Artists have been using biological and physical systems as models for artworks since time immemorial...blahblahblah. Some of the innumerable areas that have been explored (an entirely arbitrary list):
Many of these areas were originally explored by scientists and engineers searching for novel ways to approach technical/optimization problems and were later co-opted by artists. Two good examples of this are neural networks and genetic algorithms. Others are so useless that the scientists never even bothered, but the artists persisted nonetheless.
The basic idea is that you take some existing phenomenon, like the flocking behavior of birds, and try to figure out how it works. Once you have a rough idea of how it works (or you've read some of the gazillion papers out there on the topic) you start building a computer model/simulation of the system. Then you spend many, many hours tuning the system and trying to get it to look anything like the system it's supposed to be modeling. Finally you give up and decide that what you have is good enough for art. Once you have a more-or-less working system, you can then start using it to drive some aspect of your piece.
The biggest hurdle in all of this is what is sometimes called "the mapping problem." So you've got this cool model or algorithm, and it's spitting out a bunch of information; statistics, coordinates, weights, metrics, rankings. Now what do you do with that information? How do those numbers relate to the piece you're trying to make? Why are they significant or interesting? How are you going to map those numbers onto the parameters that control or create your piece? This is really the essense of the problem. Many people have made pieces that use, say, a neural network to control a piece. But just the act of using a neural network is not interesting in and of itself, any more than the fact that someone decided to use d-minor or a microtonal scale for a piece is interesting.
The tricky thing about the mapping problem, and the thing that is sort of the secret shame of the technique, is that almost all mappings end up being arbitrary. There is great power in an arbitrary mapping. Where one person might decide to map the digits of PI to chromatic midi notes and bore an audience to tears for hours on end, someone else might figure out a clever way to map the digits to the melody of a Beatles song and then go on to claim that the Beatles were somehow "tapped into the fabric of the universe." There have been several stories recently where people have mapped information from the Human Genome project (or protein folding sequences, etc.) onto musical scales. Grand claims are then made for the resulting music (which tends to be bathed in glorious reverb to hilight its profound nature.)
I'm being facetious here, and I'm not trying to say that using information from the Human Genome Project is a bad idea; I think it's a glorious idea. What I'm trying to get at is the fact that it's very easy to begin believing your own press when playing with these mappings. As anyone who's played with a random number generator for awhile can tell you, given enough massaging even a random stream of numbers can be made to sound pretty interesting with the right mapping. At the same time, a really interesting technique or dataset can easily be made to look useless given a brain-dead mapping.
I don't want to suggest that there's a right or wrong way to go about using models to create art. Sometimes an arbitrary mapping that just happens produce cool sounds is exactly what you're looking for; other times a piece doesn't make sense unless there's a more explicit and precise logic behind the mapping. What I'm suggesting is that these are important issues, and that you should, at some level, address them in your work.
There are literally gazillions of systems that people have modeled and used to make art. We'll briefly cover just three of them in class: genetic algorithms, the ant colony algorithm and neural networks. Other popular topics include artificial life, cellular automata, fractals, L-systems and grammers, Markov Chains and many, many more.
The web is a good source for information on most of these topics; magazines like Leonardo and The Computer Music Journal are also full of this stuff, as are library thesis collections.
Find some biological or physical system you'd like to work with. Do some research and find out what other people have done with similar systems. Write a proposal for a project that uses your system. Include information on how the system works, why you find it interesting, how you might implement your model, and how the model will be used in the project.