A shape grammar approach to computational creativity and procedural content generation in massively multiplayer online role playing games

Abstract

With rapid growth in both production costs and player populations over the last decade, the computer games industry is facing new scalability challenges in game design and content generation. The application of computers to these tasks – called procedural content generation – has the potential to reduce the time, cost and labor required to produce games. A range of generative algorithms have so far been proposed for procedural content generation. However, automated game design requires not only the ability to generate content, but also the ability to judge and ensure the novelty, quality and cultural value of generated content. This includes factors such as the surprise-value of generated content as well as the usefulness of content in the context of a particular game design. Studies of human designers have identified that the ability to generate artefacts that are novel, surprising, useful and valuable are facets of the human cognitive capacity for creativity. This suggests that computational models of creativity may be an important consideration for developing tools that can aid in or automate design processes. However, such cognitive models have not yet been widely considered for use in procedural content generation for games. This paper presents a framework for procedural content generation systems that use computational models of creativity as a part of the generative process. We demonstrate an example of such a system for generating instances for massively multiplayer, online role-playing games. The system combines the generative shape grammar formalism with a computational model of interest based on the Wundt curve to select new designs that are similar-yet-different to existing human designs. The approach aims to capture the usefulness and value of an existing human design while introducing novel or surprising variations through the model of interest. The system incorporates a metric that permits generated designs to be evaluated in terms of both their similarity to human designs and their novelty in the context of existing designs.