arXiv 2601.054372026-05-19cs.CLcs.AI

在大型语言模型中追溯道德基础

Tracing Moral Foundations in Large Language Models

Department of Computer Science, University of Southern California（南加州大学计算机科学系）；Department of Psychology, University of Southern California（南加州大学心理学系）；Center for Computational Language Sciences, University of Southern California（南加州大学计算语言科学中心）

Chenxiao Yu, Bowen Yi, Farzan Karimi-Malekabadi, Suhaib Abdurahman, Jinyi Ye, Shrikanth Narayanan, Yue Zhao, Morteza Dehghani

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文研究了大型语言模型中道德基础的编码、组织和表达，通过多层方法分析道德基础与人类道德感知的一致性，并发现道德结构在预训练和微调过程中自然形成，且部分解耦。

摘要：

大型语言模型常常产生类似人类的道德判断，但不清楚这种表现是内部概念结构还是表面的'道德模仿'。使用道德基础理论（MFT）作为分析框架，我们研究了14个基础和指令微调的LLM在四个模型家族（Llama、Qwen2.5、Qwen3-MoE、Mistral）和从7B到70B的不同规模上如何编码、组织和表达道德基础。我们采用多级方法结合（i）逐层分析MFT概念表示及其与人类道德感知的一致性，（ii）在残差流上预训练稀疏自编码器（SAEs）以识别支持道德概念的稀疏特征，以及（iii）使用密集MFT向量和稀疏SAE特征进行因果引导干预。我们发现模型在表示和区分道德基础方面与人类判断一致，且这种道德几何结构自然从预训练中产生，并在微调中被选择性重 wiring。在更细的尺度上，SAE特征显示出与特定基础的明确语义联系，表明在共享表示中存在部分解耦的机制。最后，沿着密集向量或稀疏特征引导会产生可预测的在基础相关行为上的变化，证明了内部表示与道德输出之间的因果联系。共同，我们的结果提供了机械证据，表明LLM中的道德概念是分布的、分层的且部分解耦的，暗示了多元道德结构可以从语言的统计规律中作为潜在模式出现。

英文摘要：

Large language models often produce human-like moral judgments, but it is unclear whether this reflects an internal conceptual structure or superficial ``moral mimicry.'' Using Moral Foundations Theory (MFT) as an analytic framework, we study how moral foundations are encoded, organized, and expressed across 14 base and instruction-tuned LLMs spanning four model families (Llama, Qwen2.5, Qwen3-MoE, Mistral) and scales from 7B to 70B. We employ a multi-level approach combining (i) layer-wise analysis of MFT concept representations and their alignment with human moral perceptions, (ii) pretrained sparse autoencoders (SAEs) over the residual stream to identify sparse features that support moral concepts, and (iii) causal steering interventions using dense MFT vectors and sparse SAE features. We find that models represent and distinguish moral foundations in a manner that aligns with human judgments, and that this moral geometry naturally emerges from pretraining and is selectively rewired by post-training. At a finer scale, SAE features show clear semantic links to specific foundations, suggesting partially disentangled mechanisms within shared representations. Finally, steering along either dense vectors or sparse features produces predictable shifts in foundation-relevant behavior, demonstrating a causal connection between internal representations and moral outputs. Together, our results provide mechanistic evidence that moral concepts in LLMs are distributed, layered, and partly disentangled, suggesting that pluralistic moral structure can emerge as a latent pattern from the statistical regularities of language alone.

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。