Simulated Art World: Artist Simulation Based on Generative Models
“模拟艺术界”：基于生成式人工智能的艺术家智能体仿真模拟
Process of Artist Simulation Based on Generative AI
Wang, Yangyu, 2025, “Simulated Art World: Artist Agent-Based Modeling by Generative Models”, Journal of Intelligent Society 4(3), 180-204+266-267. (In Chinese)
王杨聿，2025，《“模拟艺术界”:基于生成式人工智能的艺术家智能体仿真模拟》，《智能社会研究》第三期。
Abstract: Guided by symbolic-interactionist theory,sociologists of art contend that artistic creation is fundamentally shaped by interaction.Yet empirical research of how interaction translates into artistic outcomes have long been hampered by data limitations.To overcome this obstacle,I construct an Agent-Based Modeling（ABM） in which “ artist agents” interact and create artworks through the joint use of Large Language Models（LLM） and conditional Generative Adversarial Nets（cGAN）.Within the simulated environment,agents debate aesthetic opinions,construct or diminish social ties,and iteratively adjust their creative orientations.Textual traces from these debates are analyzed with BERTopic to map pathways through which shared meanings emerge.The resulting style and relationship variables then serve as conditioning inputs to a cGAN,enabling the generation of synthetic paintings that reflect postinteraction stylistic shifts.Findings show that interaction not only cultivates shared symbolic repertoires among artists but also drives a measurable convergence of visual style.Moreover,the mechanisms and speed of convergence vary across macro-level social scenarios instantiated in the simulation.The study thus provides direct,processual evidence for the “ interaction-to-art” link posited in sociology of art and proposes a scalable methodological framework for investigating elite cultural dynamics with generative AI.
中文摘要：秉承符号互动论的理论观点,艺术社会学家往往认为互动建构了艺术创作。但受数据所限,以往研究无法直接检验互动过程与创作过程的联系。针对这一研究困境,笔者借助生成式人工智能模拟了艺术家智能体的互动与创作过程。通过将大语言模型与条件生成对抗网络相结合,笔者利用艺术家的维基百科词条建构艺术家语义画像,并驱动其在模拟情境中进行风格辩论与关系构建。随后,通过BERTopic主题模型分析仿真模拟过程中的文本数据,识别共享意义的形成路径;进一步以风格与关系变量为条件,训练图像生成模型来模拟互动后的绘画变化。结果显示,互动过程能建构艺术家之间的共享意义,也会使艺术家的艺术风格渐趋相同;同时,在不同的智能体模拟情境下,互动的建构过程存在差异。本文结论不仅回应了艺术社会学中“互动—艺术”关联的理论问题,还提出了一种针对精英阶层社会过程分析的方法论框架。
Full Paper (English version) sees Here.