Recently, Microsoft announced its upcoming visual game design environment, Kodu, to be released “later this spring”.  Combined with the level editor in Little Big Planet, these two should provide fascinating case studies for the project.

Another interesting piece on formal game design is the paper “An experiment in Automatic Game Design” by Togelius and Schmidhuber (find the paper in here).  The authors created a program that would automatically create new games based on an underlying schema (or “meta-rules” or “axioms” as the authors call it).  To evaluate the games, they had a genetic algorithm with a fitness function based on the idea of using learnability as a predictor of fun. Therefore games that are easy to learn but hard to master would get high fitness values with the algorithm, thus indicating a fun game.

The paper is very interesting and the approach novel and fun. However, even though this is admittedly just an initial experiment, there are two problems that really highlight the difficulty of formalising game design. Firstly, measuring (and formalising) player experiences. Learnability may be a necessary condition for fun, but it really is not sufficient. Like Salen and Zimmerman point out, game design is a second-order design problem where the designer only indirectly affects the players’ experience. Formalising the mechanisms that turn rules into great experiences can be really hard.

Secondly, attempts at formally defining game design lead easily into really complex structures. In this case, the problem was sidestepped by providing the base schema (i.e. “meta-rules” or “axioms”) for the design – the changes in the design would only occur within the parameters of the schema. However, this leads to question (as someone did in the authors blog) whether the experiment is about game design at all. I would be inclined to say that simple schema permutations do not constitute as game design. But then again, maybe its just a question of the complexity of the schema.