A Political Theory Approach to ‘AI Ethics’

• In Event: Politics, Ethics and AI Technology: The State of the Discipline

Abstract

‘AI ethics’ has swiftly become the settled term for capturing popular and scholarly debate about the implications of artificial intelligence for society. We argue that the challenges this debate seeks to address are constrained by its framing in terms of ‘AI ethics’. The important questions are not about AI itself, but the social, economic and political contexts in which AI systems – and the institutions that develop and use them – are deployed. Nor are these questions primarily about ethics, but instead involve intractable problems of politics and economics. This paper reframes the familiar topics of the ‘AI ethics’ debate (i.e., fairness, accountability and transparency) in terms of politics and political economy. In contrast to recent scholarship that concentrates on promoting ethical decision-making and exploring AI stakeholders’ varying degrees of moral responsibility, we argue for a sharper focus on institutions, as it is ultimately their incentive structures that shape the practical ethical dilemmas that AI proprietors face in the design and deployment of their technologies.

We illustrate this argument through two case studies: Facebook’s AI-driven content moderation system; and Pennsylvania Children Youth and Family Office’s algorithm which predicts the risk of child abuse and neglect. Despite their very different contexts, each case demonstrates how determining what it is ‘ethical’ and what is ‘possible’ is constrained by both internal institutional incentives and wider economic forces.

The most challenging questions surrounding AI are, in the end, about how to structure power in modern democracies. To answer these questions, we must confront the hardest task: developing a political theory of institutions, not an ethics of AI.

Authors

• Joshua Simons, Harvard University

• Chloé Bakalar, Temple University