Skeuomorphs and Cultural Algorithms

Nicholas Gessler

UCLA Anthropology, Computational Evolution and Ecology Group
c/o 11152 Lucerne Avenue, Culver City, CA 90230-4244
gessler@ucla.edu

Abstract

Skeuomorphs are material metaphors instantiated through our technologies in artifacts. They provide us with familiar cues to an unfamiliar domain, sometimes lighting our paths, sometimes leading us astray. Our computational methods are rich with many of these structures resurrected from past practices and borrowed from analogies in non-computational domains. Failing a general theory of how evolutionary processes relate emergently with one another across hierarchical levels through the media of DNA, culture, and technology, we should be curious yet cautious when we remove them from their natural contexts to use them in the artificial world of silicon.

1 Skeumorphs are Material Metaphors

People are stirring. The session must be almost over. With any luck you’ll beat the crowds to the coffee and muffin bar. You reach towards your right front hip, slide your thumb under the gold chain, pull up and slip the watch from its pocket. Cupped in the palm of your hand it reads a quarter ‘til… "Cut!… No, no, no!…" "Everyone wears a wristwatch these days!" Sure enough, they do. But why do they still sew that pocket onto their jeans? For decoration, I suppose. But what about that one large button? It has the look of a rivet, but look more closely. The rivet’s head is covered by a thin metal cap. It hides the real rivet underneath, but in its place it presents instead the embossed image of a rivet head splayed through the center of a metal washer emblazoned with the Levi-Strauss logo.

The marketplace is full of such things. Hard plastic castings complete with screw heads and unnecessary parts. Soft plastic with molded-in textures of warp and woof and even stitching. Metered postage showing the bold anachronistic elements of the circular town mark and wavy cancellation. Copper cladding over zinc pennies. Steel cladding over copper quarters. All are efforts to make the new look comfortably old and familiar, or simply habits too deeply engrained to wash away.

Archaeologists, who make an academic practice of studying material culture change, while not the first to notice that yesterday’s functional features become today’s stylistic decorations, were at least the first to name the phenomenon:

Skeuomorph - An ornament or ornamental design due to structure… 1889: "The transfer of thong-work from the flint axe, where it was functional, to the bronze celt, where it was skeuomorphic." [1]

Skeuomorphs are material metaphors. They are informational attributes of artifacts which help us find a path through unfamiliar territory. They help us map the new onto an existing cognitive structure, and in so doing, give us a starting point from which we may evolve additional alternative solutions. They provide us with "a path" instead of "no path" at all, but as scientists we are ultimately interested in an optimal paths well suited to the problem at hand, if not simply the best solution possible. How many of the metaphors we use in evolutionary computing become skeuomorphic downloaded into our machines?

2 The World Through Pleistocene Brains

We try to understand the world through Pleistocene brains, brains not wired for science, but built to confer fitness to their owners struggling with the physical and social environments of a hunter-gatherer way of life. The scientific enterprise is a rather new development and a truly remarkable achievement. In its disciplines where truth and honesty are so highly prized that any hint of deviation from a full disclosure of methods and results brings rapid censure, the mere mention of the word "deception" arouses deep suspicion. But skeuomorphs, despite their many useful traits, may also constitute a special class of self-deception. If science is in any way the winnowing of the greater from the lesser "truths" of nature, then a closer look at deception and belief may generate that breeze which helps us in our work.

3 Spinning Out Explanations

Ian Rowland is a specialist in the more benign facets of deception and psychological game-playing which he demonstrates professionally to expose fakery, to enhance critical skills, and simply to provide entertainment. When asked recently what it is that constructs belief, his response was that (scientific) evidence had nothing at all to do with it:

People adopt the most convenient set of beliefs consistent with their needs, hopes and fears at the time. [2]

Consistency with one’s preconceptions and expectations are powerful forces acting to bolster one’s ideas. They lead to substantial economies of thought, but on occasion the price of this economy is high. Again, we scientists, through intensive investigations, become experts at teasing apart fact from fancy in our own specialized domains. Concurrently, our Pleistocene legacy continues to spin out its own explanations of our experiences. Added to this is the fact that our perceptions are further colored by the belief systems or constructions of our own culture. The result is that we carry in our heads an uneasy society of minds, all working on the problem we call science. Informed as we are, we are still not immune to error. This is the struggle of the skeptic, in Rowland’s words:

Being 'a skeptic' is simply about preferring to get things right rather than wrong. It is about being aware of the 1001 ways in which people (including skeptics!) can be taken in, or make mistakes, or delude themselves. It's about learning to ask good questions, and assess evidence intelligently, to get to the truth about a given claim or hypothesis. [3]

These constitute ways-of-thinking that are distributed as a population in our heads, a population of heads and technologies ranging across our discipline, and a population of disciplines within our distinctive culture, which is itself only one culture among many. Each way-of-thinking has its own logic, its own rules of evidence, and its own epistemology. If science is the task of building increasingly reliable representations of the world around us, then we must be constantly vigilant against confusing the world as it actually is with the world as we experience it. We tend to fashion objects skeuomorphically. Once thought is given material substance, it is not always clear what is a skeuomorph and what is not. The cover on the Levi’s button simulates a rivet, but underneath the simulation is an actual rivet. The skeuomorph here does not replace function, but merely simulates the referent, in a mirrored hall of simulations simulating their own simulations.

Not long ago, the word computer meant a person skilled in making calculations. Something changed with the proliferation of the Turing machine; such devices came to be called computers in metaphoric reference to their human progenitors. As these machines became household appliances, the associational reference to a profession ceased -- the metaphor died along with the job title. The computer became the object itself and those who did the computations became programmers. Today we question whether people really are computers in some sense. Today to talk about a human as a computer invokes the metaphor of the machine, an image that fans the flames of controversy in many crowds. In some sense we have come full circle, resurrecting the original metaphoric association of human as computer. In another sense, what it means to be a human computer has changed radically before and after the machine. The metaphoric, semantic and referential networks are configured in a constant state of flux. It is easy to lose oneself in crossed meanings.

4 Skeuomorphs in Evolutionary Computing

Evolutionary computing is now used as an umbrella term to cover more specific variations such as evolutionary programming, genetic algorithms, and genetic programming, as well as others. I have gone through the foregoing critique of skeuomorphs, because it is clear that they play a large role in the way we constitute our practice. We have learning systems based upon neural nets, which to some extent model neurons. We have evolutionary systems based upon genetic algorithms and genetic programming, which to some extent model genetics. We have evolving cultural systems based upon a hybrid of genetics and shared memory. Failing a comprehensive theory explaining why these systems should operate in silicon in the same way as they do in carbon, we have a mixed metaphor, a scrambled skeuomorphic system par excellence.

If we had wanted to develop reliable models of real neurons, we might have gone in the direction of Gerald Edelman’s neuronal group selection. If we had wanted to develop reliable models of genotype-phenotype interaction, we might have gone in the direction of protein folding. However, for our purposes, we want general working models of wider application, or failing that at least clear sets of models more suited to one purpose than another. At this juncture we are more interested in optimization than in simulation.

In the branch of evolutionary computation specifically interested in general problem solving and optimization, it is not clear why we have embraced the precise evolutionary models we currently use. We seem to be of two minds, simultaneously exploring biological processes for ideas and then extending them in non-biological realms. Genetic algorithms have been a successful hybrid of biology and machine. Moreover, they conjure up the cachet of being founded on real biology. David Fogel has repeatedly made the case that the phenotypic algorithms of evolutionary programming may be more effective than genetic algorithms for certain problem classes, even though the former lack the hallmark of biological fidelity [4]. After all, why should biological fidelity be an issue in a non-biological, silicon world? This push and pull between real biology and fanciful experiment is a healthy dynamic, since by challenging the universality of processes taking place in carbon-based life, we explore not only that which is unique about natural life, but concurrently the space of possibilities in artificial life occupied by evolutionary computation. Our agent populations already deviate quite markedly from normative (human) behaviors: they are sexual hermaphrodites. They transcend great distances on fitness landscapes to mate with others of similar accomplishment and to spawn descendants in places neither parent has visited. It is refreshing to see further bendings of the rules, like the comedic seriousness with which Gusz Eiben [5] has experimented with agent "orgiastic" sexual recombination. By this he meant sexual recombination from more than two parents. In humans this will likely be among the first kinds of gene therapy practiced. Diseases inherited from mitochondrial DNA may be obviated by replacing the cytoplasm of the "contaminated" egg with that from another mother or by transplanting the nucleus to a second mother’s egg. The result would be three genetic parents: two mothers and one father. The natural world should be mined for ideas and inspirations, but these need to be tested in the artificial worlds in which we work because they offer different constraints and possibilities.

5 Spinning up Culture

The turn of the new millennium coincides not solely with the arbitrary second cube of our base-ten numbering system; it marks a specific and significant change in our adaptation to our social and physical surroundings. From an archaeologist’s perspective, we have spent some 2,000,000 years as tool-makers, fashioning artifacts to leverage matter and energy exchanges. For 50,000 years we have created diagrammatic representations and art, originating artifacts to leverage information. We discovered evolution through natural selection a mere 150 years ago and built our first thinking machines within the last 50 years. In a wink of recent history, we have discovered our creator, evolution through natural selection. We have captured our creator and embodied this creative force in our technologies. Just as the "modern synthesis" in evolutionary theory united natural selection with genetics, what might be called the "postmodern synthesis" has united evolution with technology through computation. On the brink of the 21^st Century, we live in volatile times.

Decision rendering under duress is a process which has been studied in great detail, particularly in very large scale synthetic environment simulations developed for the military… A threshold will soon be crossed when the strategic deployment of "sentient entities" will become an [imperative], rather than an option, for commercial viability in a competitive marketplace. [6]

It is a truism in anthropology that the onset of culture accelerated human adaptation. In the traditional dualism between biology and culture, it is often argued that our biological evolution has effectively stopped, having been predominantly replaced by cultural evolution. After all, biological information, coded in DNA, reproduces only once every 15 years or so while discursive cultural information, coded in neurons, reproduces maybe once a second, about as fast as we can think, some 473 megafold faster. Culture is surely a wonderful thing with which to solve (and generate) problems. But technology is still more wonderful. We can extend this comparison of reproductive rates to culturally derived technological information, now coded in silicon. A PC, for example, reproduces information some 300 megafold faster than thought, or some 142 kiloterafold faster than the courtship for a mate with DNA. Measured in "reproductive events per second," while biological evolution has spun us from 2.14 nanoreps with DNA to 1 reps in thoughtful contemplation, our culture and technology have spun us up a gigaterafold advantage with teraflops computers. Technology has upped biology’s pace one-thousandfold in one one-hundred-millionth of the time. Sex may be more immersive, culture may be more inspirational, but technology has left them in the dust (at least as far as reproduction is concerned).

6 Spinning up Technology

There are some problems inherent in this double dualism between biology and culture, and again between culture (as ideas) and technology. Not the least of these difficulties is the growing realization that this "progress" relies upon "evolution all the way up" as Terence Deacon [7] is fond of reminding us. The genome is not a blueprint for the body; rather it is a beginning set of processes which engage each other and their environments in a developmental process which produces emergences which build one upon the other. The same nesting of emergences could profitably be argued for culture [8]. Biology and culture are further tied together through Baldwinian evolution, in which genetic change effects behavior which in turn alters the environment and thus modifies gene fitness. Biology and culture comprise a dynamic interactive process. The same holds true for the dynamic between humans and their artifacts. In an extension of Baldwin’s argument to technology, the behavioral practices creating new industries change our environments both socially and materially. They change not only how we think and act, but in so doing also change the selective pressures on our genes. The distinctions between biology, culture and technology are consequential, but their interlinkage is more profound than we might care to imagine.

One lesson often drawn from this trend is that evolution has accelerated as we shifted from biology, to culture, and finally to technology. This begs a provocative question: Wouldn’t evolutionary computation be similarly enhanced by launching our models along the same trajectory from biologically, to culturally, to technologically inspired modeling? Perhaps so, but there seems to be no a priori reason why the constraints and advantages of the natural world should map to the artificial world in the same ways. The "natural superiority" of cultural over genetic methods cannot simply be assumed for silicon. It must be demonstrated.

A recent paper at the Genetic Programming [9] conference illustrates the confusion that is often caused by removing assumptions from their original contexts and applying them where they may no longer be valid. This work attempted to demonstrate the efficacy of a culturally inspired model by using a measure for computational efficiency (i.e. computational effort) that was designed only to measure non-cultural activity. The measure scored the cultural methods inordinately high, simply because it turned a blind eye to the computational cost of all the cultural exchanges that were going on. In other words, the evaluative measure tallied up the costs of genetic exchanges and let cultural transmissions run for free. The attempt to demonstrate the superiority of culture methods was commendable. It was the measure that was flawed.

One priority should be to generate commensurate comparative data on the computing resources used to arrive at a solution for a suite of methods applied across a variety of problems. Such a plan could produce a method-problem-fitness space which we could use to begin to address both the engineers’ desire to find an optimal solution to a design problem and the scientists’ desire to understand the similarities and differences between evolutionary methods in the natural and artificial worlds. It would add to our cooperative understanding of the general question of why we choose the metaphors and skeumorphs in evolutionary programming that we do. It would help us fit the methods and the problems to the media in which we work, in this instance in both carbon and silicon.

7 The Medium Makes the Difference

The medium makes the difference. In the natural world, DNA and genes are molecules which have to wait for their surrounding tissues to mature and find a mate in accordance with some culturally established marriage rules. In silicon, these genes do not have to wait. In the natural world, culture as ideas requires a brain constructed from the emergent interaction of genes with their environment. In silicon, these ideas don’t need their organic substrate. In other words, both genes and culture, when embodied in machines in the abbreviated way we are using them in evolutionary computing, use the same medium of reproduction, namely silicon oscillating at teraflop speeds. In the natural world their speeds were constrained by the two different media they inhabited. In the artificial world their speeds are dictated by only one shared medium -- the integrated circuit. In our quest for optimization, we’ve warped the speed of genetic operators to match that of cultural operators. We have totally changed the context of our argument that "culture is faster than genetics." Why should we expect the same argument to hold when we have leveled the playing field, when we’ve normalized the competition? Why should culture outstrip genetics when both run at the same speed?

If we had been intent on simulation rather than optimization, we should expect the differences to hold in both worlds. If we had accurately simulated the relationship between genes and culture in silicon by including all the levels of emergence in between, then we should expect to see the differences maintained. This, however, was not our purpose.

8 Cultural Algorithms

A caution, however, is not a prohibition. Much the same caution could be raised (and has) against an unexamined use of genetic methods. Cultural methods may well prove advantageous. Their domain is of interest to both the scientist and engineer. In this respect, it is particularly rewarding to see Bob Reynolds’ work on cultural algorithms [10]. He has conscientiously measured his work against the computing resources used.

Since cultural algorithms beg the same questions concerning their relationships to their natural namesakes as genetic algorithms do, it might be enlightening to examine some of the similarities and differences between these artificial cultures and their natural cultural referents as an anthropologist or archaeologist may see them.

"Culture" is one of those handy words which can mean almost anything, or for the same reason, almost nothing. It’s a wild-card or place-holder for a myriad of meanings, one or more of which may be invoked somewhere down the road, implied from the context of the discourse, or simply left to the imagination. Culture to the public is the "arts." Culture, to Leslie White, is "man’s extrasomatic means of adaptation," and so by definition, is that which biology is not. This dualism has strong roots in our own culture, in the nature/nurture controversy and in arguments over genetic determinism and free will. To the anthropologist concerned with ways of thinking, culture is a body of "shared beliefs." But what remains if we remove everything from everybody which is not held in common? Shared beliefs, and therefore culture, by this definition. But what sort of culture would this be with no individual differences? Homogenized and flat, dimensionless for sure? A more dynamic path to adaptation might be to entertain the patterns of thoughts and deeds and things distributed throughout the population. To the anthropologist with more holistic interests, culture includes thoughts, behaviors and artifacts as well, whether they are shared or not.

In cultural algorithms, the belief space constitutes the canonical shared beliefs of anthropology. But it does so with a uniformity found rarely in the pluralistic natural world. In the cultural algorithm, the shared belief space is the foundation on which the efficiency of the society depends. Real world cultures only approximate this efficiency through the structures of authoritarian hierarchies. In many ways, the belief space is more like the fictional worlds envisioned among the Borg, the efficient collective consciousness of the cyborg inhabitants of the Star Trek universe. It echoes the psychic unity and power of the denizens of the cinematic worlds of City of Lost Children and Dark City. Not intended as a criticism of cultural algorithms, these comparisons merely highlight the usefulness for optimization of approaches now censured in the natural world. They seem to corroborate the notion that natural cultures do not tend towards equilibria, but rather towards dynamic coevolutionary stand-offs. Individual humans rarely wish to give up their freedoms for a common cooperative cause. But within the domain of optimization where cultural algorithms are implemented, the lives of its individual agents are justifiably subordinated to finding the best solution. This social structure, which has been tried and failed so many times in natural human history, has a valued place in the artificial world of computing.

9 Technological Algorithms

The use of a belief space also raises questions about its cultural status. Is culture simply information or an object? In the natural world, the transmission and inculcation of beliefs is often achieved technologically through artifacts. Hashishans instilled a common purpose in their initiates with drugs, the church through cathedrals, the polity through rallies and demonstrations. In contrast, the belief space is more reminiscent of recent innovations though print media, broadcast radio and television. Indeed, the belief space could be taken as a metaphor for the Internet itself. It has the same single point dynamic. It is equilocal to every user.

I would suggest that Reynolds’ cultural algorithms are already moving beyond culture in the narrow sense of an unmediated common pool of information and ideas. In accordance with a broader, more productive view of culture, one which includes technology and artifacts as part of an intelligent distributed cognitive system, he is exploring a computational space of value to both the goal of optimization, and its flip side, simulation.

Conclusion

Computer science is still in its infancy. The days when computers were people calculating figures by hand, or by mechanical and electromechanical devices, are still within our memories. As individuals, creating and being recreated by this unprecedented technology, the changes we have witnessed were barely imaginable. The highways unfolding before us are uncharted. The milestones, if ever they were clearly set, have been removed. The routes and accesses have all been changed. Intersections lie ahead for which we have no signposts, lights or warnings. We cling to the maps we have at hand. We are only beginners poking and prodding our creations, not quite seeing them for what they are, and always unconsciously trying to understand them in terms of the old and familiar.

In like manner a beginner who has learnt a new language always translates it back into his mother tongue, but he has assimilated the spirit of the new language and can freely express himself in it only when he finds his way in it without recalling the old and forgets his native tongue in the use of the new. [11]

References

[1] The Compact Edition of the Oxford English Dictionary. Oxford University Press, 1971. Volume II, page 4064.

[2] Rowland, Ian. Personal communication, 28 March 1998.

[3] Rowland, Ian. http://www.irowland.demon.co.uk/faq.htm, April 20, 1998.

[4] Fogel, David. "On the Philosophical Differences between Evolutionary Algorithms and Genetic Algorithms." In Proceedings of the Second Annual Conference on Evolutionary Programming. The Evolutionary Programming Society, San Diego, 1993. Pages 23-29.

[5] Eiben, Gusz. Personal communication, San Diego, 27 March 1998.

[6] Ostman, Charles. "Synthetic Sentience as a Strategic Commodity Resource." In press, 1998. Page 10.

[7] Deacon, Terrence. The Symbolic Species – The Co-evolution of Language and the Brain. WW Norton and Company, New York, 1997.

[8] Gessler, Nicholas. "Book Review: Growing Artificial Societies by Joshua Epstein and Robert Axtell, MIT Press, Cambridge 1996." Artificial Life, 1997. Volume 3, number 3, pages 237-242.

[9] Spector, Lee and Sean Luke. "Cultural Transmission of Information in Genetic Programming." In John Koza, et al, editors, Genetic Programming 1996: Proceedings of the First Annual Conference. MIT Press, Cambridge, 1996. Pages 209-214.

[10] Reynolds, Robert, et al. "Using Cultural Algorithms for Constraint Handling in GENOCOP." In John R. McDonnell, et al, Proceedings of the Fourth Annual Conference on Evolutionary Programming. MIT Press, Cambridge, 1995. Pages 289-305.

[11] Marx, Karl. "The Eighteenth Brumaire of Louis Bonaparte." 1852. In Karl Marx and Frederick Engels - Selected Works. Progress Publishers, Moscow 1966. Volume 1, page 398.