[3O1-OS-44a] OS-44

[3O1-OS-44a-03] Developing an Open-Ended Cultural Evolution Model Using LLM Agents

〇Mahiro Kato¹, Reiji Suzuki¹, Takaya Arita¹ (1. Nagoya University Graduate School)

Keywords:LLM, Cultural Evolution, emergence

In this study, we aimed to propose a method for generating new ideas and stories using large language models (LLMs). To achieve this, we developed a platform to enable creative cultural evolution through a group of LLM agents and analyzed the properties of the generated stories as well as the dynamics of cultural evolution. Specifically, we focused on science fiction (SF) novel plots and conducted evolutionary experiments to examine the effects of combining two mechanisms of cultural transmission—individual learning and social learning (e.g., Random, Attraction-Based, Conformist, Anti-Conformist)—on the dynamics of cultural evolution.The results revealed the emergence of diverse cultural evolution patterns, including plots resembling the renowned SF work Paprika by Yasutaka Tsutsui. Furthermore, we found that combining learning strategies enabled the sustained creation of diverse stories, while relying solely on social learning tended to result in thematic convergence. These findings not only contribute to understanding the dynamics of cultural evolution but also highlight the potential of LLM-based platforms as practical tools for continuously generating open-ended creative ideas.