A while back I was reading David Cope’s book on the idea of computer created music. Cope is using computers to create music that sounds like it was written by humans, and that’s a laudable, worthwhile goal. But I find myself wondering if the real value of computer music would be creating music only computers could write. Could computers create music that would tax the compositional abilities of mere humans?

What would this music be like? I suppose really long pieces – songs that go on for hours or days might be an example. So too could music that requires an incredible attention to detail, like music with very precise rules about how notes vary their frequency. Or perhaps a kind of variations-on-a-theme process that could generate endless variations on a melody.

Such music might be interesting, but I’ll grant you it might be quite boring; some people these days can barely pay attention to a ten minute song, much less a day long one. I think this music would be probably fall into the category of “furniture music”; music meant to be in the background.

Today I came across the website electricsheep.org. This site has a collection of computer created artwork (non-representational art) generated by computers. The computers use a genetic algorithm which is essentially a software process that duplicates the process living creatures go through to evolve. As I understand it, genetic algorithms introduce mutations into output and if those mutations are beneficial, they are adapted as traits. If the mutations are not beneficial it drops them.

I know that’s a bit hard to understand so let me explain with a practical example. Let’s say the electric sheep computer(s) start(s) off with a big red circle. Some possible variations could be “Make the circle blue,” “make the circle more square,” “fill in the circle with polka dots” etc. The traits that succeed get added to the artwork’s “DNA” and carry forth into new generations with additional mutations. But what constitutes the idea of success in this realm? People rank the artwork on the site. Top ranked artwork is more successful that lower ranked art.

Could such a process be applied to computer music? What’s electric sheep doing exactly? It’s taking shapes – circles, squares, grids, cloud type shapes, and changing them. Could something similar happen with music? First we’d have to find some corollaries to shapes in music. This could be chords, melodies, maybe even rhythms. Could a genetic algorithm be applied that morphed these music elements, and then tracked listener preferences, adopting the high scoring mutations into the music’s DNA? That’s the kind of music only a computer could compose.

One final point here. I suspect one other area worth exploring in computer generated music would be microtonal music. This is…

…music using microtones—intervals of less than an equally spaced semitone. Microtonal music can also refer to music which uses intervals not found in the Western system of 12 equal intervals to the octave.

Manipulating music on that level seems like something computers would be good for.