Date of Award

8-1-2023

Language

English

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

College/School/Department

Department of Political Science

Dissertation/Thesis Chair

Julie Novkov

Committee Members

Virginia Eubanks, Patricia Strach, Stephan Stohler

Subject Categories

Political Science

Abstract

This dissertation is centrally concerned with understanding how algorithmic harms evolve into legal claims in the context of automated public benefits determinations. Data collection consists of multiple source of data including semi-structured in-depth interviews, court opinions, expert witness reports, deposition transcripts, and investigative news reports. Analysis was conducted in parallel with data collection through iterative process of open coding.Results indicate that algorithmic errors are not merely technical glitches, they lead to cascading harms. In-depth analysis of algorithmic errors identified in legal cases reveals the type of errors that top-down formal audit mechanisms fail to capture. As statistical tools automated systems operate under the logic of categorization and cannot account for individual circumstances. Data shows that in automated fraud determinations statistical correlation is not equivalent to legal attribution. Additionally, the investigation underscores multiple challenges in identifying algorithmic errors before the onset of litigation. These challenges are a labyrinth of lack of explanation, lack of understanding, and futility of fair hearings as a forum for questioning, each contributing to the formidable task of determining whether automated determination is erroneous and what the error may be. Analysis suggests that claimants past experiences color their responses to algorithmic harms, perceiving automated determination as an anomaly that contradicts their previous experiences. Piercing the opacity of automated determinations and achieving legal recourse hinge on claimants’ persistence in the face of causal ambiguity, and critically contingent upon the availability of due process claims as a key legal opportunity structure. These findings present a significant leap in our understanding of how algorithmic harms evolve into fully-fledged legal disputes, despite the challenges associated with opacity and complexity of automated decision-making.

Download

Share

COinS