From Fair Representation to Just Recognition in Generative AI

Generative AI requires a different fairness framework than predictive systems: Unlike traditional AI that makes specific decisions about resource allocation, large language models are fundamentally about expressing meaning and shaping how people understand the world, making representational concerns central rather than peripheral to fairness.

Accuracy alone cannot solve representation problems: The paper argues that focusing on "descriptive accuracy" as the solution to fair representation has critical limitations—for many social groups, there is no stable definition of what "accurate" representation even means, and accurate representations can still reinforce harmful social patterns.

The real problem is misrecognition, not just misrepresentation: Drawing on political theory, the authors distinguish between getting the facts right about a group and according them proper social status and respect. Even factually accurate portrayals can fail to recognize groups as deserving equal dignity and standing in society.

Current approaches lack clarity on authority and legitimacy: Existing fairness strategies don't adequately address who has the authority to decide what counts as misrepresentation or whose values should guide AI systems, raising questions about whose voices shape how AI represents different groups.

Participatory parity should guide AI governance: The paper proposes that fairness in generative AI should be grounded in ensuring all social groups have genuine voice and standing in decisions about how they are represented, rather than relying on technical fixes alone.