General Game-Playing Systems

Yngvi Björnsson, Reykjavik University

Abstract: The aim of General Game Playing (GGP) is to create intelligent agents that can automatically learn how to play a wide variety of different games at an expert level without any human intervention. This requires that the agents be capable of learning diverse game-playing strategies from basic game rules without any game-specific knowledge being provided by their developers. A successful realization of this task poses interesting research challenges for artificial intelligence sub-disciplines such as knowledge representation, agent-based reasoning, heuristic search, computational intelligence, and machine learning. This talk gives an overview of the state-of-the-art of general game playing systems, including different design models, and discusses some open research challenges.

Yngvi Björnsson is an associate professor at the School of Computer Science at Reykjavik University, as well as the co-director of the CADIA research lab. His research interests are in artificial intelligence, more specifically heuristic search and search-control learning, and the applications of such techniques in computer games. He is a co-author of the general game-playing agent CADIAPLAYER, which won the 2007 and 2008 AAAI GGP competitions.

Playing Machines: Machine Learning

Applications in Computer Games

David Stern and Joaquin Quiñonero Candela

Abstract: The tutorial will give an introduction to the emerging area of applying machine learning to computer games and of using computer games as test beds for machine learning. Since this is an application area, the tutorial will focus on past and recent applications, open problems and promising avenues for future research.

Tutorial Webpage

David Stern studied Engineering at Cambridge University and then continued in Cambridge to complete a PhD entitled ‘Modelling Uncertainty in the Game of Go’. In his thesis he investigated how machine learning techniques can help build an Artificial Intelligence for playing the ancient Chinese game Go, a game which has defeated traditional AI techniques. After completing his PhD in late 2007 he joined Microsoft Research Cambridge where he has been continuing his work on Go and also applying probabilistic modelling to online services.

Joaquin Quiñonero Candela studied Telecommunications Engineering at the Carlos III University of Madrid and received his PhD in machine learning from the Technical University of Denmark in 2004. During his post-doc at the Max Planck Institute for Biological Cybernetics he worked on non-parametric Bayesian machine learning, Gaussian processes, sparsity and predictive uncertainty. Joaquin worked during 2006 at the Technical University of Berlin on dataset shift in machine learning, a topic on which he has recently published a book (MIT Press, 2009). Joaquin joined Microsoft Research Cambridge in early 2007, and he has since been working on applications of machine learning and probabilistic modelling to computer games and to the Web.

Artificial Intelligence in Racing Games

Stefano Lecchi, Milestone

Abstract: A key aspect in the development of computer games is the behavior of non-player characters. Each type of game poses different challenges for the development of a successful artificial intelligence. In racing games, this translates into the programming of an AI which can adapt to the driving style and to the driving capabilities of the human player so as to improve its gaming experience. In addition, in racing games, the behavior of non-player characters should be plausible, challenging throughout the game, adaptive, and it should also lead to realistic group behaviors.

Stefano Lecchi has been working in the game industry since 1989. He is currently the head of the development at Milestone, an Italian game developer company specialized in racing titles for the console market. He was previously the lead programmer on titles such as Superbike 2001 and Screamer 2. His main focus is on physics simulation and artificial intelligence applied on racing games.

AI Isn’t Just for Players: AI-based Authoring Tools

Michael Mateas, UC Santa Cruz

Abstract: Game AI research has successfully focused on improving the player experience, creating better tactical and strategic opponents, more convincing non-player characters, better path-finding approaches, and so on. However, as we create richer AI-based experiences for players, we can not forget authors. Human authors must be able to craft game experiences, creating the richness and nuances that make games compelling, while still taking advantage of the generativity and adaptability that is the hallmark of next generation game AI systems. This will require new AI-based authoring support tools. In this talk I will describe the authoring problem, and present a number of authoring
support tool projects currently taking place in the Expressive Intelligence Studio at UC Santa Cruz.

Michael Mateas Michael Mateas’s research in AI-based art and entertainment combines science, engineering and design to push the frontiers of interactive entertainment. He is currently a faculty member in the Computer Science department at UC Santa Cruz, where he holds the MacArthur Endowed Chair. He runs the Expressive Intelligence Studio, an AI-based interactive entertainment research group, and helped create the undergraduate Game Design degree program, the first such program in the University of California system. With Andrew Stern, Michael released Façade, the world’s first AI-based interactive drama. Façade has received significant attention, including top honors at the Slamdance independent game festival. He received his Ph.D. in Computer Science from Carnegie Mellon University.