DarwinTunes

In our student experiment we had three evolving populations of loops (or channels). Each student was assigned, at random, to one of them. (See “Population Memetics of DarwinTunes” (PDF)) The public channel is a fourth quasi-replicate. We had three channels because we wanted to know whether or not music evolution is repeatable. To put it another way, is the evolution of music deterministic or contingent?

We have a lot of analysis to do – but just listening to the channels suggests some really interesting behaviour. Two channels (#1 & 3) evolved rather nicely. After ~10k selective events, their songs sound a lot better than they did at the start. They also sound quite similar. The public channel also seems to evolving in much the same way.

After one or two generations	After 75+ generations under selection
Populations 1,2&3	Population 1	Population 2	Population 3
Audio snippets are made with loops that two or more students rated, on average, better than "I like it".

That’s all rather deterministic. But one student channel (#2) seems to be very different! If anything, it’s gone backwards – the loops have become less attractive and less complex than they were at the start! (At least that’s what our ears are telling us – though we really have analyse the songs formally to find out.) A plot of mean genotype complexity (actually pseudo-complexity, measured as the compressed size in bytes) vs. generation confirms our suspicions:

What’s going on here? We’re not sure, but one possibility is that the fitness landscape of DarwinTunes is complex. Three of the four channels have marched smoothly up the same big musical hill, but something bumped population #2 off course so that it has found its way onto a quite different local, lower, optimum.

That’s just speculation. We are going to have to crunch a lot of numbers to really find out what’s going on. Excitingly, the public channel is now at 20k selective events which should give us a look at a longer evolutionary trajectory. We’re pretty sure that it’s still evolving – but for how long?