Computational Modeling in Games (16w5160)
Michael Mateas (University of California, Santa Cruz)
Andrew Nealen (New York University)
1. In sessions of 30 minutes, participants will present their recent research in the fields of systems and/or player modeling, thereby leveling the knowledge between everyone in the workshop. This will cover topics as diverse as human motion tracking, modeling human preferences, system abstraction and simplification, logical-probabilistic calculus, complex systems theory, game theory, and machine learning.
2. Game designers will share their most difficult design challenges when working with complex systems that will be played by millions of people with entirely different backgrounds and cultures. We envision that this will lead to sustainable collaborations between the research and practice of game design.
3. Researchers and game designers will form break-out groups to summarize open problems in game design, after which these will be presented to all participants. From this we will develop a clear and concise framework that will allow us to better guide computational modeling research, and synchronize our efforts across disciplines, departments, and cultures.
Game design, as a professional discipline, has fully established itself in contemporary culture. But despite this development, most design challenges are solved in a heuristic manner. This strategy works well for experienced designers, but makes it extremely difficult for entry-level designers, who would likely benefit from a coupling of traditional, heuristic design, and some formal elements guided by scientific research into both game systems as well as human perception and cognition. Unlike design fields such as architecture, in which the artistry of design is complemented by mathematically rigorous analysis and predictive modeling carried out by CAD (Computer-Aided Design) systems, mainstream game designers make use of no formal modeling and mathematical analysis during the design process, instead depending on developing rapid prototypes and gathering informal feedback from players and the design team playing the prototypes. This results in the game designer being able to explore only a small fraction of the design space, missing out on innovative new rule systems and approaches. Even worse, it means that completed games often exhibit unexpected consequences of rules and/or the sensory presentation of the game, such as players finding exploits that break the game balance or destroy the desired dynamics of the game, or result in the game being completed unexpectedly quickly, or players failing to experience desired content, or challenges in the game being unexpectedly easy or difficult; in general, games often fail to achieve their design goals with the target audience.
Bringing together researchers who are well-versed in devising, implementing, and verifying computational models of complex systems and human perception, and pairing them with working game designers, will allow us to isolate the pertinent design problems, and solve them formally and algorithmically.
Computational models are ubiquitous in everyday life. Economists use computational models to predict the dynamics of the stock market, automotive vehicles are full of predictive systems, and even archaeology predicts the presence of archaeological features using proxies such as soil types, elevation, and vegetation. At the same time, the field of machine learning has only started to scratch the surface of predicting human behavior. This is not to belittle the accomplishments made, but rather to state that there is still much work left to be done. Furthermore, it is quite obvious that, in order to advance the field of computational modeling, interdisciplinary efforts will be necessary; predictive models rely on context, and without context there can be no model. And while our focus is on game design as a discipline, we envision our system/player centric approach to design - the creation of form - to have repercussions in all other design fields. Formal design theory as established in the 20th century, originating in the Bauhaus and carried on through the work of designer engineers such as Dieter Rams and design theorists such as Christopher Alexander, has inspired much scientific work in formal methods and tools for designers, but these advances have suffered from the unfortunate fragmentation of theory and practice that is still a reality in many disciplines. Game designers have developed a vocabulary and variety of heuristic methods to tackle their design challenges, while the scientific community has made great strides in both modeling probabilistic systems and the variation of human behavior through statistical machine learning. In this workshop, we will gather experts from both theory and practice, thus forming a community that can isolate previously unseen, real-world problems and devise realistic, implementable algorithmic solutions.
The time is ripe to begin creating a design science for games. As a cultural form, games have become a part of mainstream culture, and are quickly growing to become a dominant media form in the 21st century. Yet game design is still primarily a poorly understood craft practice, with game designers operating almost entirely based on rules of thumb and intuition.
However, a small but growing community of researchers has begun applying formal modeling techniques to game design. This scattered community has been developing multiple techniques, including:
- Logic models in which dynamic consequences of rules are discovered via finding stable Herbrand models.
- Petri-net modeling capturing resource flows and feedback loops in rule systems.
- Machine learning techniques applied to temporal traces of human game players, resulting in the classification of different play styles (player types), predictive modeling of player action, predictive modeling of the relationship between game design changes and player retention (how long a player will continue playing the game), and automated matchmaking of play styles and skill levels for multi-player games.
- Framing of game design as a complex optimization model, solvable by optimization techniques such as evolutionary programing.
- Formal verification techniques used to verify that a game implementation actually meets the requirements of the abstract rule system.
- Automated content and game generation driven by a formal model of the game design space, sometimes adapting automatically to an online-model of the player.
- Predictive, automated selection of camera positions in a game, based on perceptual, statistical modeling of human viewpoint preference.
People working in these areas tend to be working in isolation, to not know about the other areas, or to be developing little sub-communities that lack an understanding of the shared goals and formal techniques of the broader enterprise. We intend to host the first gathering that unites this emerging and fragmented community under the shared banner of computational modeling for games, jumpstarting a new international community that will make fundamental advances in the computational modeling of rule systems and players, driven by the unique challenges and requirements of contemporary games.