基于大语言模型智能体的社会认知模拟

doi:10.16182/j.issn1004731x.joss.25-0612

摘要/Abstract

摘要：

随着生成式LLM能力的持续演进，其在社会认知模拟中的应用逐渐展现出范式级变革的潜力。传统社会仿真方法多依赖静态规则和简化行为模型，难以捕捉人类社会行为的动态演化和文化复杂性。基于LLM驱动的智能体因具备上下文理解与自然语言生成能力，成为建模社会认知机制的新兴工具，能够模拟诸如身份建构、价值判断、意图推理等复杂社会心理过程。简要介绍LLM的技术基础，指出使用LLM智能体与社会认知模拟的适配性；构建了一个涵盖属性建模、记忆管理、规划与行动的智能体社会认知建模框架；在模拟流程层面，提出“数据收集-智能体协同-多维评估”的技术链条，并探讨认知可解释性与模拟现实对齐等挑战；总结了当前在社会学、经济学、军事学等方面的应用进展，并讨论LLM社会认知模拟的前沿趋势与未来发展方向。

关键词: 大语言模型, 社会认知模拟, 智能体建模, 模拟-现实对齐

Abstract:

With the continuous evolution of the capabilities of generative LLMs, their application in social cognition simulation is demonstrating paradigm-shifting potential. Traditional social simulation methods predominantly rely on static rules and simplified behavioral models, making it difficult to capture the dynamic evolution and cultural complexity of human social behavior. LLM-driven agents, equipped with contextual understanding and natural language generation capabilities, are emerging as novel tools for modeling social cognitive mechanisms, enabling the simulation of complex socio-psychological processes such as identity construction, value judgment, and intentional reasoning. This paper briefly introduced the technical foundations of LLMs and highlighted their suitability for social cognition simulation. It constructed a framework for agent-based social cognition modeling, encompassing attribute modeling, memory management, planning, and action. At the simulation process level, the paper proposed a technical pipeline consisting of “data collection, agent collaboration, and multidimensional evaluation,” while delving into challenges such as cognitive interpretability and simulation-reality alignment.It summarized the current application progress in fields such as sociology, economics, and military science and discussed emerging trends and future directions for LLM-based social cognition simulation.

Key words: LLM, social cognition simulation, agent-based modeling, simulation-reality alignment

中图分类号:

TP391.9

张明新,伍瑾轩,朱睿等 . 基于大语言模型智能体的社会认知模拟[J]. 系统仿真学报, 2026, 38(2): 261-277.

Zhang Mingxin,Wu Jinxuan,Zhu Rui,et al . Social Cognition Simulation with Large Language Model-driven Agents[J]. Journal of System Simulation, 2026, 38(2): 261-277.

图/表 6

表1

图1

表2

表3

表4

表5

参考文献 71

[1]	Wang Lei, Ma Chen, Feng Xueyang, et al. A Survey on Large Language Model Based Autonomous Agents[J]. Frontiers of Computer Science, 2024, 18(6): 186345.
[2]	Gao Chen, Lan Xiaochong, Li Nian, et al. Large Language Models Empowered Agent-based Modeling and Simulation: A Survey and Perspectives[J]. Humanities and Social Sciences Communications, 2024, 11(1): 1259.
[3]	Yehudai A, Eden L, Li A, et al. Survey on Evaluation of LLM-based Agents[EB/OL]. [2025-03-20]. .
[4]	Masterman T, Besen S, Sawtell M, et al. The Landscape of Emerging AI Agent Architectures for Reasoning, Planning, and tool Calling: A Survey[EB/OL]. (2024-04-17) [2024-06-04]. .
[5]	Jinghua Piao, Yan Yuwei, Zhang Jun, et al. AgentSociety: Large-scale Simulation of LLM-driven Generative Agents Advances Understanding of Human Behaviors and Society[EB/OL]. [2025-02-12] .
[6]	Wang Xintao, Wang Heng, Zhang Yifei, et al. CoSER: Coordinating LLM-based Persona Simulation of Established Roles[EB/OL]. [2025-05-28]. .
[7]	Hu Tiancheng, Collier N. Quantifying the Persona Effect in LLM Simulations[C]//Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics. Stroudsburg: ACL, 2024: 10289-10307.
[8]	Lewis P, Perez E, Piktus A, et al. Retrieval-augmented Generation for Knowledge-intensive NLP Tasks[C]//Proceedings of the 34th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2020: 9459-9474.
[9]	Gamage Gihan, Mills Nishan, Daswin De Silva, et al. Multi-agent RAG Chatbot Architecture for Decision Support in Net-zero Emission Energy Systems[C]//2024 IEEE International Conference on Industrial Technology (ICIT). Piscataway: IEEE, 2024: 1-6.
[10]	Wei J, Wang Xuezhi, Schuurmans D, et al. Chain-of-thought Prompting Elicits Reasoning in Large Language Models[C]//Proceedings of the 36th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2022: 24824-24837.
[11]	Yao Shunyu, Yu Dian, Zhao J, et al. Tree of Thoughts: Deliberate Problem Solving with Large Language Models[C]//Proceedings of the 37th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2023: 11809-11822.
[12]	Guo Daya, Yang Dejian, Zhang Haowei, et al. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning[EB/OL]. [2025-01-22]. .
[13]	Pan Jianfeng, Deng Senyou, Huang Shaomang. CoAT: Chain-of-associated-thoughts Framework for Enhancing Large Language Models Reasoning[EB/OL]. [2025-11-13]. .
[14]	Shen Yongliang, Song Kaitao, Tan Xu, et al. HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in Hugging Face[C]//NeurIPS 2023 Conference. Red Hook: Curran Associates, Inc, 2023: 38154-38180.
[15]	Yao Shunyu, Zhao J, Yu Dian, et al. ReAct: Synergizing Reasoning and Acting in Language Models[C]//ICLR 2023 Conference. New York: ICLR, 2023: 1-33.
[16]	Wang Zekun, Peng Zhongyuan, Que Haoran, et al. RoleLLM: Benchmarking, Eliciting, and Enhancing Role-playing Abilities of Large Language Models[C]//Findings of the Association for Computational Linguistics: ACL 2024. Stroudsburg: ACL, 2024: 14743–14777.
[17]	Xu Rui, Wang Xintao, Chen Jiangjie, et al. Character is Destiny: Can Role-playing Language Agents Make Persona-driven Decisions?[EB/OL]. [2024-11-18]. .
[18]	Tu Quan, Chen Chuanqi, Li Jinpeng, et al. CharacterChat: Learning Towards Conversational AI with Personalized Social Support[EB/OL]. (2023-08-20) [2024-06-04]. .
[19]	Chen Nuo, Wang Yan, Jiang Haiyun, et al. Large Language Models Meet Harry Potter: A Dataset for Aligning Dialogue Agents with Characters[C]//Findings of the Association for Computational Linguistics: EMNLP 2023. Stroudsburg: ACL, 2023: 8506-8520.
[20]	Li Cheng, Leng Ziang, Yan Chenxi, et al. ChatHaruhi: Reviving Anime Character in Reality via Large Language Model[EB/OL]. (2023-08-18) [2024-06-04]. .
[21]	Li Yifan, Fan Shixuan, Wei Wei, et al. Improving Personality Consistency in Conversation with Commonsense Knowledge[J]. IEEE Transactions on Audio Speech and Language Processing, 2025, 33: 262-271.
[22]	Lee S, Peng Taiquan, Goldberg M H, et al. Can Large Language Models Estimate Public Opinion About Global Warming? An Empirical Assessment of Algorithmic Fidelity and Bias[J]. PLoS Climate, 2024, 3(8): e0000429.
[23]	Bisbee J, Clinton J D, Dorff C, et al. Synthetic Replacements for Human Survey Data? The Perils of Large Language Models[J]. Political Analysis, 2024, 32(4): 401-416.
[24]	Qu Yao, Wang Jue. Performance and Biases of Large Language Models in Public Opinion Simulation[J]. Humanities and Social Sciences Communications, 2024, 11(1): 1095.
[25]	Atkinson R C, Shiffrin R M. Human Memory: A Proposed System and its Control Processes[J]. Psychology of Learning and Motivation, 1968, 2: 89-195.
[26]	Ritter S, Wang J, Kurth-Nelson Z, et al. Been There, Done That: Meta-learning with Episodic Recall[C]//Proceedings of the 35th International Conference on Machine Learning. Chia Laguna Resort: PMLR, 2018: 4354-4363.
[27]	Ji Shaoxiong, Pan Shirui, Cambria E, et al. A Survey on Knowledge Graphs: Representation, Acquisition, and Applications[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(2): 494-514.
[28]	周棪忠, 罗俊仁, 谷学强, 等. 大语言模型视角下的智能规划方法综述[J]. 系统仿真学报, 2025, 37(4): 823-844.
	Zhou Yanzhong, Luo Junren, Gu Xueqiang, et al. Survey on Intelligent Planning Methods from Large Language Models Perspective[J]. Journal of System Simulation, 2025, 37(4): 823-844.
[29]	Wang Lei, Xu Wanyu, Lan Yihuai, et al. Plan-and-solve Prompting: Improving Zero-shot Chain-of-thought Reasoning by Large Language Models[C]//Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics. Stroudsburg: ACL, 2023: 2609-2634.
[30]	Sun Linzhuang, Dong Yao, Xu Nan, et al. Rational Sensibility: LLM Enhanced Empathetic Response Generation Guided by Self-presentation Theory[EB/OL]. [2024-08-23]. .
[31]	Cox M T. Metacognition in Computation: A Selected Research Review[J]. Artificial Intelligence, 2005, 169(2): 104-141.
[32]	Light J, Cai Min, Shen Sheng, et al. AvalonBench: Evaluating LLMs Playing the Game of Avalon[C]//NeurIPS 2023 Workshop FMDM. Red Hook: Curran Associates Inc., 2023: 1-28.
[33]	Lin Jiaju, Zhao Haoran, Zhang Aochi, et al. AgentSims: An Open-source Sandbox for Large Language Model Evaluation[EB/OL]. (2023-08-08) [2024-06-04]. .
[34]	Wang Guanzhi, Xie Yuqi, Jiang Yunfan, et al. Voyager: An Open-ended Embodied Agent with Large Language Models[C]//NeurIPS 2023 Workshop IMOL. Red Hook: Curran Associates Inc., 2023: 1-43.
[35]	Minsky M. The Society of Mind[M]. New York: Simon & Schuster, 1986.
[36]	Chen Weize, Su Yusheng, Zuo Jingwei, et al. AgentVerse: Facilitating Multi-agent Collaboration and Exploring Emergent Behaviors[EB/OL]. [2023-10-23]. .
[37]	Elliott E, Kiel L D. Exploring Cooperation and Competition Using Agent-based Modeling[J]. Proceedings of the National Academy of Sciences of the United States of America, 2002, 99(S3): 7193-7194.
[38]	Dai G, Zhang Weijia, Li Jinhan, et al. Artificial Leviathan: Exploring Social Evolution of LLM Agents Through the Lens of Hobbesian Social Contract Theory[EB/OL]. [2024-07-01]. .
[39]	Wang Junqi, Zhang Chunhui, Li Jiapeng, et al. Evaluating and Modeling Social Intelligence: A Comparative Study of Human and AI Capabilities[EB/OL]. (2024-05-20) [2024-06-04]. .
[40]	Liu Xuan, Zhang Jie, Shang Haoyang, et al. Exploring Prosocial Irrationality for LLM Agents: A Social Cognition View[C]//ICLR 2025 Conference. New York: ICLR, 2025: 1-30.
[41]	Ren Siyue, Cui Zhiyao, Song Ruiqi, et al. Emergence of Social Norms in Generative Agent Societies: Principles and Architecture[C]//Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence. California: IJCAI, 2024: 7895-7903.
[42]	Park J S, O'Brien J, Cai Jun, et al. Generative Agents: Interactive Simulacra of Human Behavior[C]//Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology. New York: ACM, 2023: 2.
[43]	Zhao Kaiya, Naim M, Kondic J, et al. Lyfe Agents: Generative Agents for Low-cost Real-time Social Interactions[C]//ICLR 2024 Conference. New York: ICLR, 2024: 1-31.
[44]	Gyevnár Bálint, Lucas C G, Albrecht S V, et al. Integrating Counterfactual Simulations with Language Models for Explaining Multi-agent Behaviour[EB/OL]. [2025-10-28]. .
[45]	Yang Hailong, Gu Mingxian, Zhao Renhuo, et al. XAgents: A Framework for Interpretable Rule-based Multi-agents Cooperation[EB/OL]. [2025-09-22]. .
[46]	Ju Tianjie, Wang Bowen, Fei Hao, et al. When Disagreements Elicit Robustness: Investigating Self-repair Capabilities Under LLM Multi-Agent Disagreements[EB/OL]. [2025-10-02]. .
[47]	Joon Sung Park, Q Zou Carolyn, Shaw Aaron, et al. Simulating Human Behavior with AI Agents[EB/OL]. [2025-05-20]. .
[48]	Ashery A F, Aiello L M, Baronchelli A. Emergent Social Conventions and Collective Bias in LLM Populations[J]. Science Advances, 2025, 11(20): eadu9368.
[49]	Macy M W, Willer R. From Factors to Actors: Computational Sociology and Agent-based Modeling[J]. Annual Review of Sociology, 2002, 28: 143-166.
[50]	Cui Jiaxi, Ning Munan, Li Zongjia, et al. Chatlaw: A Multi-agent Collaborative Legal Assistant with Knowledge Graph Enhanced Mixture-of-experts Large Language Model[EB/OL]. (2024-05-30) [2024-06-04]. .
[51]	Fu Yao, Peng Haochun, Khot T, et al. Improving Language Model Negotiation with Self-play and In-context Learning from AI Feedback[C]//NeurIPS 2023 Conference. Red Hook: Curran Associates Inc., 2023: 1-11.
[52]	Akata E, Schulz L, Coda-Forno J, et al. Playing Repeated Games with Large Language Models[C]//ICLR 2024 Conference. New York: ICLR, 2024: 1-18.
[53]	Guo Jiaxian, Yang Bo, Yoo P, et al. Suspicion-agent: Playing Imperfect Information Games with Theory of Mind Aware GPT-4[C]//ICLR 2024 Conference. New York: ICLR, 2024: 1-27.
[54]	Wu Zengqing, Peng Run, Zheng Shuyuan, et al. Shall We Team Up: Exploring Spontaneous Cooperation of Competing LLM Agents[C]//Findings of the Association for Computational Linguistics: EMNLP 2024. Stroudsburg: ACL, 2024: 5163-5186.
[55]	El-Sayed A M, Scarborough P, Seemann L, et al. Social Network Analysis and Agent-based Modeling in Social Epidemiology[J]. Epidemiologic Perspectives & Innovations, 2012, 9(1): 1.
[56]	Gao Chen, Lan Xiaochong, Lu Zhihong, et al. S3: Social-network Simulation System with Large Language Model-empowered Agents[EB/OL]. [2025-06-3]. .
[57]	Gur I, Furuta H, Huang A V, et al. A Real-world WebAgent with Planning, Long Context Understanding, and Program Synthesis[C]//ICLR 2024 Conference. New York: ICLR, 2024: 1-28.
[58]	Nasrinpour H R, Friesen M R, McLeod R D. An Agent-based Model of Message Propagation in the Facebook Electronic Social Network[EB/OL]. (2016-11-22) [2024-06-04]. .
[59]	Liu Yuhan, Chen Xiuying, Zhang Xiaoqing, et al. from Skepticism to Acceptance: Simulating the Attitude Dynamics Toward Fake News[C]//Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence. California: IJCAI, 2024: 7886-7894.
[60]	Hou Yupeng, Zhang Junjie, Lin Zihan, et al. Large Language Models are Zero-shot Rankers for Recommender Systems[C]//Advances in Information Retrieval. Cham: Springer Nature Switzerland, 2024: 364-381.
[61]	Qian Chen, Liu Wei, Liu Hongzhang, et al. ChatDev: Communicative Agents for Software Development[C]//Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics. Stroudsburg: ACL, 2024: 15174-15186.
[62]	Zhang An, Chen Yuxin, Sheng Leheng, et al. On Generative Agents in Recommendation[C]//Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: ACM, 2024: 1807-1817.
[63]	Huang Xu, Lian Jianxun, Lei Yuxuan, et al. Recommender AI Agent: Integrating Large Language Models for Interactive Recommendations[EB/OL]. (2024-01-30) [2024-06-04]. .
[64]	Domenico Delli Gatti, Desiderio Saul, Gaffeo Edoardo, et al. Macroeconomics from the Bottom-up[M]. Milano: Springer Milan, 2011.
[65]	Zhao Qinlin, Wang Jindong, Zhang Yixuan, et al. CompeteAI: Understanding the Competition Dynamics of Large Language Model-based Agents[C]//Proceedings of the 41st International Conference on Machine Learning. Chia Laguna Resort: PMLR, 2024: 61092-61107.
[66]	Nascimento Nathalia, Alencar Paulo, Cowan Donald. GPT-in-the-loop: Supporting Adaptation in Multiagent Systems[C]//2023 IEEE International Conference on Big Data (BigData). Piscataway: IEEE, 2023: 4674-4683.
[67]	Xia Tian, He Zhiwei, Ren Tong, et al. Measuring Bargaining Abilities of LLMs: A Benchmark and A Buyer-enhancement Method[C]//Findings of the Association for Computational Linguistics: ACL 2024. Stroudsburg: ACL, 2024: 3579-3602.
[68]	Filippas A, Horton J J, Manning B S. Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus?[C]//Proceedings of the 25th ACM Conference on Economics and Computation. New York: ACM, 2024: 614-615.
[69]	Han Xu, Wu Zengqing, Xiao Chuan. "Guinea Pig Trials" Utilizing GPT: A Novel Smart Agent-based Modeling Approach for Studying Firm Competition and Collusion[EB/OL]. (2024-01-31) [2024-06-04]. .
[70]	Hua Wenyue, Fan Lizhou, Li Lingyao, et al. War and Peace (WarAgent): LLM-based Multi-agent Simulation of World Wars[C]//ICLR 2025 Conference. New York: ICLR, 2025: 1-41.
[71]	Mukobi G, Erlebach H, Lauffer N, et al. Welfare Diplomacy: Benchmarking Language Model Cooperation[C]//ICLR 2024 Conference. New York: ICLR, 2024: 1-61.

项目	智能体核心能力	LLM智能体对应能力
观察	快速感知环境、多源信息采集	工具调用；短期记忆
判断	动态分析信息；构建态势认知	角色扮演能力；长期记忆；知识库
决策	敏捷生成可行方案；权衡风险与收益	推理能力；规划能力；反思与优化
行动	精准执行指令；实时反馈闭环	工具执行能力；动态响应

领域	数据集	类型	来源
个体	Final Dialogue Dataset	对话	维基百科
	P-weibo Dataset	对话、描述	微博
	P-Ubuntu dialogue corpus	对话、描述	corpus
	LISCU Dataset	描述	失业率
	FoCus Dataset	描述	选举
	LaMP Benchmark	描述	党派偏见
	Multimodal Persona Chat	图像、对话	reddit
	RoleBench	对话	电影、剧本
	Character-LLM Dataset	描述	—
	PersonaChat Dataset	描述	—
	Synthetic Persona Chat	对话、描述	LLM生成
	RoleEval Dataset	描述	维基百科、百度
	Life Choice Dataset	图像、对话、描述	维基百科、百度
	RP Dataset	对话	小说、剧本
	MPI dataset	描述	—
人口统计特征	Who is GPT3 Dataset	—	—
	Dataset Movielens 1M	—	—
	EmotionBench	—	调查
	CultureLLM Dataset	对话	调查
	PersonaHub Dataset	描述	LLM生成

场景	数据集	类型	模拟目标
经济相关	2018.U.S.population	profile	宏观经济活动
	政府公开数据	租赁信息	资源分配
	names-dataset 3.1.0	profile	传染病建模
	Bureau of Labor Statistics	劳工数据	失业率
社会学与政治学	ANES	profile, answers	选举
	Pigeonholing Partisans	profile, opinion	党派偏见
	Twitter	用户言论	选举投票
	Milgram Shock Experiment	行为记录	服从行为
	collective decision-making	用户决策意见	集体决策行为
	Becker-2019	profile, answers	群体智慧
网络社交平台	Gender Discrimination	profile, opinion	观点动力学
	Metoo	用户言论	观点动力学
	Twitter	新闻评论	谣言传播
	Reddit	社交评论	羊群效应
	Amazon Beauty	用户-项目交互	用户-项目交互

阶段	核心要素	分类	描述/内涵
模拟前(数据与建模)	建模维度	个体	聚焦智能体自身的静态属性与动态言行数据
模拟前(数据与建模)	建模维度	情景	聚焦智能体交互所处的环境、场景与人物数据
模拟中(协作机制)	协作模式	辩论模式	通过观点交流与知识共享，激发集体智慧，达成共识
		反思模式	个体通过审视过往决策进行自我批判与学习，优化行为
		合作模式	多智能体通过任务分解与协调机制，协同完成集体目标
		竞争模式	在资源或利益冲突下，基于博弈论与强化学习进行策略博弈
		社交模式	模拟基于信任、模仿与角色扮演的日常社会互动
模拟后(评估验证)	评估维度	社交智能	评估智能体理解他人意图、形成社会关系的能力
		模拟准确度	评估行为与理论/数据的契合度及对多场景的适应性
		记忆提取能力	评估记忆提取是否受时效性、相关性、重要性影响
		可信性与可解释性	可信性：行为是否自然拟真；可解释性：决策逻辑是否透明可追溯
		鲁棒性	评估在不确定、冲突环境下的系统稳定性与自我修复能力
		社会效用	评估仿真在社会科学研究与现实应用中的价值与预测能力

应用场景	任务/设置	指标 / 评估维度	基线/对照	LLM智能体结果
社会行为—谈判协作^[51]	买方-卖方多轮议价，引入“第三方批评者/裁判”提供反馈	成交价趋势、达成率、破局率	初始回合/无反馈	强模型在多轮博弈中能“利用经历与AI反馈持续改进”，但“破局风险升高”；不同模型/角色学习能力差异显著
社会网络—假新闻传播^[59]	带个体画像、短/长时记忆与反思的LLM代理网络；比较干预策略	感染人群流行度曲线拟合、干预频率效果	无/低频干预	早期且适度频次的官方辟谣最有效；“每日vs每3日”发布在控制传播上差异不显著
推荐系统—用户模拟/对齐^[62]	用1 000代理在MovieLens等场景做“页级互动”	偏好识别准确率/召回率	随机/干扰项	准确率≈65%，召回率≈75%，在引入干扰项数量变化下仍保持稳定
经济行为—竞争与市场动态^[65]	GPT-4餐馆-顾客博弈小镇	价格收敛、马太效应、服务质量/策略演化	初始阶段/无竞争压力	观察到价格收敛趋势与马太效应；竞争促使经营策略差异化与服务改进
经济行为—合谋/串通^[69]	“Guinea Pig Trials”智能体模拟企业竞价	均衡价格相对关系(与伯特兰/垄断/卡特尔比较)	理论均衡	无沟通下价格收敛于“高于伯特兰、低于垄断”；允许沟通时逼近卡特尔价
军事推演—历史战例复现^[70]	WWI/WWII/战国多国体制仿真	仿真有效性(与史实进程对齐)、触发因素分析	—	提供多表多图对仿真有效性与开战触发因素的系统评估框架