【作者简介】刘河庆，华中科技大学社会学院

【文章来源】《社会学研究》2026年第2期

【内容提要】大语言模型作为准社会主体融入社会生活，正在重塑社会互动模式与社会秩序。基于人工智能社会化视角，本研究使用四个大语言模型模拟不同角色对“知乎”平台上的海量提问进行回答，进而对大语言模型的行为特征及其对内容生态的冲击等进行分析。研究发现，大语言模型在互动中兼具精确性与机械性：它们能灵活识别场景，但在输出中会机械地强化性别、年龄等方面的角色刻板印象。同时，大语言模型生成的内容语义高度集中，趋近于中立或高点赞的人类回答，进而可能引发社会符号系统的收缩和同质化。研究人工智能参与社会互动及内容生成的过程，为理解数智时代社会形态的变迁提供了重要视角。

【关键词】人工智能社会化；准社会主体；社会符号系统；内容生成；大语言模型

【项目基金】本文为国家社会科学基金一般项目“数智社会大语言模型的知识生产模式及其社会影响研究”（24BSH024）的阶段性成果。

【全文链接】https://shxyj.ajcass.com/Magazine/show/?id=122488

AI Socialization: Content Generation and the Reconfiguration of Symbols

Abstract: As “ quasi-social agents”, large language models (LLMs) are integrating into social life, reshaping social interaction patterns and social order. From the perspective of AI socialization, this study employs four LLMs to simulate different Social roles n answeringa vast number of questions on the Zhihu platform, thereby analyzing the behavioral characteristics of large models and their impact on content ecosystems. It finds that LLM exhibit both precision and mechancal rigidity in interactions. They are capable of flexibly identifying contexts, yet they tend to mechanically reinforce stereotypes related to gender, age, and other roles in their outputs. At the same time, LLM-generated content exhibits semantic convergence, clustering around generic or highly upvoted responses, which may lead to a contraction and homogenization of social symbolic systems.The involvement of AI socialization in social interactions and the process of content generation offers a crucial perspective for understanding the transformation of social structures in the digital-intelligence era.