Learning Algorithms and Discrimination

Abstract

In most accounts of Artificial Intelligence (AI), humans are pitted against the machines, and concerns about relative physical and mental strength populate the discussion. There are, however, numerous processes and shifts that AI stimulates within the human context that are equally essential to uncover as AI technology gradually penetrates various markets. In this chapter, we focus on data-driven discrimination, explaining its origins, scope, and current treatment under the law. We explain how errors, biases, and opaqueness are endemic to AI systems, which rely heavily on data collected and analyzed. Nevertheless, it is unclear whether and to what extent existing anti-discrimination doctrines such as disparate impact apply to biases resulting from algorithmic systems. We also discuss examples of discriminating AI technologies in five dominant fields: finance, employment and labor, health and healthcare, education, and the legal system. These examples demonstrate that while AI enjoys a scientific glory of improving human performance, streamlining operations, and saving costs, AI also suffers from inherent biases and inaccuracies that risk perpetuating social injustices by systemizing discrimination. To capitalize on the innovative benefits that AI has to offer and mitigate discriminatory concerns, proper scholarly awareness, industry cooperation, and regulatory attention must accompany the introduction of AI technologies.

Keywords: Algorithms, Discrimination, Black Box, Choice-making, Machine Learning, Artificial Intelligence, Data Accuracy, Big Data, Disparate Impact, Due Process, Opaqueness, Equal Credit Opportunity Act, General Data Protection Regulation, Social Biases, Financial Service, Employment, Health, Education