Discussion and future work

This work has shown that it's possible to apply evolutionary algorithms to evolve music patterns in video games: it requires adaption to make the game playable. Simplifying the search space is necessary to avoid long runs with unwanted patterns, to make it possible to create a simpler interface to interactive authoring and thus producing a more compelling experience.

While this work just provides a foundation framework for music evolution in a playable game, its development is still embryonic. More work has to be done starting from the creation of an enjoyable user interface, with fancy graphics and more studied interactions. The genetic algorithm can be improved by studying other representation schemes, trying to enlarge the search space while maintaining the game enjoyable (we don't want to deal with loads of ugly patterns), and adding different styles of music, obviously following the respective composition rules.

Another interesting perspective is the user interaction with the genetic algorithm: the possibility to add patterns during the gameplay will certainly make the game more compelling. This could bias the search space with the human abilities to create patterns, but I believe the rewards system and the other genetic approaches would balance this. Adding new ways of interacting with the system is an interesting problem by itself: one could envision integrating pitch recognition to feed the genetic algorithm nice melodies to jam with, or add interaction directly with the body like in the air-guitar style to produce new rhythms.