Clone-Robust AI Alignment

Ariel D. Procaccia, Benjamin Schiffer, Shirley Zhang
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:49903-49926, 2025.

Abstract

A key challenge in training Large Language Models (LLMs) is properly aligning them with human preferences. Reinforcement Learning with Human Feedback (RLHF) uses pairwise comparisons from human annotators to train reward functions and has emerged as a popular alignment method. However, input datasets in RLHF can be unbalanced due to adversarial manipulation or inadvertent repetition. Therefore, we want RLHF algorithms to perform well even when the set of alternatives is not uniformly distributed. Drawing on insights from social choice theory, we introduce robustness to approximate clones, a desirable property of RLHF algorithms which requires that adding near-duplicate alternatives does not significantly change the learned reward function. We first demonstrate that the standard RLHF algorithm based on regularized maximum likelihood estimation (MLE) fails to satisfy this property. We then propose the weighted MLE, a new RLHF algorithm that modifies the standard regularized MLE by weighting alternatives based on their similarity to other alternatives. This new algorithm guarantees robustness to approximate clones while preserving desirable theoretical properties.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-procaccia25a, title = {Clone-Robust {AI} Alignment}, author = {Procaccia, Ariel D. and Schiffer, Benjamin and Zhang, Shirley}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {49903--49926}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/procaccia25a/procaccia25a.pdf}, url = {https://proceedings.mlr.press/v267/procaccia25a.html}, abstract = {A key challenge in training Large Language Models (LLMs) is properly aligning them with human preferences. Reinforcement Learning with Human Feedback (RLHF) uses pairwise comparisons from human annotators to train reward functions and has emerged as a popular alignment method. However, input datasets in RLHF can be unbalanced due to adversarial manipulation or inadvertent repetition. Therefore, we want RLHF algorithms to perform well even when the set of alternatives is not uniformly distributed. Drawing on insights from social choice theory, we introduce robustness to approximate clones, a desirable property of RLHF algorithms which requires that adding near-duplicate alternatives does not significantly change the learned reward function. We first demonstrate that the standard RLHF algorithm based on regularized maximum likelihood estimation (MLE) fails to satisfy this property. We then propose the weighted MLE, a new RLHF algorithm that modifies the standard regularized MLE by weighting alternatives based on their similarity to other alternatives. This new algorithm guarantees robustness to approximate clones while preserving desirable theoretical properties.} }
Endnote
%0 Conference Paper %T Clone-Robust AI Alignment %A Ariel D. Procaccia %A Benjamin Schiffer %A Shirley Zhang %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-procaccia25a %I PMLR %P 49903--49926 %U https://proceedings.mlr.press/v267/procaccia25a.html %V 267 %X A key challenge in training Large Language Models (LLMs) is properly aligning them with human preferences. Reinforcement Learning with Human Feedback (RLHF) uses pairwise comparisons from human annotators to train reward functions and has emerged as a popular alignment method. However, input datasets in RLHF can be unbalanced due to adversarial manipulation or inadvertent repetition. Therefore, we want RLHF algorithms to perform well even when the set of alternatives is not uniformly distributed. Drawing on insights from social choice theory, we introduce robustness to approximate clones, a desirable property of RLHF algorithms which requires that adding near-duplicate alternatives does not significantly change the learned reward function. We first demonstrate that the standard RLHF algorithm based on regularized maximum likelihood estimation (MLE) fails to satisfy this property. We then propose the weighted MLE, a new RLHF algorithm that modifies the standard regularized MLE by weighting alternatives based on their similarity to other alternatives. This new algorithm guarantees robustness to approximate clones while preserving desirable theoretical properties.
APA
Procaccia, A.D., Schiffer, B. & Zhang, S.. (2025). Clone-Robust AI Alignment. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:49903-49926 Available from https://proceedings.mlr.press/v267/procaccia25a.html.

Related Material