In this study, we investigate how interactions with large language models (LLMs) can lead to emergent behaviors, allowing players to shape game narratives. Using a text-adventure game and GPT-4, we observed 28 gamers discovering new narrative nodes through interactions with non-player characters. Surprisingly, these emergent nodes were not part of the original narrative but enhanced gameplay engagement. Players who enjoyed discovery, exploration, and experimentation in games were more likely to create these unique nodes. This study sheds light on the potential of LLMs in evolving game narratives and the player’s role in co-creating content.
https://arxiv.org/abs/2404.17027