Public tastes may exert a kind of "natural selection" that improves music's appeal—up to a point—finds a study published this week in Proceedings of the National Academy of Sciences. The research supports the theory that culture and art are shaped by processes similar to those in biological evolution.

Whereas past research using computer models has probed whether popular songs could evolve by selecting for particular musical attributes, "the real difference here is the selection process," says Armand Leroi, professor of evolutionary developmental biology at Imperial College London and a co-author of the paper. Instead of using a computer program or an individual to select which songs "reproduce," "we just let public taste decide," Leroi adds.

He says that people are very comfortable with the idea of natural selection in organisms, but when it comes to music they fail to recognize the creative power of consumer preferences on which songs survive. There is a perception that only the composer and performer are the innovators.

In the experiment, the researchers first populated their "tune world" with short clips of sound encoded by a computer program. They mimicked sexual reproduction by exchanging code between two "parent" clips, analogous to genetic recombination, and by adding random mutations to create "daughter" clips, after which the parent clips would "die." For each generation, they kept 100 songs.

After the researchers used random processes to develop a "gene pool" with sufficient diversity, members of the public were invited to the DarwinTunes Web site and rated the songs on a five-point scale from "I can’t stand it" to "I love it." The computer program then recombined the codes of the top-rated songs, and new songs emerged to be rated in the next generation.

  • Click arrows for audio
  • Darwin Medley at 0 Generations
  • MP3 file
  • Darwin Medley at 150 Generations
  • MP3 file
  • Darwin Medley at 600 Generations
  • MP3 file
  • Darwin Medley at 3000 Generations
  • MP3 file

As the generations "evolved" and progressed, the clips started sounding like music. "What we're making here sounds kind of like electronica dance music," Leroi says. In part, the style might have been influenced by the type of people who would choose to visit a Web site such as DarwinTunes. "I wouldn't be surprised if it's a whole lot of little ravers generating the kind of music that they really like," Leroi says. The results have proved popular among the scientists, too: "We've had some informal little parties in which we've played some DarwinTunes," he says. "Well, you don't want to listen to them all night, but it's definitely listenable to, and certainly danceable to."

After DarwinTunes had evolved for about 2,500 generations, users rated songs from different generations side by side to produce data about the way the songs' musical appeal changed over time. The researchers found that the average likability increased rapidly for about 600 generations and then settled into a kind of equilibrium, or stasis.

Using previously existing music-analysis technology, researchers compared two musical traits, chord clarity and rhythmic complexity, with the users' ratings of the clips' appeal. They found that both were correlated with maximal listener appeal, and that all three characteristics entered stasis around the same time. Researchers estimated an "evolutionary peak" of maximal appeal based on chord clarity and rhythmic complexity and found that the tunes had not reached that peak. Rather, using a variety of statistical techniques, the group determined that the most likely cause of stasis was "low transmission fidelity"—in other words, the children were "too different" from their parents, so the favorable musical qualities were not being passed down intact.

Further analysis indicated a couple of likely reasons for the low transmission fidelity. Recombination, the process by which the genetic material of the two parents is blended, might break up genes that positively influence each other. Alternatively, the delicate structures that make music pleasing could be very sensitive to mutation. The researchers could change the computer programs to overcome these difficulties. Leroi says, "We want to fiddle with the mutation rates, fiddle with the recombination rates." He wants to see if they can get closer to a peak in musical appeal.

The paper notes that although this study suggests that individual consumers' tastes shape music produced by DarwinTunes, the evolutionary process in human-composed music is more complex. Composers and musicians themselves are responsible for a large amount of musical development, and social groups influence individual ratings as well. The researchers want to incorporate these other influences into musical software. "This could be much, much bigger," Leroi says. He imagines modifying DarwinTunes, even turning it into an iPhone game with not thousands but millions of users. "You could produce a whole new musical ecosystem, as it were."