Centring the Margins: Mapping AI Systems as Systems of Power

Abstract

This paper introduces a method of critically mapping AI as an assemblage of social relations. This approach is rooted in the principles of centring the margins and highlighting power structures. Viewing an AI or algorithmic system as a wide-reaching network of social relations, including their impacts on different groups and contexts, enables consideration of how AI is embedded within different discourses and domains of power while emphasising the impact on those most affected. The paper provides a discussion of the critical framing of the project, the principles, processes and templates for mapping AI in this way, and three examples of algorithmic systems in public sector contexts that have been discontinued, are in use now or are being proposed for future use. A discussion of uses and limitations is provided, situating the method beyond a descriptive or analytical tool towards a critical approach to identifying locations for intervention in harmful or unjust uses of algorithmic systems.