Composing via a Collaborative Genetic Algorithm

Genetic algorithms are an interesting and powerful problem-solving tool. They can be used for some of the same applications as Monte-Carlo simulation, but the fundamental operating mechanism between the two is different: a Monte-Carlo method throws random values at a problem and arrives at a solution via the law of large numbers; a genetic algorithm, although also reliant upon multiple iterations, seeks out a solution through random mutation and selection–modeled after biological genetics. In other words, a genetic algorithm is a more structured and directed way of using a stochastic method to arrive at a solution. A pseudo-code algorithm (courtesy of Wikipedia) can be written as:

Choose initial population
Evaluate the individual fitnesses of a certain proportion of the population
Repeat
Select best-ranking individuals to reproduce
Breed new generation through crossover and mutation (genetic operations) and give birth to children
Evaluate the individual fitnesses of the children population
Replace best-ranking individuals
Until terminating condition

This works surprisingly well for many science and engineering applications. But what about music? The “fitness” criterion is generally expressed mathematically, but a subjective human opinion of good/bad music could also determine fitness. My preliminary idea is as follows:

1) Randomly generate a set of music clips (probably just one voice to begin with, such as a bassline)
2) Listen to the clips, and vote on which ones are good (“fit”) and which ones are not
3) Keep the good clips, and produce children via crossover and random mutations
4) Repeat the process until a desirable musical progression is produced
5) Repeat to allow for multiple voices (bass, melody, rhythm, etc.)

I think this could produce some interesting results. I’m going to play around with this idea in my head some more, and then maybe develop the idea. If I do, the listening/voting process will likely be through a web interface so anyone can contribute to the selection process.