Date sent: Sun, 12 May 96 05:18 MDT
From: Pit Schultz (by way of Pit Schultz )
Subject: nettime: The Darwin Machine: Artificial Life and Art - Simon Penny
Send reply to:

The Darwin Machine: Artificial Life and Art

Simon Penny


Complexity, Artificial Life and Art

Although I will cite specific artistic and scientific works in this paper, my goal is to establish a framework of cross-related artistic, scientific and philosophic ideas as a base for the discussion of the subject.

I want to examine the historical relationship between these new disciplines and the traditions from which they arose, the better to point up continuities and discontinuities with those traditions. I believe these new fields have within them certain very radical ideas which challenge basic premises of the scientific method and the enlightnement world view. I will assume that my audience has a nodding familiarity with the development of chaos theory and fractal geometry, as these fields were in some ways the precursosrs of the topics under discussion.

I'll begin and end my paper with the subject of behavior. Between these parentheses, I'll discuss the relevance of Alife and complexity to art, both philosophically and technically, and in particular to the question of behavior.

Esthetics of interactivity

I'll start by claiming that Western visual arts have no tradition of an esthetics of interactivity, and I'll defend this claim with an inverse taxonomic calculus:

If real time interaction is our starting point, then we can say that video or film is a linear record of interaction, and painting or photography a still frame.

We have a well established esthetics of the still image, of color and line, shape and area, of representational geometry and perspective. We have an estehtics of time based image, of camera angle and movement, wipe and cut. But we do not have an esthetic langugage of real time interaction.

The performing arts are, from this perspective, only different from video and film in that the viewer replaces the camera as the perceiving device. Drama or orchestral music is not 'interactive' in the sense of real time reaction, though improvisation is. Changes of form which occur in the moment, according to the forces of the moment, define improvisation. In conventional drama a director induces the actors to reproduce a version of a script. From this perspective, I am interested in what we might call a meta-script, a script that is the director, that defines in real time, the scripts or roles or behaviors of the 'actors'. To use the improvisation analogy, it is a system which understands something of the context, and responds to the situation at hand.

Such a script can clearly be assessed according to systems of esthetic analysis. It might be anthropomorphic or biomorphic, it might be literal or associative, compulsive or schizophrenic, aggressive or retiring. If we are serious about autonomous machine based interaction, then 'compared to what' is a central question. Certainly the famous Eliza, and the Turing test before it, are unselfconsciously anthropomorphic and mimetic in their behavior. In their case, 'compared to people' is the answer. When we posit synthetic agent doing work in cyberspace, such as locating references on a certain subject at various sites, the environment of the agent is quite alien to us. It is digital, with little equivalence of the geography and geometry which we inhabit. And yet we must understand it and it must understand us. So the interface becomes the glass of the aquarium in a more dramatic way: we are looking at an alien species in its environment. The interface is the zone of translation. As with neural nets, we will never know how it thinks, how it evolved, we could never assemble it from component parts. We understand only our image of it, which is to say, we extrapolate from our cultural experience examples which carry some traits which seem to have an analogous relationsship with what it is we think we're seeing. What we're seeing is of course constituted by our cultural experience. This regress leads those who have faith in any sort of cultural universals spiraling out of control down an objectivity dissolving vortex.[1]

The Darwin Machine

Prior to my brief excursion into the shallows of epistemology, I leapt across the question of intelligent agents and artificial life in general, and I should backtrack. Just what do we mean by artificial life? The name has been claimed by a group of interdisciplinary scientists: biologists, roboticists and computer scientists who have held several conferences by that name at the Santa Fe Institute for Non-Linear Dynamics since 1988. Similar events have since occurred at MIT, in Europe and in Japan. Alifers are peripherally associated with the related, perhaps less deterministic fields of non-linear dynamics and complexity theory by virtue of their common interest in self-organising systems and emergent order, ideas which arise from the study of chaos. The applications of these techniques vary from the building of digital ecologies in which the dynamics of evolution might be studied, to the shaping and control of these systems to breed algorithms to do particular arithmetic, graphical or informational tasks.

The Alife community includes:

1. Computational biologists. Until now, natural selection, the mechanism of evolution, has been limited to the organic. The realization of evolving, reproducing digital species in silicon using genetic algorithms [2] prompts the question: "Is it alive?" This question divides Alifers into two groups:

1a. Hard Alifers hold that self replicating digital organisms are alive in every sense, and that biology must include the study of possible life, and must arrive at some universal laws concerning wet life and digital life.

1b. Soft Alifers claim only that genetic and evolutionary simulations are useful in understanding biological dynamics, but remain simply simulations.

Around this central group cluster several others:

2. Builders of procedural systems, like Craig Reynolds' Boids and Jessica Hodgin's robot flocks. More recently, these systems are self evolving, such as Karl Sims' recent work on evolving 3D morphology and behavior by competition, and Jeff Ventralla's evolving animated characters.

3. Subsumption and 'bottom up' roboticists who utilise ethological analogies to create bottom up emergent behavior in mobile machines.

4. Builders of autonomous digital agents to do work in the digital realm.

5. Wet Alifers. Wet Alifers are molecular biologists who are breeding or constructing replicating or behaving groupings of proteins, enzymes and nucleic acids. the instrumentalization of natural selection carries not only for the digital alifers, but equally for the Wet-Alifers, the closeness in attitude between Alife and the new genetics and reproductive technologies, and nano-technology, should not be elided. [3]

Eliza's children, Frankenstein's grandchildren.

Tom Ray, a biologist and designer of the Tierra system, recently made a proposal to promote biodiversity in the net, a distributed digital wildlife preserve on the internet, in which digital organisms might evolve, circumnavigating diurnally to available CPU's. [4] These creatures would evolve good net navigation and cpu sensing abilities, among other things. Predators and parasites would emerge. Ray notes that "Evolution just naturally comes up with useful things". [5] He argues for the proposal in the following way: you couldn't imagine a silk worm, even if you could, you couldn't guide evolution to make it. But evolution did make it, we can take it, cross breed it, neuter it, delete its poisonous properties, domesticate it.

This proposal is emblematic of paradigm shifts which characterize Alife. According to the traditional christian outlook which functions as a foundation for the ideology of industrial capitalism, we humans, (and particularly westerners) could harvest the products of biodiversity and harness them as components of the industrial machine. In the post-industrial, Alifers are harnessing the mechanism of biodiversity itself. A somewhat insidious example of this tendency is a Japanese project to build Artificial Brains for the internet. This conception sees the internet as a nervous system, and draws upon the evolutionary narrative to validate its claims that, as nervous systems developed prior to brains, so it is only logical that the internet will grow a brain. The central technique in this research is referred to as the Darwin Machine and has more than passing resemblance to the Frankenstein theme. Like Dr Frankenstein, the developers of the Darwin Machine are seeking to cobble together a quasi-human machine. [6] Like most cutting edge 'techno-science' research projects, including the exercise of Alife in general, the researchers conveniently and almost unnoticeably omit any mention of just what this technology might be used for.

Anthropologist Stefan Helmreich has noted that the Alife community is statistically 30-40 years old, straight white males who are agnostic or atheist of judeo christian backgrounds. Within this community, subjective and value laden assumptions of the researchers themselves are disguised as axioms. As an example of these assumptions, Helmreich quotes Tom Ray as saying "I'm omniscient to the physics of the world I create" and notes the similarity of this position to that of the judeo-christian notion of the omniscient creator. The evolutionary narrative chosen implicitly supports social darwinism, and other less tasteful social models, such as racism.

Alife avoids the aspects of cell dynamics and evolution in which the informational and the material are "deeply entangled" [7], thereby enforcing a simplistic DNA=3Dalgorithm generalisation. Alife is predicated on the computer-science inspired dictum that the informational content of life can be separated from the material substrate in the same way that software can be separated from hardware. This induces the assumption that modern computational techniques are structurally similar to the deep structure of biological life. We must be clear that this is a rhetorical device, a validation by back-formation with reference to a presumed natural or authentic condition. It is one example among many, of computer technology functioning as the paradigmatic technology or our era, to use J. David Bolter's term. [8] Proposing a division between matter and information in biological systems is a very old- fashioned and familiar narrative construction rooted deeply in Enlightenment precepts. It serves to reinforce other such contrived dualistic structures as form and content and ultimately mind and body.

Elsewhere I have discussed the similarity between the attitudes of St.Augustine and Descartes to the body, and those of cyberpunks, epitomized by Gibson's words "the body is meat" [9] It is through examples such as these that we can see just how clearly so called 'objective science' can be haevily value laden, perpetuating dualistic and colonialising ideologies. High tech enterprises, such as Artificial Intelligence and Top-down robotics validate and reinforce these dichotomies with the rhetorical power they derive from being high tech and futuristic.

Paradigm Busters

Scientific ideas have been a powerful influence in shaping western culture. In many cases, the power of influence that the hard sciences have had, has encouraged social sciences and humanistic disciplines to become more 'scientific' (and therefore, by definition, more rigorous, more respectable) by the adoption of scientific tropes. The theory of relativity and quantum theory are examples which have been ludicrously mis-applied in the popular science and the social sciences. It is arguable that the modernist tradition in art itself is a highly scientized world view, privileging as it does ideas of experiment and progress.

As I noted at the beginning of this paper, ideas arising in complexity and Alife challenge some traditonal scientific ideas and the Enlightenment frame in general. In some cases they also reinforce traditional attitudes.

The ideas that Chaos theory brought: strange and chaotic attractors, bifurcation and fractality, and particularly 'sensitive dependance on initial conditions' revealed vast jungles of unpredicability in the heart of newtonian physics. [10] The adage that a butterfly flapping its wings in the Amazon will cause a typhoon in India has achieved the status of a clich=E9, but it underlines the oft-edited fact that classical physics can deal with only a small subset of physical phenomena and ignores the rest.

Fractality. The significance of fractals will not be found in any number of computer renderings of the Manderlbrot set, nor in their application to computer graphic simulations of fictitious valleys, islands and planets. Fractals show us a geometry which approximates the logic of natural growth: recursive, multi-scaled, infinitely detailed, with symmetry across scale [11]. This idea not only replicates the generative and recursive geometries of biology, but exposes the roots of Euclidian geometry in Platonic abstraction. The geometry of Euclid, premised on lines infinitely thin and points infinitely small, is steeped in intellectual abstraction, predicated on the notion of an 'ideal'. Newtons mechanics is itself predicated on this style of abstraction.

Entropy and Self-organization. Since the mid C19th, the second law of thermodynamics has held western culture in its nihilism inducing grip. This in itself indicates just how powerful the grip of science and particularly physics has been in the last century. It's strange because experientially we know life is anti-entropic. New science, in the form of the ideas of self-organization and emergent order has validated this intuition and liberated us from the defeatism of the 2nd law. That is not to say that the 2nd law is no longer valid, but that extrapolation of its implications into the life sciences and humanities has been shown to be misplaced. As Beckers, Holland and Deneubourg have persuasively demonstrated, random behavior amongst simple animals or machines can result in an anti-entropic outcome. [12]

Emergence and Reductivism. Perhaps the most far reaching implication of self organisation and emergent order in complex dynamics is the demise of the entire method of reductivism. Reductivism is a keystone of the scientific method.[13] It is premised on the assumption that to understand a complex object, one breaks it into component parts and examines those parts in controlled settings, then adds the results of those examinations together. The basic principal of emergence is that organisation (behavior/order/meaning) can arise from the agglomeration of small component units which do not exhibit those characteristics. Emergent order implies that the whole is indeed greater than the sum of its parts, that higher level behaviors cannot be disassembled into their component lower level building blocks. Simple examples include the generation of mind form individual neurons and the complex behaviors of colonial insects and organisms. Inherent in this pheonomenon is a critique of reductionism, the major tool of science, which is premised on the assumption that to examine a complex object, one breaks it into component parts and examines those parts in controlled settings. Complete understanding arises when those parts are added together. Emergence throws that method in the trash. As De Landa puts it: "The road to reductionism has been permanently blocked. If the properties of matter and energy at any given level of organisation cannot be explained by the properties of the underlying levels, it follows that biology cannot be reduced to physics or anthropology to biology" [14] Or one might add, psychology to physiology.

The Top-Down Artificial Intelligence paradigm has come in for its fair share of bashing in recent years, one of the earliest and loudest whistle blowers being Hubert Dreyfus [15] who refers to the paradigm as 'good old fashioned artificial intelligence' or GOFAL. Its inability to deal with real world problems without formally bounded domains led to the development of Brooksian Subsumption Architecture, the entire Bottom-Up trend and the exploration of emergent order. There is a substantial political force in this trend, as Alife opposes authoritarian power structures. The Top-Down paradigm, on the other hand, exactly replicates and reinforces very traditional tropes of lord and serfs, boss and workers and more abstractly, body/mind, form/content and hardware/software. Distributed and parallel computing, connectionism and subsumption all point to demise of the Cartesian dualism as a useful analytic idea.

Another techno-scientific paradigm is Claude Shannon's Communication theory. Ported into the humanities and particularly into telematic arts, this technologically validated paradigm entirely ignores the question of interpretation in communication. Hors Hendriks Jansen, in his discussions of situated robotics and what he refers to as interactive emergence, argues that the methods of ethology which emphasize the importance of observation in the environment rather than the reductivist methods of controlled experiments in the lab, offer new insights into the complexities of human communication. Several of his examples from early childhood psychology indicate that early childhood actions trigger responses in adults by appearing to be intentional. This 'bootstraps' the child into meaning. [16] The significance of such 'exchange' is that the message received was never sent! Such real life examples suggest that Shannon's communications theory is not particularly relevant to the study of human communication.

Emergent Behavior in Art

So what can complexity and Alife offer us as tools for an esthetics of behavior? The first hint might come from the now well established field of procedural modeling. Craig Reynold's Boids are perhaps the best known example, in which, rather than draw a flock of birdlike images and animate them, he equipped a virtual space with a terrestrial physics and equipped the boids with some basic flocking behavior and voila! a virtual flock flew through the space and around obstacles. More recently Jessica Hodgins has used similar techniques to generate a 'herd' of cyclists. Jeff Ventrella, Karl Sims and others have combined the notion of procedural modeling with simulated evolution, 'breeding' new characters to the strictures of their environment.

Of course, you can breed anything, a physics, a joke, if you establish the payoffs and taxes, the right rugged fitness landscape and manipulate your mutation rate to avoid becoming stranded on local fitness peaks. You could have a self evolving user interface, in which the environment it adapted to was the habit and interests of the individual user. Such an interface might hide little used functions in sub-menus or mutate a new floating palette. Use of the palette would reward that mutation, allowing its progeny a better rate of survival in the next generation. This could be fun, you'd never know quite what you interface would look like, though I guess the mutation would be fairly conservative, given the trained and habit bound nature of user activity. Radically novel mutations would be nipped in the bud because the user wouldn't understand them.

These examples suggest that not only will procedural modeling replace figure animation, but perhaps genetic algorithms will breed not only creatures, but storylines as well! If this computation were to occur in real time, this self evolving storyline which is responsive to its environment sounds very like the meta-script I proposed at the beginning of this paper. And indeed, this is possible now, not simply in screen-based simulation, but in the more difficult case of mobile robotic platforms where power and computation are limited, as Luc Steels has recently demonstrated. [17] This presents a curious prospect that art practice might become a kind of virtual horticulture as evolutionary models are adopted.

When the 'work' is making choices to which the user responds, the relationship between the user and the work is reversed, the work might 'play' the player, deriving its own pleasure from the exchange. Why, we might ask, is the nature of the artwork being so reprocessed? Erkki Huhtamo, myself and others have pointed out that the flurry of artistic interest in interactivity has followed the emergence of interactive technologies in general. As Huhtamo has noted: "One might argue that the proliferation of forms of computer mediated interactivity in our everyday lives has already given rise to a new subject position in relation to modes of audio- visual experience." [18] That art should trail behind cultural/economic developments is to be expected. It took almost a century before the phenomenon of the industrial machine was brought into art by such people as Takis and Tinguely in the '60s.

One great value of Alife to artists, as I see it, is that it offers an alternative to the current all too deterministic paradigm of interactivity as pre-set responses to user navigation through an ossified database. This paradigm is firmly within the Top-Down camp. Emergent interactive behavior would not be derived from a map of pre-set responses. Rather, as in the Subsumption architecture of Rodney Brooks or more radically in the PDL system of Luc Steels, behaviors would arise from the context the system found itself in. In robotic terms, no mapping would occur, the world would be the map. In subsumption architecture, as in the child psychological examples cited by Hendriks Jansen, and the behaviors of termites described by Grass=E9, behaviors arise through what we might characterize as a contingent and unconnected chain of triggers. In the case of the termites, highly complex building behavior arises when a simple behavior produces an action or a product which then triggers a higher level behavior. [19] This is a new paradigm of interactivity, radically different from the notion of a pre-linked database. Simply regarding this method as a possibility points up the presence of deterministic top down strategies in current interactive art practice. Hendriks Jansen notes that the message received by the mother was never sent (see footnote 15). This must lead us to consider just how much of interactive art theory (such as it is) is predicated upon Shannonesque and Top-Down approaches which are quite questionable in our context. An interactive work, like any work, consciously or unconsciously embodies a value system. Particularly in the realm of computer art, we are always subject to the insinuation of the very C19th value system of engineering. Considering an emergent approach to interactivity is a way of bringing those value laden basic assumptions into visibility.

Behavior and mimesis

In the past I have discussed the question of mimesis with respect to art history, modernism and technological media. [20] In these papers, I noted that mimesis has been historically, the focus of western art since the Greeks, and has advanced by utilising the most advanced technologies and techniques available. The abstraction of modernism is an aberration in that historical flow. Certainly the history of popular media followed another path, from automata of the seventeenth century, such as Vaucanson's famous Duck, a mechanical automaton which could flap its wings, eat and excrete foul smelling waste matter, through photography, C19th visual novelties such as Wheatstone's stereograms and Daguerre's Diorama, to cinema, TV, computer animation and theme park attractions, mimesis has been the guiding principle. [21]

Now interactive media and artificial life offer a quite new type of mimesis, one which combines the trajectory of technological mimesis with ideas influenced by the 'systems art' of the 70s. The representationalism here tends to be not so much optical as systematic. The dynamics of biological systems are modeled more than their appearance. These works exhibit a new order of mimesis in which nature as a generative system, not an appearance, is being represented. This change of order is akin to the move from harnessing the products of biodiversity to harnessing the mechanism of biodiversity which I discussed eralier. Numerous works employ 'nature' not as a representation but in the structure of the systems: biological growth algorithms, simulated ecosystems or communities, genetic algorithms, neural networks. [22]

The parallelism inherent in Paul Cezanne's early modernist dictum that 'Art is harmony parallel to nature' is very like the goals of Alife research. This makes Alife researchers 'artists' in Cezannes terms. The discussion of mimesis is complexified by the intrusion of this goal of a parallel order. If the Alife researcher seeks a condition parallel to nature, then this is in a sense, very like the goals of the modernist artists following on from Cezanne, whose goals were not to represent the world, but to 'render visible'. One of the active issues in an esthetics of interactivity, then, is the question of mimesis and modernism: Whether the artist is concerned with simulating or interactively representing an existing being, or inventing another possible being. [23]

Science, Complexity and Art

I want to revisit the discussion of the relationship between science and art in the light of these new ideas. Science and art can be regarded as similar to the degree that both involve invention, but at that point the methodologies diverge. I teach in a cross disciplinary program which includes students in the fine arts and in the techno-sciences. One of the problems I see scientifically trained students encountering is that their training requires of them a singular, clear and unequivocal answer. The unresolved, the internally contradictory, worries them. Yet these conditions energize art. The scientific craves closure, a fixed result. Closure kills art. Art must remain volatile in order to colonize the mind of the user.

The notion of phase transition is commonly applied in discussions of complexity, ice/water/steam is the conventional example. The crystal form has a fixed regular crystalline order, the gaseous, chaotic motion. Water, the medium of biological life, is in between. Roger Lewin notes that life exists only at this cusp between frozen order and chaotic disorder. [24] We find that the solid form is too rigid to allow development, and the gaseous form , too chaotic. It is the liquid mode where complexity, and hence life, can be supported, on a margin. The fecundity of the coral reef likewise occurs at the cusp of the contintental shelf, between the sun- warmed shallows and the yawning, mineral rich depths. Similarly, an artwork is energised to the extent that it inhabits this cusp between the determinism of the rigid grid structure and the free- for-all in which meaning dissolves in a chaos of signs. More prosaically, while internal contradiciton, like a fault line, breaks the rigid grid of scientific knowledge, artwork is invigorated by internal inconsistency. [25] But like a fault line through sedimentary beds, the order of the beds is necessary so that the rupture can be seen, and seen to work as a connector between previously isolated pockets of meaning.


The new ideas in Complexity and Alife seem to have great potential in art, both as techniques and as food for philosphical thought. But several cautions must be issued. I've tried to show that while the disciplines I've discussed do take radical positions with respect to traditional ideas, in other ways they perpetuate a view of the world which is deterministic and instrumentalizing, and are thus themselves, ripe for critical examination.

New scientific ideas powerfully inform the value systems and world view of our culture. New technologies are almost always clad in utopian rhetoric. Any technology which is trumpeted as a 'liberation' should be examined extra-carefully. Historically, those technologies have transpired to be the most oppressive. Artists must be careful not to unconsciously endorse the paradigms hidden in scientific discourses, where they often lie hidden, disguised as axioms. That said, I think there is ample reason to be very excited about genetic programming and related techniques, as they offer versatile new modes for interactivity.


[1] cf. my discusison of the fallacy of the universally intuitive interface in 'Consumer Culture and the Technological Imperative: The Artist in Dataspace' in Critical Issues in Electronic Media, Ed. Simon Penny. SUNY Press 1995.

[2] made famous by Richard Dawkins, in his book The Blind Watchmaker. More recently the notion of co-evolution has emerged, in which several 'species' or a species and its environment, mutate in parallel, the species getting better at its environment, but its environment getting harder as well.

[3] Stefan Helmreich, Artificial Life IV conference, July 1994, personal notes.

[4] There is a clear similarity here between Ray's notion of the internet as 'wilderness' and the idea of cyberspace as a 'frontier'.

[5] Artificial Life IV conference, personal notes.

[6] Similarly, at MIT Rodney Brook's current project is to produce a robot with the mind of a two year old human.

[7] see Moreno et al, 'Universality without matter?' Alife 4 proceedings.

[8] Other examples of computer technology functioning in this way include current cognitive science and certain (slightly dated) neurological and psychological models. See J. David Bolter's Turings Man... see also my 'The Intelligent machine as anti-christ', SISEA proceedings 1990.

[9] see my 'Virtual Reality as the end point of the enlightenment project', in Culture on the Brink, Eds. Bender and Druckrey, Bay Press 1994.

[10] The scientific method hinges on the ability of researcher to perform reproducible experiments... But what can one do when the slightest error in reproducing an experiment's initial conditions can lead to a vastly different outcome? ... certain physical systems - governed by nothing more complex than Newton's law of motion - can display just such pathology ... No randomness is built into the model, yet the final state of the system cannot be predicted with certainty if there is any error (no matter how small) in the measurement of the initial condition... even qualitative reproducibility in simple classical systems cannot be taken for granted". 'Finding Riddles of Physical Uncertainty', Science News, vol 144, #2, Sept 18 1993.

[11] It is worth noting that what have come to be called Fractals were a late C19th mathematical oddity which languished as a mathematical curiosity until the vast arithmetic ability of the digital computer bacame available. The propensity of the computer for large scale