"I really like to swarm in harmony space," Tim Blackwell says as he stares at a few brightly colored shapes darting across a computer monitor. We're sitting in the small studio in Blackwell's house in the northwest London community of Harrow on the Hill. Blackwell is a computer-science researcher and a jazz musician, and he looks the part: shaggy shoulder-length hair and an affable demeanor. As we watch the shapes on the screen, Blackwell plays a series of chords on a music keyboard at his side. With each chord change, the shapes swoop together toward a new spot on the screen, like a flock of birds suddenly spotting a food source. It's a delightful effect, like a musical score in which all the notes have come alive and started dancing across the sheet music. "You could really lose a whole day just playing with this," I say. Blackwell laughs. "More like a whole year," he says.
For almost as long as we've had digital computers, enterprising programmers have been trying to teach them how to play music. Perhaps the most challenging remaining hurdle is the spontaneous back-and-forth flow of improvisation. Machines are quick to learn when it comes to rolling out standard chord progressions and following predictable rhythms. But they turn out to be lousy at riffing, precisely because riffing is a much more chaotic sort of pattern, one that relies on intuition more than structure. A computer might be able to convincingly replicate the bass line for Jimi Hendrix's "Are You Experienced?" but a machine-simulated version of a Hendrix solo would be a dead giveaway. The problem gets more complicated when the music in question is "free improv," a contemporary movement—inspired by the "free jazz" recordings from the 1960s of Ornette Coleman and Cecil Taylor—in which all the musicians play improvised solos at the same time, following each others' evolving cues rather than a pre-existing score.
As daunting as it sounds, free improv may yet become a part of the computer's musical repertoire, thanks to sophisticated software programs like the one developed by Tim Blackwell. The key to this breakthrough does not lie in teaching the computer how to think like Coltrane. The key is teaching it to think like a bee. Bees, you see, display an increasingly influential behavior in their everyday interactions. They swarm. In other words, they move in loosely organized groups that form higher-level patterns without relying on a central leadership or master plan. Flocks of birds show similar bottom-up organization.
In June 2001, Blackwell, a computer-science graduate student at University College London, was looking for a master's degree project. He started to think about the connection between flocking animations and his own experience as a free-improv saxophonist. So he contacted Peter Bentley, a professor at the college who had already built a name for himself applying biological models to computing. "I've used evolution to generate novel designs or to detect fraud, immune systems to find hackers, or neural networks to control robots. When Tim came to see me about doing his master's project, I was looking for an interesting application for swarms. As we chatted, we both realized just how similar swarms and jazz musicians are, as strange as it sounds."
Picture a swarm of bees moving toward a summer garden in full bloom. There is structure in that movement, both in the coherence of the group itself and its ability to home in on a specific target. But there is also variation. Swarms of bees don't look like F-14s flying in formation; they look like swarms. This complex behavior, somewhere between order and chaos, derives from relatively simple local rules followed by each bee: Move toward the center of the swarm, avoid colliding with other bees, and follow gradients of scent toward the pollen source. As long as each bee plays by these rules, the swarm will do its magic.
To mirror the swarm behavior of bees, Blackwell and Bentley decided to translate the language of music into a 3-D space, with a dimension each for pitch, duration, and volume. A "particle" moving through that space changes its musical parameters as it changes its position, growing louder, higher pitched, or longer. At any given time, multiple particles explore the space, following the rules of swarm logic: Avoid collisions with other particles, gravitate toward the center of the swarm, and track toward targets that have been defined by the system. Collision avoidance keeps the software from generating too much repetition, such as having the same note played at the same volume by several different particles. Swarm-centering keeps the musical riffs in a coherent zone. For example, you don't want one note that's consistently three times louder than all the other notes.
Targeting may be the most interesting element. A target can be defined by a pre-existing score, by a human musician playing in real time with the software, or by another particle swarm. Imagine a flesh-and-blood pianist who begins to play staccato notes in C minor; a particle swarm playing along in real time would move through the musical space toward compatible notes. The swarm wouldn't directly mimic what the human was playing; it would take the general properties of the human-created music and riff off them. "The striking characteristic of the swarm is its sensitive responsiveness," says Robin Higgins, a pianist and singer who has played more than a dozen pieces with the "swarm music." "Like a good improvising partner, it picked up musical structures I gave it."
Blackwell and Bentley have also experimented with a kind of hall-of-mirrors system, in which two particle swarms follow each other's lead: Swarm A conjures up a certain musical palette, which becomes a target for swarm B. The response generated by swarm B becomes a new target for swarm A, and so on. Thus far, the performance that Blackwell and Bentley are most proud of involves precisely this feedback structure but with a twist: two swarms playing off each other, while Blackwell "conducts" by tweaking variables as the composition unfolds.
"In this case, I was periodically fixing swarm B to produce sustained chords," Blackwell says. "I was also changing instruments for swarms A and B—like orchestral sections. Swarm A was always playing melody, and I fixed the interpretation to a small range of beats per minute. This means that swarm B moves toward these targets in a more restricted part of its music space. And then swarm A is attracted to where swarm B is, and so on. The result was an astonishing high-speed chase through music space."
It's an arresting image—musical swarms chasing each other through a mathematical space—but what does the music sound like? You can hear the results for yourself at Tim Blackwell's site (www.timblackwell.com). Be forewarned: It's a challenging sound. The closest thing on the pop charts is probably the electronica dabbling on recent Radiohead albums. But it grows on you, particularly if you're already versed in the unstructured play of improvisational music. "I'm not a musician," Bentley says, "but I've worked with them in the past, so I understand how the musical mind works, and sometimes they can find rapture in the most bizarre of noises. But when it comes to swarm music, even my humble musical appreciation abilities are sufficient. The way the notes chase each other, causing cascades of melodies, countermelodies, and harmonies, is wonderful."
To my ears, the swarm compositions, even when judged by free-improv standards, have an unfinished feel to them. Of course, Blackwell and Bentley have only just begun to explore the possibilities of their software. Blackwell thinks it may be most useful as a compositional tool, not as a replacement for the creative process. Just as the flight of a real-world bumblebee inspired Rimsky-Korsakov to write his now-ubiquitous melody, the flight of these virtual swarms may inspire a new generation of composers, creating passages of music that would then be shaped and refined into final renditions. The swarm doesn't write songs; it suggests new avenues to explore.
I find this idea enticing precisely because it swarms in the face of the preconception that computers are there to store our files and record our keystrokes and nothing more. Swarm music suggests that this virtual-stenographer role may be an artifact of computing's early days. Simple digital machines are good at capturing and storing information. Complex machines are good at generating new kinds of information and triggering new connections, even if that information must eventually be polished by humans. What they produce is more launching pad than archive. That's why I suspect the rough-edged quality of the swarm compositions may turn out to be, as they say, a feature—not a bug. Blackwell and Bentley's creations don't yet sound like finished products, but they do sound like a tantalizing place to start.