful impact on disabled people, and inform the development of more inclusive technologies. Recent research has examined the impact of algorithmic systems on disability communities via fairness, structural and methodological lenses. However, most prior work focuses on a particular algorithm system or disability community. We lack a cohesive summary of harm across individual (micro), community (meso), and societal (macro) levels, as well as across different types of algorithmic systems. To bridge this gap, we conducted a systematic review of literature (n=76) from human-computer interaction, accessibility, and responsible AI venues. Applying the taxonomy proposed by Shelby et al., we annotated 175 instances of harm and present a synthesized summary across the original categories of representational, allocative, quality-of-service, interpersonal, and societal harms. Additionally, we identify three patterns of harm specific to the intersection of disability and algorithmic systems. We connect these harms to existing manifestations systemic ableism, concluding with a discussion of challenges and potential pathways toward more equitable algorithmic systems.