Jackpot! Alignment as a Maximal Lottery

TL;DR

This paper proposes using maximal lotteries, a probabilistic social choice rule, as a new approach to align large language models with human values, addressing limitations of existing reinforcement learning methods from human feedback.

Contribution

It introduces maximal lotteries as an alignment mechanism and demonstrates that existing techniques like Nash Learning from Human Feedback approximate these outcomes, improving alignment robustness.

Findings

Handles preference majority better than RLHF

Provides principled handling of non-transitivities

Increases robustness to irrelevant alternatives

Abstract

Reinforcement Learning from Human Feedback (RLHF), the standard for aligning Large Language Models (LLMs) with human values, is known to fail to satisfy properties that are intuitively desirable, such as respecting the preferences of the majority \cite{ge2024axioms}. To overcome these issues, we propose the use of a probabilistic Social Choice rule called \emph{maximal lotteries} as a replacement for RLHF. We show that a family of alignment techniques, namely Nash Learning from Human Feedback (NLHF) \cite{munos2023nash} and variants, approximate maximal lottery outcomes and thus inherit its beneficial properties. We confirm experimentally that our proposed methodology handles situations that arise when working with preferences more robustly than standard RLHF, including supporting the preferences of the majority, providing principled ways of handling non-transitivities in the preference data, and robustness to irrelevant alternatives. This results in systems that better incorporate human values and respect human intentions.