Exploring the aesthetical qualities of scaled game maps through Human-AI Collaboration

Abstract: The primary objective is to explore the scalability of two-dimensional game maps while preserving certain aesthetical qualities in scaled representations. By either upscaling or downscaling the maps, features of the map inducing these aesthetical qualities may diminish. For instance, there could be alterations in the layout of corridors, rooms, characters, and treasures, as well as variations in the quantity of them. To address this, AI technology has been used as a means of preserving the feature soriginally introduced by the designer to create an alternatively scaled representation. The explorationis made possible by utilizing a game designer tool, Evolutionary Dungeon Designer (EDD), to designmaps - scale them - and generate AI-based solutions through an evolutionary algorithm. Furthermore, evaluations through both a user study and a controlled experiment were performed to analyze thescalability of game maps and the AI-generated representations. The user study showed some divisive results regarding whether the scaled or the AI-generated maps were superior. Often the AI-scaled maps were regarded as dissimilar compared to the original map. However, the AI could to some extent provide the same prevalence of some of the wanted features, but in a different design. This was also evident in the controlled experiment, where the AI managed to contain a specific feature to the same degree, but lacked the capability of making the maps similar.