From Fair Representation to Just Recognition in Generative AI

Abstract

The fair AI/ML literature has long recognized two distinct normative issues raised by increasing automation: issues of distributive fairness, on the one hand, and issues of so-called “representational” fairness on the other. The former—distributive issues—pertain to challenges automated systems create for the distribution of resources and opportunities across individuals and social groups. The latter—representational issues—refer to challenges automated systems create for the way individuals and social groups are perceived, understood, and accorded social status. While distributive issues have attracted more attention from researchers and advocates, the rise of generative AI—fundamentally expressive systems whose ultimate function is to convey meaning rather than to automate domain-specific decision procedures—has made representational issues increasingly urgent. Yet, as many have pointed out, we lack adequate normative theories for diagnosing and addressing them. These problems are also being discussed within the rapidly developing field of value alignment, particularly through research on what and whose values and perspectives AI systems should represent. These discussions frequently appeal to standards of descriptive accuracy, asking whether generative AI systems accurately depict specific social groups. This strategy, we argue, falls short on several counts. For many (if not most) social groups, there are no stable or bounded referents against which to judge claims of representational inaccuracy. It is unclear who has the authority to decide what counts as misrepresenting any particular group. And even accurate representations can reproduce harmful patterns that are detrimental to group outcomes. The problem, at bottom, is not misrepresentation; the problem, we argue, is misrecognition. Drawing from relevant work in political theory—especially Nancy Fraser’s account of participatory parity—we show how moving from “representational fairness” to “recognitional justice” equips us with better conceptual and normative tools for understanding and addressing these challenges.