DRAFT ONLY - not for publication or redistribution!

tom ray's hammer:
emergence and excess in a-life art

mitchell whitelaw
mitchellw@spin.net.au

> tom ray's hammer

During an interview, Rod Berry recounted a story about Tom Ray, noted a-life researcher. Apparently when he was coding the first version of Tierra, an open-ended computational a-life experiment, Ray smashed the serial ports off his computer with a hammer, to make sure none of his self-replicating code entities escaped. This image stayed with me in the following months. Like throwing a brick through a television, the gesture has the visceral appeal of a "hardware" solution to a "software" problem - a transgressive, category-error brutality. More significantly, the excess of the gesture indicates the excess it tries to contain. Ray's digital replicators were, for him, creatures of such potential virulence that they needed to be constrained in hardware. Ray's hammer blow protects against, and simultaneously projects, a basic transgression; a-life goes "out of control", it escapes its isolated virtual pond and makes its way in the world. This is a basic moment in a-life and a-life art, the moment where the system exceeds itself, breaches its initial boundaries, surprises us; the moment of emergence.

As it turns out the story is untrue - probably derived from Stephen Levy's account of Ray's plan to enclose the Tierra computer in a "containment facility", preventing access to disk drives and ports.[1] While the image is less dramatic, the intent is the same. The cage defines a highly-charged boundary between the computational inside and the worldly outside; the prison walls make us think of escape.

This paper posits this will to escape, the desire for emergence, as a basic drive in contemporary a-life art. Not the only drive, by any means, but a basic, common dynamic in a-life systems. Here it's isolated, extrapolated outwards and followed; it provides a useful theoretical perspective on current a-life work, and at its limits it provokes more abstract questions about the edges of art and intentionality.

> emergence in a-life

A-life works have a common structure, when viewed as designed emergence-systems; exactly the same structure can be applied to a-life experiments in the sciences. There are two interconnected planes; a designed framework or substrate, the hardware and software system, and the emergent phenomena generated by the system. Different works display these planes in various ways. In the case of a "breeder" work, such as Karl Sims' interactive evolving images or more recent 3-d morphological breeders, the "system" consists of the programmed evolutionary engine that mutates and renders the symbolic genotypes, as well as the external input that selects genotypes to breed for the following generation. The emergent phenomena corresponds to the (singular) object of this process, the image or 3d model; this is the emergent "organism". Other types of work split into two in a similar way. In those simulating a population and its environment (for example Sim Earth), the emergent phenomena are not only individual phenotypes but individual and collective behaviours, population fluctuations, symbiotic relationships between phenotypes, genetic drift, and so on. In robotic (or "real") a-life, the system is designed in hard- and software, and the emergent phenomena are behavioural, largely arising from interactions with other entities, robotic and/or human.

In each of these cases there is a movement outwards or upwards, from the system-substrate to its emergent results. This movement can be identified in most, if not all, a-life based work. It is simple enough; the focus of the work, the thing it produces, is the emergent phenomenon. This is the result the artists are concerned with. This is particularly clear in software-based simulation works that present a "world" or "organism" - the hardware and software substrate all but disappears. Even in robotic a-life work where the hardware substrate functions sculpturally and gesturally, the focus is on the emergent interactions of the robot entities; the a-life related "smarts" of these systems is what distinguishes them from earlier sculptural kinetics. The movement from system to emergent phenomena within each work is part of a macro-scale drive shared with "hard" a-life work, the drive towards increasingly emergent results, outcomes that increasingly exceed their substrates.

Larry Yeager, discussing his 1993 Polyworld simulated ecosystem, lists emergent features of his simulated creatures, including response to visual stimuli (the creatures have simulated vision and neural net "brains"), food-seeking and flocking. In projecting the system's future, Yeager anticipates that the move to a more powerful computational platform would allow "larger populations,... more complex neural models, thus fostering greater speciation and more complex behaviours." [2] Kenneth Rinaldo, asked in an interview about his interactive robotic piece The Flock, similarly anticipates its future 'evolution', including a switch from a Brooksian subsumption-architecture approach to behaviour to a genetic algorithm method that would "allow a new Flock to evolve its own behaviour." [3] When I interviewed Rod Berry about his Feeping Creatures musical breeder he described his goal as a "perpetual novelty machine" which maintains a diverse and evolving ecosystem, avoiding the homogenising "hillbilly effect" of inbreeding. [4] Paul Brown describes a lineage in cybernetic and a-life related art in pursuit of an emergent, "computational aesthetic".[5] These system-builders are all moving in the same direction, aiming for more complexity, more diversity, more evolution, a greater emergent excess.

> emergences

A single definition of emergence is elusive, but its various aspects are worth investigating. The history of the concept extends back to J.S. Mill, who in A System of Logic (1843) talks about "heteropathic causation"[6] - the case where a joint effect of several causes cannot be reduced, or traced back, to its component causes. Mill, G.H Lewes (who seems to have coined the term) and early twentieth century proponents of "emergent evolution" C. Lloyd Morgan and Samuel Alexander develop a notion of emergence that is basically a kind of hierarchical holism: elements interact to form a complex whole, which cannot be understood in terms of the elements; the whole has emergent properties which are irreducible to the properties of the elements. For Morgan and Alexander in particular, emergence becomes a universal rule which explains the formation of life from matter, and of consciousness from life; the cosmos as a hierarchy of material levels, each emergent from the last.[7]

Later, more reserved treatments argue that emergence is simply a property of a complex whole such that our models, or theories, about the properties of its components are inadequate to describe the properties of the complex whole. Ernst Nagel argues: "the doctrine of emergence ... must be understood as stating certain logical facts about formal relations between statements rather than any experimental or even 'metaphysical' facts about some allegedly inherent traits of properties of objects."[8] So emergence is the unpredictable, or un-modelled, result of complex interactions, but as soon as we have a model or theory that accounts for this result, it ceases to be emergent and becomes predictable.

Peter Cariani pursues the implications of this epistemological approach for artificial life research. He presents a notion of emergence-relative-to-a-model comparable to Nagel's; it regards emergence as "the deviation of a physical system from an observer's model of it."[9] Particularly interesting here is the result that under this definition, computational artificial life simulations must be non-emergent. Cariani argues:

"All computer simulations can be described in terms of finite-state automata, as networks of computational state transitions, as formal manipulation systems. As observer-programmers we can always find a frame which will make our simulation appear nonemergent."[10]

In other words given the deterministic nature of computation (even with chaotic mathematics and pseudo-random numbers), a given initial state will always progress through the same succession of states and produce the same results. The key to Cariani's notion of emergence is the idea of the observational "frame", which constitutes the model that we have of the system. Cariani's point is that however difficult it might be to access practically, given a model of the computer as a closed, deterministic symbol processor and complete knowledge of its (finite, symbolic) states, its activity will never deviate from that predicted by that model. Cariani refers to the emergent properties exhibited by such simulations as instances of "computational emergence", where emergence is simply the production of diverse, complex macro-scale phenomena from a few simple, micro-scale rules. Like earlier ontological accounts of emergence, computational emergence is a hierachic holism, where the emergent excess of the macro-level results from complex interactions (compound causation) at the micro-level.

Cariani sets out a typology of devices that clarifies the structure of systems that meet his definition of emergence, as well as those that don't. The key features that the adaptive (emergence-capable) devices share are openness to the environment, some means of measuring and/or effecting the world 'outside', and reflexive flexibility, the device's capacity to adaptively alter itself. The types of alteration define two of the device types: "syntactically adaptive" devices can alter the rules that define their internal computations, while "semantically adaptive" devices can alter the mapping between environmental input and the internal symbolic representation. The latter corresponds to, for example, the formation of a new sensory organ. A device which is both syntactically and semantically emergent Cariani calls a "general evolutionary" device - capable of altering the "meaning" of its environmental measurements as well as the symbolic operations it performs on them. The criteria for the evolution of these categories might be predetermined and fixed, but Cariani anticipates devices with even greater levels of autonomy, capable of "constructing their own performance-measuring apparatuses" and hence attaining "motivational autonomy". Cariani muses, deadpan: "Such devices would not be useful for accomplishing our purposes as their evaluatory criteria might well diverge from our own over time, ...".

We could cast Cariani's typology in the less formal terms of a general drive or dynamic of emergence. Computational emergence - which corresponds to a nonadaptive, "fixed computational" device in this typology - is a single-stage model: there is a fixed relation between a computational substrate and its emergent complex phenomena. The initial state and simple rule-set are a deterministic "seed" that grows a complex plant (this analogy is literalised in the case of Lindenmayer systems), but the same plant will grow from the same seed time and time again. There is a single moment of excess, a single jump from the micro-computational to the macro-phenomenal level, and no more. The device-types that Cariani describes as capable of emergent behaviour under his model-relative definition are capable of ongoing emergence. The two levels, substrate and emergent result, mentioned earlier, form a feedback loop here: the emergent result loops back to effect changes in the infrastructure, which in turn alters the emergent result. The device types are determined by which parts of the infrastructure are included in this loop.

>imaginary meta-mutants

It is interesting to look at current a-life projects (both 'art' and 'science') through Cariani's typology. Many of the systems we think of as 'evolutionary' or adaptive begin to seem non-emergent. Breeder works, such as those mentioned earlier, seem to be what Cariani calls "fixed robotic" devices: they take an environmental input (choice of genotype to breed), subject it to computational processes, feed this back to the environment (or user) in the form of images of the mutant offspring, and the cycle continues. The system is open to its environment, through the tiny portal of boolean logic opened by interface decision-making, but neither the rules of its computation nor the semantics of its mapping of environmental input are subject to mutation. Even more embodied, robotic a-life systems fall into this category whenever they use fixed computational rules and fixed sensor-effectors.

So why bother with Cariani's formal, epistemological definition of emergence, if it won't allow us to describe these apparently evolutionary or emergent computational and robotic systems in such terms? Because it clarifies, even formalises, the fulfilment of the emergence-drive that was argued for earlier. Cariani's motivationally autonomous, general evolutionary device fulfils this desire for increased autonomy, increased emergent excess; it is the absolute endpoint of this drive. As well, Cariani's definition allows us to locate the frustration evoked by many (particularly simulation-based) a-life works. Once the initial stage of computational emergence becomes familiar, we are impatient for the next stage, another excess, another surprise. The tendency in a-life system designers discussed earlier reflects the same desire; and "hard" a-life, the goal of which is the creation of living things in a technological medium, shares this goal.

How can we resolve the disjunction between a-life practice and this supposed drive for emergence? What currently prevents this urge for excess and autonomy from being fulfilled?; and what might a-life art be if it were?

These questions can be considered in parallel as we perform some imaginary modifications on a-life art systems. Consider a breeder type system, one that produces a virtual three-dimensional "phenotype" through simulated genetic processes. These works have a simple, functional imperative of "exploration" and the generation of novel, appealing phenotypes - they are utilitarian. As such, applying the criterion of absolute emergence (as continual excess, continual variation) seems absurd. An object-breeder that decides to become an autonomous solipsistic sphere-generator fails to fulfil the functional brief of the system. However even within the bounds of functional object-breeding, existing systems are constrained by their fixed object-elements and combinatorial grammar. These might have built-in zoomorphic biases or more abstract vocabulary (such as William Latham's tentacle-shell-spirals[11]) but in each case this grammar limits the outcomes in a way that produces a kind of familial style. This emergent style is readily accepted in an art context as the style of the individual artist, but from the perspective of the exploratory, generative aims of these breeders it's a boundary, a limit.

Imagining a breeder that is emergent by Cariani's definition seems simple enough, initially: instead of simply evolving phenotypes using a predetermined genotype grammar, allow variation and evolution in the grammar and the rules of expression that link it to the phenotype. Mutant syntax, mutant rules, and meta-mutant phenotypes. The spiral-sphere tentacle-shell becomes a string of cubes, a lofted procedural "skin" covering the cubes, intersects with a 3dmf file sucked from the web, re-uses the recursive structure that generated the initial spiral but applies it to transformations on vertices of the imported file, a trait which becomes linked to surface characteristics of the lofted skin. A mutant syntax evolves that can generate its own singular permutational consistency (or style).

But programmers reading this will be crying foul, and rightly so. These syntactical changes would have to have been prepared, in advance, anticipated by the system's designer/programmer. Build in the capacity to be open to 3d content from the web and it's no surprise when the system fulfils it. The same applies to all the other mutations - the system must have been designed to accommodate them: this is not emergence, just a larger permutational repertoire. For these traits to be emergent, they must have emerged spontaneously from a general computational substructure, and it's difficult to imagine a computer program, however self-organising, happening upon the necessary code to deal with a network protocol, let alone the correct data structure for a 3d file. And if we withdraw all our handy bits of a-priori knowledge (about how to write to video memory, make a window, access the disk) and instead hope for a completely spontaneous result, it would seem ludicrously optimistic to expect the result to even be an object-breeder, no matter how many cycles of guided evolution we subject it to. Other lines of becoming branch off everywhere; the breeder looks more like a virus or a network worm, or a disk utility or most likely a spastic system-crashing monster, unexecutable code that takes its infrastructure down with it.

Mutant variability meets the formal brittleness of computational processes: to come into being, the mutant code must fulfil the (static) requirements of the interpreting and executing formal system. An analogous problem faces robotic a-life; how can we imagine a robotic device that evolves sensors and effectors? The rigidity or viscosity of the structure of our robots is such that any variation is more likely to be non-functional than functional: imagine a robot randomly evolving a sensory organ that transmits a video signal of the format that its vision-system is expecting.

It seems likely that any such radically emergent device would be operating outside the technological and formal grammars that define human-designed robotics and computation. We could expect an evolved infrastructure to be able to support variation with relative robustness, like biological life; its structures, like biological life, might not conform at all to the anthropomorphic metaphors which shape our technologies. Recent work on evolvable hardware indicates that the results function in a way completely unlike human-designed systems: the evolved devices use their circuit boards as physical artefacts with particular electrostatic properties, rather than as transparent instantiations of formal computational processes.[12] Looking for more efficient circuit designs, engineers using these processes get something else, a machinic alterity, an "other" functioning that works beautifully, but in a completely unexpected way.

>becoming-other

Artificial life begins to peel away from design, intent and human conceptual models, and becomes alien, as alien as our own bodies and those of our fellow biological creatures. We attempt to codify them, grasp them as knowledge, formal relations, biological causalities, but they continually slip out of our grasp. If it succeeds, a-life will involve the same relation, but in reverse: the known, formal, designed, modelled structures of our single-stage, holistically emergent systems, will give way to autonomous, mysterious, open systems. A-life as a scientific epistemological project relies, as Katherine Hayles says, on a kind of reverse-reductionism; the creation of the mysterious, excessive, ungraspable from its knowable components.[13] Instead of dissecting the frog, it tries to build one - but the goal, absolute knowledge of the living thing, is the same.

Artists, by comparison, embrace a-life with more synthetic, creative aims. As suggested earlier, the emphasis here is on the emergent result, the excess, rather than the known relation between the formal infrastructure and the emergent phenomenon. In as much as it is driven by a desire for absolute emergence, endless excess, a-life art is a meta-creative endeavour: it wants to create creation, variation, otherness. In Deleuzean terms, it wants (in a very literal way) to begin a line of becoming, a line of flight. If a-life science is about knowing and understanding, a-life art is very basically about making and being, that is being- and becoming-other, becoming-unknown. This becoming puts a-life art in tension with itself; currently, paradoxically, making increasingly sophisticated a-life systems demands increasing technical knowledge, an increase in willed design, control, intentionality; all towards an end which aims to exceed that design and knowledge. This approach leads to an a-life art which follows the explorations of scientific a-life, implementing its techniques in a different aesthetic guise. If it is to get what it wants, a becoming-other, an endless excess, it has to surrender at some point in this process. The question is whether this point of surrender, the point of emergence, will arrive when the technological infrastructure and formal innovation reaches a certain crucial point, or whether it might appear in another domain, on another axis altogether.

Isn't emergent excess already all around us? Stepping the emergence-frame back a few notches, it's possible to see art in general as a system producing emergent phenomena. A work is only its concrete material self, paint, canvas, steel, electricity, plastic, but it is open to its environment in rich and multiplex ways: it effects individuals in ways that we can anticipate (cultural theory) and ways we cannot (the myriad particularities of an encounter or interpretation); it transmits itself to societies and individuals (themselves involved in complex feedback relations) and might reproduce or persist in various ways (quotation, critique, imitation, style). Considered diachronically, art gives rise to emergent phenomena (culture, discourse) which inform the production of subsequent works - the feedback loop of ongoing emergence.

A-life art begins to seem like a strange reflexive involution of this commonplace - it seeks to formalise emergence, wrap it up into a finite system (within the bounds of which it is bound to fail) and then presents it in a context where its results are truly part of something open, emergent and unpredictable. Cariani, intriguingly, allows for this possibility when talking about computational emergence. Emergence here is in the mind of the observer: "The interesting emergent events that involve artificial life simulations reside not in the simulations themselves, but in the ways that they change the way we think and interact with the world."[14] The frame shifts; these simulations are not emergent in themselves but like all other artefacts of human culture (and all other formed matter, for that matter) are part of the complex system of all complex systems.

The edges of all these categories start to blur. Art (and culture) are reflexive emergent phenomena, and a-life art is embedded in its own loops of cultural and technological feedback and discourse. Evaluating it against its own implicit drive towards emergent excess is less a critique than a tactic for tracing the wider implications of that drive, following it out beyond the edges of the individual system. Emergence exceeds the capacity of its host-system to contain it, it moves across domains. A-life (art?) that `succeeds' might be memetic, or cultural, as much as robotic or computational, it might be imperceptible, subsisting within and across existing structures but changing, adapting itself and them. If the coevolutionary processes observed in biological life are any indication, emergent a-life would sustain itself in a niche formed by the transverse ecologies of media, culture, technology and biology. There is no reason why it should stay in the gallery, or in the computer - Tom Ray's urge to contain his replicating code followed that intuition. If a-life art were to fulfil its desire for excess, it would cease to be art, as currently defined, becoming unbounded and unintentional. It would be more like a-human engineering: purpose and design are overtaken by raw, persistent functionality. "The functional question in ... desiring-production is not what is it for but simply whether it works: Does it make something happen?"[15]

Current a-life art is contained by the rigidity of its technological substrates, its emergence channelled along designed paths. It only hints at the kind of abstract, transverse, machinic emergence considered above. But as much as it is involved in a desire for emergent excess, it projects that threshold, beyond which the system continually exceeds itself. This point, as a trope or a cultural value, is fraught with difficulty. It seems to participate in the worst kind of techno-utopianism or mystical apocalyptic storytelling - it's reminiscent of Vernor Vinge's "Singularity"[16], or Frank Tipler's "Omega Point". Within new media theory this kind of emergent threshold appears in Roy Ascott's telematics, as a Noospheric network-consciousness. These tropes sit easily with the dominant drives of the affluent West, increased prosperity and technological progress, and tend to treat the technological means to their cyber-transcendent ends as unproblematically transparent and accessible. A-life artworks, however much they subscribe to such notions, pull back from their abstract hyperbole simply by being concrete implementations. They involve a machinic pragmatics, a functionality; they deal with the real conditions of technological production, easily neglected in rhetoric. They are situated in a crucial gap, between the desire for excess and its implementation. At a time when post-human rhetoric is running high, a-life art can be seen as an empirical study in constructing a machinic, a-human "outside" from within the cultural and technological present. Extropian posthumanism leaps easily into a distant techno-future, abandoning worldly things. In contrast, a-life art provides an instructive example, in an odd microcosm, of the real difficulties involved in pursuing an absolute becoming-other.

Endnotes

[2] Larry Yeager, "Polyworld: Real Life in an Artificial Context" in Karl Gerbel and Peter Weibel (ed.s), Ars Electronica 93 (Linz, Austria: Ars Electronica, 1993), p.126.

[3] PDQuick, "The Emergence of ALife" (interview with Kenneth Rinaldo) in Switch vol. 1 no. 3 <http://cadre.sjsu.edu/switch/alife/pdq/rinaldo.html> (March 1998)

[4] Rodney Berry, interview with the author, 14 October 1997.

[5] Paul Brown, "Emergent Behviours, Towards Computational Aesthetics" in Artlink vol. 16 no.s 2&3 (1996), 16.

[6] John Stuart Mill, A System of Logic (London: Longmans, 1843 repr. 1961), vol. 1, pp. 410-413.

[7] see for example C. Lloyd Morgan, Emergent Evolution (London: Williams and Norgate, 1923)

[8] Ernst Nagel, The Structure of Science: Problems in the Logic of Scientific Explanation (London: Routledge & Kegan Paul, 1961), p. 369. (italics in original)

[9] Peter Cariani, "Emergence and Artificial Life" in Christopher Langton, Charles Taylor, J. Doyne Farmer and Steen Rasmussen (ed.s) Artifcial Life II (Redwood City, CA: Addison-Wesley, 1992), p.779.

[10]Ibid. p.789.

[11] see Stephen Todd and William Latham, Evolutionary Art and Computers (London: Academic Press, 1992)

[12] see Clive Davidson, "Creatures from Primordial Silicon" in New Scientist Vol. 156 No. 2108 (November 1997), 30-34.

[13] Katherine Hayles, "Narratives of Artificial Life", in Jon Bird, Barry Curtis, Melinda Mash, Tim Putnam, George Robertson and Lisa Tickner (ed.s) FutureNatural (London: Routledge, 1996), p.153.

[14] Cariani, p.790. (italics in original)

[15] Ronald Bogue, "Art and Territory", South Atlantic Quarterly vol. 96 no. 3 (Summer 1997), p.479.

[16] see Vernor Vinge, "The Singularity" <http://kuoi.asui.uidaho.edu/~kamikaze/documents/vinge.html> (March 1998)