It has also given us the splendours of human culture. This may seem like a bold claim, but it is self-evidently true. People copy cultural artifacts – words, songs, images, ideas – all the time from other people. Copying is imperfect: there is “mutation”. Some cultural mutants do better than others: most die but some are immensely successful; they catch on; they become hits. This process, repeated for fifty thousand years, has given us all that we make, say and do; it is the process of “cultural evolution”.

However, the underlying mechanisms are poorly understood. For example, how important is human creative input compared to audience selection? Is progress smooth and continuous or step-like? We set up DarwinTunes as a test-bed for the evolution of music, the oldest and most widespread form of culture; and, thanks to your participation, we’ve shown that reasonably complex and pleasing music can evolve purely under selection by listeners.

How DarwinTunes works
DarwinTunes is a computer program developed by Dr Bob MacCallum, a bioinformatician at Imperial College London.

In the first instance DarwinTunes does three things. It stores a lot of computer-generated songs on a server, presents the songs to the world via a web interface, and allows people to listen to and rate them.

So far this sounds rather like the online music services on offer today – but there is an important difference. Whereas the songs on Last.fm, Pandora, Spotify, etc are made by singers, songwriters and bands, DarwinTunes makes its own songs, and the songs get better through time.

DarwinTunes does this by evolving new songs. Each song in DarwinTunes is based on a bit of computer code. These bits of code periodically “reproduce” to form a new population of daughter songs. However, the daughters are not the same as their parents. First, parent songs have “sex” – they exchange bits of their code so that their daughters are unique mixtures of their parents’ codes. Second, daughters “mutate” – they contain random changes in their code. The result is that every generation of songs is subtly different from the one that preceded it.

These processes are analogous to those make genetic variety in living things; we think that that they are also at work when humans make new songs. They are not, however, enough for evolution. Many daughter songs will not sound very good; in fact, they may sound worse than their parents. (In the same way many mutant animals are very unhealthy.) In order to get evolution we need a way of identifying attractive songs – and ensuring that they increase in the population. In short, we need a selective force. And we have one: you.

Visitors to the website are asked to listen to and rate songs randomly plucked from the current population. The songs are actually short (four bar) seamless loops – anything longer is impractical at the moment. After being auditioned, the score of each song is averaged (over multiple visitors) and is used to determine whether or not it will reproduce and so contribute to the following generation. Songs with high scores have many progeny; songs with low scores have few; songs with “nul points” go extinct.

This process, continued for many generations, is nothing less than evolution itself. We think it is very similar to the musical evolution that happens in the real world: the world in which a thousand bands bang away in a thousand garages, mutate the musical idiom of the day, and test the results in the market. Most fail, but a few suceed and popular music continues its relentless evolutionary march. In the same way, DarwinTunes will rapidly result in songs that are radically different from the songs in the starting population – indeed, that may sound like nothing ever generated before, but that should be immensely appealing. These songs will be a uniquely democratic kind of art – they will be the result of thousands of individual choices made by hundreds, or perhaps thousands, of people.

For more details
Please see the attached documents below:

• The_population_memetics_of_DarwinTunes.pdf 112.05 KB
• The_technology_behind_DarwinTunes.pdf 95.53 KB