A few things seemed to have aligned themselves recently, so here’s an attempt to cobble them together into a meaningful form. This is an edited version, with more references, but without a proper conclusion.
Richard D James has answered 25 questions from musicians, DJs and producers in an article for the German magazine Groove. In response to a question from Mate Galic, a founder of Native Instruments, about the use of hardware and software, RDJ reveals his involvement in a new form of evolutionary software for making music:
But I’ve actually recently hired a Chinese programmer to make a music software for me. It’s taking the concept of mutation into music software. You give the program some sounds you made and then it gives you six variations of it and then you choose the one you like most and then it makes another six and it kind of keeps trying to choosing the variations by itself. It’s a bit like that, but more advanced, but basically it starts with a sound, analyzes it, then does different versions of variations. It randomizes, it compares all of them to the original and then it picks the best one. It sounds totally awesome, but it needs to be tweeked a little bit. I will continue with this. I have a whole book full of ideas for software and instruments.
It seems to me that this is an example of an approach to music making that I’ve written about previously in the post Computer Fatigue and the Rise of Sonic Complexity, which discussed a trend of moving away from digital music making as a way of seeking richer and more complex sounds. These musicians (mainly those involved in electroacoustic music) are turning away from the computer-generated sounds because they are too clean, precise and uniform. RDJ expresses the same attitude when he says that he favours analogue synths because “everything on the computer just sounds perfect”. But some musicians have decided to stick with the computer and face the challenge of creating more complex sounds. Computer musicians have at their disposal a method described by Manuel De Landa as ‘The Virtual Breeding of Sound’ (2008, in: Miller (ed.) Sound Unbound, pp.219-226). De Landa describes “topological” transformations of sound as a means of exploring the search space of musical possibilities, in a process analogous to evolutionary variation and selection.
Clearly, current uses of genetic algorithms display only the tip of an iceberg, the exploration of which will perhaps take decades. This is a sobering thought, preventing us from being overenthusiastic about our current capabilities of breeding sound, but simultaneously it is a source of excitement at all the unknown domains waiting to be discovered.
Genetic algorithms were popularised by John Holland’s (1975) book Adaptation in Natural and Artificial Systems. A friend of mine, Honar Issa, gained a PhD by developing genetic algorithms to find the optimum construction of steel portal frames for industrial buildings. Richard Dawkins also wrote about this type of evolutionary mechanism in The Blind Watchmaker (1986) with the idea of “Biomorphs” – simple virtual creatures where one organism gives rise to 8 offspring with slight variations. Selecting one of these offspring as the new parent then generates 8 new variations on that theme, and so on. There are a few examples of interactive Biomorph applications, such as this one at EmergentMind.com. By picturing a (real) organism, you can fairly quickly arrive at something that resembles and ant or a crab, for example (these virtual creatures have bilateral symmetry, as do most higher lifeforms). Thus the system offers a means to explore genetic space through a process of artificial selection.
It is important to note that evolutionary search space is not pre-defined. In the book Investigations (2001), Stuart Kauffman explored how this space of possibilities expands and develops as its constituent organisms themselves evolve and multiply with variations. Kauffman called this co-evolving space of emerging possibilities the “adjacent possible”. This is the space of possibilities that opens up with each new genetic variation. For example, ornamental feathers that covered the skin of prehistoric reptiles opened up the possibility of flight. As the function of the feather changed, so did the potential for new behaviour. Kauffman also suggests that this describes the development of human economies. I think it also applies to the development of art and music. More recently, the mathematician Steven Strogatz and others published a paper called The dynamics of correlated novelties that proposes a mathematical model for the way we explore the adjacent possible to find new things. Strogatz et al. suggest that this models the way in which we find new music, for example, where we find new things that are similar to or related to the music we already know of. In this way, our cultural capital (Bourdieu) continually evolves and expands. The path taken by evolutions from one form to another
There is a subtle but significant difference between finding new music and making new music, however. In the first case, it is a matter of exploring a space that already exists; in the other, new possibilities are created with each new development. The difference is similar to the distinction that Margaret Boden draws between two types of creativity in The Creative Mind: Myths and Mechanisms (1990). In a summary of that book (PDF), Boden says:
What you might do, and what I think you should do in this situation, is to make a distinction between “psychological” creativity and “historical” creativity. (P-creativity and H-creativity, for short.) P-creativity involves coming up with a surprising, valuable idea that’s new to the person who comes up with it. It doesn’t matter how many people have had that idea before. But if a new idea is H-creative, that means that (so far as we know) no-one else has had it before: it has arisen for the first time in human history.
P-creativity is similar to finding new music because it involves personal, but not historical, novelty. In contrast, H-creativity describes the creation of new music. Algorithms for the P-creative discovery of new music are now quite widespread, such as those offered by Last FM, Spotify, and iTunes Genius. But algorithms for the H-creative creation of new music are less common. Aphex Twin’s music mutation software is one example, which realizes De Landa’s idea of genetic algorithms for topologically transforming and breeding sounds. Autechre and Rashad Becker also appear to use something like this kind of approach to creating new music by exploring the adjacent possible through the evolution of sound. Maybe there’s potential for a crossover between the creative use of these algorithms by musicians and the analytical use of statistics and information theory (e.g. by Strogatz and others) to describe the creation of new music.