Erik McClure

Genetic Game Generation

So theoretically, one could take most of the known game mechanics and give them a genetic identity, and combine this with content generation in such a way you could create an algorithm that pumped out entire randomly generated games with their own (weird) mechanics. If you had a sufficiently advanced library, the genetic code of a given game wouldn't actually be particularly complex, similar to Spore's procedurally generated creatures. By taking advantage of this one could craft a game using the genetic code and then send it around as nothing more then a highly compressed ASCII character sequence.

Naturally this would be very difficult to pull off, because there would have to be very complex interactions between the genetic codes of various game mechanics and rules. However, it has also been demonstrated that you can combine just about any type of game with any other type of game, so in practice it should be possible to generate a majority of fairly workable games after a bit of training. Once the genetic algorithm gets trained enough, it would probably start producing much more reasonable games, although it could also collapse into a single genre if there isn't enough mutability in the algorithm. Seeing whether it collapsed into a popular genre or not would be a rather interesting thing to study.

What's even weirder is that because everything is generated, the game should be able to improvise the game forever, if there is no need of an end. This naturally ignores all the issues with plot, which is an entirely different AI related task.



  1. 2020
  2. 2019
  3. 2018
  4. 2017
  5. 2016
  6. 2015
  7. 2014
  8. 2013
  9. 2012
  10. 2011
  11. 2010
  12. 2009