Content area
In games, non-player characters (NPCs) are central to bringing the designer's vision to life for players. NPC behavior must be robust to unexpected situations, contextual to the game world and to players' actions, and constrained to the designer's goals. This means that designers, especially those who work solo or in small teams without specialized game AI engineers, need tools to flexibly implement and tune NPC behavior. This leads me to ask: how could creativity support tools for NPC behavior design assist independent game designers in bringing their visions to life? This question calls for direct collaboration with game designers to make a tool that works for their needs.
In this dissertation, I describe a multi-phase project in which I: investigate designers' requirements for an NPC behavior tool via participatory design engagement with two experienced game designers; implement and test the tool, EvolvingBehavior, in a popular game engine; engage in long-term co-design to embed it in the context of an independent game developer's RPG game project; and broaden the resulting insights in participatory design collaboration with four more independent game designers. Through this process, I evince designers' needs for interpretable, controllable algorithms to support their iterative workflows, and show how these can be implemented with the EvolvingBehavior tool.
In conjunction with the technical contributions of the tool implementation, I present detailed thematic analyses, refined over the course of the participatory project, that describe how researchers and developers can make creativity support tools that are situated in and responsive to designers' needs. There is much to learn and translate from these insights for other kinds of creative work. They also represent a compelling alternative to current black-box, big-data paradigms in machine learning research, showing how we can move beyond single-objective optimization towards more flexible, exploratory tools. Finally, I contribute methodological recommendations that describe how future participatory research could better shift power to designers through collaboration.