5   Can Evolutionary Art Provide Radical Novelty?

Margaret A. Boden

5.1   The Question

Many computer artists engage in evolutionary art, or Evo-art, in the taxonomy of chapter 2. They include Jon McCormack, Karl Sims, William Latham, Ashley Mills, Lise Autogena, and Joshua Portway (for others, see Whitelaw 2004; Romero and Machado 2007). Rather than relying on symbolic artificial intelligence (AI) programming techniques, as Harold Cohen (for example) does, they draw on A-Life (artificial life) methods inspired by evolutionary biology.

A-Life scientists employ several very different computational methods, including cellular automata, situated robotics, and genetic algorithms (Boden 2006, chap. 15). And artists inspired by A-Life use all these, as is clear from Mitchell Whitelaw’s (2004) wide-ranging interviews with this community. But our interest here is in artworks produced by evolutionary methods. More accurately, our interest is in the creative potential of that general class of artworks.

Evo-artists work with programs containing self-altering rules—genetic algorithms, or GAs (Boden 2006, 15.vi). They rely on the GAs to effect random mutations in the code that generated the current artwork. Then, they select the preferred items from the descendants for use as parents of the next generation—continuing, perhaps, for many hundreds of generations.

Although the variations are generated by the program itself, the selection is usually done by the human artist or sometimes by the audience, in the gallery or on the Internet (e.g., Graham Nicholl’s Living Image). That is because it is not normally possible to express aesthetic values clearly enough for them to be stated as an automatic fitness function.

A few Evo-artists do employ automatic selection, because the evolution is not driven by aesthetic criteria. For instance, Autogena and Portway’s Black Shoals Stock Market Planetarium, which was exhibited—and evolving—for three months at Tate Britain in 2001, is an artificial ecosystem whose evolving creatures adapt their behavior to real-time financial input from stock markets.

The types of mutation that are allowed vary from case to case and are decided by the human artist. The general types of change, then, are broadly predictable (although which type will be seen next is unknown). But the actual variations are not. Often, they surprise even the originating artist. And the more fundamental the types of mutation that are allowed, the more unpredictable the observed results will be.

So much is beyond dispute. What is much more controversial is whether these mutations and variations could ever be truly radical. Could they do more than merely tweak the current style (the current program)? Could they ever jump from one style into a fundamentally new one? Could Evo-art go beyond highly unpredictable exploratory creativity to achieve radically transformational creativity (Boden 2004)? Or is that a capacity enjoyed only by human artists?

If it is, then the lack of interest in computer art that is shown by most of the art world, as described in chapter 3, would be largely justified. Besides their skepticism with respect to computers and emotion (see chapter 6), most people think that computers could not come up with anything interestingly new.

It would be possible, of course, for a human artist to transform a style and then express the new (post-transformation) style in an Evo-art program—which could proceed to tweak and explore it. Indeed, a process of successive transformations effected by the originating artist can already be seen in the versions of Cohen’s (nonevolutionary) AARON program, each of which draws in a recognizably different style (McCorduck 1991; Boden 2004, 150–166, 314–315). But the question here is whether the program itself could do this. Can computer art and, in particular, Evo-art provide radical creativity on the part of the computer?

More accurately: Can computer art at least appear to provide radical creativity? As explained in chapter 1, whether this would count as real creativity is a philosophical issue that is here ignored (but see Boden 2004, chap. 11; 2014).

Most people seem to believe that the answer is obviously no. Given a style, they may admit, a computer can explore it. But if you want it to come up with a new style, don’t hold your breath! After all, they say, a computer does what its program tells it to do—and no more. The rules and instructions specified in the program determine its possible performance (including its responses to input from the outside world), and there is no going beyond them. Those rules and instructions can change as a result of GAs. But because the initial Evo-program includes specifications of the GAs, those changes themselves lie within its computational power. Going outside the limits set by the program is therefore impossible. So radical creativity simply cannot happen, QED.

We see later that this assumption is too quick. In principle, Evo-art can offer radical transformations. Why it does not do so every day is another question, discussed briefly in section 5.5. First, let us ask why artists choose to do Evo-art in the first place; more specifically, whether they do so in the hope of arriving at radical stylistic change.

5.2   Why Do Artists Choose to Do Evo-art?

There are three main reasons why some computer artists have decided to use this particular A-Life technique. First, they may want to celebrate the marvel of life itself. That this is a common justification is clear from interviews with many Evo-artists (Whitelaw 2004).

A few of these life-celebrating Evo-artists go even further. They have been inspired by the handful of maverick A-Life scientists who hope to create life in computers. They see their own efforts as linked with, even contributing to, this general project. In other words, some Evo-artists believe in the possibility of life in cyberspace: virtual life, or strong A-Life.

However, they are mistaken. Briefly (for a full discussion, see Boden 1999), the reason they are mistaken is that metabolism is a criterion of life, and computers do not metabolize. Metabolism is a form of self-organization that cannot be understood in purely computational terms but only by reference to physical energy. Computers consume energy, of course. But metabolism in living organisms is the self-creation and maintenance of a physical unity (the cell or the body) by means of interlocking biochemical cycles of some necessary complexity. There is none of that in computers.

So the idea that Evo-art could be a first step on the road to virtual life is misguided. It does not follow that art inspired by this fond hope is aesthetically worthless. After all, even atheists can appreciate a Madonna or a Pieta. But one does have to know about these (few) Evo-artists’ faith in virtual life—just as one has to know about the Renaissance artists’ faith in the scriptures—to understand what they are up to.

The second reason for getting involved in Evo-art is to piggyback on the creative power of biology. Biological creativity, whether in evolution or the developing embryo, differs from what is normally meant by creativity. Unlike artistic (or scientific) creativity, it does not generate ideas or artifacts. Nor is it goal driven—still less consciously monitored. But it does produce structured phenomena that are new, surprising, and valuable.

“Valuable”, here, does not mean valuable to us but valuable in terms of the likelihood of survival or of finding a mate. Sometimes biological evolution does favor structures that are highly valuable to us—the male peacock’s tail, for example. And in that case, the feathers we find so extraordinarily beautiful are attractive also to the peahen (although not necessarily with respect to exactly the same visual features). But there are many biological structures, evolved because of their value for survival or attracting mates, that most humans find uninteresting or even repulsive. This can act as a reminder that what humans are willing to count as valuable can be highly problematic, not least in art (see section 5.6).

GAs mimic the creative power of Darwinian evolution, being modeled on the sorts of mutations that happen in living organisms. Biological evolution results from many small changes, not sudden saltations. A-Life simulations have proved that the saltations sometimes observable at the phenotypic level—so-called punctuated evolution (Eldredge and Gould 1972)—can be produced by a gradual accumulation of small changes at the genetic level. Very few, if any, of those individual (genetic or phenotypic) changes count as transformations. Even small mutations can be damaging for a living organism, and larger ones are very likely to be lethal. Nevertheless, over a vast period of time, the evolutionary process has delivered an unimaginable variety of flora and fauna, many of which we find delightful.

It is not surprising, then, that some computer artists have sought to enrich their work by means of evolutionary programming. Indeed, there is even more likelihood that GAs will produce a fair proportion of acceptable (that is, nonlethal) changes than that biological mutations will do so. The products of Evo-art are not subject to the many fierce constraints posed by metabolism, food supply, climate, and predator-prey ecology.

In short, the second reason for getting involved in Evo-art is well founded. It is entirely reasonable to expect that Evo-art will often produce aesthetically acceptable novelties, including many that could not even have been imagined, still less anticipated, by the human artist concerned. What is more, that expectation has already been satisfied on countless occasions—for instance, in the work of the Evo-artists previously named.

The third reason for doing Evo-art follows on from the second. It is the hope that truly radical stylistic change might result, as has often occurred in the evolution of life. If that were to happen, then computer art would have achieved a strong form of originality: transformational creativity, comparable to that celebrated in histories of art. Not a new style to compare in value with Giotto or Pablo Picasso, perhaps. (Then again, why not?) But not mere run-of-the-mill exploration or even modest tweaking, either.

Not all Evo-artists cleave to that third reason. Indeed, most of them allow only relatively superficial changes. Their seeming lack of adventurousness is grounded in the fact that—like most artists, most of the time—they have found a style that interests them and are content to explore it. Occasionally, they may switch to a different style, but then they devote their effort to exploring that one. Moreover, as professional artists they are usually happy for their Evo-art work to display their personal signature (Boden 2010) and so consolidate their professional reputation. (An interesting exception is discussed in section 5.5.) The Evo-artist Latham, for instance, generates images that, despite their literally unimaginable variations, are instantly recognizable as his (Todd and Latham 1992).

Even if an Evo-artist does start out partly inspired by this third reason—that is, the hope for radical stylistic transformations—he or she will very likely be disappointed. An artistic style is a sustained pattern of activity. If radical mutations, and therefore huge structural changes in the phenotypic image, are possible at every generation, no pattern will be sustained for long. There will often be no visible family resemblance between ancestors and descendants, even if they are separated by only one or two generations. To give a biological analogy, it is almost as though a cat had given birth to a puppy or a fish or even a buttercup.

This is evident in Sims’s (1991) work, wherein the mutations include nesting one entire image-generating program inside another or concatenating two such programs. These radical changes can occur on many hierarchical levels and so may produce programs or images of significant complexity. The system has been exhibited at the Pompidou Centre and elsewhere around the world. The gallery visitors are initially amazed and fascinated by the transformations taking place before their eyes. But they soon become bored and frustrated, precisely because of the unsatisfying lack of sustained stylistic exploration. When they try (by choosing the parents for the next generation) to steer the system toward certain colors or shapes, they are rapidly disappointed: sooner rather than later, unwanted features will appear and previous advances will be lost. This is exasperating for anyone seriously interested in the aesthetics of the evolving images. For artistic purposes, the program is too transformational. It is surely no accident that Sims is not a professional artist seeking to advance his oeuvre but a computer scientist concerned to show the potential of evolutionary technology.

So the third reason for engaging in Evo-art is something of a poisoned chalice. There is an antidote for this poison, however. Namely, the third-reason Evo-artist could choose to freeze the program on its attainment of an especially interesting or attractive image and then go on to explore the stylistic possibilities inherent in that. Evo-art techniques might still be used, but these would not now allow mutations as radical as those that enabled the system to come up with the favored transformation in the first place. This schedule of work would mirror the activity of mainstream artists, who—as suggested earlier—normally follow transformation, if they ever achieve it at all, with exploratory consolidation of the new style. Consider, for instance, the chronological sequence evident in a retrospective exhibition of a highly transformational painter such as Picasso.

This antidote to the “poison” threatening third-reason Evo-art suggests that our key question can be restated. Instead of asking whether Evo-programs could ever generate radically new styles, we can ask whether they could generate radically transformed structures, which might then be explored (with or without GAs) so as to found a new, sustained style. Prima facie, this appears to be possible; Sims’s work is enough to show that.

One might ask whether it would be possible for Evo-art to achieve stylistic transformation directly if radical mutations were allowed that were not as unconstrained as Sims’s are but that were more adventurous than Latham’s. For example, suppose a GA technique could lengthen the programmed genome (as Sims’s system can) but do so more gradually. Lengthening the genome is potentially radical because it enables greater structural complexity in the resulting image. In biology, to be sure, length of genome does not neatly match morphological complexity: some lowly creatures have many more genes than we do. Nevertheless, organisms that are more complex do tend to have more genes, which allows for a greater variety of gene-gene interactions. In other words, longer genomes can provide an explosion in the possibility space.

A GA technique like that just suggested was developed more than twenty years ago. Inman Harvey’s (1992) SAGA (Species Adaptation Genetic Algorithm) varies the length of the genome in a gradual—and therefore relatively sustainable—manner, enabling (as he puts it) new “species” as well as new varieties to arise. So perhaps SAGA could lead third-reason Evo-artists to a sustained stylistic transformation, not just to a single structural transformation?

The many SAGA-evolved forms that already exist include some impressive quasibiological examples. One is a simple feature detector in a robot’s “brain,” broadly comparable to the orientation detectors found in visual cortex (Harvey, Husbands, and Cliff 1994; Husbands, Harvey, and Cliff 1995). This mini-network, which is sensitive to a light-dark gradient at a particular orientation, evolved unexpectedly from a controller network whose connections were initially random.

One can say that it was evolved by the GA, not merely (like Sims’s surprising images) generated by it, because all its robot ancestors were automatically selected by the same fitness function. This was intended by the programmers to result in the ability to navigate within a toy environment. That environment happened to contain a white cardboard triangle, and the feature detector evolved as part of an adaptive visuomotor mechanism. Connections to visual and motor units within the whole network enabled the robot to use the triangle as a navigation aid. This ability, once it had emerged, was selected and maintained by the fitness function.

At first sight, then, this example proves that GAs can indeed provide radical novelty. But the skeptics whose argument is summarized at the end of section 5.1 may disagree. In particular, those critics who take the biological inspiration for evolutionary programming seriously may still insist that GAs cannot get beyond exploratory creativity. Even varying the length of the genome, they will say, does not avoid the problem. For the possibility space, albeit huge, is still inherently fixed by the program.

Nor is the problem avoided, some objectors say, by human intervention at the stage of selection, as happens in almost all Evo-art. It is true that such intervention can vary the fitness function (which is rarely explicit anyway), whereas an automatic fitness function, like the one used in evolving the orientation detector, is fixed. This is relevant because biological fitness functions change too as evolution proceeds. But even interactive Evo-art, these objectors would insist, is limited by the initial program, so that truly transformational creativity is impossible.

Third-reason Evo-art, on this view, is doomed. In section 5.5, we see that this common objection is misguided and that the significance of the white cardboard triangle in the robots’ playpen was not fully appreciated.

But first, we must ask how biology, and also human artistry, manages to produce radical changes. If we understood that, we might—despite the seemingly cogent objection previously outlined—be able to achieve genuinely radical (third-reason) Evo-art after all. We consider biology in section 5.3 and artistry in section 5.4.

5.3   Open-endedness and Openness in Biology

Biological evolution exemplifies transformational creativity whenever it generates forms radically unlike what existed before: vertebrates from invertebrates, for instance. Such morphological changes are deeper, more surprising, than mere iterations such as segmentation, wherein a structure is repeated many times without significant variation; think of millipedes and earthworms or the vertebrae and spinal nerves of human beings. They are also deeper than slight improvements in preexisting functions, such as those that enable creatures to run a little faster or to see a little better than their parents.

For example (as skeptics regarding GAs often point out), biological evolution has resulted many times in fundamentally new types of sense organs. Besides producing improved eyes, ears, and noses, it has created first-time eyes, ears, and noses. And, of course, it has generated new types of communication and representation, including (in Homo sapiens) language and thought. These are all examples of what is meant by saying that biological evolution is open ended.

As we have seen, it is not obvious that evolutionary A-Life can do the same sort of thing. Are the transformations produced by GA programs really transformations at all? Or are they merely deceptively impressive cases of exploratory creativity, in which the possibility space being explored has been implicitly predefined by the programmer? This issue worries not only skeptics about Evo-art but also the scientific A-Life community. Indeed, they have identified open-ended evolution as the most important of fourteen grand challenges for the field (Bedau et al. 2000; Rasmussen et al. 2003).

An enthusiast for Evo-art might reply by saying that, likewise, an organism’s DNA (plus the mechanisms of meiosis) defines a set of possibilities. So if DNA genomes can create genuine novelties, why not artificial genomes?

The answer often given by critics of GAs is that embodiment and environmental embeddedness are crucial. That is, biological form does not result only from DNA, which exists in a richly varying physical environment that can affect it in unpredictable, evolutionarily fruitful, and sometimes even transformative ways. It follows, on this view, that purely programmed A-Life (and by implication Evo-art) cannot give us what we want—namely, genuine transformations. Only physical models could do that.

This is now an increasingly popular position within A-Life circles, but it dates back to an A-Life pioneer of the mid-twentieth century (Ashby 1952; Boden 2006, 4.viii.c–d). William Ross Ashby’s “law of requisite variety” stated that the variety in a system’s regulator must be at least as great as the variety in the system being regulated. So if the system is to outstrip the possibilities inherent in its regulator, it must be informationally open. That is, it must be capable of being affected by events that were not specified within the regulator.

There is a potentially confusing pun here. The open-endedness of an evolving system is its ability to come up with radical novelties of many different kinds. That is different from its openness to information from outside the system itself. Openness to external influences is what the general systems theorist Ludwig von Bertalanffy (1950, 1962) had in mind when he contrasted closed and open systems—although he spoke of openness to energy and physical causation rather than information.

Although the two notions are distinct, they are closely related. The first is made possible by the second. Open-endedness of evolution, as we will see, depends on informational openness in the members of the evolving population. However, the reverse is not true: informational openness need not result in open-endedness. That is why, even though all organisms are open systems, not all biological evolution is transformational. If the external events in question are of a type to which the organism is already adapted, then evolution is likely to be more a matter of incremental or exploratory improvement than of radical transformation. But if they are not, then open-ended evolution may come into play, and the irrelevant may suddenly become relevant.

To a species without light sensors, for instance, light is irrelevant. Nevertheless, given some fortuitous mutations, it may turn out to be highly relevant, leading to the emergence of sensory capacities of a radically new kind.

Specifically, some mutations must occur within the creature such that new molecules are generated that are sensitive to light and are near enough to the surface of the creature to receive light. However, that is not enough for even the crudest form of vision. In addition, the photochemical changes must affect other molecules, eventually resulting in changes that have survival value for the organism. Suppose, for example, that the molecules generated at some point in the chain of light-triggered chemical reactions were to stimulate the motor organs, causing movement toward or away from the light. This could bring the animal within reach of better food or a safer hiding-place. In that case, natural selection could take over to maintain—and, given further mutations, even to improve—the new sensor. (For fascinating accounts of the evolution of various types of eyes, see Cronly-Dillon 1991; Land and Nilsson 2001.)

In this context, the robotic minicontroller whose evolution is described earlier is an interesting case. Strictly speaking, perhaps, it should not be counted as the evolution of a wholly new sense organ, because it started off with some very simple photoreceptors. But it did evolve a wholly new sensorimotor circuit and an equivalent behavioral capacity. Initially, its light sensor had purely random connections to any motor neurons. The new sensorimotor circuits were able to evolve only because the robots’ environment happened to contain a piece of white cardboard, one of whose edges happened to match an angle of light-dark contrast to which the photoreceptors could respond.

So openness is crucial for open-ended evolution. And this applies to evolutionary programs no less than to biology. Indeed, Ashby explicitly applied his law of requisite variety to both natural and artificial systems. Admittedly, he was thinking of physical artifacts (as in his Homeostat), not of computer programs. But the same principle applies.

Where the regulator is DNA (plus natural selection), physical interactions with external events can sometimes prompt radically new senses and forms of behavior, which can then be honed and improved by further adaptations. Where the regulator is a program, informational openness would be satisfied if—analogously—the physical states of the system being regulated by the program were sometimes affected by causal interactions with physical events in the external world. Some of these might appear, beforehand, to be utterly irrelevant to any potentially adaptive change in the system concerned. But if they satisfied the fitness function being used by the program, they could be maintained (and even improved) likewise.

The problem with (most) artificial evolution, then, is not that the system is artificial but that it is informationally closed. Artificial systems that are open to environmentally triggered change may show transformations that are truly radical.

This was exemplified half a century ago by the exceptionally creative cybernetician Gordon Pask (1959; Fernandez 2008; Boden 2006, 4.v.e). One of his evolutionary chemical machines grew threadlike crystalline structures—dynamically balanced between the deposition and re-solution of ions—supposedly analogous to concepts because they could discriminate sounds of different pitch (fifty or one hundred hertz), or the presence or absence of magnetism, or differences in pH level. These chemical threads were not initially designed to discriminate pitch (or magnetism or pH) but naturally arose in a way that made this discrimination possible. They were a primitive example of the self-organization of new perceptual abilities, creating new perceptual dimensions, which happens from time to time during phylogenetic evolution (Cariani 1992, 1993, 1997).

In recent decades an A-Life experiment unexpectedly led to a comparable result. Jonathan Bird and Paul Layzell (2002), to their amazement, found themselves confronted with a primitive radio antenna (a radio wave sensor) that picked up and modified the background signal emanating from a nearby PC monitor. They had been using a technique for evolving circuits in hardware (Thompson 1995) to evolve an oscillator, not a radio. But to their surprise, some of their devices exploited unconsidered or accidental aspects of the physical environment to produce radio reception and seemingly irrelevant features of the environment had unexpectedly become relevant.

These aspects would never have been specified in a program that was intended to evolve a radio antenna. They included the aerial-like properties of printed circuit boards, the proximity of the PC, the order in which the analogue switches had been set, and a soldering iron left on a nearby workbench that happened to be plugged in.

This is a mixed bag. Bird and Layzell concluded (2002, 1841) that the open-ended evolution of novel sensors requires a physical device, whether natural or artificial, whose primitives are sensitive to a wide range of environmental stimuli; not just light (which in this case remained irrelevant) but many other physical properties too. If the specific physical situation in which this research was done had been slightly different, arcane chemical properties of the paint on the surrounding wallpaper might have played a role.

Informational openness has contributed to transformative biological evolution and to some artificial versions too. In all cases, this evolution has occurred entirely within the overall possibility space. But the skeptic cannot therefore argue that there is no genuine novelty involved. Any novel structure, whether biological or not, is made possible by—and constrained by and confined within—the totality of preexisting conditions, environment included. That is just to say that creativity does not happen by magic. Creation ex nihilo is an illusion, but creativity is not.

We have seen that artificial evolution can sometimes emulate biology’s power to originate (e.g., sense organs) as well as to improve. But in the cases described so far, it does so by physical interaction with the environment. That is not normally attempted in Evo-art. Indeed, it is not normally possible, because Evo-art’s evolution usually occurs in cyberspace. If Evo-artists really want to generate radical change, will they have to arrange for physical interactions too? Will they have to forsake pure programs, turning instead to electronic circuitry or robots?

Before we can answer that question (in section 5.5), we must ask (in section 5.4) whether art in general achieves its transformations as a result of physical influences from outside. Have the radical novelties found in painting, music, or literature arisen thanks to serendipitous physical interactions—like those that took place in Pask’s glass containers, in Harvey’s robot’s environment, or on Bird and Layzell’s laboratory bench?

If so, then presumably third-reason Evo-art would have to do likewise: Evo-art only as evolutionary robotics. But if not, then not. Evolution in cyberspace, with not a robot in sight, could suffice.

5.4   Openness in Art

For members of Homo sapiens, the environment is not only physical and biological but sociocultural too. Some of the external factors that affect the organism—or better, the organism’s mind, the person—are driven by or cause physical events that implement or are interpretable as thoughts. Consider Marcel Proust’s memory-triggering madeleine, for instance, or Immanuel Kant’s being (as he put it) “awoken from [his] dogmatic slumbers” by reading David Hume.

It may seem unlikely that a merely physical stimulus could lead to interesting (valuable) changes in art. But there are some examples. It is fairly common, for instance, for sculptors to be strongly influenced by uncovering hidden veins in a block of marble. In general, art involves stigmergy: changes in the overall plan that are prompted by unexpected events during its execution (Harrison 1978). Those events are sometimes psychological in nature, a remark made by a bystander, perhaps. But often they are intrinsically meaningless physical features (like veins in marble) that are interpreted by the human artist as semantically or aesthetically significant.

Some such cases involve transformational creativity. Proust’s madeleine is one example: the ingesting of the cake triggered a host of memories and productive thoughts in Proust’s mind, which led in turn to a radically new form of novel writing. Another concerns a jazz drummer with Tourette’s syndrome (Sacks 1985), who was a fine enough musician to be able to use the random noises caused by uncontrollable tics in his hands as the seeds of improvisations that could not have happened otherwise. And it is highly probable that some of the deeply surprising (transformationally creative) leaf arrangements and ice sculptures created by the artist Andy Goldsworthy were grounded in his serendipitous perceptions of fallen leaves or melting ice in their natural state.

As for scientific creativity, think of Alexander Fleming’s accidentally abandoned petri dish. The unexpected physical stimulus (the clear areas visible in the dish) led him to infer that there might be a bactericidal substance there—the very first to be discovered. Even Friedrich von Kekule’s chemically revolutionary benzene ring, according to his own report many years later, was triggered by the specific shapes of flames dancing in his hearth (Boden 2004, 25–28, 62–71).

Clearly, then, the various forms of unpredictability that characterize creative thought (Boden 2004, chap. 9) include some serendipitous events that are physical rather than meaningful. However, we must remember Louis Pasteur’s observation that “fortune favors the prepared mind.” Only a highly accomplished jazz drummer could see how to relate random taps to musically meaningful schemas for improvisation. Similarly, only an expert chemist, already aware that the relations between neighboring atoms within a molecule are important, could see the potential chemical relevance of a topological change from string to ring. In other words, the physical event must be understood as meaningful or assimilated into meaningful schemas if it is to have its creative effect.

Typically, however, the unexpected stimuli that lead to creative thought—whether transformational or not—are “thoughts” themselves. (This term is enclosed in quotation marks because it includes thoughtfully produced artifacts, such as paintings or music, as well as purely conceptual examples.) Remember Hume’s “awakening” of Kant, again. Or think of Henry Adams’s (1918) highly creative response to the contrasting artifacts he encountered in the cathedral and the turbine hall.

Some of these thoughts come into the mind from the external world: other people’s remarks, sentences read in books, items heard in concert halls, or sights seen in art galleries or turbine halls or even fireplaces. Their transformational effect in the relevant person’s mind may be almost immediate (recall Kekule’s reaction “But lo! What was that?” on seeing two flames suddenly linking themselves together). Alternatively, they may be stored away in the recesses of the memory, to be resuscitated only when the relevant creative process (the painting, poetizing, composing, theorizing, and so on) gets under way.

Yet other creativity-triggering thoughts come not from the external (physical or sociocultural) world as such but from that part of the person’s internal world that appears to be external to the specific task in mind. They are generated within the person’s own mind or brain on the basis of already existing thoughts and mental associations, many of which would previously have been regarded—even by the artist—as utterly irrelevant to the artistic project in question.

These thoughts and mental associations include highly general combinational principles (such as synonymy, antinomy, and superordination) and exploratory and transformative heuristics (such as iteration and consider the negative). But even in those cases, previous outside influences—the person’s past social interactions, education and reading, and cultural experiences in general—are crucial.

In brief, human minds are open systems. And what they are largely open to are thoughts. No man, after all, is an island—and no man’s memory, either. Our language and cultural communication are guarantors of that.

5.5   What Sort of Openness Does Radical Evo-Art Need?

The implication of section 5.3 is that if Evo-art were to produce artworks existing in the physical world—robots, for instance (R-art, in the taxonomy of chapter 2)—then it might show true open-endedness. Indeed, critics who take biological embodiment especially seriously would probably say that any truly radical (third-reason) Evo-art must be a subclass of R-art.

The implication of section 5.4, however, is that physical openness is not strictly necessary for artistic open-endedness. Radical Evo-art could also result if the program were open to semantically interpretable representations from outside or to internally generated representations (thoughts) that are prima facie irrelevant. And if nonevolutionary computer art were open to a semantic environment, it too could avoid being trapped inside the space of possibilities defined by the program.

Let us consider these implications in turn.

It might seem revolutionary to recommend that Evo-art turn toward physical artifacts such as robots, given that R-art is only a small corner within C-art as a whole. It would be better, however, to call it counterrevolutionary. The seminal Cybernetic Serendipity exhibition, held in London nearly fifty years ago, was dominated by a bevy of simple robots (Reichardt 1968, 2008; Fernandez 2008). That was before the development of hugely powerful general-purpose computers, not to mention computer graphics. Today, thanks to those technological advances, relatively few C-artworks are robotic.

Even fewer are both robotic and evolutionary. One such example, however, is especially interesting here. The C-artist Paul Brown, who was awakened to the artistic potential of computers by his visit to Cybernetic Serendipity, has been trying for many years to lose his personal signature.

An artist’s personal signature is not his or her autograph on the canvas but consists of a range of features, including many of which the artist is not even conscious, that make the artwork recognizable as having been done by that individual. Personal signatures exist because of the need for computational efficiency in human minds (see Boden 2010). So Brown is an exception to the generalization made in section 5.2 that professional artists want to preserve, and even highlight, the characteristic individuality of their work. The reason lies in his commitment, as a very young man, to the impersonal program of modernism—which attracted him (and Ernest Edmonds; see chapter 9) to computer art in the first place.

Thus far, he has not succeeded in doing this. To the connoisseur of C-art, Brown’s work is still instantly recognizable as his. Simply using computers (usually to generate cellular automata) has not enabled him to bypass his own psychology. So he recently decided to set up a project in Evo-art in which—because of the random mutations—the results would not be entirely under his control. Specifically, he hoped to evolve pen-carrying robots to draw aesthetically acceptable marks, drawings that would not betray his authorship as the originating artist.

This example counts as R-art, wherein robots are used for artistic purposes. As explained in chapter 2, however, it is an unusual case. The aesthetic interest here is not in the robots themselves but in the drawings made by them.

Of course, he would be the arbiter and designer of the (automatic) fitness function. So it might turn out that even these robot-drawn marks would betray Brown’s idiosyncratic style. Then again, it might not. The more that our aesthetic preferences are rooted in very low-level visual features, perhaps even in biologically inherited perceptual mechanisms, the more likely that culturally and personally idiosyncratic preferences are not necessary for positive aesthetic valuation. The extent to which this is so is an empirical question, whose answer is unknown (Boden 2010).

The reason Brown used robots, instead of evolving images in cyberspace, was not mere gimmickry or a fascination with intriguing gizmos. Rather, it is the issue explored in section 5.3. The robots would be physical things moving in a physical world and therefore open to factors such as friction (between pen and paper or wheels and floor), temperature (affecting the density of the ink on the pen), and obstacles (causing the robot to stop or to move away from or around the impediment). Thus, seemingly irrelevant physical contingencies (such as a soldering iron’s being plugged in) might become relevant, by producing features in the drawings that were then selected by the fitness function. With luck, this openness to external influences might result in a radically transformed style, wherein—at last—Brown’s personal aesthetic preferences were not identifiable.

Thus far, Brown’s R-art project has not progressed long enough to see whether he can succeed in losing the mark of his own hand in the robots’ drawings. (Early reports on the work are given in Bird, Bigge, et al. 2005; Bird, Husbands, et al. 2008; Bird and Stokes 2006a, 2006b, 2007; Brown et al. 2007; Stokes and Bird 2008.) For present purposes, however, Brown’s (modernist) self-negating aim is irrelevant. More to the point is whether the evolved robots will be able to draw aesthetically acceptable marks. If they can, Brown would have shown that an automatic fitness function can sometimes suffice for artistic success. Even more to the point, he would have shown that third-reason Evo-art can—as implied by section 5.3—be achieved by focusing on physical openness.

Now, let us turn to the implications of section 5.4. The first thing to note is that we do not need to be overly literal in our interpretation of “interpretation.” Just as we need not ask whether computers can really be creative, so we need not ask whether they can really interpret representations as having meaning. Indeed, the latter question concerns one of the highly controversial philosophical issues that, as remarked in chapter 1, underlie the former question. For our present purposes, the mere appearance of semantic interpretation on the part of the computer will suffice.

A C-art program that was open to semantically interpretable or aesthetically significant representations from outside might take them directly from some human being. That is what happens in interactive computer art (CI-art, in chapter 2 terminology) including virtually all Evo-art, in which the selection at each generation is done by artist or audience. So the objection given at the end of section 5.2 is misguided. Any CI-art program is, by definition, an open system. Although the CI-artworks it generates are indeed limited by it, they do not result from the program alone.

But that does not address the basic worry here, which is that an unaided C-art program could never generate radical stylistic transformations. What are we to say about that?

A C-art program that was open to outside (semantic) influences might be able, for instance, to trawl the web, maybe even searching for texts, music, or images that could turn out to be helpful for the task at hand. Some current (nonevolutionary) computer art incorporates items taken randomly from the Internet so as to increase the unpredictability of the artwork. Examples include Christa Sommerer and Laurent Mignonneau’s installation The Living Room (see chapter 2).

Future C-art programs might also be able to communicate with other programs carrying out broadly comparable tasks but with a different educational or cultural experience. In addition, to satisfy the implication regarding internally generated representations (thoughts) that are prima facie irrelevant, the C-art program might spend its downtime in generating novel representations in a relatively unconstrained manner for possible use in as-yet-unforeseen creative tasks. Similarly, future versions of the machine-discovery programs described in chapter 3 might, as suggested long ago by the AI researcher Herb Simon, be able to read and learn from scientific papers, as human scientists can (Langley et al. 1987).

This is a tall order. If C-art programs are ever to respond fruitfully—and sometimes transformationally—to serendipitous thoughts rather than predetermined keywords, or to accidentally encountered musical structures or techniques rather than canned note sequences, or to meaningful images rather than simple lines and curves, huge progress will be needed in AI. Specifically, this will require significant advances in research on associative memories, recognition of analogy, natural-language processing, computer vision, musicology and comparable domain-specific studies, and knowledge representation in general.

It is not at all certain that this will be feasible in practice, except in toy and largely preordained examples. Some of the problems are mentioned in chapter 4 (see also Boden 2006, 7.v, 9.x–xi, 10, 12, 13). In fact, there are very good reasons for doubting whether AI will ever attain the full richness and subtlety of human thought and therefore of human creativity (Boden 2006, 7.iii.d, 9.x.e). But it is not impossible in principle. This claim assumes that human thought is a matter of complex computation, or information processing (for a host of arguments and evidence to this effect, see Boden 2006).

It follows that radically transformational Evo-art is possible in principle, provided that its programs are (physically or semantically) open systems. But it also follows that success in third-reason Evo-art is highly improbable in the short run. It may even be unlikely in the long run. These caveats are especially apt with respect to noninteractive Evo-art, in which the fitness function is applied automatically by the program. In such cases, the artist must express his or her aesthetic preferences as a clearly defined fitness function—which is very much easier said than done.

5.6   Evo-values?

If third-reason Evo-art is ever achieved, it will have to face the problem that is inherent in all radically transformational creativity. Namely, if the change is too radical, it may be rejected outright.

By definition, transformational creativity breaks existing stylistic rules. That transgression has to be accepted, and even appreciated, if the new style is to be valued at all. If it is not, the change will be seen as chaotic, or at best incompetent, rather than creative.

These matters are difficult enough in the case of human-generated art. Even the community of practicing artists, never mind the commercially influenced art world (see chapter 8), may reject the new creation as valueless. The transformative Demoiselles d’Avignon, for example, was scorned even by Picasso’s artist friends and languished unseen in his studio for several years.

The idea that an Evo-art program, written prior to 1907, might have produced (a cruder version of) the Demoiselles, perhaps after importing some African images from the (anachronistic) Internet, is not utterly absurd. But the idea that its fitness function, programmed to reflect the aesthetic preferences typical of artists (whether avant-garde or Impressionists or Academicians) before 1907, would have allowed the new image to breed future generations is surely fanciful.

For that to happen, a higher-level fitness function would have to be defined that could recognize, and favor, specific continuities and contrasts between the old style and the new one. It is stylistic continuities and contrasts—among other things (such as highly topical cultural themes, the bombing of civilians in war, for example)—to which art critics refer when they try to persuade doubters that the new style, although shocking, is actually valuable.

Again, this is a tall order. Structural continuities and contrasts (including the brushwork employed in applying the paint; see chapter 4) are difficult to identify and still more difficult to define clearly enough for expression in an AI program. As for topical themes, these are a conceptual quagmire. What noncheating fitness function could lead an Evo-art program to select a neo-Guernica partly because of its successful depiction of the horrors of war? Even a cheating fitness function, written with the political theme of potential neo-Guernicas explicitly in mind, is barely conceivable.

Perhaps we shall always need human artists, and art critics, to persuade the public to value the new styles. Although, to be sure, a mere increase in familiarity can help. Whether an Evo-art program could ever take on, or even assist, that delicate task is an intriguing question. But to quote from the skeptic imagined in section 5.1, “Don’t hold your breath!”

References

Adams, H. 1918. “The Dynamo and the Virgin.” In H. Adams, The Education of Henry Adams, 379–390. Boston: Houghton and Mifflin.

Ashby, W. R. 1952. “Can a Mechanical Chess-Player Outplay Its Designer?” British Journal for the Philosophy of Science 3 (9): 44–57.

Bedau, M. A., J. S. McCaskill, N. H. Packard, S. Rasmussen, C. Adami, D. G. Green, T. Ikegami, K. Kaneko, and T. S. Ray. 2000. “Open Problems in Artificial Life.” Artificial Life 6:363–376.

Bertalanffy, L. von. 1950. “An Outline of General Systems Theory.” British Journal for the Philosophy of Science 1:134–165.

Bertalanffy, L. von. 1962. Modern Theories of Development: An Introduction to General Systems Theory. New York: Harper.

Bird, J., B. Bigge, M. Blow, R. Brown, E. Clive, R. Easton, T. Grimsey, G. Haslett, and A. Webster. 2005. “There Does Not, in Fact, Appear to Be a Plan: A Collaborative Experiment in Creative Robotics.” In Proceedings of the Symposium on Robotics, Mechatronics and Animatronics in the Creative and Entertainment Industries and Arts, edited by T. Hirst and A. Green, 58–65. Society for the Study of Artificial Intelligence and the Simulation of Behaviour Hatfield, UK: Study of Artificial Intelligence and the Simulation of Behavior (AISB).

Bird, J., P. Husbands, M. Peris, B. Bigge, and P. Brown. 2008. “Implicit Fitness Functions for Evolving a Drawing Robot.” In Applications of Evolutionary Computing: EvoWorkshops 2008, edited by M. Giacombini, A. Brabazon, S. Cagnoni, G. Di Caro, R. Drechsler, A. Ekart, A. Esparcia-Alcazar, M. Farooq, A. Fink, J. McCormack, M. O’Neill, J. Romero, F. Rothlauf, G. Squillero, S. Uyar, and S. Yang, 473–478. New York: Springer.

Bird, J., and P. Layzell. 2002. “The Evolved Radio and Its Implications for Modelling the Evolution of Novel Sensors.” Proceedings of Congress on Evolutionary Computation CEC-2002, May, Institute of Electrical and Electronic Engineering (IEEE), 1836–1841.

Bird, J., and D. Stokes. 2006a. “Evolving Fractal Drawings.” In Generative Art 2006: Proceedings of the 9th International Conference, edited by C. Soddu, 317–327. Milan.

Bird, J., and D. Stokes. 2006b. “Evolving Minimally Creative Robots.” In Proceedings of the 3rd International Joint Workshop on Computational Creativity (ECAI ‘06), edited by S. Colton and A. Pewase, 1–5. Riva del Garda.

Bird, J., and D. Stokes. 2007. “Minimal Creativity, Evaluation and Pattern Discrimination.” In Proceedings of the 4th International Joint Conference on Computational Creativity, edited by A. Cardoso and G. Wiggins, 121–128.

Boden, M. A. 1999. “Is Metabolism Necessary?” British Journal for the Philosophy of Science 50: 231–248. Reprinted in M. A. Boden, Creativity and Art: Three Roads to Surprise, 235–254. Oxford: Oxford University Press, 2010.

Boden, M. A. 2004. The Creative Mind: Myths and Mechanisms. 2nd ed. London: Routledge.

Boden, M. A. 2006. Mind as Machine: A History of Cognitive Science. 2 vols. Oxford: Clarendon/Oxford University Press.

Boden, M. A. 2010. “Personal Signatures in Art.” In Creativity and Art: Three Roads to Surprise, 92–124. Oxford: Oxford University Press.

Boden, M. A. 2014. “Artificial Intelligence and Creativity: A Contradiction in Terms?’ In The Philosophy of Creativity, edited by E. S. Paul and S. B. Kaufman, 224–246. Oxford: Oxford University Press.

Brown, P., B. Bigge, J. Bird, P. Husbands, M. Perris, and D. Stokes. 2007. “The Drawbots.” Proceedings of Mutamorphosis 2007: Challenging Arts and Sciences, November, Prague.

Cariani, P. 1992. “Emergence and Artificial Life.” In Artificial Life II, edited by C. G. Langton, C. Taylor, J. D. Farmer, and S. Rasmussen, 775–797. Redwood City, CA: Addison-Wesley.

Cariani, P. 1993. “To Evolve an Ear: Epistemological Implications of Gordon Pask’s Electrochemical Devices.” Systems Research 10 (3): 19–33.

Cariani, P. 1997. “Emergence of New Signal-Primitives in Neural Systems.” Intellectica 2:95–143.

Cronly-Dillon, J. R. 1991. “The Origin of Invertebrate and Vertebrate Eyes.” In Evolution of the Eye and Visual System, edited by J. R. Cronly-Dillon and R. L. Gregory, 15–51. Basingstoke, UK: Macmillan.

Eldredge, N., and S. G. Gould. 1972. “Punctuated Equilibria: An Alternative to Phyletic Gradualism.” In Models in Paleobiology, edited by T. J. M. Schopf, 82–115. San Francisco: Freeman Cooper.

Fernandez, M. 2008. “Aesthetically Potent Environments, or How Gordon Pask Detourned Instrumental Cybernetics.” In White Heat Cold Logic: British Computer Art 1960–1980, edited by P. Brown, C. Gere, N. Lambert, and C. Mason, 53–70. London: MIT Press.

Harrison, A. 1978, Making and Thinking: A Study of Intelligent Activities Hassocks, UK: Harvester Press.

Harvey, I. 1992. “Species Adaptation Genetic Algorithms: A Basis for a Continuing SAGA.” In Toward a Practice of Autonomous Systems: Proceedings of the First European Conference on Artificial Life, edited by F. J. Varela and P. Bourgine, 346–354. Cambridge, MA: MIT Press.

Harvey, I., P. Husbands, and D. Cliff. 1994. “Seeing the Light: Artificial Evolution, Real Vision.” In From Animals to Animats 3: Proceedings of the Third International Conference on Simulation of Adaptive Behavior, edited by D. Cliff, P. Husbands, J.-A. Meyer, and S. W. Wilson, 392–401. Cambridge, MA: MIT Press.

Husbands, P., I. Harvey, and D. Cliff. 1995. “Circle in the Round: State Space Attractors for Evolved Sighted Robots.” Journal of Robotics and Autonomous Systems 15:83–106.

Land, M. F., and D.-E. Nilsson. 2001. Animal Eyes. Oxford: Oxford University Press.

Langley, P. W., H. A. Simon, G. L. Bradshaw, and J. M. Zytkow. 1987. Scientific Discovery: Computational Explorations of the Creative Process. Cambridge, MA: MIT Press.

McCorduck, P. 1991. AARON’s Code: Meta-Art, Artificial Intelligence, and the Work of Harold Cohen. New York: W. H. Freeman.

Pask, G. 1959. “Physical Analogues to the Growth of a Concept.” In The Mechanization of Thought Processes, edited by D. V. Blake and A. M. Uttley, 2:877–922. London: Her Majesty’s Stationery Office.

Rasmussen, S., L. Chen, M. Nilsson, and S. Abe. 2003. “Bridging Nonliving and Living Matter.” Artificial Life 9:269–316.

Reichardt, J., ed. 1968. Cybernetic Serendipity: The Computer and the Arts. London: Studio International.

Reichardt, J. 2008. “In the Beginning. …” In White Heat Cold Logic: British Computer Art 1960–1980, edited by P. Brown, C. Gere, N. Lambert, and C. Mason, 71–82. London: MIT Press.

Romero, J., and P. Machado, eds. 2007. The Art of Artificial Evolution: A Handbook on Evolutionary Art and Music. New York: Springer.

Sacks, O. 1985. The Man Who Mistook His Wife for a Hat: And Other Clinical Tales. New York: Simon and Schuster.

Sims, K. 1991. “Artificial Evolution for Computer Graphics.” Computer Graphics 25 (4): 319–328.

Stokes, D., and J. Bird. 2008. “Evolutionary Robotics and Creative Constraints.” In Beyond the Brain: Embodied, Situated and Distributed Cognition, edited by B. Hardy-Valle and N. Payette, 227–245. Newcastle, UK: Cambridge Scholars.

Thompson, A. 1995. “Evolving Electronic Robot Controllers That Exploit Hardware Resources.” In Advances in Artificial Life: Proceedings of the Third European Conference Artificial Life, edited by F. Moran, A. Moreno, J. J. Merelo, and P. Chacon, 640–656. Berlin: Springer.

Todd, S. C., and W. Latham. 1992. Evolutionary Art and Computers. London: Academic Press.

Whitelaw, M. 2004. Metacreation: Art and Artificial Life. Cambridge, MA: MIT Press.