[3Yin-A] Poster 5

[3Yin-A-55] Balancing Shared Experiences and Autonomous Exploration Through the Integration of Generative AI and Fixed Scenarios

〇Genki Shintani¹, Yoshimasa Ohmoto¹ (1. Shizuoka University)

Keywords:

Shared Experience,Hierarchical Task Network,LLM

Generative AI significantly enhances game development efficiency but introduces challenges regarding content inconsistency and the loss of "shared experience." When generated content varies too widely, the common ground for player interaction diminishes. Furthermore, relying solely on probabilistic generation risks narrative divergence and logical incoherence. To address these issues, this study proposes a neuro-symbolic development method integrating Large Language Models (LLMs) with Hierarchical Task Networks (HTN).Our approach distinguishes narrative structure from realization. We define fixed events (e.g., boss battles) as high-level goals within an HTN, using deterministic planning to decompose them into logical prerequisites. The LLM is then tasked strictly with the concrete realization of these subtasks. This architecture strictly enforces causal consistency in the narrative backbone while leveraging AI's diversity for local expression. This design aims to satisfy both "autonomy" and "relatedness" as defined in Self-Determination Theory. We have developed a prototype and will subsequently verify its effectiveness in balancing logical integrity with generative freedom.