Given: The Universe. Shown: Every Artwork.

Essay for Deep Storage, Haus der Kunst, Munich, Ingrid Schaffner and Matthias Winzen, curators, 1997.

Jon Ippolito

The universe (which others call the Library) is composed of an indefinite, perhaps an infinite, number of hexagonal galleries, with enormous ventilation shafts in the middle, encircled by low railings. From any hexagon the upper or lower stories are visible, interminably. The distribution of the galleries is invariable. Twenty shelves--five long shelves per side--cover all sides except two. One of the free sides gives upon a narrow entranceway, which leads to another gallery, identical to the first and to all the others.1 ...[I]ts shelves contain all the possible combinations of the twenty-odd orthographic symbols (whose number, though vast, is not infinite); that is, everything which can be expressed, in all languages. Everything is there: the minute history of the future, the autobiographies of the archangels, the faithful catalogue of the Library, thousands and thousands of false catalogues, a demonstration of the fallacy of these catalogues, a demonstration of the fallacy of the true catalogue, the Gnostic gospel of Basilides, the commentary on this gospel, the commentary on the commentary of this gospel, the veridical account of your death, a version of each book in all languages, the interpolations of every book in all books.2

--Jorge Luis Borges, "The Library of Babel," 1941.

If Borge's mythical library contains every possible book, the World Wide Web often seems to contain every possible page. 3 While the Web does not literally encompass every possible combination of words and images, it has grown to such unmanageable proportions that it often feels that way. To make matters worse, there is no Editor-in-Chief of the Web, no silver-throated museum director's accoustiguide to steer the uninitiated toward the Rembrandts and Flauberts and away from the scanned-in dog photos and uploaded jello recipes. Not surprisingly, the incoherent and bad artworks encountered in most Web surfs far outnumber the good ones: instead of an embarrassment of riches, you are left with simply an embarrassment. Without an editor or curator as intermediary, how do you cull the good stuff from the bad? What are the selection criteria?

What follows is a guided tour through this electronic Library of Babel, making stops at Web sites that embody different answers to these questions. After test-driving a few of the more conventional means of information delivery, the tour will try out some experimental models made possible by the extraordinary engines of automation available on the Internet.

1. Evaluative criteria
There are official searchers, inquisitors.4

Probably the most expedient way to locate a book in Borges' library, as in any other, would be to ask the librarian. The drawback of this approach, of course, is that different librarians have different tastes and will recommend different books. On the Web there are already scores of curated sites for visual art--sites such as ”da'web *(, the Thing ***(, and Talkback! ***( essentially function as librarians. The connoisseurs who program these sites point users to texts and images they deem worth browsing; simply adding a link on, say, ”da'web to an outside artist project confers upon that project an air of legitimacy.

Some web-savvy archivists, however, have created their own thematic sites rather than rely on someone else's official imprimatur. These archives document everything from disasters (Cati Laporta's deadpan Almanac of Disasters at * to celebrity deaths *( to bad art (the deliciously egregious Museum of Bad Art at * Many of these archives share the expressed purpose of collecting what would otherwise be lost to history. Artist Ben Kinmont's Web project We Both Belong ***(, for example, shows us his dirty dishes--and those of his collaborators. In past projects, Kinmont has polled Wall Street workers to solicit ideas for a sculpture, given away his paintings to people who signed a testament to this act of generosity, and washed dishes at the Wadsworth Atheneum in a public performance. Up until his Web project, the only residue of these attempts to draw ordinary people into the process of making art has been the artist's physical archives--signed statements, t-shirts, and other ephemera. For We Both Belong, Kinmont solicited photographs of people washing their dishes; in return, he offered participants a diptych of their photograph framed together with a photo of himself washing dishes. Kinmont then uploaded to his Web site the various images, along with the letters that accompanied them and sundry notes about the project (the "finances" page lists his rent and telephone expenses associated with the project), thus helping to realize his utopian ideal of distributing the art-making process to as broad an audience as possible.5

The documentation of what has hitherto remained concealed is the sole purpose of Antonio Muntades' File Room ***( The File Room documents cases of censorship, making it a sort of card uncatalogue, a list of what cannot be found in the library. A user can search the File Room by date, location, medium, or grounds for censorship (political, religious, or sexual). The over-500 cases cited range from the Oliver Cromwell's condemnation of John Milton's "Areopagitica, A Speech for the Liberty of Unlicence'd Printing" in London in 1644 to the debate over whether to ban Madonna's Sex book in the small town public library of Downer's Grove, Illinois in 1992.

Other Web archives, meanwhile, focus on specific artistic subcultures whose practices are unlikely to appear on the white walls of the Museum of Modern Art or the glossy pages of ArtNews. Susan Farrell's Web site devoted to graffiti, Art Crimes ***(, documents an ephemeral--and illegal--form of communication. In addition to featuring digitized photos of graffiti from around the world, the site's newest venture, The World Wide Wall, allows anyone with a Java-enabled browser to select a digital "surface"--White and Red Brick, Stucco, and Train are among the offerings--and spraypaint on them with the help of a mouse. Of course, the transgressive thrill you get from spraying a few pixels onto the screen of your personal computer hardly compares to the risk of being arrested or run over by a train taken by real graffiti artists. And the practice of tagging territory doesn't make much sense in a territoryless cyberscape. If there is a digital soulmate to the graffiti artist, it is not the casual Web surfer, but another virtuoso trespasser: the cypherpunk. Both the graffitti artist and the cypherpunk employ flamboyant noms-de-plume ("Futura 2000" v. "Lex Luthor"), a specialized vocabulary ("tagging" v. "phreaking"), and cryptic language for communicating with their peers (illegible handwriting v. the "Pretty Good Privacy" encryption algorithm).6

Another project to archive the work of an invisible culture is Jumping Lines *(, a Web site that documents examples of the art of the Maisin people of Papua New Guinea. Although the Maisin rainforest is about as far as you can get from the subway tunnels of the graffiti artists, the curvilinear patterns of the Maisin's tapa paintings bear an odd resemblance to Keith Haring's more abstract images.7 This being said, the systematic working method they employ is something of a cross between the Surrealists' exquisite corpse and Sol LeWitt's combinatorial line drawings. The Maisin fold a cloth made from pounded Mulberry bark in four sections so that only one of the four is visible. They then paint one section at a time, trying to reproduce the same pattern from memory on each. When the cloth is unfolded, the slight asymmetry of the overall image betrays the inexact memory of its maker, but at the same time it lends the composition a jaunty rhythm.

In the spirit of dedicated librarians, the creators of the above sites diligently catalogue their subjects, cross-referencing cases or works by geography, time period, and participants. Blast, on the other hand, is a collaborative archive that deliberately avoids a catalogue. Each edition of Blast takes two forms. First there is the physical "vehicle," ranging in shape from a finely crafted wooden box to a bean-shaped plastic pod, which contains artistic multiples created by its contributors. Second, there is a Web site that bills itself as a "conversional archive" *( The idea of packaging together artistic multiples in various media dates at least as far back as 1968, when William Copley published the SMS portfolios, featuring such curiosities as lunch menus by Claes Oldenburg, game cards by John Giorno, and phonograph records of Bernar Venet reciting astrophysical data. What distinguishes the Blast vehicle from SMS is not merely that it has a "conversional" electronic counterpart, but also that it subsumes the identities of individual contributors into an overriding theme such as "drama" or "bioinformatica." The rolled-up pieces of paper, miniature vials of liquid, or computer disks in the vehicles bear no names or titles to identify their authors. Meanwhile, the conversional archive Web page shows nothing more than a screen of colored orbs you can click on to access texts vaguely related to the collaborative venture. As its name suggests, Blast's disorganized presentation resembles a library that has exploded, leaving a field of atomized fragments for the user to poke through and try to identify.

2. Categorical criteria
In some shelf of some hexagon, men reasoned, there must exist a book which is the cipher and perfect compendium of all the rest.8

One problem with relying on a librarian to suggest which books to read is that the Library is too vast for any single librarian to be expert in the entire universe of books. Another is that a user's tastes may not concur with those of the librarian at hand. That's why libraries have card catalogues. Unfortunately, there is no single card catalogue for the Library of Babel, for no single librarian could possibly have ranged far enough through the hexagons to catalogue all of the books therein. Nevertheless, it is possible that individual librarians have composed different, perhaps only partly overlapping, catalogues based on the information they have gathered. On the Web these librarians have names like Lycos, Infoseek, Alta Vista, and Hot Bot. While none of these search engines by itself could encompass the entire Web, the larger ones claim to have catalogued tens of millions of Web pages. (There are even megasearch engines like Magellan or SavvySearch that query more than one search engine at a time--like a librarian who consults other librarian's card catalogues.9) Unfortunately, the scale of the Web is so huge that a user can only search by key words ("abstract" or "expressionism"), date restrictions (1945 < %date% <1954), or simple numeric calculations (%price% < $250). If you're looking for something less quantifiable, it's hard to do a focused search based on more nebulous criteria; a Lycos search on the keyword "gesture," for example, returns almost 10,000 entries.

While most of these search engines return verbal excerpts from pages matching the user's criteria, artist John Simon10 has pioneered a more schematic approach to cataloguing pages on the Web. For the Space of Information exhibition held at the Bannf Centre for Art and Technology in the spring of 1996 ***(ttp://, Simon created an "Archive Mapper" that plots each Web site developed for the exhibition as a different point graphed according to criteria selected by the user. For the graph's horizontal axis, the user can choose from objective variables such as file size or date. For the vertical, the user can choose from subjective variables Simon assigned in consultation with the exhibition's curator, Laura Trippi. Among these latter choices is the ratio of physical to digital presence; for example, Laura Kurgan's plotting of the Bannf environs using Global Positioning Satellites might rank higher on this scale than Sophie Tottie's Web pages showing comments about war superposed on images of city streets. Once the user has chosen the x- and y-axes, Archive Mapper draws a scattershot cluster of colored icons representing the Webspace; clicking on any icon brings the user to the page corresponding to that artist's project. Unlike the home pages of such evaluative sites as Talkback! and the Thing, Archive Mapper offers a nonhierarchic view of artist projects, since no artist is given stylistic emphasis in the design. Without a central list of links, no artist gets top billing over another.

3. Generative criteria
A blasphemous sect suggested that all searches be given up and that men everywhere shuffle letters and symbols until they succeeded in composing, by means of an improbable stroke of luck, the canonical books....[Men] would hide out in the privies for long periods of time, and, with metal disks in a forbidden dicebox, feebly mimic the divine disorder.11

Archive Mapper is egalitarian because it represents every existing site equally; the extreme of such an egalitarian approach would be to represent every possible site equally. The best way to do this is to write the books yourself--to become not just the librarian of all of the archive's texts, but also their author. This seemingly impossible task can be begun--if not finished--by creating a self-assembling archive.12 To build this archive, you need only expand on the technique used by the heretics in Borges' Library. Of course, it would take years of manipulating metal disks in a dicebox to derive an entire book at random; on the other hand, it only takes a few minutes of manipulating the metal disk in a computer's hard drive to generate a book-sized chunk of random text. In fact, to write a bot--a mini-program--that automatically outputs every mathematically possible combination of letters and spaces would require only a few lines of computer code. Load this generative algorithm onto a laptop, place it in a briefcase along with a portable printer, and you have the self-assembling equivalent of Marcel Duchamp's portable museum, a sort of Bot en Valise. What might you see if you opened this valise?

One possible answer to this question, John Simon's Combinations (1995, *ttp://, would have made any of Borges' heretics envious. Simon's Java applet starts from a reduced vocabulary--all the possible combinations of four line segments in a square--that recalls the LeWitt wall drawings mentioned above in connection with the Jumping Lines Web site. Once the user has chosen the angles and placement of the four segments, Simon's applet produces a grid of combinatorial drawings on the computer screen or in a plotter printout. LeWitt's assistants may be just as envious of Simon's program as Borge's heretics: a LeWitt drawing can take a team of draughtspeople weeks to complete, while Simon's program spits out the grid of permutations in a matter of seconds.13

Even more than Combinations, Simon's Every Icon (1996, *ttp:// reveals the power--and limitations--of a self-assembling archive. Simon's description of the archive--its card catalogue, as it were--is simplicity itself:

Given: A 32 x 32 Grid
Allowed: Any element of the grid to be black or white
Shown: Every Icon

The word "shown" is somewhat misleading. Once triggered by the user, Simon's applet will in good faith begin to display every possible combination of black and white elements; yet even at a typical desktop computer's rate of 100 new icons per second, it would take over 10^298 years to draw every icon. Like Borges' library, there is always the potential of finding a meaningful artifact--a copy of Moby Dick, say--but in practical terms the user is likely to be exhausted long before the icons are. In fact, Simon estimates that the first recognizable image wouldn't appear for several hundred trillion years. (What would the first one be? A "no smoking" icon? A bitmapped Mona Lisa?)

4. Learned criteria
I know of districts where the youth prostrate themselves before books and barbarously kiss the pages, though they do not know how to make out a single letter.14

There is a middle ground between relying on someone else's exclusive, evaluative criteria and the more egalitarian--though somewhat desperate--search through every possible variation on some fundamental generative criteria. Archives organized by learned criteria are not exclusive, for they offer each user a completely different set of links; their aim is to home in on exactly what each individual user will find most interesting. Imagine that there was a librarian in Babel who had no taste of her own; when you asked her to select books for you, she would deliver ten at random to your desk. At that point you would review them, dismissing most of them as uninteresting but pointing out one or two that seemed remotely interesting. The librarian would then return to the stacks to select another ten at random. You would repeat the evaluation again and again, until the librarian learned to choose books that held some meaning for you. Each visitor would have to train a different librarian, for a visitor who read Aramaic might enjoy proto-Christian gospels while to another they would be complete gibberish. Of course, the Library of Babel would have to have a budget as large as its collection to employ a different librarian for every visitor. Fortunately for visitors to the Internet, however, nonhuman librarians are much more economical. In fact some actually exist on the Web right now.

An example that has become popular lately is the Firefly agent ***(ttp://, developed by Pattie Maes at the MIT Media Lab. Upon first arriving at the Firefly music site, you see a list of ten musicians, which you rate from " the best" to "it's alright" to "hate it!". Clicking a button marked "more" brings up ten more albums; again you rate these and go on to the next list. As the Firefly agent learns your preferences, it gradually suggests albums you find more and more interesting. (If you rated Tracy Chapman highly on a previous list, Firefly is more likely to suggest James Taylor than AC/DC.) Since its original debut as a music filter, Firefly has launched a movie database as well as an agent, developed in collaboration with Yahoo, that recognizes and downloads preferred Web pages--presumably allowing you to filter the online artworks you would prefer from the ones you don't. Like similar services offered by Sun Microsystems *(ttp://, Firefly makes its decision not from any innate aesthetic sense, but from comparing an individual user's tastes to those of like-minded users whose tastes it has already sampled. Firefly is essentially a statistical correlation program running on a database; it assigns each user a vector of preferences and then finds close fits to that vector in its archive of tastes. Unlike conventional search engines, Firefly doesn't know any categorical criteria, such as the fact that James Taylor is "soft rock." It merely knows that other users who liked Tracy Chapman also enjoyed James Taylor.

Interactive genetic art is a sort of half-breed--rather literally--between learned criteria and generative criteria. Although Karl Sims is the best-known exponent of this work, one can also find examples of interactive genetic art online. At the International Gallery of Genetic Art ***NOTE: THIS SERVER IS VERY TEMPERAMENTAL, BUT IT IS THE ONLY EXAMPLE OF INTERACTIVE GENETIC ART ON THE WEB: (ttp://, for example, the user is presented with five examples of simplistic computer art. As in Firefly, the user is prompted to evaluate these images and then wait for new ones to appear. As in Firefly, each new screenful of images tend to be more interesting than the previous ones. Unlike Firefly, however, the subsequent images are not selected by running statistics on a database of past users' preferences, but by breeding successful images together. To be more specific, each image is created by a simple computer algorithm; the program codes for these algorithms can be interspliced to produce new images that are in a genetic sense the offspring of the older algorithms. The higher the user rates an algorithm's image, the more of that algorithm's code is likely to be found mixed into the algorithms represented in the next batch of images. It is as though one bred a brooding portrait by Edgar Degas with a brushy abstraction by Willem de Kooning to make one of Francis Bacon's contorted visages. Of course, interactive genetic art is a bit cruder--the home page offers circles and squiggley lines rather than 20th-century masterworks--but fortunately the results are also more subtle and less predictable than this example.

5. Evolutionary criteria
The impious speak (I know) of "the febrile Library, whose hazardous volumes run the constant risk of being changed into others, and in which everything is affirmed, denied, and confused as by a divinity in a delirium."15

On the horizon lies the future of the archive, one hinted at by the interbreeding algorithms of interactive genetic art. There the Library lives, breathes.

Research into our own bodies reveals that organic archives have already been around for millions of years. The human immune system has a memory; chromosomes carry junk DNA from generation to generation, organized in ways that approach a language; the brain stores a vast array of mutually contradictory thoughts. According to leading biologists, what makes these organic archives different from their physical counterparts is their evolutionary dynamic. The agents that comprise organic archives, be they t-cells, genes, or groups of neurons, compete according to learned selection criteria for representation in the overall population. DNA for a fast antelope is more likely to survive in the gene pool than DNA for a slow one; neurons that fire often will build more connections than neurons that fire rarely. There is a further difference between body and book: organic building blocks contain no information individually. Genes and cells aren't like books or files that can be consulted individually, but more like the letters from which the library's books are written; they are meaningless except in large-scale combinations. Knowledge is a property of the whole, so an organic archive can only accrue knowledge by evolving, by adapting to the information provided by its environment.

A quick reality check reminds us that we are a long way from creating an organic archive ourselves.16 Nevertheless, the literally burgeoning field of artificial life offers a glimpse at a living library--though its lifeblood pulses across modem lines rather than capillaries. A good example is the Tierra project ***(ttp:// by Tom Ray of the University of Delaware and the Advanced Telecommunications Research laboratory in Tokyo. Ray describes Tierra as a "wildlife refuge" for artificial organisms. Ray creates this refuge by copying a population of self-replicating computer algorithms onto a computer network. These algorithms are designed to reproduce themselves in a manner similar to their notorious cousins, the computer viruses--except that they are confined for security reasons to a "sanctuary" on the network. To this promiscuous mix Ray adds two ingredients that make the network more of a Darwinian jungle than a sedate server: mutations in the program code that produce new algorithms from time to time, and a reaper subroutine that weeds out and erases any mutated algorithms that fail to function properly or cause errors. Once Ray lets the system go, it proceeds of its own accord, generating new algorithms that are most fit to survive in the population. Unlike the self-assembling archives of John Simon, Tierra's potential isn't easily summed up in prophetic encapsulations like "Given a 32 x 32 grid, show every icon." Since its inception, the autonomous evolution of Tierra's digital creatures has created new algorithms that even Ray himself could not have foreseen. Among these are parasitic algorithms, snippets of program code embedded in other algorithms that are automatically reproduced when their hosts reproduce. Ray's sanctuary has also produced nocturnal nomads: populations of algorithms that migrate, via a sort of guerilla e-mail, to the dark side of a global computer network in order to take advantage of computers that are available because their users are asleep.

If the World Wide Web--or whatever its successor as the cultural archive of the 21st century may be--ultimately turns to such evolutionary models, then the Library of Babel will no longer be an appropriate allegory. The approaches suggested by the metaphor of the Library emphasize either selection or definition. To base an archive on evaluative or categorical criteria is to focus on selecting the right items (the card catalogue), while to base an archive on generative or learned criteria is to focus on defining the inventory from which that selection can take place (the stacks). In an organic archive, however, it is meaningless to consider selection or definition independent of the other. The archive of the future may not be a Library but a Landscape: to be specific, the Epigenetic Landscape.17

The Epigenetic Landscape is a metaphor used by evolutionary biologists to suggest the way the development of an organism can be subject to genetic and environmental forces at the same time. Imagine a stretch of land in which dramatic peaks and valleys have been formed by powerful seismic forces; different populations live in the various valleys of this landscape, out of touch with each other due to the intimidating ridges that separate them. In this visualization, seismic forces represent genetic influences, which tend to segregate species into incompatible gene pools; hence birds and reptiles can no longer mate, even though they evolved from a common ancestor. Nevertheless, the behavior of the population is not determined solely by this seismic topography, for rain and wind can erode previously impassable peaks down into humble hills more easily traversed by the landscape's inhabitants. In this metaphor, the wind and rain represent environmental influences, which tend to encourage the evolution of new species through dramatic climatic change. (Paleontologists hypothesize such a cataclysm to explain the sudden extinction of dinosaurs and diversification of mammals 70 million years ago.)

To recast this metaphor in terms of the organic archive, the valleys of the epigenetic landscape define the possible repositories of knowledge at a given time--such as all of the types of computer algorithms currently on Tierra's network. Left to themselves, these algorithms might just evolve toward a steady state in which a few species of reproducible code come to dominate and the others die out. To interject diversity into this system, however, Ray can unleash some wind and rain--change the mutation rate, for example--and the complacent established algorithms may have to evolve anew in order to survive. If the Library becomes a Landscape, selection criteria will become a selection force, and "the hazardous volumes" will indeed run the constant risk of being changed into others. The librarian, meanwhile, will become a sort of divinity watching over her creation--a divinity who is perhaps a bit delirious at the thought of her archive responding animately to the world around it.


1 Jorge Luis Borges, "The Library of Babel," Ficciones (New York: Grove Press, 1962), translated by Anthory Kerrigan, page 79.

2 Borges, page 83.

3 Various authors have noted a connection between Borge's library and the Internet, most recently Daniel Rubey, "Meditating on the Library as Archive: From Alexandria to the Internet," Talkback! 1, December 4, 1995, ***ttp://

4 Borges, page 84.

5 Before We Both Belong, Kinmont had been suspicious of the Internet's capacity to reach people; see Laura Trippi's comments in "The View from the Street," World Art (Newark), no. 3 (Summer), p. 55.

6 Compare the proliferation of encryption in the digital age with its parallel in the myriad languages afforded by the Library of Babel: "I cannot combine certain letters, as dhcmrlchtdj, which the divine Library has not already foreseen in combination, and which in one of its secret languages does not encompass some terrible meaning. No one can articulate a syllable which is not full of tenderness and fear, and which is not, in one of these languages, the powerful name of some god." Borges, p. 86.

7 Larry Rinder, one of the curators of the Jumping Lines site, goes so far as to claim "it's hard to believe that Haring didn't see works like these at some point." *(

8 Borges, p. 85.

9 "Someone proposed a regressive approach: in order to locate book A, first consult book B which will indicate the location of A; in order to locate book B, first consult book C, and so on ad infinitum...." Borges, p. 85.

10 Most of John Simon's projects can be accessed from his home page at *

11 Borges, page 84.

12 As long as the items to be archived can be broken down into discrete building blocks, such as the twenty-odd letters and punctuation marks that make up the alphabet or the 256 pixel colors that make up an 8-bit digital image, then there is a way to list all the possible combinations of these building blocks, though the list may not be finished in your lifetime (or anyone else's). Even if the number of combinations is infinite, you can devise a procedure that will eventually generate any given item in the set, as long as the fundamental variables are discrete. If, however, the variables are continuous--such as pixels that can vary across a continuous spectrum of color--then the list cannot be begun even in principle.

13 Simon's work plays havoc with the notion that conceptual art is "pure information," for in the computer an artist of the 1990s has a much better tool for creating works composed simply of data than an artist of the 1960s ever had. Nevertheless, Simon's work is not necessarily intended to devalue or upstage LeWitt's. The excitement of LeWitt's wall drawings in fact stems from the contrast between the trivial instructions for making them and the remarkable presence they can achieve when executed in the right architectural frame at the right scale. That the content of conceptual art lies not in the idea alone but in the tension between idea and execution is the subject of my article "Where Did All the Uncertainty Go?" in Flash Art (New York) 29, no. 189 (Summer 1996), pp. 83-87.

14 Borges, p. 87.

15 Borges, p. 86.

16 Perhaps the DNA computer represents a step in this direction. See my essay "The Work of Art in the Age of Mechanical Sex," originally presented at the 1997 College Art Association panel "Cyberspace: Trojan Horse or Roman Holiday?" !*** THIS SERVER IS NOT RESPONDING-I WILL CONTACT THE WEBMASTER TO CONFIRM.

17 The term was coined by the embryologist Conrad Waddington in the 1950s. A good description for the lay person is found in Jack Cohen and Ian Stewart, The Collapse of Chaos (NY: Penguin, 1994), p. 93.