Using a Large Language Model to turn Explorations of Virtual 3D-Worlds into Interactive Narrative Experiences
Thomas Rist · 2024
In this contribution, we report on work towards the development of an “Inner Voice” that accompanies and talks to the player while exploring a 3D-game world. We have developed a prototype for experimentation and testing that relies on off-the-shelf LLM-based chatbots from OpenAI [1]. Fundamental to our approach is that both prompts passed to the LLM and the responses received from the LLM are anchored in the game world and the course of gameplay. In a gaming context, using an inner voice opens up various possibilities. Firstly, it can be parameterized with regard to its linguistic realization, including, among other things, message length, language style and tone, word choices, lexical frequency of words, or the use of certain rhetorical figures. Secondly, when triggered by a game world event, the message content can be biased to influence the player’s decision-making. Thirdly, an inner voice can provide clues about a background story. Variations in the background story and the provided clues can condition the player’s interpretation of observed game world events as well as decision-making, which in turn impacts the construction of a self-narrative. We describe our approach to prototype development and discuss experiences and findings made through various trial runs of our prototype.