AI Alignment from Social Choice Perspectives

Alignment from human feedback uses human judgments about model outputs to steer how language models behave after pretraining. When those judgments reflect conflicting views of desirable behavior, the learned objective becomes an aggregate judgment about what the model should prefer. Recent work has studied this aggregation through the lens of social choice theory. The goal of this paper is to survey applications of this perspective to alignment from human feedback. Taken together, this body of work reframes alignment as collective decision-making under disagreement, uses that framing to identify failure modes in the feedback aggregation layer, and highlights a broader design space for handling disagreement in more explicit and principled ways.