Core Concepts
This scoping review examines the current applications of large language models like GPT in game research, identifying five key areas: procedural content generation, mixed-initiative game design, mixed-initiative gameplay, playing games, and game user research. The review highlights emerging trends, unexplored opportunities, and future research directions for leveraging GPT's capabilities to enhance game development and player experiences.
Abstract
This scoping review explores the current applications of large language models like GPT in game research. The authors identified five key areas where GPT is being utilized:
Procedural Content Generation:
GPT is used to generate game-related text, such as stories, quests, levels, characters, and even music.
Newer GPT models like GPT-3, ChatGPT, and GPT-4 demonstrate superior capabilities compared to earlier versions like GPT-2, requiring less fine-tuning and prompt engineering.
However, smaller GPT models still have value due to their lower operational costs, fewer ethical constraints, and more flexibility for customization.
Mixed-Initiative Game Design and Development:
GPT is used to assist designers and developers in the game creation process, generating scenarios, game mechanics, and rules through iterative collaboration.
The review suggests integrating various GPT models to create a comprehensive game development support system, improving context understanding and increasing efficiency.
Mixed-Initiative Gameplay:
GPT is used to support mixed-initiative storytelling, where players and the AI system take turns contributing to the narrative.
GPT is also employed to assist dungeon masters and game masters in tabletop role-playing games by generating enemy descriptions, summarizing game situations, and brainstorming narratives.
Playing Games:
GPT has been used to autonomously play language-based games, such as word-guessing and debate games.
The review suggests exploring GPT's capabilities in playing games with more complex mechanics, strategies, and hidden game states.
Game User Research:
GPT is used to assist in game user research, such as analyzing player reviews, categorizing audience comments during live streams, and comparing human and AI responses to interview questions about voice interaction.
The review highlights the need for further research in applying the latest GPT models to non-story generation tasks, integrating GPT's functionalities for comprehensive game development support, exploring diverse player-GPT interaction modes, and leveraging GPT in game user research beyond text analysis.