[4G3-GS-2] Machine learning: General

[4G3-GS-2-05] Research on Ideation Applications Using LLM-based Multi-agent Systems and Idea Evaluation Methods

〇Takaaki Tanaka¹, Shun Otsubo², Kotaro Ito², Takuya Hatakeyama¹, Yuji Anzai¹, Tomoaki Nagasaka¹, Takashi Matsui¹, Nobuyuki Ishikawa¹ (1. HAKUHODO TECHNOLOGIES INC., 2. NTT DATA Mathematical Systems Inc.)

Keywords:Multi Agent, Ideation, Large Language Models, Creativity Support Tools

Large Language Models (LLMs) have recently been applied to multi-agent systems. LLM-based multi-agent systems are platforms where multiple AI agents cooperate or compete to accomplish complex tasks. These agents are designed to interact with each other efficiently using natural language. Applications of such systems include improving the accuracy of question-answering (QA) systems, simulating real-world interactions, and automating software development workflows. One significant research area involves optimizing the roles and forms of communication for individual agents. Influenced by recent developments in LLMs, there has been an increase in the use of LLMs in Creativity Support Tools (CSTs) within the field of Human-Computer Interaction (HCI).
In our study, we developed CSTs employing an LLM-based multi-agent system for ideation in new business and product development. We varied the input conditions to the CST (such as the personality traits of the agents and the diversity of agent professions) and measured the diversity of the generated ideas and the differences between these ideas and those created by humans. It was found that certain trends exist in the diversity of the outputted ideas, which provide insights into new methods for idea evaluation using LLM-based multi-agent systems.