Does anybody have some knowledge about the current state of art in terms of computer generated music?

I came across this today: (… it unfortunately seems to work only with the internet exploder)

But this seems quite bad in the sense that even very simple harmonic “rules” seem to violated almost constantly (at least for the couple of times I tried to generate pieces of Latin music).

Personally, I was thinking about a simple approach to tune suggestion as follows. First you feed the system a few hundred tunes, ideally already broken up into musical bars where each bar more or less corresponds to a certain accord. Then you tell the system about basic harmonic rules, e.g., which accords fit together and which can follow one another given that the whole piece is arranged in a certain musical key. Then you try to recombine the pieces that you already have and maybe also try to insert “random” notes as long as they fit with the key of the piece and the accord for the current musical bar. Then you suggest this (short) tune to a human and get some sort of feedback. For example, for a fixed tune you could tell the system to experiment with different rhythmic patterns.

I could imagine that this simple tune suggestion could give some help to people composing popular music. It would not be adequate to compose complex tunes or pieces with many voices but it could serve to give you the “main” catchy tune of a standard pop song.

Wolfram’s approach (see link above) I don’t find very promising. Generating the whole piece without any intermediate feedback is simply bound to fail.