Games playing themselves, enjoying themselves and designing themselves

NYU Community Event

Speaker: Julian Togelius, IT University of Copenhagen


Ever since Turing, AI researchers have been concerned with playing games. Games are regularly used as AI benchmarks and important algorithms were invented in the context of game-playing. But what about designing games? Could we develop a system that automatically designs complete games, games that are novel and that people would want to play? This is a question which is interesting both from an AI perspective and from a design perspective. Even solutions to small parts of the problem - such as how automatically generate game levels - have important industry applications. I will discuss some recent attempts at generating both complete games and parts of games. I will also argue that any solution to this problem will require search through a space of game variants, and the ability for the generating system to play any of the games it comes up with. Furthermore, it needs to have an idea about what constitutes "fun" for a human, which could be achieved through modelling the preferences of actual humans or through implementing a computational curiosity mechanism. Finally, I will discuss some examples of incorporating game generation capacities into mixed-initative tools for game designers.


Julian Togelius is Associate Professor at the Center for Computer Games Research, IT University of Copenhagen, Denmark. He works on all aspects of computational intelligence and games and on selected topics in evolutionary computation and evolutionary reinforcement learning. His current main research directions involve search-based procedural content generation in games, game adaptation through player modelling, automatic game design, and fair and relevant benchmarking of game AI through competitions. He is a past chair of the IEEE CIS Technical Committee on Games, and an associate editor of IEEE Transactions on Computational Intelligence and Games. Togelius holds a BA from Lund University, an MSc from the University of Sussex, and a PhD from the University of Essex.