Beyond Bias: The Ethics of Risk-Based Artificial Intelligence Decision-Making in the Criminal Justice System

This dissertation explores the ethical implications of using fully automated algorithmic decision-making based on risk assessment instruments in the United States criminal justice system, envisioning a hypothetical future where the human decision-maker is removed from the process altogether. While acknowledging the important study of algorithmic bias, the work instead examines these ethics based on its reliance on statistics. It traces the historical development of the statistical normal, beginning with its origins in the late 18th and early 19th centuries, and examines its influence on criminology and the eugenics movement. In contrast it examines the pluralistic normal, a concept based on Hannah Arendt's expansion of Kantian judgment, which emphasizes both understanding individuals in their particularity and the inherent equality of human life. The dissertation argues that artificial intelligence is inherently unsuitable for making decisions about human life and freedom due to its inability to meet the demands of human dignity as considered through utilitarian, deontological, and virtue ethics perspectives.