Survival-of-the-Catchiest: How Music “Evolves”

DarwinTunes lets computer-generated music evolve naturally according to user ratings.

  • Share
  • Read Later
Digital Vision / Getty Images


We all know the pop music industry can be dog-eat-dog, but could Darwin’s theory of “survival of the fittest” actually create the next pop hit? A novel experiment is attempting to compose popular hits by letting computer-generated music clips “evolve” into music that listeners like.

You might be familiar with Pandora, the online music streaming service that uses listeners’ like/dislike ratings to generate personalized radio programming. In its ambitious “Music Genome Project,” Pandora seeks to identify the attributes that distinguish the “DNA” of musical styles. With its knowledge of the “family trees” of artists and genres, Pandora uses the preference data its users provide to recommend music for each particular listener.

(LIST: Listen to the All-TIME 100 Songs on Pandora)

DarwinTunes uses a similar “DNA” approach to understanding why we like music—and pushes it further by letting new music evolve through the feedback loop. DarwinTunes throws computer-generated musical clips into a simulated environment in which user popularity mimics the force of natural selection.

DarwinTunes starts out with the musical equivalent of primordial bacteria: short, computer-generated, grating, ugly sound sequences. These primitive specimens are turned loose in a program that generates random variations on each sequence of audio wave forms. Then these audio sequences go out into the wild, to kill or be killed: The researchers had more than 6,000 study participants listen to the diverse audio loops and rate how much they liked each on a five-point scale.

(MORE: The Ever Evolving Theories of Darwin)

Audio sequences that are highly rated go on to further levels of mutation, and even “mate” with other popular audio loops to create next-generation clips. Unpopular sound bytes die off.

The end products (after some 500 generations) are congenial little tunes that are not far from the opening bars of an electronic-style pop song. Click here to listen to one of these musical creatures at 30 distinct stages of its evolution.

The research team, led by U.K. computational biologists, hopes to expand the experiment to include millions of listeners sampling the tracks. Researchers predict that a greater-scale version of the same project could produce music comparable in quality to today’s human-composed electronic dance music. A paper explaining their research is published in the Proceedings of the National Academy of Sciences.

Interestingly, the study’s authors note, DarwinTunes’ model of “pop evolution” may not be that novel after all. Human culture, they point out, is the result of a millennia-long, process of evolution—with people copying and “mutating” each other’s cultural artifacts. DarwinTunes merely provides a micro-sized, sped-up, controlled version of the natural selection process of cultural evolution.

(MORE: Pandora Boasts 100 Million Users; Redesign Incoming)