Who Beyls Us Out?
Kalvos & Damian’s New Music Bazaar, Show #75, 26 October 1996. On the road at the St. Lukas School for the Arts in Brussels. Listen to the interview from the original broadcast [0:35:55–1:24:26].
Beyls explores computer programming as a medium for artistic expression and develops generative systems in music, the visual arts and hybrid formats. His approach views computers as cognitive partners in the process of artistic creation and borrows methods from the science of artificial intelligence. Beyls studied music and computers science in Brussels, Stockholm and London. He published extensively on various aspects of digital art. His work was widely shown and performed at conferences like Siggraph, ICMC, Imagina, ISCM and ISEA. He currently teaches Aesthetics of New Media and coordinates research at the Interaction Lab, University College Ghent. Beyls is also a member of the evolutionary computing team at the Interdisciplinary Centre for Computer Music Research, University of Plymouth, UK.
[Kalvos] Well, after an hour and a half of searching, we have found the composer of the day, on the Kalvos & Damian New Music Sesquihour, and here in an echoing studio at the St. Lukas School for the Arts in beautiful downtown Brussels…
[Damian] You’re sure that’s where we are?
[K] I think in the amount of time, that we might have been back in Amsterdam, I’m not sure, but anyway, our guest today Peter Beyls. Tell us about what you are, because you do a lot of different things.
[Peter Beyls]. Well, I teach here at the school, but for many years I’ve been trying to find out what computers really mean to the artist, and how computers can be used as a means to express yourself as an artist. The only way to do it in a good way, I think, is to write programs, to not just use off-the-shelf commercial software. So the essence of me, for using computers… the strongest argument I would say, is that when you’re using a computer as a medium to program, you can touch your ideas. Computers are a very flexible means to find out what you’re really after.
[K] Are computers today at a level of power where you can do that? Can you really do that, or are we just touching that now?
[PB] Well, we certainly have, at this point. I’ve been using computers for over 20 years, but at this point, I think we have very small and power home computers that are indeed powerful enough to do very complex things. But there is no real relationship between the complexity of the computer, or the performance level or sophistication of the machine, as a function of the ideas, the sophistication of the ideas you try to express as an artist. The computer doesn’t guarantee any quality from the æsthetics point of view.
[K] Right, but it would certainly be hard to perform a contemporary, let’s say, clarinet concerto on an instrument that was still made up exclusive of holes without keys.
[PB] Yes, exactly. But the objective is different, of course. And the things I do are really on two levels, from the musical point of view. I also developed a couple of graphic expert systems, so I’m also active as a visual artist. But in music, I’m interested in two things. One is autonomy, that is how can you build computer programs that contain some of the intelligence which is needed to solve æsthetic problems and to make musical decisions in an intelligent way. The other idea I’ve been working on for many years is the idea of collaboration, or what I’ve been calling “social computing.” That is, where you have a man and a machine, and you’re trying to build structures that allow easy understanding from man to machine and the other way around. So, how can you build an ear, for instance? How can you simulate certain human perceptual capacities, I could say, into a machine, so as to build bridges between two things?
[K] And how do you set about doing that?
[PB] One program I’ve been working on for many years and that exists in various incarnations is a program called Oscar, which is an acronym for Oscillator Artist. The first version of Oscar was written over ten years ago, and the key idea here is that the whole thing is based on a social model, not a scientific or purely artistic or perceptual model. A social model in the sense that Oscar is a computer program that is designed explicitly to interact with a single human being, so the machine tries to express its own character, its own personality, while at the same time it tries to accommodate the external world, the human interactor. So we have some kind of a conflict here, a conflict between expression and integration, and that is really the motor of the program. That’s the beginning of the idea, and that takes us to various studies on perception, and how can you simulate parallel processing, which is a natural thing in a human being? And how do you do that from the musical point of view, into a computer program?
[K] An American audience would be what we would call “product oriented,” so maybe you can address the question in terms of what it does, and what results one can hear from what you have done, and the experience of interacting with this. For example, Henning Berg [see interview in this issue of eContact!] actually personifies his computer, he thinks of it as another person.
[PB] Well, me too in a sense, because that’s the whole thing with Oscar. I’ve always been thinking of computer programs as kind of computer assistants, or artificial creatures that coexist with me. So I’m not interested computers as slaves, but as collaborators, as partners in the process of æsthetic decision-making. So what you try to do is delegate, in fact, the power collaboration and decision-making to a machine. To do that, we hit the key problem, and that is creativity. How can you formalize and implement certain aspects of human creativity into a computer program? This is, of course, pretty difficult to do, because nobody really knows what creativity is, and there are many different theories on that.
On the other hand, also, we like to be pragmatic, and again we hit the unique possibilities of the computer, because typically in the creative process, the problem is still very ill-defined or even incompletely understood. So, how can you program something you don’t understand? This is where help from the machine comes in, because in contrast to what many people think — that your idea has to be finished, in a sense, before you can even program it — I think this is not true. I think you can develop ideas by implementing them and seeing the results, so you get feedback from incomplete ideas, which takes you to the next step, and so on. So, I liken the computer to a kind of conceptual microscope, so you can look into yourself. It is a vehicle for introspection.
[K] As creative artists, we also are always at an incomplete level, and get feedback from ourselves and from others. Is that what you’re trying to emulate with this?
[PB] Not really, what I’m trying to say is that when you implement even a simple idea and you get feedback, the feedback is always different, so you get things that were not anticipated by the programmer. So this, in fact, sheds light on your true objective. As it were, in A.I. (artificial intelligence), which is my background, we talk about search and exploration. So, you find yourself in a search space or a conceptual space, and what you’re trying to do is to go to this point, where in that space, that is in fact the solution to your problem. One thing is that the space changes through the act of searching, you see?
We listen to an excerpt of The Headless Horseman by Peter Beyls [0:45:40–0:47:30].
[PB] The space changes through the act of searching itself, because in every step you take in a search process, new territory is exposed, and again, that was not anticipated. So, one of the fundamental characteristics of a search process is that it is not linear. You cannot say, “Well, I have this problem, I will use that machine as an assistant and go there and see what happens.” No, because you will jump in that space according to what you find, and this is typical to the creative process, I think. This is also one of the very dangers of working with computers, because you can get yourself trapped into what I call “the eternal design cycle.”
[K] [Laughter] Well, some of us have that problem when we compose, too.
[PB] Yes, yes, but you know, a computer amplifies this. And you get feedback. A pretty complex computer program is a system you make, and Oscar consists of about 60 modules that run in parallel. At a certain moment, you find yourself as if you are losing control over the machine or the computer, because the complexity becomes too big. And this is a general problem with any rule-based system, and this is what led to the use of complex dynamics and self-organization as an alternative to knowledge. I’ve been exploring these two approaches in-depth in many different applications.
[K] There’s a popularization of the idea in America of fuzzy logic. Are you using that as your second approach?
[PB] Well, the two approaches are growing closer together these days, and the first experiments with Oscar was that it was a purely rule-based system, a classical kind of real-time expert system, what’s also called a pattern-directed inference system, you know? But the problem with that is that if the machine is faced with a situation that it does not understand, the system breaks down, so you need something that allows for a graceful degradation in performance. And you can find the solution for that in nature, because nature has these abilities of natural adaptation. In fact, the problem is that we want to get to a truly creative system, that is a system that can both interact and express itself on an autonomous level, right?
So, the first thing you need is responsiveness. We want to make something interactive, so it has to be responsive. So, if I do something, it has to respond. On the second level, it has to be adaptive, which means that if large swings in contexts occur, the system has to adapt flexibly. On this next step, you could say, “Well, intelligence is at fault,” which means for me that the system has to express a certain personality. At the next step, you would expect a learning system, so that it can learn from experience and learn from examples from the outside world. Learning is the key to intelligence, I think.
[K] Does it take on your personality? Is it you? And if I can continue that, does it require a new start each time with a blank slate, or does it keep a history?
[PB] Well, I have one version of Oscar that keeps a history, yes. The personality question, well obviously, since I am both the conceiver of the program and the programmer, it is clear that Oscar has certain characteristics of my own character, but I’m really trying to have a system that is complex enough to surprise me. Which is a difficult question, but that takes us back to your other question of the relationship between rule-based systems and complex dynamics. On the one hand, if you rely on explicit knowledge that you can formalize in rules, “Well okay, that will work given certain conditions,” et cetera. On the other hand, if you rely on self-organization like is the case in so many behaviours in nature, you find yourself having an open-ended system that can eventually evolve into something pretty different.
And that’s the way to go, I think, because with explicit instruction (rule-based systems), the search space or the conceptual space in which you work is limited. It works for that domain, but if you start using complex dynamical systems — and I’m not really talking about chaos or fractals and things like that, but truly open-ended systems at the molecular level, or instead of molecules I should say “agents,” which is better (I’m doing a lot of work with agents) — what you have is another problem. You have critical mass, you’re trying to solve a problem by defining certain relationships, or what I’ve been calling affinities between agents. So, an agent can be a musician having its private MIDI channel, and you can devise some kind of computer-animated interface, so that you can both see and hear how the agents behave. So what you really do in the second application, with complex dynamics, is invent your own physics.
We listen to an excerpt of The Headless Horseman by Peter Beyls [0:54:40–0:57:05].
[PB] You’re inventing an alternative world, in which your music is supposed to grow, rather than be constructed. So, instead of building architectures of time, what you do is create a favourable environment for compositions to grow, as it were. And the ultimate start in that direction is, of course, genetic algorithms.
[K] So, at what stage are you? Are you at the stage where you can perform with the program, or can it perform on its own satisfactorily? Or is it…
[PB] Oh, I’ve been performing with Oscar for 10 years. Oscar is designed explicitly to interact. It has its own character, it does not have pre-stored sequences, so everything is invented on the spot. But it has memory, both short-term and long-term memory, so it does automatic memory management, it knows when to forget something, things like that.
Yeah, on the question of autonomy, I have another program which is called Louise, and Louise is named after Louise Brown, the first artificial, or as you call it, “test-tube baby” in England. So Louise is a pianist, plays the piano only. So I like to restrict, that is, having my private musician sitting there in the corner of the room, playing from time to time. The program, again, is based on the whole collection of generated transformers, critics, it has a number of critics.
[D] For the last couple of days, we’ve had conversations with Tango and Oscar and Louise, and in America, we have another program, called Yanni. I don’t know if you’re heard of it.
[K] Damian had a question about 2 of the same, or 2 of these programs interacting.
[PB] Well, Henning’s work is really truly interesting, I know it very well. But his idiom, as it were, is the jazz idiom. So he starts off from a very rich culture, as it were, and tries to work from that, and trying to have a number of harmonic progressions, et cetera, and make that adaptive. My approach is very different, although I have a program called Harmonic Navigator, that allows you to navigate in harmonic space. I have lots of programs that I’ve never actually used to compose music. That’s a problem with writing computer programs, you know.
[D] You mentioned that sometimes Louise is off in the corner, just playing by herself. Is this also true of Oscar?
[D] But can Louise play off of Oscar, or do a duet?
[D] So that you can go out and get something at a café or something?
[PB] No, not really. Louise is not interactive, it’s an autonomous agent.
[K] But if Louise is playing, can Louise be you, in effect, and play, and Oscar then reacts to her?
[PB] Oh, that would be possible, to have Oscar react to what Louise plays. Yeah, I’ve never tried it, actually, but it would be possible.
[K] How does it sound?
[PB] How does it sound? Well, good question. I don’t have any tapes here, but I will have to send them. Some people say Oscar sounds pretty aggressive. [Laughter] Also, there are so many levels of collaboration, you know? Oscar can be, say, building some parallel line delayed from what has played in the past, or it can try to make counterpart structures in real-time, it can repeat a certain sequence that it has in memory and transform from that. It also has a nice feature that’s called exploration. If Oscar considers that nothing really interesting is happening at some point, it goes back in long-term memory and tries to explore it, and come up with something interesting that has happened in the past.
[D] Is it strictly your program, or have others worked on it?
[PB] No, this is my own baby, as it were. [Laughter]
[K] Turning back to something that you had said earlier, and sort of slid by, let me ask in terms of a question: has it ever truly surprised you?
[PB] Oh yes, yes. There were some really dramatic moments, really. There are three major versions now of Oscar. When I was at the A.I. lab I was working on the second version. I should also say that we built a number of alternative input structures, I haven’t said anything about that yet. Like a violin with a computer inside, wind controllers…
[K] Does the sound go in by MIDI controllers, or does it go in by sampling?
[PB] Well, it goes in by MIDI, because that’s the easiest way. I have a number of programs that are truly an artificial ear and you can connect a microphone, that have four buffers to compare in space and compare in time, and you know, some simple spectral analysis. But at the time I was working, that was about eight years ago or something, we didn’t have the ISPW workstation. I abandoned that, because they simply didn’t have the technology available at that time.
The element of surprise… the second version, one evening late at night, I got the program running after a couple of months of simply having written a number of basic drivers… easy, nothing very special, nothing sophisticated, but suddenly, it worked, you know? And it was so much expression that I was really surprised. It did more than I had ever anticipated, even from this very simple thing, you know.
[K] Did it simply surprise you, did it make you happy, did it make you angry?
[PB] It made me… well, I can tell you, in a very emotional way — without being romantic (it has nothing to do with being romantic or not) — I realized that I had built something which was purely artificial, yet it had these features that speak to the soul, rather than to the intellect that was put into it to build it, you see? So it was kind of an emotional shock I had, yes.
[K] What, in the course of all this, would you say was the single thing that you did that was the most important? For yourself, or for the world?
[PB] For myself or for the world?
[K] A pioneering act.
[PB] The pioneering act… well, I’ve been using computers since 1970–71 maybe… what did I do at that time? Well, I was trying to (as were so many other people) control randomness. I was trying to build computer programs that… well, you could have a handle, or a means, to control randomness. And that is still a very up-to-date idea, actually, because if you look at some of the work going on in artificial life, like the work of Chris Langton and other people, they’re trying to build controllers, software or hardware controllers, with which you can control a very complex world, but using very few controls.
We listen to an excerpt of The Headless Horseman by Peter Beyls [1:07:18–1:08:35].
[PB] And that was one of the key problems in any control structure. Another challenge is that, given the extremely sophisticated topology of the human body, how can you map that to a machine, you know? It’s very simple to map expression from human being to another human being, but the machine has trouble understanding the most basic of human gestures. That was the thing that led to the creation of the infrared violin. We built three controllers many years ago, a collaboration with engineers at the artificial intelligence lab. One is a wind controller, the second is a violin, and the third one was a bandoneón. It has a computer inside. That’s another guy, an American composer in Amsterdam, Nick Collins.
[K] Yes, we’re going to speak with him.
[PB] Well, ask him to show his bandoneón, it’s wonderful. It’s actually more sophisticated than mine, but his software is also very different. My software, for many of the controllers actually, is modeled after, or very strongly inspired by, Marvin Minsky’s book Society of Mind. There is this wonderful idea of having many different simple processes that interact on the local level and give rise to global phenomena that we call intelligence or music. So I call my violin the “infrared violin” because it has infrared sensors built into the strings. So basically, you have four strings, which means four continuous controllers, so the bow has another infrared sensor. So that’s five continuous controllers, plus a number of special function keys, and the neck of the instrument has an eight-bit input strip that you can send 256… like having a switch into the software, and the software interprets the gestures of what I do, and stores them in memory.
[K] Does it store exclusively physical gesture?
[PB] It doesn’t store any samples. Well, it stores samples of the physical activity that was happening on the instrument, and that of course, is translated and mapped to MIDI, and I can play the solo instrument, and the whole symphonic orchestra will sound. So it does automatic orchestration in real-time. I’ve been performing a lot with this violin, up to a number of years ago.
[K] Do you use to interact with Oscar?
[PB] No, no. The wind controller originally was developed to interact with Oscar. But the new version of Oscar, which I’m still working on right now, can also listen to more than one musician. So it can compare, it can be some kind of a social sensor, as it were, and make an opinion of what’s going on in a small society of interacting musicians.
[K] Do you think that will increase or decrease the level of surprise? Will the averaging, as it were, of the two, decrease that level? Or will it come with something even newer?
[PB] Well, it’s another problem altogether. I had a paper at ICMC in San José on “Dynamic Models for Musical Interaction in Virtual Reality.” That was the title of the paper, and then I have three or four different strategies. You yourself can consider from being in the outside world: how can you be present inside an algorithm, or inside a program? There are many ways, or degrees of intimacy, with which you can be connected to your algorithm. You can control a parameter, which is the most simple thing you can do, like having a knob, but you can also be present inside an algorithm by, for instance, using cellular automata. If you have a cellular automaton, you can consider yourself a cell inside the string of the automaton. It is as if it were that you’re present inside this automaton, which is something completely different. Or, you can say, “Well, I want to interfere with a social climate,” and then you’re an agent, expressing a certain affinity towards other agents that happen to be in this space. You could consider yourself a conductor. I have one implementation of the agents where I’m using a Power Glove, so I can put my hand inside this collection of agents and move one, say, and then the whole society will have to re-arrange itself to re-establish harmony in this society. That is translated into music.
[K] So speak, if you will, to the listener of popular music, and say, “Here is what I’m doing, what I do now, what it means to you now or will mean to you in the future.” How will this affect how I hear, perhaps, popular music?
[PB] I consider myself a product of popular music. I grew up with, you know, David Bowie and Frank Zappa, which I love.
[K] But clearly, most listeners of popular music don’t associate it cellular automata. How would you translate that?
[PB] Popular music has this wonderful physical presence, which I like. Maybe it’s because I’m working all the time with my brains, I’m squeezing my brains, you know? And anyway, I grew up in the 60s and the 70s, so what would you expect? I have something else to say about that too, which is what I call the “physical parameters,” or the “sensualility” and the “sensual parameters,” they, of course, are over-developed in popular music. In the avant-garde, they are under-developed, and they are extremely well-developed in ethnical music. So we have different degrees of sophistication there. But, the idea of physicality (like building this violin) to tackle that kind of problem: how do you channel the physicality, how can you map the complexity and energy in the human body towards a machine? How can you make a machine understand, and do something useful with that information? And that’s also one of the reasons to perform, you know?
[K] So that’s the reason why, when we asked you the question about Louise playing with Oscar, it didn’t seem to interest you.
[PB] Not really. Because it’s another problem altogether, and that is the case with much of my work. It fits into different general approaches to creating intelligent and creative machines. One of my most recent efforts is that I’m building a workbench using genetic algorithms, and that, of course, I would say is a mental effort, or intellectual effort, to try to explore what genetic algorithms really mean. It’s definitely one of the means to escape from the dominance of ideas you happen to have at a certain moment, because genetic algorithms are open-ended, and can push you to another level in search space, and you can work from there
[K] If you had all the time, all the computer power, and all the help you needed, what would be the program that you would create?
[PB] Well, first of all, it would be difficult… it would not be a trivial thing, to find a good group of programmers. You know, there are lots of good programmers, but…
[K] Assuming you could snap your fingers, and they would be before you, would you create… a person? In other words, would you create a musician?
[PB] One of the things I’d be interested in is to have a society of agents that live on the Internet, and that would require extremely fast Internet links that are not really available yet today. Fiber optics would be necessary. At that point, you would have this society of maybe hundreds or thousands of artificial musicians living on the Internet, that could jam together, and at the same time accommodate human input. So there would be a respect from both sides, I would say. Because respect and responsibility are also very important things in my work, in the sense that… I’m not naïve, of course, and thinking that music will change the world. I’m absolutely convinced that any musician has a certain responsibility, because it’s so easy to forget the outside world, as it were.
We listen to an excerpt of The Headless Horseman by Peter Beyls [1:20:30–1:21:27].
[PB] I mean, programming can be as bad as watching T.V.
[K] Or playing a videogame? [Laughter]
[PB] By that, I mean that you can see the world from the wrong perspective, in the sense that you can have the illusion of power, which is not there. That’s important to know. [Laughter] What I do is I teach computer graphics here, and æsthetics, and a number of introductory courses on computer programming for graphical applications. Although we don’t do any desktop publishing here, or things like that, what I do is creative approaches to using computers as yet another tool, aiming to be as free as possible. I have a course on the history of science and technology in the arts, which I like to do. What else? I do music, research, write computer programs, do some consultancy work with other software companies, especially computer graphics and virtual reality applications.
[D] Do you and Oscar go to coffeehouses any time and sit down when there’s an open mic, and just jam? Do you have open mics in Brussels?
[PB] No, no, we don’t have any open mics. Not in the same sense that you mean.
[K] But seriously, have you ever, just among a group, almost a coffeehouse type setting…
[K] You have?
[PB] Yeah, yeah, a long, long time ago. I also did some concerts at openings of exhibitions, and that’s a long time and worked very well. That was at a certain time in the late 70s, early 80s, when there was a kind of a culture of having concerts in coffeehouses, especially in the U.K. Actually, I’ve lived in London for 3 years, studied there at UCL. After that, I started traveling, worked with Bill Buxton in Toronto, then did some teaching at CalArts with Morton Subotnick. I lived in Brooklyn, then I went to Japan, so anyway… I’ve travelled a lot, and now I find myself back in Brussels.
[K] And in just a few minutes you’ve got to go back to teaching, so we’re going to let you do that, and we thank you very, very much for joining us today on the Kalvos & Damian New Music Sesquihour.
[PB] You’re very welcome.