The marriage of computers and music has spawned digital instruments that sound "real," an interface-the musical-instrument digital interface-that's now an industry mainstay, and composition software that helps generate ideas, assemble phrases, and analyze existing works.

But now machines are actually learning to compose music of their own. Eduardo Miranda at England's University of Plymouth has developed software that generates music from scratch. Other composition software tools rely on high-level mathematical rules, but Miranda's approach is "bottom-up," he says. His software, which grew out of his research at the Sony Computer Science Lab in Paris, France, adapts ideas from the field of artificial intelligence to create a sort of virtual orchestra. Ten "players" get together and generate simple sequences of musical notes. Each player is programmed to listen, evaluate, imitate, and generate variations. After running for a few days, the artificial society produces haunting melodic streams.

These melodies are still ploddingly crude. "This is a beginning to getting a computer to create something new," Miranda says. The next step, he adds, is to evolve rhythms and dynamics.

 It's too soon to say whether creative machines will supplant their flesh-and-blood counterparts, but "the technology is here," says Rodney Waschka II, a North Carolina State University researcher. Initially, such technology will help human composers by speeding the process and providing new ideas. In the long run, boy bands beware.