Werewolf: A Straightforward Game Framework with TTS for Improved User Engagement
Project Overview
The document explores the application of large language models (LLMs) in education through an innovative framework designed around the Werewolf social deduction game, utilizing text-to-speech (TTS) technology to enhance user engagement. By enabling LLMs to operate autonomously, the system showcases their ability to strategize and adapt to social dynamics, thereby enriching the learning environment. The integration of generative AI facilitates immersive audio experiences and low-latency interactions, making educational activities more engaging and interactive. This approach not only highlights the potential of LLMs in educational settings but also emphasizes their role in creating captivating experiences that blend learning with entertainment. Ultimately, the findings suggest that generative AI can transform traditional educational paradigms by fostering dynamic interactions and deeper engagement among learners.
Key Applications
Werewolf game framework with LLMs and TTS
Context: A social deduction game designed for both educational entertainment and AI research, targeting players of various ages who enjoy interactive gameplay.
Implementation: The framework integrates LLMs that simulate player roles and a TTS module for voice output, allowing for both textual and auditory gameplay. It employs parallel processing to minimize latency during interactions.
Outcomes: Enhanced user engagement through immersive audio experiences, autonomous reasoning in gameplay by AI agents, and a more dynamic and interactive game flow.
Challenges: The framework relies solely on in-context learning without external modules, which may limit the depth of reasoning over longer interactions. It also currently supports only a limited set of game roles, impacting gameplay complexity.
Implementation Barriers
Technical Barrier
The system's reliance on in-context learning limits the depth of reasoning and can lead to truncation of earlier dialogues, affecting gameplay continuity.
Proposed Solutions: Future versions could integrate structured knowledge bases or retrieval-augmented generation to enhance reasoning capabilities.
Complexity Barrier
The current implementation supports only a subset of standard game roles, limiting the possible interactions and strategies.
Proposed Solutions: Expanding the system to include more complex roles could introduce deeper gameplay mechanics and improve agent adaptability.
Interaction Limitation
The absence of voice input restricts real-time spoken interactions, reducing the naturalness of gameplay in multiplayer settings.
Proposed Solutions: Implementing voice recognition capabilities could enhance user experience and allow for more fluid interactions.
Project Team
Qihui Fan
Researcher
Enfu Nan
Researcher
Wenbo Li
Researcher
Lei Lu
Researcher
Pu Zhao
Researcher
Yanzhi Wang
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Qihui Fan, Enfu Nan, Wenbo Li, Lei Lu, Pu Zhao, Yanzhi Wang
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai