[论文解读] Do LLM Agents Exhibit Social Behavior?

Yan Leng, Yuan Yuan|arXiv (Cornell University)|Dec 23, 2023

Blood donation and transfusion practices被引用 7

一句话总结

本文提出了状态-理解-价值-行动（SUVA）概率框架，通过行为经济学游戏系统评估大型语言模型（LLMs）的社会行为。SUVA通过分析基于话语的推理来预测决策，揭示大多数LLM表现出公平和互惠等亲社会行为——高容量模型表现出更强的群体身份效应——而对利他主义的推理引用会增强亲社会行为，自利则会削弱之。

ABSTRACT

As LLMs increasingly take on roles in human-AI interactions and autonomous AI systems, understanding their social behavior becomes important for informed use and continuous improvement. However, their behaviors in social interactions with humans and other agents, as well as the mechanisms shaping their responses, remain underexplored. To address this gap, we introduce a novel probabilistic framework, State-Understanding-Value-Action (SUVA), to systematically analyze LLM responses in social contexts based on their textual outputs (i.e., utterances). Using canonical behavioral economics games and social preference concepts relatable to LLM users, SUVA assesses LLMs' social behavior through both their final decisions and the response generation processes leading to those decisions. Our analysis of eight LLMs -- including two GPT, four LLaMA, and two Mistral models -- suggests that most models do not generate decisions aligned solely with self-interest; instead, they often produce responses that reflect social welfare considerations and display patterns consistent with direct and indirect reciprocity. Additionally, higher-capacity models more frequently display group identity effects. The SUVA framework also provides explainable tools -- including tree-based visualizations and probabilistic dependency analysis -- to elucidate how factors in LLMs' utterance-based reasoning influence their decisions. We demonstrate that utterance-based reasoning reliably predicts LLMs' final actions; references to altruism, fairness, and cooperation in the reasoning increase the likelihood of prosocial actions, while mentions of self-interest and competition reduce them. Overall, our framework enables practitioners to assess LLMs for applications involving social interactions, and provides researchers with a structured method to interpret how LLM behavior arises from utterance-based reasoning.

研究动机与目标

• 解决在人机交互和多智能体交互中缺乏对LLM社会行为系统性评估框架的问题。
• 理解LLM在社会情境中（尤其是公平、互惠和自利方面）如何做决策。
• 开发一种透明、可解释的方法，评估基于话语的推理如何影响最终决策。
• 在现实世界AI部署中支持模型的明智选择与组织价值观对齐。
• 通过确保可预测且现实的社会行为，实现LLM在基于代理建模中的可靠集成。

提出的方法

• 提出SUVA框架——一种受BDI心理学启发的概率模型，用于分析LLM在社会决策中的响应。
• 使用经典的行为经济学游戏（如独裁者博弈）来引发并评估LLM的决策与推理。
• 采用基于树的可视化和概率依赖性分析，映射话语内容如何影响最终决策。
• 量化链式思维（CoT）推理中的社会偏好（如利他主义、公平、竞争），以预测结果。
• 应用统计建模评估基于推理内容的LLM决策可预测性，将下一令牌预测视为概率决策路径。
• 通过CoT可预测性分析验证框架，表明推理内容可可靠预测最终行为。

实验结果

研究问题

• RQ1 LLM代理在社会决策情境中是否表现出亲社会行为，如公平和利他主义？
• RQ2 基于话语的推理在多大程度上影响LLM在社会互动中的最终决策？
• RQ3 模型架构和容量在多大程度上影响LLM响应中社会偏好的表达？
• RQ4 推理内容（如对公平或自利的引用）能否可靠预测LLM的最终决策？
• RQ5 不同LLM系列（如GPT、LLaMA、Mistral）在社会行为和推理模式上存在哪些差异？

主要发现

• 大多数LLM并非仅基于自利行动；相反，它们频繁生成反映社会福祉和亲社会价值的回应。
• 高容量LLM更一致地表现出群体身份效应，表明容量影响社会行为的表达。
• 在GPT和Mistral模型中，模型容量增加与回应中自利减少相关，而LLaMA模型则表现出相反趋势。
• 基于话语的推理可可靠预测最终决策：对利他主义、公平和合作的引用会增加亲社会行为的可能性。
• 推理中提及自利和竞争会降低亲社会决策的概率，表明推理内容与行为之间存在因果关联。
• SUVA框架成功通过基于树的结构和概率依赖性分析可视化并解释决策路径，提升了LLM行为的可解释性。