Cascading Effects of AI-Mediated Decisions on Wealth-Building for Communities of Color

• In Event: Social Equity and Economic Policies

Abstract

As artificial intelligence (AI) becomes embedded in the systems that determine who can rent a home, secure employment, or access credit, scholarly and policy attention has largely focused on algorithmic bias within individual domains. This paper advances a different argument: the most consequential harms of AI-mediated decision-making for communities of color emerge not within any single system, but in the cascading effects that link systems together through shared data infrastructure. Drawing on qualitative interviews conducted as part of a study of AI’s impact on the racial wealth gap, this paper examines how algorithmic decisions in one domain trigger compounding financial harms across others, and how the resulting patchwork of state-level protections produces sharply uneven outcomes for the communities most affected.The study draws on semi-structured interviews with ten adult participants from communities of color across the United States who had navigated housing, lending, or employment processes involving AI-driven tools within the prior three years. Interviews were transcribed and analyzed in NVivo using a combined inductive–deductive thematic analysis approach: open coding surfaced emergent themes from participant narratives, while axial and selective coding mapped these themes against the study's research questions. Analysis yielded 683 coded excerpts across 496 unique codes organized into eight thematic domains.Findings reveal a consistent mechanism we term cascading algorithmic harm. The same credit infrastructure (e.g., credit bureaus, background check databases, and income verification systems) feeds into consequential decisions across housing applications, employment screenings, and loan underwriting. A single adverse marker in one domain therefore cascades across all three, compounding financial harm and foreclosing wealth-building pathways. Participants described extended AI-mediated job searches depleting savings, triggering housing denials when income verification failed, and producing credit deterioration that further restricted access to employment and housing. A pervasive transparency deficit compounded these harms: participants reported not knowing whether AI was involved in decisions affecting them, nor how to appeal algorithmic determinations. State-level variation in protections, such as New York City's bias audit law for hiring tools to divergent fair lending enforcement, means that functionally identical algorithmic systems produce sharply different protections for residents depending on jurisdiction.The paper argues that algorithmic accountability policy focused on single-domain audits misses the shared-infrastructure mechanisms that drive the most severe harms, and offers implications for state-level algorithmic accountability, cross-domain auditing requirements, and transparency standards for automated decision-making in consequential settings.

Authors

• Nadiyah J Humber, University of Connecticut
  Presenting Author

• Darlis Pantoja-Benavides, University of Connecticut
  Non-Presenting Co-Author

• Yvette Pappoe, University of the District of Columbia David A. Clarke School of Law
  Non-Presenting Co-Author