Pairwise Calibrated Rewards for Pluralistic Alignment

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Halpern, Daniel, Micha, Evi, Procaccia, Ariel D., Shapira, Itai
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908397353304064
author Halpern, Daniel
Micha, Evi
Procaccia, Ariel D.
Shapira, Itai
author_facet Halpern, Daniel
Micha, Evi
Procaccia, Ariel D.
Shapira, Itai
contents Current alignment pipelines presume a single, universal notion of desirable behavior. However, human preferences often diverge across users, contexts, and cultures. As a result, disagreement collapses into the majority signal and minority perspectives are discounted. To address this, we propose reflecting diverse human preferences through a distribution over multiple reward functions, each inducing a distinct aligned policy. The distribution is learned directly from pairwise preference without annotator identifiers or predefined groups. Instead, annotator disagreements are treated as informative soft labels. Our central criterion is pairwise calibration: for every pair of candidate responses, the proportion of reward functions preferring one response matches the fraction of annotators with that preference. We prove that even a small outlier-free ensemble can accurately represent diverse preference distributions. Empirically, we introduce and validate a practical training heuristic to learn such ensembles, and demonstrate its effectiveness through improved calibration, implying a more faithful representation of pluralistic values.
format Preprint
id arxiv_https___arxiv_org_abs_2506_06298
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Pairwise Calibrated Rewards for Pluralistic Alignment
Halpern, Daniel
Micha, Evi
Procaccia, Ariel D.
Shapira, Itai
Machine Learning
Artificial Intelligence
Current alignment pipelines presume a single, universal notion of desirable behavior. However, human preferences often diverge across users, contexts, and cultures. As a result, disagreement collapses into the majority signal and minority perspectives are discounted. To address this, we propose reflecting diverse human preferences through a distribution over multiple reward functions, each inducing a distinct aligned policy. The distribution is learned directly from pairwise preference without annotator identifiers or predefined groups. Instead, annotator disagreements are treated as informative soft labels. Our central criterion is pairwise calibration: for every pair of candidate responses, the proportion of reward functions preferring one response matches the fraction of annotators with that preference. We prove that even a small outlier-free ensemble can accurately represent diverse preference distributions. Empirically, we introduce and validate a practical training heuristic to learn such ensembles, and demonstrate its effectiveness through improved calibration, implying a more faithful representation of pluralistic values.
title Pairwise Calibrated Rewards for Pluralistic Alignment
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2506.06298