Using AI to free players' agency from the limits of predefined langua...

This paper explores how language in video games can restrict player agency and how the utilization of generative artificial intelligence via large language models, such as those used in ChatGPT, could help in mitigating problems like these. More explicitly, we analyse how predefined choices in a selection of video games are phrased and potentially (mis-)interpreted by players, and what other choices are potentially not available in these contexts. Giving players agency by allowing them to act according to their own intentions in the virtual world of a game is an important part of game design. However, most contemporary implementations of (dialogue) choices often lack nuance and do not allow players to pursue their actual intentions, as they are limited to the tools that the designers have provided to them, which in turn are oftentimes limited by language. After this analysis, we discuss the potential use of AI-assistance tools to grant players more agency by allowing them to input their own choices, wishes, and decisions. These could then be interpreted by language models, to allow the game to respond appropriately and dynamically, both in regards to text or dialogue, as well as by creating new game states. Players proposing new options could help in creating a new form of emergent storytelling and gameplay. This can already be seen in text-based adventure games that utilize similar mechanics, or in-person role-playing games. Potential dangers and problems that arise from the use of such technology, as well as the many ethical implications of using generative AI, are also discussed, before drawing a conclusion about this potential future of video game design.