Tag
The paper reformulates AI alignment as a social choice problem using linear optimization over a convex impact space, applying welfare economics and mechanism design to derive alignment protocols. It empirically illustrates welfare implications using real human preferences in scenarios like kidney allocation and trolley problems.
This position paper argues that large language models should learn from personalized rather than aggregated human preferences, highlighting theoretical limitations from social choice theory and practical issues from demographic diversity. It proposes bounded personalization frameworks that respect individual autonomy while maintaining universal safety constraints.
Introduces a stratified framework to identify the minimal aggregated preference information needed to compute disagreement measures, proposing the plurality matrix and showing that pairwise comparisons are insufficient; designs elicitation protocols that trade off participant numbers and cognitive load.
This paper introduces a new embedding model designed to capture preferential similarity rather than just semantic similarity, improving preference prediction for collective decision-making systems.