Gordon Monro - Second Iteration conference

Second Iteration Conference

I attended this conference, and here are some comments, with a focus on music-related presentations. Usual disclaimer: this is a personal view of a complex event.

The conference, described as the "second international conference on generative systems in the electronic arts", was held at the Caulfield campus of Monash University, Melbourne, on 5th – 7th December 2001. The conference was ably organised by Alan Dorin. He was assisted by Jon McCormack and Troy Innocent, but the main burden fell on Alan, so there were no exhibitions or performances, just the conference itself. However, for me it was very interesting and worthwhile.

At most sessions there were around 50 people; since some could only stay for part of the conference, there were about 80 participants in total. About a dozen of these were from overseas; they made a very significant contribution, so the label "international" was justified.

The presentations covered a wide area: animation, music, architecture, robotics, biology, and theoretical issues.

The conference Web site.

Generative systems and generated art

"Generated art" involves some process which is created or set up by the artist but then proceeds to operate more or less independently of its creator. Furthermore, the actual process is an important part of the artwork. This conference had a sub-theme "emergence", which is discussed below.

The commonest technique used in the conference was that of "genetic algorithms" (GAs for short) or "simulated evolution". Suppose we want a system to generate musical phrases. We treat each musical phrase as an organism, which has "genetic material" (just a string of numbers) and a "body" which is the actual musical phrase. We select two phrases as "parents" and breed them to produce "children" (whose genetic material is a mixture of that of their parents). We can also have "mutation", by making random changes in the genetic material. We need a process to convert genetic material into "bodies" (musical phases), and then we subject the resulting phrases to selective pressure: we need a "fitness function" which determines how good the musical phrases are. We keep the best phrases, breed from them, and so on.

In scientific and engineering problems, there is often a fairly clear idea of what a good solution is, so the fitness function can be straightforward. In an artistic context we may want to create "interesting" musical phrases, without any idea of how to measure interestingness. One solution is to make an interactive system. The user selects the most "interesting" musical phrases of each generation and allows them to breed, while the rest are discarded. If we want a fully automated process, it seems that getting a good fitness function is the hardest part.

Apart from genetic algorithms, it seemed that half the people at the conference had had a go at simulating ant swarms.

Musical presentations

There were three presentations on breeding music. Tatsuo Unemi (Japan) presented work on composing short jazz pieces (for medium-sized band), Andrew Brown (Brisbane) described work on breeding continuations of melodies from a short initial seed, and David Birchfield (USA) described and demonstrated an elaborate real-time composition system.

Tatsuo Unemi's system had a recursive patterning process built-in, so it produced successful results within a limited range of variation. Andrew Brown's process was more open-ended and less immediately successful. His fitness function was a weighted average of various characteristics, including note density, rhythmic variety and repetitions of pitch patterns. Random initial melodies did evolve to something more coherent. However, Andrew also had a process for generating reasonably structured melodies. If these were used as the initial population, the genetic algorithm was unable to improve them. Rob Saunders (of whom more below) suggested that instead of just one fitness function (Andrew's weighted average), it might be better to use several functions, and keep melodies with a high score in one or more functions.

David Birchfield's system used the interesting idea of genetic information at several levels. Each note had its own genetic material for pitch, duration and timbre, but in addition each phrase had independent extra genetic material controlling the phrase as a whole, and then there was still further genetic material controlling a section (group of phrases), and so on for about two more levels. David played a video of himself performing (as percussionist) live with the system as it evolved from an initial random state to a coherent structure of block chords.

An audio-visual work involving genetic algorithms was Rodney Berry's (Japan) "Listening Sky". Alan Dorin also worked on this piece, in which objects, represented by pointillist swirls of dots, move and breed on the surface of a sphere. Each produces a melody, which affects attractiveness for mating purposes, though the colour of the object is also involved in some way. Rod was eventually able to demonstrate the work running on two laptops (one for the visuals and one for the sound).

There were two other presentations concerned with sound and music which did not involve genetic algorithms. Garth Paine (Melbourne) described several of his interactive installations, which in general produce sounds which in some way respond to the movement of people in the installation. His latest work "Gestation" involves realtime generation of graphics as well, utilising ultrasound images of his friends' unborn children. Ananda Poernomo (Melbourne) described a system intended to assist composers who work with so-called set-theoretic composition. This is strongly associated with 12-tone serialist composition, and takes sets of pitch-classes as a fundamental concept. The software presented by Ananda can carry out transposition, inversion and so on, and can apparently help with the organisation of large-scale structures.

In the open session at the end of the conference there were two short music-related presentations. Jonathan Crane (Sydney) talked about some experiments in. breeding timbres, using FM and additive synthesis; I talked about my piece "Peer Pressure", which is an aural representation of a process which causes fireflies to flash in synchrony. Additionally the only actual performance not in a presentation was music generated by Andrew Garton (Toy Satellite). I have to plead guilty to treating this as muzak; I was too busy talking with other conference participants and getting into the grog that Alan Dorin provided. So I can't report on what Andrew's system was doing.

Other presentations

Of course, music was only one of the topics of the conference. Here are summaries of some of the other presentations.

The opening presentation was by Katherine Hayles (USA). Katherine was well-known to a number of conference participants through her book "How We Became Posthuman". Her base is in literary theory and criticism, and she is concerned with the relationships between literature, new media, science and technology. In her presentation she described and commented on an alliance between the United States military, the entertainment industry and academic computer scientists to construct elaborate simulations for the military. The one Katherine described was set in a Bosnian village, where a riot is developing near a military armoury, but also a soldier has been involved in a car accident in which a child has been injured; additionally a news cameraman is present. The participant has to decide on priorities and allocate forces. The simulation had computer game elements combined with sophisticated voice recognition software and motion tracking. It raises questions about ideological biases and portrayal of a foreign culture as well as questions about the reactions of the participants: did they become immersed, or did they treat it as just another artificial test? (There was certainly a "correct answer".) Also, a staggering amount of money was spent.

Maria Verstappen and Erwin Driessens (Netherlands) presented an impressive body of work which encompassed both screen-based works such as their "IMA traveller", a constantly changing cellular automaton landscape, and construction of actual physical objects. "The Factory" was an automated installation. In it, a dipper scooped up a quantity of molten wax from a heated container and dropped it into a bath of water. The wax formed an irregular blob of interesting (and often suggestive) shape, which was transferred to a conveyer belt and passed in front of a video camera, which recorded the blob. It was then returned to the wax container and melted down. Then another blob was created, and so on. Each blob stayed in existence for about 6 minutes, the only permanent record being the videotape.

Christa Sommerer (Japan) also presented a body of work, mostly done in collaboration with Laurent Mignoneau. Some of it concerned a system where Japanese mobile phone users could have little "artificial life" creatures downloaded to their phones. The phone owners could then nurture these creatures, tamagochi style, and even breed them with other phone-owners' creatures. There is also an option to get the phone number of the other creature's "owner"... . The most engaging piece that Christa presented was "A-Volve", a sort of virtual aquarium. Visitors to the installation could draw a profile and a cross-section; the software would then create a 3D creature. The creatures were projected from below into a tank of water, so they appeared to be swimming in the tank. The creatures could prey on each other, and also mate. In addition, there was a camera above the tank, so by putting one's hand in the water one could, for example, immobilise a predator and allow one's own creature to get away.

These talks illustrate what became two sub-themes of the conference. Some artists seem to have become frustrated with screen-based works and want to make real objects, though Maria and Erwin appear to move happily between exploring impossible worlds on screen and working with real artefacts. The other sub-theme concerns the nature and extent of interaction with an installation. The works of Maria and Erwin generally have little or no user input, whereas Christa's depend heavily on user interaction, as do the works of Garth Paine, John Tonkin and other conference participants.

Maria made an interesting comment: some visitors to her studio who at first did not know that some of the graphic images were produced by a computer program expressed relief when they found out: they did not have to guess the artist's "meaning"!

Stuart Bunt (Perth) described the "SymbioticA" project, which uses tissue culture techniques for artistic purposes. This year the group exhibited at Ars Electronica (Linz, Austria). The installation, entitled "Fish and Chips", consisted of some fish neurons (part of a goldfish brain); the electrical activity of the neurons drove two robot arms which in turn drew on sheets of paper. There was also a sound output. The neurons apparently produced patterned white noise, and the drawings were mostly crosshatched lines. Stuart said that the team originally intended to have a real fish and chips shop there so that visitors could eat fish and chips while watching the installation. This was vetoed, ostensibly on health and safety grounds, but in fact because the organisers were trying to improve relations with the conservative citizens of Linz (whom they had thoroughly offended the previous year), and it was felt that eating fish while watching fish neurons performing would be too confronting.

Rob Saunders (Sydney/UK) presented his work on "curious agents". Each agent (a small part of a computer program) produced "artworks" and made them available to other agents. (The artworks were in fact images produced by genetic algorithms.) An agent would take an artwork from another agent according to a criterion of "creativity", which Rob took as "appropriately novel": novel, in that the artwork differed from the agent's current production, but appropriately novel, i.e. not too different. Agents with similar tolerance for novelty tended to form cliques; sometimes Rob also observed fashion cycles.

This work provoked considerable discussion, and also, I felt, some misunderstanding. By the way, Rob's model of a closed society of artists consuming each others' work appears to correspond pretty well to the situation of modern music in Sydney. According to a survey carried out some years ago by Fiona Allan, about half of the people attending modern music concerts regularly are themselves active producers, either performers or composers (and presumably most of the rest are spouses and girl/boyfriends).

Jon McCormack and Alan Dorin (Melbourne) presented a paper discussing the somewhat slippery notion "emergence", and there was some discussion during the conference. In the context of generative systems, it appears that "emergence" refers to: behaviour that is not "built-in"; behaviour when the system exceeds expectations; behaviour when the system surprises its creators. In discussion there was reference to some experiments about using genetic algorithms to configure arrays of electronic components to perform various tasks. In one case, it turned out that the resulting circuit was in part acting as a radio receiver and was collecting and making use of radio waves emitted by a computer in the same room. This is surely emergent behaviour!

There were plenty of other fascinating things at the conference. In general the presentations were of a high standard and the atmosphere was of friendly and constructive discussion, though I think there were some ideological fault lines lurking beneath the surface. A couple of the overseas people commented that there appears to be a relatively high concentration of algorithmic or generative artists in Australia. I don't know if this is true, or an illusion created by Alan Dorin's skilful organisation of the conference. In any case, Alan is to be congratulated.