This project explores a radical new theoretical and computational model for creativity, casting it as a rational pursuit of curiosity, and in particular develops a model of expressive artifact creation as a knowledge seeking effort. A discovery agent engages in the rational pursuit of curiosity when, given an experimental situation and a set of discovery actions, the agent takes the action it predicts will result in the maximum knowledge gain. This is a key feature of our proposed model of transformational creativity: creative agents in a design domain are creative precisely because they seek to increase their design knowledge in the domain. This new discovery system approach will be built on the foundations of the Co-PI's earlier contributions in formalizing the domain of game mechanics and generating a wide variety of human-playable games through logical reasoning. In doing so, the project will bridge a gap in the creativity literature? recasting the creation of aesthetic artifacts as a knowledge building process, rather than the product of a limited grammar or the knowledge-free exploration of a generative space. We call this approach expressive discovery.
The proposed game design system will be focused in the creation of the underlying models of the mechanics that drive game play with the goal of providing a new game development platform. This area of computational creativity will support game design that moves beyond the production of static outputs (e.g., musical compositions or fictional stories) toward the creation of artifacts that are inherently self-reflective and interactive. This project consists of three core activities: 1) creating the first formalization that captures a large design space for interactive artifacts in a computational creative system; 2) develop a new model for discovery systems that embodies the Co-PI's model of transformational creativity as the rational pursuit of curiosity.; and 3) preliminary evaluations of the game design model formalization and the larger creative process it supports to inform future work in the area.
This project adapts AI models of creativity to the creation of new kinds of playable experiences. Games are a powerful new medium, capable of improving our intuition about complex systems such as the economy, climate, and social organizations, capable of providing highly personalized and engaging presentation of educational materials, and capable of letting us explore our humanity and place within the universe. However, to fully realize the expressive potential of games, we need a better understanding of the science of game design, supporting the creation of tools that multiply the creative powers of game creators, much the same way that computer-aided design has allowed engineers and architects to create designs that would have been impossible to conceive of without computer support. To improve our understanding of the science of game design and work towards this vision of intelligent design tools, we begin by building on computational models of creativity. The computational creativity community has, over the last few decades, begun to build models of the creative process, building systems that exhibit creativity in domains such as painting, music, and mathematics. However, the vast majority of such systems have focused on the creative production of static artifacts, such as an image, a musical score, or a mathematical lemma. Games, on the other hand, are complex dynamic systems – they communicate through the interaction of a player with this dynamic system. So developing a science of game design, and developing next-generation design tools, requires developing computational models of creativity that understand dynamic artifacts. In this project we pursued two major branches to this goal. The first involved figuring out how to formally represent games using a modern form of logic programming called Answer Set Programming (ASP). We have built ASP-based systems that can reason about the interaction between different game rules, and generate content such as levels and puzzles for games. This later work is having impact in educational games through the dynamic generation puzzles for the fractions-learning game Refraction created by University of Washington. They adapted our ASP-based puzzle generation techniques in order to dynamically generate puzzles that track the current performance of the learner. Thus the work supported by this grant is already beginning to make good on the promise of radically adaptive and customized games for education. Second, in the area of computational creativity, we have taken one of the most-cited but least evaluated approaches (that of the famous Minstrel system) and undertaken evaluation work that revealed unexpected information about its characteristics and apparent limitations. We were able to use this information to develop a new approach to using the same underlying creativity model, and adapted it for interactive use. We have also made a public release of our reconstruction -- which will allow others to learn more about this approach and potentially use it as a foundation -- and have developed new methods for quantitatively evaluating computational creativity in the process of doing our work with the system.
