Where Virtual Dialogue Robots Begin. State-of-the-art of AI Implementation in Video Games


  • Bálint Gusztáv Mosóczi University of Vienna


While building non-playable video game characters, developers focus on creating life-like physical behavior, while they leave the dialog to be handled by writers and voice actors. Large and accessible natural language processors have not been around for long enough to be implemented in major video game titles; however, these technologies could be used to generate characters that inhabit the words of open-world video games. NLP powered non-playable characters combined with real-time transcription tools and voice synthesizers could mean the future of immersive gaming, but how close are we to seeing something like this widely available? What tools need to be developed to make something like this work? What are the possible implications of making these tools widely available? What are the possible limitations and challenges that still need to be overcome? My research is exploratory in the intersecting realms of game design and emerging AI technologies. Based on the project conducted at ÖFAI “Charming – Character mining and generation”[1] that mainly focused on extracting character types from movies, I will examine these questions, see what is already there and where the industry is headed. Charming developed a character mining tool that they used to determine different personality markers from immense movie dialogue data. They studied how companion robots could be trained based on film characters, but this proved to be insufficient due to the lack of technicality in film dialogue. However, video games have been implementing this character type for decades, built for technical correctness to guide players in the diegetic world and show them proper control usage. The investigation of this phenomenon is currently a highly prominent subject of scholarly inquiry, captivating the attention of both academic scholars and industry professionals, leading to a continuous influx of novel findings and models daily. In my research I am going to find the state-of-the-art implementations, models, and research on character generation, and I am also going to review the limitations the other researchers have found, trying to contact them and cooperatively try to find ways to overcome them. An example for this can be seen on a TikTok posted by AI enthusiast and coder @Tamulur who already made significant contributions [2] to making AI implementation experiments accessible and have found that at the moment delay [3] is one of the largest problems to overcome. 


[1] Charming – Character Mining and Generation, Project leader: Robert Trappl, Project number: P27530-N15

[2] Tamulur: “ChatGPT -driven NPC experiment 1” April 15, 2023 Available: https://www.tiktok.com/@tamulur/video/7222341218460585222 [Accessed: 10.05.2023].

[3] Tamulur: ”Well the by far biggest problem right now is the awful delay. [...], Personal message on Tiktok (April 18, 2023)