生成式人工智能实践下的社会观念分化探析--基于自主行动者建模的仿真模拟实验研究  

Differentiation of Societal Perspectives in the Social Context of Generative Artificial Intelligence:A Simulation Experiment Research Using Agent⁃Based Modeling

作　　者：梁玉成[1] 马昱堃 LIANG Yucheng;MA Yukun

机构地区：[1]中山大学社会学与人类学学院,广东广州510275

出　　处：《济南大学学报（社会科学版）》2024年第6期120-132,193,共14页Journal of University of Jinan:Social Science Edition

基　　金：国家社会科学基金重大项目“基于大型调查数据基础上中国城镇社区结构异质性及其基层治理研究”(项目编号:15ZDB172)。

摘　　要：当前,以大语言模型为代表的生成式人工智能在短时间内迅速发展,其对社会观念分化产生了影响。采用自主行动者建模方法,借鉴文化传播模型,研究大模型对社会观念分化的影响。实验结果表明,大语言模型对不同观念之间的交流和讨论具有激活作用;不同大语言模型之间的内在观念差异越大,社会观念越难以形成稳定的分化格局,并且分化程度会逐渐加剧;此外,在已经形成了一定观念分化格局下,大语言模型可能会进一步减少社会观念之间的差异,有助于缓解社会观念“多中心化”。研究表明,减小不同生成式人工智能输出结果的观念差异,能够有效降低社会观念的“多中心化”、遏制局部“信息茧房”的自我强化。长远来看,大语言模型存在造成“整体茧房”的可能性。Currently,generative artificial intelligence(GAI)represented by large language models(LLMs)has seen rapid advancements with tangible impact on societal perspective differentiation.This study explores the societal implications of these technologies using agent-based modeling with reference to cultural dissemination models.Results show that LLMs can stimulate discourse and interaction among diverse societal perspectives.Greater heterogeneity in the inherent perspectives of LLMs correlates with increased instability in societal perspective differentiation patterns,with these divergences amplifying over time.Additionally,in environments where societal perspective differentiation patterns are already established,LLMs tend to reduce the heterogeneity of societal perspectives,thereby mitigating poly-centrification.The findings indicate that decreasing the heterogeneity of inherent perspectives in GAI effectively curbs societal perspective poly-centrification and inhibits the self-reinforcement of local information cocoons.However,from a long-term perspective,there may be a potential emergence of a comprehensive information cocoon under the influence of LLMs.

关 键 词：大语言模型 ChatGPT 生成式人工智能 社会观念分化 自主行动者建模 

分 类 号：C912.6[经济管理]