Emergence of Scale-Free Networks in Social Interactions among Large Language Models
Giordano De Marzo1,2,3,41234{}^{1,2,3,4}start_FLOATSUPERSCRIPT 1 , 2 , 3 , 4 end_FLOATSUPERSCRIPT, Luciano Pietronero11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPT and David Garcia2,525{}^{2,5}start_FLOATSUPERSCRIPT 2 , 5 end_FLOATSUPERSCRIPT 11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPTCentro Ricerche Enrico Fermi, Piazza del Viminale, 1, I-00184 Rome, Italy. 22{}^{2}start_FLOATSUPERSCRIPT 2 end_FLOATSUPERSCRIPTComplexity Science Hub Vienna, Josefstaedter Strasse 39, 1080, Vienna, Austria. 33{}^{3}start_FLOATSUPERSCRIPT 3 end_FLOATSUPERSCRIPTDipartimento di Fisica Università “Sapienza”, P.le A. Moro, 2, I-00185 Rome, Italy. 44{}^{4}start_FLOATSUPERSCRIPT 4 end_FLOATSUPERSCRIPTSapienza School for Advanced Studies, “Sapienza”, P.le A. Moro, 2, I-00185 Rome, Italy. 55{}^{5}start_FLOATSUPERSCRIPT 5 end_FLOATSUPERSCRIPTUniversity of Konstanz, Universitätstraße 10, 78457 Konstanz, Germany
Abstract
Scale-free networks are one of the most famous examples of emergent behavior and are ubiquitous in social systems, especially online social media in which users can follow each other. By analyzing the interactions of multiple generative agents using GPT3.5-turbo as a language model, we demonstrate their ability to not only mimic individual human linguistic behavior but also exhibit collective phenomena intrinsic to human societies, in particular the emergence of scale-free networks. We discovered that this process is disrupted by a skewed token prior distribution of GPT3.5-turbo, which can lead to networks with extreme centralization as a kind of alignment. We show how renaming agents removes these token priors and allows the model to generate a range of networks from random networks to more realistic scale-free networks.
中文速览
大规模语言模型（LLM）能否像真实人类一样，在群体互动中自发涌现出复杂的社会网络结构？研究者用GPT-3.5-turbo驱动多个生成式智能体（generative agents）模拟在线社交网络的增长过程，让新加入的智能体自主选择关注对象，观察整体网络的拓扑形态。实验发现，当智能体能看到其他节点的连接数且对名称标记（token）的先验偏好被消除后，网络会自发涌现出无标度网络（scale-free network）特征，度分布符合幂律，幂指数约为1.93，与真实在线社交网络高度吻合；而未消除先验偏好时，模型对特定字符串的偏好会造成极端的"赢家通吃"轮辐式网络，这实际上是一种模型对齐（alignment）带来的失真。通过在每轮迭代中随机重命名智能体来消除这一标记先验，研究者成功复现了从随机网络到无标度网络的完整谱系，证明LLM驱动的多智能体系统不仅能模仿个体语言行为，还能在集体层面涌现出人类社会固有的网络结构，这对理解AI社交生态系统的潜在风险与构建更真实的基于智能体的社会模型具有重要意义。
原文 arXiv:2312.06619;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2312.06619v1