Five emerging research opportunities for deep learning in video games: A curated overview

Video games are a natural and synergistic application for deep learning, offering the potential to enhance player experience and immersion, while also providing valuable benchmarks and virtual environments for advancing artificial intelligence more generally. This work offers a compact, high-level overview of five technically diverse research pathways that have recently emerged at the intersection of modern deep learning and digital gaming. We highlight the potential of these areas for impactful research in the near future within the context of the current literature and provide effective entry points for their investigation. We concisely discuss (i) large language models as core engines for game agent modelling, (ii) neural cellular automata for procedural game content generation, (iii) the acceleration of computationally expensive in-game simulations via deep surrogate modelling, (iv) self-supervised learning of useful video game state embeddings, and (v) generative models of interactive worlds derived from unlabelled video data. We also address current technical challenges associated with the integration of advanced deep learning systems into video game development and briefly indicate key areas where further progress is likely to be impactful.