Dito. Although I got a bit weary of people using Genetic Algorithms to do X. Usually, using a GA is saying "I have no flipping idea how my search space looks like", when the truth is "I've got a good intuition what I'm looking for, but I'm too lazy to formalise this and GAs sound cool". And then you go on and constrain your search space on the wrong level by manually tweaking the Genotype-Phenotype-Mapping.
GAs are good a good optimization tool once you've broken down your search space to a domain that is so absurdly small that manually fiddling with parameters produces no noticeable change. For exploring interesting patterns and areas of a search space, the algorithm you're looking for is called "random()".
Wow, that's really awesome; hadn't seen that before. While it's not nearly at that level yet, I'd love to get to the point where I'm generating content as nice as he managed to. Hopefully once I get the genetic algorithm attached I'll start seeing some magic happen.
http://www.karlsims.com/papers/siggraph91.html