Reddit reviews Music and Probability (The MIT Press)

Wow, thats an interesting question. I have a book on my 'to-read' stack called "Music and Probability" - http://www.amazon.com/Music-Probability-David-Temperley/dp/0262515199 - maybe you could draw some connections on your own?

> I don't think music really can be randomly generated like this, or at least not with something so primitive.

Obviously, to write quality generative music, you need to be able to write quality music. I.e. you need to understand compositional structure, music theory, what makes a good melody, etc. You can't expect to generate good music without this understanding. Writing quality music is difficult in any medium or compositional domain.

That said, using an Nth order Markov model and a lot of analyzed music from the same style (e.g. the works of Bach, Mozart, etc) as input, you can create quality music and have absolutely no idea why or how it's quality. This would be more along the techniques successfully employed by David Cope.

However, these analysis-bases techniques don't deal with meaningful interaction or note/beat-level changes, both primary issues in interactive software (like a game). To get quality procedural/generative/interactive music in this domain, you need to employ a more directly-compositional approach. The best efforts in this area are those that use probabilistic techniques and compose music in a way that is quite analogous to traditional composition, but instead of composing particular rhythms, melodies, or structure, you compose these elements using [probability distribution functions] (http://en.wikipedia.org/wiki/Probability_distribution). Until recently, there haven't been too many books detailing this approach, but now there are.

Edit: One more thing -- there may be more music than you realize that employs this sort of probabilistic approach. When it's good, you don't notice it. It just sounds like music.