Computer Games - Boon or Bane for AI Research?

Will applications like computer games finally help break down the walls of the ivory tower? Or is it simply a silly distraction, undermining the seriousness of our research field in the public view? In this article, I present my opinion. As a precaution: I believe that points of view can rarely be advanced by harmonious speeches accompanied by general nodding. Instead, a friendly but spicy discussion can help find and clarify positions in a much better way. Therefore, please do not be offended if you feel your adrenaline levels rising. At first, let me clarify what I mean by “computer games”. This article is not about board/card/puzzle games, like chess or Rubik’s Cube. Games like chess are actually good examples of what I would like research to steer away from. While chess has served as a kind of Drosophila for AI research, and has led some great advances in search, nowadays, most chess research is so over-specialized that I can hardly call it research anymore — it is application development. Does computing a rebel’s behavior in the galactic civil war in a Star Wars computer game require sophisticated technology? You bet! Suddenly, one has to deal with optimization in real time, a highly dynamic and complex environment, incomplete knowledge, very restricted computation power and memory, ... and forget about all these nice things like closed-world assumptions. Completeness properties and complexity classes below undecidability? Hah, are you kidding? Tackling the features above may sound difficult to begin with, and I do not recommend that anyone tries to realize them all at once. However, the important thing is to keep these features in mind when initially striving for a subset. Algorithm design must take place in a holistic view of the complete set of target features, and not for single properties in isolation. I refer to this as “multiobjective algorithm design”. There are lots of trade-offs to be considered when striving for these features. Trade-offs between speed and memory are well