CMCA Front Page
  .Home | Workshop |Dana Centre Event | Speakers | Archive | Links | Locations

Welcome to the CMCA Website

Position Statements

From Systems Art to Artificial Life: The DrawBots* Project at Sussex University
Paul Brown

As an artist I first started using computer systems in the early 1970’s. At that time I was influenced by the European Systems Art movement and American conceptualists like Sol LeWitt. In my work since I attempted to make processes that could produce artworks that were not stamped with my personal signature. Back in the early ‘70’s using symbolic languages like computer-programming seemed a good way to go about this. However, by the mid-90’s I had realised that I hadn’t accomplished this aim. My work was clearly identified with me just as much as if I had been pursuing a form of personal self-expression.

In 2000 I was artist-in-residence at the Centre for Computational Neuroscience and Robotics at the University of Sussex in England. I discussed this problem (of autonomous artworks) with colleagues there and we hatched what we called the DrawBots Project. It’s an attempt to use evolutionary and adaptive systems to make an embodied robot that can draw in interesting (maybe creative?) ways and, for me especially, ways which do not carry the signatures of its creators.

The project was funded by the AHRC in December 2004 as Computational Intelligence, Creativity and Cognition: A multidisciplinary investigation and involves a group of researchers including Margaret Boden (Sussex, UK); Ernest Edmonds (UTS, Australia), Phil Husbands (Sussex, UK); Charlie Gere (Lancaster, UK), Mitchell Whitelaw (Canberra, Australia), Jon Bird (Sussex), Dustin Stokes (Sussex), Simone Gristwood (Lancaster), Bill Bigge (Sussex) and myself. The project commenced in May 2005 and will run for three and a half years.

We are using robots controlled by a neural network. The genome is a string of values that modulate the interconnections within this network. This string is evolved using a genetic algorithm and phenotype behaviour is measured using a number of fitness functions. Put simply this can be seen as a complex system where the fitness functions define attractors which (hopefully) converge on the desired behaviour.

It’s my belief that within the set of all possible fitness functions there is a bar. Above the bar the functions carry the aesthetic bias of their creators. However below this bar there exist functions that are simple –or- abstract enough that they will enable an independent, autonomous behaviour to emerge.

* Since naming our project the DrawBots I have become aware that another workshop presenter – Rubin Coen Cagli – is also using this name for his research and I must apologise for any misunderstanding this may cause.

Net Work: Using Computational Systems in Art
Mark D'Inverno (School of Computing, University of Westminster) and Jane Prophet (Visual Art and New Media, University of Westminster)

Mark d’Inverno is a computer scientist with a background in multi-agent systems and formal modelling, who has increasingly become interested in working in interdisciplinary applications including art, music, biology and design. Jane Prophet is a conceptual artist with a strong background in using computational techniques in her practice such as those from ALife, multi-agent systems and tree branching algorithms. For several years we have been collaborating on several projects involved around the use of computation in art.

For example, we have just come to the end of the running of a cluster called designing physical artefacts from computational simulations and building computational simulations of physical systems. This is part of the AHRC/EPSRC Design for the 21st Century Initiative. This cluster has been concerned with exploiting the techniques from multi-agent and cellular automata in art and design.

Within this cluster Prophet proposed to build a prototype for the artwork entitled Net Work. The proposed work consists of a grid of around 2500 autonomous, computational, light-emitting buoys, sensitive to their environment and the activity of other neighbouring buoys. The algorithms used to drive the behaviour of this artwork are based on the multiagent and cellular-automata approaches to building software. In fact, many of these algorithmic techniques have been used in an attempt to understand the collective behaviour of stem cells in the adult human body. Interdisciplinarity is a strong and recurrent theme in our work to date.

Interactive Causation and Computer Music
Peter Rawbone (Cardiff School of Creative Industries, University of Glamorgan)

This paper combines four approaches to interaction with an environment: interactive computation (interaction between computational devices and an environment); causality (interaction between an observer and a causal environment); artificial life (interaction between an agent and an environment); neurosymbolic integration (interaction between representation and an environment).

Turing Machines, as closed systems which cannot accept external input while they compute, shut out the environment and the external passage of time; algorithms do not consider the environment during computation, and are insufficient for modeling interactive systems in which communication between the environment and the computer occur during computation (Wegner 1997).

Interactive computation, particularly agent-orientated models, is interrelated with the notion that causality is observer-related (Pearl 2000): modeling the interaction of an observer with a causal environment is an approach to an agent-environment framework. Neurosymbolic integrations lie at an interface between symbolicism and connectionism, arguably combining the benefits of both approaches

Finding Aesthetic Pleasure on the Edge of Chaos: A Proposal for Robotic Creativity

Ron Chrisley (Centre for Research in Cognitive Science, University of Sussex)

I have been developing a robot neurocontroller called the CNM since 1990. I propose to extend it in such a way that it allows a robot to exhibit creative behaviour, in the sense that human observers may find the products of the behaviour both of value, and novel. Thus, no direct modeling of human creativity is intended. However, I reserve the right to opportunistically and shamelessly exploit what is known about creativity in humans, when and as is expedient. Also, I suppose that there is always the possibility that insights into the human case may accrue from the artificial case by accident, as it were.

The CNM is, at root, a learned, forward model of the environment that allows a robot to anticipate or predict the sensory input it will receive if it performs a given action in a given context. This, in conjunction with built-in motivators (such as a "pain" signal when colliding with an object, or a "pleasure" signal when the robot's battery charge increases) can enable the robot to engage in adaptive behaviour by selecting actions that result in an expectation of pleasure or absence of pain.

The heart of the creativity-inducing extension to the CNM would be additional motivators, corresponding roughly to the two dimensions of creativity mentioned before: Value and novelty.


An assumption behind this approach is that a good way to make an artificial system be capable of creating works of value is to make it be capable of appreciating works of value. Thus, the ability of the CNM to evaluate its own output is crucial; but of at least equal importance is its ability for it to experience and evaluate the works of others, and the world in general. A way to avoid the representational bottleneck of having to "feed" a system digital encodings of others' work is to let it experience those works in the real world, as we do. Suppose, for example, that the CNM is operating within the domain of music. Then the proposal is that instead of giving the CNM a MIDI score of Beethoven to ponder, let it hear Beethoven coming out of your stereo, or better yet, out of your piano. But if it experiences others' works in the real world, then it should experience and evaluate its own works that way too. This double embodiment of appreciating and creating in the real world, in conjunction with a judicious choice of hard-wired prefrences akin to ours (e.g., for low integer relationships between frequencies), will go a long way to ensuring that what the CNM ends up valuing is the kind of output that we, too could value. It also allows for the possibility of serendipity, in the space between what the sensory inputs the CNM expected its actions to actually produce, and the sensory inputs the actions actually produced: the space we call "the world". These can augment more familiar ways of ensuring connections to human value, such as built-in preferences for human attention. For example, a preference for the sounds of human presence (especially applause!) would yield a preference for human proximity. Alternatively or additionally, one can formalise human approval by putting buttons on the robot that would allow listeners to provide approval or disapproval feedback.


Another kind of motivator will have the effect of encouraging novel, open-ended behaviour by pitting the creative and predictive capacities of the system against each other in a salutary co-evolutionary arms-race. Saunders and Gero have suggested motivating an agent in a way that can be characterised by a Wundt curve. The pleasure of a stimulus is a hump-shaped function of its novelty: the agent receives little pleasure out of non-novel experiences, maximal pleasure out of mildly novel experiences, and reduced, zero, or even negative pleasure for experiences that are "too novel". The CNM aims to improve on this idea with the inclusion of a motivator that is not dependent on novelty itself, but on something that often results in novelty: complexity. One can summarize the motivator as a preference for experiences that are on the subjective "edge of chaos": sensory inputs that almost, but do not quite, elude analysis or understanding, for that agent at that time. This motivator can be described with the same Wundt curve, but the x-axis of the "hump" curve is not novelty, but rather: effort (e.g., time) required to be able to understand. In the case of the CNM, understanding is a case of subsuming under prediction. If an experience is too simple/predictable, it will offer little pleasure from the complexity motivator (although it may result in pleasure because of one of the inbuilt motivators). If an experience is harder to predict (e.g., some time passes before prediction error can be held below a certain threshold), then more pleasure will be derived. However, if the error cannot be reduced, or only at great cost in time and/or resources, little or no pleasure will result. Thus, real-time dynamics will be crucial to the CNM's appreciation of a work.

This is enough for the CNM to be an appreciator of novel works. But where there is evaluation, there can be creation: In the case of sensory experiences that were the result of perceiving the CNM's own creations, the reward signal from the motivator will have the result of changing the action selection policy in such a way that both overly simplistic, and also "uncompressable", incomprehensible works are less likely to be produced in future. What will be left are works on the edge of chaos, where valuable novelty lives.

As a result, the CNM will be aiming at a moving target. What once was pleasurable because of its elusiveness will become boring, because now easily understood. However, there does remain the theoretical possibility that conceptual development is not monotonic. Especially in connectionist systems, learning to understand something new may make one forget something previously learned, potentially making it of interest again. Preventing such limit cycles may require the introduction of some kind of explicit autobiographical memory as a more conventional means of calculating novelty, if the cycles are not prevented by some other means (such as the bored faces of other agents that do have a memory of the CNM's earlier, and now repeated, output).

A further extension of this approach can be achieved by making the CNM's own states, particularly the states of its motivators, the subject of "perception" and prediction, and therefore a potential source of meta-motivation. Experiences that have an effect on the CNM's motivators such that a pattern is produced that the CNM's model of its own motivator activity initially fails to predict, but eventually succeeds in predicting, will be preferred (and will be therefore more likely to be produced). Perhaps there could be hard-wired preferences for certain motivator patterns as well. But one must be careful to not give the CNM too much access to its own states: If the self becomes easily predicted, any object produced by that self will tend to be rendered easily predictable, reducing its subjective aesthetic value. The only option then would be for the system to seek out regimes of activity that evade the predictive capacities of its own self model. Perhaps this could even serve as a way to ground what is meant by "too complex" in this context: Whatever can be predicted only by modelling oneself. However, such prediction will also require modelling of the environment, so this bound will not be inconveniently low. On the other hand, being able to predict one's own perceived bodily movements (toe-tapping?) may itself be a source of pleasure via various motivators, including the complexity motivator.

The intended robotic platform for this system is a pair of AIBO ERS-7 robots. Having more than one CNM agent will dramatically foster the co-evolution of the generative and evaluative capacities. But the AIBOs are motorically challenged; they can't draw, for example. They can produce sound, but it would hardly be in the embodied spirit of this proposal if the rich source of creativity that is the gap between conception and realisation were not made available to the CNM. So although the initial stages of the project may involve simple auditory output, a further stage will explore the following possibility: Impose a mapping between joint angles and sounds produced, at least when the CNM AIBO is in "creative mode". Thus, every slight move the AIBO makes (or is imposed upon it by the world, or another AIBO) will result in a change in its sound output. In order to make more pleasurable sound outputs, the AIBO will have to move. The resulting music improvisation through dance, and observation of such by humans and the other AIBO, should open up dramatic new possibilities for creative output.

I look forward to hearing from others about the undoubtedly enormous amount of work along these lines that has already been done!


Saunders, R., Gero, J.S. (2001) Designing for Interest and Novelty: Motivating Design Agents. In de Vries, B., van Leeuwen, J., Achten, H. (eds.), CAADFutures 2001. Dordrecht: Kluwer, pp 725-738.

On the CNM architecture:

Chrisley, R. and A. Holland (1995) "Connectionist synthetic epistemology: Requirements for the development of objectivity," Niklasson, L. and Boden, M. (eds.) Current Trends in Connectionism: Proceedings of the 1995 Swedish Conference on Connectionism. Hillsdale, NJ: Lawrence Erlbaum, pp 283-309; ISBN 0 8058 1997 5.

Chrisley, R. (1993) "Connectionism, Cognitive Maps, and the Development of Objectivity," Artificial Intelligence Review 7, pp 329-354.

Chrisley, R. (1990) "Cognitive Map Construction and Use: A Parallel Distributed Processing Approach," in Touretzky, D., Elman, J., Hinton, G., and Sejnowski, T. (eds.) Connectionist Models: Proceedings of the 1990 Summer School. San Mateo, CA: Morgan Kaufman. pp 287-302.

Wall of Woven Sounds
Janis Jefferies (Digital Studios, Goldsmiths College) and Tim Blackwell (Computing Department, Goldsmiths College, London)

In ‘A Sound You Can Touch', we used woven sound to explore visual and sonic texture. Woven sound, in an analogy with warp and weft weave, is a real-time mapping of linear sound into two dimensional textile design. A virtual swarm flies over the design, attracted to regions of high texture, rendering sonic improvisations through the unweaving of small tiles of woven sound captured during live performance. We are continuing to develop this project to produce a Wall of Woven Sounds, a patterned texture of movable Jacquard tiles that, when touched, trigger the recordings generated by live performance. In Wall of Woven Sounds we are extending sonic texture into the third dimension.

In the first phase of our project micro-textures are explored by clicking the mouse at various points on the pattern, causing a small tile of image texture to unweave into a grain of sonic texture, which is immediately heard. Our aim is to stage the users experience between the screen and the wall through touch and aurality. Patterned, Jaquard (double weave) tiles are overprinted with chaotic mutation generated from the mutated screen image. This mutation, stumbled upon by chance during coding, generates monstrous detail.

The tiles are moved around and reassembled on the wall, forming a new collaged and patterned multi-sensory aesthetic. The whole hand touches the tile to trigger the first recording (hence the need for a double woven structure to hold the sensor) and moves the tiles to reconfigure textiles and sound. The pieces mimics and reference that which was generated in our earlier work but by a series of different physical actions, from the click of the mouse to the stroke of the textured surface, the hand acting in place of the attractors on a screen that senses previously visited sites. The swarm in our previous work might wander, at times, aimlessly over the image; in Wall of Woven Sounds we might also aimlessly move the Jacquard tiles in a search for what kinds of meaning the sound produces. At moments we are unable to find anywhere to hold our attention, patterns on the screen and in the Jacquard are dense and play with optical stability.

Sound and Image System
Alan Sutcliffe (Computer Arts Society)

The sound module composes using a hierarchy of levels from overall structure to details of the waveforms. The same general method is used to structure and generate the details at each level, modified in ways appropriate to the context. A parallel module composes graphics to be shown with the sound using the same approach. There is no close coordination between the sound and images: they are simply produced by similar generative processes which are adjusted to the differences between hearing and seeing.

Use of the same generative process does not lead to self-similar structures at different levels and different times, though a generated pattern may influence patterns generated elsewhere in the process.

Initially the system is autonomous. A user interface will be added for interaction with real-time and offline composition. Later it is proposed to add sound output to Midi files


White Noise: Abstraction and Interaction-The Question of Economy

Ernest Edmonds (Department of Information Systems, University of Technology, Sydney)

White Noise was exhibited at the Australian Centre for the Moving Image (ACMI) in 2005. It was a visually seductive and sensory journey into the ongoing relevance of abstraction. Early twentieth-century abstraction was a vehicle for artists to explore and universalise ideas and sensations; White Noise continued this exploration in the digital age of the twenty-first century.

Devoted to the contemporary revival of interest in abstraction by digital artists, White Noise featured large-scale installations by renowned artists Ryoji Ikeda (Japan) and Ulf Langheinrich (Germany), and six Australian and international artists who are leading the current resurgence of abstraction. The exhibition also included the work of 18 software artists in a web content program, as well as an historic film program that traces the earliest experiments in abstract filmmaking.

The visual and artistic strategies of early abstraction have changed and expanded with new media technologies. Like the pioneers of early abstraction, the artists in White Noise share a strong desire to evade narrative frameworks, to streamline representation and to depict 'nothing'. Showcasing works that invite the viewer into experiences that are at once, spectacle and transcendence, White Noise explored ways in which artists test the material form of the moving image.

Adapted from Australian Centre for the Moving Image

Society of Neurons
Robert Zimmer (Digital Studios and Department of Computing, Goldsmiths College) and Warren Neidich

The development, and ultimate structure, of the brain influences the art we create, while the art we create influences the brain we develop. The first observation has led to studies of aesthetics based on neuro-physiology; the second to explorations of the brain through artistic and cultural phenomena. The society of neurons brings out these connections, in a new way, by creating an artwork that embodies contemporary ideas of brain development as a process informed by social and aesthetic interactions. The web-brain, like a natural brain, is a product of an open ended, unbounded creative act built from selection based on creativity and noise. The artwork uses the computer as a model of the ways that brains that create themselves.

The Society of Neurons is a web-based artwork whose changing shape is used to generate a series of physical sculptures. The web-site and the sculpture series tell the story of the internet user interactions and hence the project partakes in a history of time based sculptures, such as Jean Tinguely's Metamatic No. 9, 1958 — a series of exhibited drawing machines that produced works of art on paper — and Pol Bury's 1110 Dots Leaving a Hole, 1964 in which nylon filaments move through a canvas and then twitch in front of the viewers eyes. The society of neurons’ changing shape documents the moment-to-moment expressivities of its thousands of user/producers.

Harnessing Mid-level Vision for Artistic Stylisation
John Collomosse (Department of Computer Science, University of Bath)

John Collomosse is a lecturer at the Department of Computer Science, University of Bath. His research interests lie in the convergence area between Computer Vision and Computer Graphics, specifically in video analysis and the semantic representation of scene content and dynamics. Application areas focus on information retrieval (video search and summarisation), immersive and augmented environments and in particular the problem of non-photorealistic rendering (NPR) from two-dimensional content.

The Media Technology Research Centre at Bath is interested in the convergence of Graphics and Computer Vision. This includes video effects processing, non-photorealistic rendering and depiction, and augmented/immersive environments. The domain of processing an image so that it can be passed off as artwork is a particular interest, because it challenges us to understand human perception. This naturally leads to our interest in machine learning, so that humans teach the machine which lines are appropriate to draw, for example.

Aikon: Artistic Automated IKONograph
Patrick Tresset (Digital Studios, Goldsmiths College) and Frédéric Fol Leymarie (Digital Studios, Goldsmiths College)

We consider the problem of generating face sketches with a computer system, based on reality as it relates to the understanding of a drawing style and aesthetics. Our research goal is for the system to mimic the interplay between the ensemble of processes — perceptual, cognitive, motor — active when a draughtsman is sketching.

One hypothesis we are exploring is that the aesthetics of a sketch is dependent on each step involved in the collection of processes leading to the creation of a stylised portrait, as well as in the interactions between such processes. A second hypothesis we formulate is that during a sketch session each limitation, approximation, shortcut performed by the draughtsman due to his/her physical, perceptual, or cognitive impairments or qualities are what constitute the final mood, style, aesthetic of the sketch. Interestingly, the simulation of a stylised portrait sketching session via a computational model remains approximate, and as such the computerised system might be thought to draw in its own limited style. The decision to study this specific drawing discipline is motivated by the third hypothesis that creativity mainly precedes the execution.

Knowing Computer Creativity

Dustin Stokes (Centre for Research in Cognitive Science, University of Sussex)

Dustin Stokes specializes in the philosophy of mind and art. He works on the philosophy and psychology of creativity, imagination, propositional attitudes, and perception, and their implications for engagement with and creation of art. Stoeks also has a continued interest in the metaphysics and epistemology of the arts, with a new interest in interactive and digital arts. He is a member of an interdisciplinary research team targeting creativity, cognition, computational intelligence, and aesthetics.

The Museum as Laboratory: A Performative Approach to Complex Systems Research
Hans Diebner (Institute for New Media, Frankfurt)

Three concrete examples of media installations shall be presented to eventually support a general methodological approach to research called “performative science” which seeks to open the usually stationary methods in science to the process. Process, thereby, is added twofold through artistic creativity and through the spectators interaction. Artists take part in the interface design and the spectators' creativity is seen as a valuable social resource for the purpose of research. The museum is regarded as a laboratory in this approach, and the media installation as a critical interface.

The three examples are “Liquid Perceptron”, “EyeVisionBot”, and “AI-Arena”. The first one is a simple model of a neural network. The neurons of this model brain are excitable by the “external” world, i.e., by the spectators in front of the video projection of the simulation through their movement. “EyeVisionBot” is an image searching interface based on an eye tracking technique. The searched category is controlled by the user's gaze behaviour. “AIArena” is a simulation of species that can be endowed with survival strategies by the users.

Art and the Mirror of Nature
Charlie Gere (Institute for Cultural Research, Lancaster University)

Charlie Gere is Reader in New Media Research and Director of Research at the Institute for Cultural Research, Lancaster University

His main research interest is in the cultural effects and meanings of technology and media, particularly in relation to art and philosophy. Gere's current book project, Art, Time and Technology: Histories of the Disappearing Body (to be published by Berg in 2005), concerns artistic and theoretical responses to the increasing speed of technological development and operation, especially in terms of so-called ‘real-time’ digital technologies. It draws on the ideas of Jacques Derrida, Bernard Stiegler, Jean-François Lyotard and André Leroi-Gourhan, and looks at the work of Samuel Morse, Vincent van Gogh and Kasimir Malevich, among others.

Technology: The Very Idea!
Michael Punt (School of Computing Communications and Electronics, University of Plymouth)

Between 1992 and 1995 I led a small group of academics, artists and designers who surveyed the developments in the emerging digital industry and evaluated the products. Although the public rhetoric suggested that we were engaged in the progressive sophistication of interactive design and the convergence of analogue media in a digital culture, in fact we noticed almost the reverse.

In 1995, after an examination of the discourses that converged on CD Rom and digital media I argued that the cause of its spectacular failure to deliver its promise was the rhetoric: the vapour ware that surrounded digital storage and distribution media. Information and product designers were misled by a critical discourse that, with few exceptions, was characterised by four oversights: (i) the idea that technology has agency is a substantial claim that for the better part of three decades has been largely refuted, (ii) the relationship between technology and culture is a complex interaction that (as film scholars) we reduce at our peril, (iii) those people who use the inventions of scientists and technologists seldom, if ever, interpret the products in ways that the inventors anticipated (or in the case of cinema could even have anticipated). Finally, and this was perhaps a more generalised oversight in all interaction design, that the meaning of any reality is contingent on the convergence of local and universal forces in a moment of consciousness.

Contingent Realities: the role of the creative arts

Whatever position one eventually takes in the spectrum of the technology and culture argument will invariably be the consequence of the texture and resolution of the evidence. In our research a decade ago we found that reductive approaches to the question of technology and history simply failed to account for the way that designers failed to understand CD Rom then, nor did it explain the subsequent interpretations of related inventions such as the PC or the cellular telephone. Moreover, we noticed that a refusal to distinguish between technology and technological system collapsed pubic attitudes to technology-as-a-concept into a melange of confusions. A less reductive approach suggested that the lay interpretation of technology was better accounted for as a stable coalition of discrete contradictions, and following this we were more fully able to understand the relationship between technology and its users as a interpretative response to malleable hardware. From this conclusion we made a number of claims including the identification of a specific function for creativity and the interpretation and uses of technology.

DrawBot: A Bio-Inspired Robotic Portraitist
Ruben Coen Cagli (Department of Physics, University of Napoli, Italy)

The human being has been conceived in several different ways in the course of history, ranging from the special result of divine creation, the only rational being or the most evolved animal to a complex and robust machine or a huge collection of input data for computational systems. Today, all individual gestures and activities, as well as collective processes, are potentially regarded also as data fluxes.

As a visual artist, my research aims at creating (art)works that portray, and bring to the focus of conscious attention, the subjective experience of this condition of contemporary human beings and collectives: the condition of ”being immersed in data” and even of being themselves data.

As a PhD student in Physics, I am currently doing research on visual creativity in a cognitivistic and computational framework. The focus of my research is on visuomotor coordination in the activity of realistic portrait drawing.

Brain Wave Gaming for Learning
Paras Kaul (Electronic Publications and Department of Art and Media Technology, George Mason University)

Computer games have generated a gaming culture with millions of people willing to invest time and money in the pursuit of fictionalized violence. Commercially available games and those available for free downloading vary in degree of intellectual challenge. The challenge of most computer games is based on an aggressive approach to gaming, which includes a high degree of violence. For this reason, it is time to consider becoming more responsible to the gaming community by developing games that have a healing impact on gamers, as opposed to games that increase agitation and glorify violent conflict.

The time is right for the introduction of neurological gaming environments that enable gamers to increase their brainpower by relaxing and refocusing mental attention. This type of game emphasizes mind over matter and not might. The brain wave game that has been developed combines a 3-D gaming environment with the Interactive Brain Wave Visual Analyzer. This brain wave interface system, referred to as IBVA4.2.4, has been developed by Masahiro Kahata, and facilitates the use of brain waves to animate 3-D digital objects. The goal of this new type of gaming is to lower brain wave frequencies and amplitudes and to increase overall coherence of neural signaling in order to animate objects in a predetermined manner. IBVA4.2.4 uses a psychic feedback multimedia control system.