Computational Authority in Platform Society: Dimensions of Power in Machine Learning

A growing body of literature maintains that AI and machine learning (ML) technologies have—or exert—some kind of power: power to act in ways that do not mechanically reflect developers’ intentions and values, as well as power over the individuals, social groups, and cultural manifestations subjected to automated predictions and classifications. This chapter aims to dissect the notion of ML power, intended both as power to and power over. First, two complementary scholarly perspectives on ML power to, here labeled “networked” and “cultural” views, are presented and critically discussed. Second, drawing on both classic and contemporary social theories, the chapter outlines four main dimensions of ML power over—as opaque coercion, computational authority, structural conditioning, and data-driven governmentality.