Masaaki Kunigami, Takamasa Kikuchi, Takao Terano

Proceedings of the 18th International Multi-Conference on Society, Cybernetics and Informatics: IMSCI 2024, pp. 62-67 (2024); https://doi.org/10.54808/IMSCI2024.01.62

Masaaki Kunigami
School of Computing, Tokyo Institute of Technology, Yokohama, Japan / CUC Research Institute, Chiba University of Commerce, Ichikawa, Chiba, Japan

Takamasa Kikuchi
CUC Research Institute, Chiba University of Commerce, Ichikawa, Chiba, Japan

Takao Terano
CUC Research Institute, Chiba University of Commerce, Ichikawa, Chiba, Japan

Cite this paper as:
Kunigami, M., Kikuchi, T., Terano, T. (2024). A Layered Architecture to Apply Large-Scale Language Model for Simulation-Based Virtual Business Case Generation. In N. Callaos, J. Horne, B. Sánchez, M. Savoie (Eds.), Proceedings of the 18th International Multi-Conference on Society, Cybernetics and Informatics: IMSCI 2024, pp. 62-67. International Institute of Informatics and Cybernetics. https://doi.org/10.54808/IMSCI2024.01.62

This paper proposes a constructive architecture for using large language models (LLMs) to generate virtual business cases from agent simulations. This architecture consists of a layered structure for case generation and the concept of operational evolution. The layered structure consists of a Fundamental Layer consisting of typified facts extracted from simulations, a Narrative Layer generated from the typified facts by LLM, and a Presentation Layer utilizing LLM to enhance the integrity as a business case. In the operational phase, this layered structure enables easy modifications and reflection due to feedback from the field and the changing environment. It is expected to enhance the accuracy and explainability of the cases generated and evolved by the LLM. This layered architecture is also a design-oriented approach that pushes the role of the human case creator from "writer," which is easily substituted for artificial intelligence (LLM), to "designer," which is less likely to compete with AI. This paper also discusses the generalization to quantitative persona creation.