arXiv 2410.236302024-10-31cs.LGcs.AI

自适应对齐：面向多元AI的基于多目标强化学习的动态偏好调整

Adaptive Alignment: Dynamic Preference Adjustments via Multi-Objective Reinforcement Learning for Pluralistic AI

Deakin University（迪肯大学）；Federation University（联邦大学）；CSIRO（澳大利亚联邦科学与工业研究组织）；University of New South Wales（新南威尔士大学）

Hadassah Harland, Richard Dazeley, Peter Vamplew, Hashini Senaratne, Bahareh Nakisa, Francisco Cruz

备注：: Accepted for the Pluralistic Alignment workshop at NeurIPS 2024

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

针对多元AI对齐中适配多样动态用户偏好的问题，提出基于多目标强化学习的学习后策略选择调整的动态对齐框架，分析其优势、假设与实现细节，并探讨该回溯对齐方法的社会技术影响。

摘要：

多元人工智能（Pluralistic Artificial Intelligence，AI）对齐领域的新兴研究旨在解决如何依据多样化的人类需求与价值观设计并部署智能系统的问题。我们为此方向提出了一种动态对齐方法，通过多目标强化学习（Multi Objective Reinforcement Learning, MORL），借助学习后策略选择调整，让AI适配多样且不断变化的用户偏好。本文介绍了该方法的拟议框架，概述了其预期优势与假设，并讨论了实现层面的技术细节。我们还从社会技术系统视角，探讨了采用回溯式对齐方法的更广泛影响。

英文摘要：

Emerging research in Pluralistic Artificial Intelligence (AI) alignment seeks to address how intelligent systems can be designed and deployed in accordance with diverse human needs and values. We contribute to this pursuit with a dynamic approach for aligning AI with diverse and shifting user preferences through Multi Objective Reinforcement Learning (MORL), via post-learning policy selection adjustment. In this paper, we introduce the proposed framework for this approach, outline its anticipated advantages and assumptions, and discuss technical details about the implementation. We also examine the broader implications of adopting a retroactive alignment approach through the sociotechnical systems perspective.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。