Tunes without composers: music naturally evolves on DarwinTunes

ByEd Yong
June 18, 2012
6 min read

The tunes embedded above weren’t written by a composer, but fashioned through natural selection. They are the offspring of DarwinTunes, a program which creates bursts of noise that gradually evolve based on the preferences of thousands of human listeners. After hundreds of generations, tracks that are boring and grating soon morph into tunes that are really quite rhythmic and pleasant (even if they won’t be topping charts any time soon).

DarwinTunes is the brainchild of Robert MacCallum and Armand Leroi from Imperial College London. “We suspected that musical styles evolve through Darwinian natural selection,” says MacCallum. “They are copied and modified from artist to artist and generation to generation, with popular styles more likely to be copied as they get more exposure. “ The duo created DarwinTunes to see if music could actually evolve in this way.

The DarwinTunes tracks are all 8-second-long loops, each encoded by a ‘digital genome’ – a program that determines which notes are used, where they’re placed, the instruments, the tempo, and so on. The genomes of two parent loops can shuffle together in random ways to produce daughter loops, which also develop small random mutations. This mimics the way in which living things mate and mutate. It also mimics the way in which composers merge musical styles together, while inventing new motifs.

The experiment began with 100 randomly generated loops. On the DarwinTunes website, listeners could listen to these and rate them on a five-point scale, from “I can’t stand it” to “I love it”. Every time 20 loops were rated, the top 10 pair off, mate with each other to produce two daughters, and die. At any time, there are only 100 loops in the total population.

To date the loops have been evolving for 3,060 generations, and over 50,000 of them have been born. By taking loops from DarwinTunes’ entire history and asking volunteers to rate them, MacCallum and Leroi showed that they became more appealing with time. For example, they were more likely to contain chords found in Western music and they contained more complex rhythms. “We hoped for slightly more “advanced” music, but were very happy with the results,” says MacCallum.

This upward rise in appeal only lasted for 500 or 600 generations. After that, the loops hit a plateau and apparently stopped evolving. MacCallum and Leroi think that this is because the loops become so complex that their intertwining melodies and rhythms don’t merge very well. The act of mating, rather than combining the best of both parents, ends up splitting up elements that work well together.

Alternatively, it may be that as the loops become well adapted to the tastes of their listeners, it becomes harder to change them without messing something up – they become trapped in an adaptive peak, unable to reach a new peak without first crossing into a valley. Both of these processes have their counterparts in the world of real genetics. MacCallum and Leroi argue that this might explain why many old musical styles tend to be very conservative, changing little over thousands of years.

LIMITED TIME OFFER

Get a FREE tote featuring 1 of 7 ICONIC PLACES OF THE WORLD

DarwinTunes is the latest in a line of digital evolution programs, where computer code copies itself, mutates, evolves and adapts. For example, in The Blind Watchmaker, Richard Dawkins describes a programme of the same name that can evolve complex shapes from initially simple collections of lines. These programs never fully reflect the reality of evolution, but they allow scientists to ask basic questions about evolution in a controlled way. They can set up controlled experiments, repeat them, replay evolution from specific points, and analyse how specifically their artificial creations have changed. It’s incredibly hard (but not impossible) to do that with actual living things.

But Michael Scott Cuthbert, who works on computer-aided musical analysis at MIT, is sceptical that the approach tells us anything about the evolution of music. “They have shown that people can sense a glimmer of the things they like about music even when most of it consists of sounds they hate,” he says.  “But it doesn’t give any information about why music sounded differently in the past, why people like different things today, or how music might evolve in the future.”

“Suppose you randomly threw car parts into piles and asked people to rate those they’d most like to buy,” he says. “Then you took parts from the highest-rated heaps, and rearranged them into new heaps.  People might hate all of them at first, but they’d probably rate the ones with four tires or a trunk in the back or a steering wheel in the drivers’ seat higher than the rest. Do that long enough and I wouldn’t be surprised that you’d eventually get something that looked like a 2011 Honda Civic.  But that doesn’t mean that that’s how a car is made.”

MacCallum and Leroi acknowledge that real music changes in a more complex way than DarwinTunes currently captures. Composers write music with their own intentions, while listeners choose music based not just on what it sounds like, but on whether other people like it too. DarwinTunes could be changed to include these dynamics – volunteers could combine the loops themselves, and listeners could see earlier ratings.

“The big question for me is can we bring the quality up a level where you don’t have to be curious about the science to take part?” says MacCallum. “We can do that if we had millions of users, and segregated them based on musical genre preferences.  It’s a chicken and egg problem though!”

Reference: MacCallum, Mauch, Burt & Leroi. 2012. Evolution of music by public choice. PNAS http://dx.doi.org/10.1073/pnas.1203182109

Image by Pedro Sanchez

More on music:

Go Further