Mario Carpo

Publication History:

“Every Dataset is a Canon.”  AD 94, 3 (June 2024), 14-19

The text posted here is a preprint draft and it is significantly different from the published version. Please only cite from copy in print

Who remembers structural linguistics?  I do, but that’s because I’m old.  Like social housing, marxism, and progressive rock, structural linguistics was popular in Europe when I was coming of age.  And as it happens, some basic tenets of that new science were permanently etched in my memory by a bizarre incident that occurred in a foggy evening in the early fall of what must have been 1974, or 1975.  I was then a high-school student and, with a group of similarly minded friends, I regularly attended the weekly sessions of a highfalutin ciné-club (film society) in the backwater, provincial town where I grew up.  The screening on that particular night was Antonioni’s Red Desert, by that time already a classic—we could not have said a “modernist” classic back then because we had no notion that there was life outside of modernism, and that modernism itself was about to come to an end.  In the debate that followed the screening (also, a fixture of the time) one of my older friends stood up to vent his discontent.  That movie, he claimed, was conventional and bourgeois.  Its reactionary stance showed, evidently, in the decadent lifestyles it depicted, but even more so in the conventional, sequential, and fully predictable narrative storyline it followed.  That narrative predictability was worse than reactionary, went on my friend, more and more inflamed; for, the science of structural linguistics already positively demonstrated that the “beauty of a work of art is in reverse proportion to its semantic predictability.”  In that context, and delivered by a shaggy looking, callow and barely bearded college student, that tagline sounded like perfect mumbo-jumbo, and the audience erupted in a roar of laughter.  And even we, the young revolutionaries, made fun of my friend (today a well-known art critic) on our way home later that evening—and, in fact, for years to come; which is why I still remember that phrase, verbatim, as it was delivered that night.  Yet now, almost half a century later, if I think of it, I can confirm that my friend was—technically—almost all right.  To be entirely right, we would just need to edit one word; for structural linguistics—and even more so the information theory that derived from it—does not deal with semantic, but only with syntactic predictability.

Starting with De Saussure’s Cours de Linguistique Générale (1906-11) modern linguistics has argued that all use of languages depends on the dialectics between code and creativity, convention and invention (in Saussure’s terms, langue and parole).  In the artsy version of structural linguistics popularized in the 1970s by Umberto Eco and Roland Barthes, the amount of information conveyed by a message grows in proportion to its quirkiness (or statistical unpredictability), but only within limits: creativity and invention must not hinder the legibility of the code on which a message is based; when that code is lost, so is all meaning—because no one will know which language you are speaking.  In other terms, breaking rules is only meaningful in so far as one can tell which rules are being broken; all revolutions presuppose constitutions.  A few years after the ciné-club anecdote I recounted, as a student at the University of Florence, I had to toil on a monumental handbook of architectural semiology by Giovanni Klaus Koenig,[1] who had been a friend and associate of Umberto Eco during Eco’s brief Florentine stint.  This is how I learned that all architectural creation, or invention, must first single-out, acknowledge, and formalize the code embedded in the tradition to which it refers, and into which it inscribes itself: the rules of that code are the structure that underpins that language, and makes communication in that language possible; only when that code is spelled out and stated may any creative use of that language ensue—hence contestation or transgression, and the creation of additional meaning.  In short: in order to say something meaningful in any artistic language, you must first get acquainted with the rules inscribed in some precedent which you realize is out there, and which you set as your precedent of choice: that’s the language you choose for your voice, and the context you give to your text.  In architecture, this precedent can be a single model, or, more often, a corpus of models—a tradition, or canon.  In the 1970s, the only canon we had in mind was that of architectural modernism; but that’s irrelevant.  This structural logic applies to all content, linguistic medium, and creation of meaning.

Structuralism always had an innate tendency to vast ahistorical generalizations, but in this instance it does appear that, well before modern structuralism formulated it in quasi-scientific terms, a somewhat similar logic had already been intuited by many artists and scholars.  I trust that comparable ideas of precedent must have surfaced and blossomed outside of the European tradition—that is a part of the story I do not know.  In the European tradition, however, at one point in time—known today as the Renaissance—reference to precedent famously became a universal mandate, and almost an obsession.

For reasons too long to explain, and which some found and still find inexplicable, at the end of the Middle Ages Renaissance scholars concluded that one period in the history of Europe—the period we now call classical antiquity—was the climax and zenith of all civilizations, past present and future; and that modern scientists and artists, and even society at large, should revive the lore of their classical predecessors by imitating them.  As a side effect, Renaissance architects were often tasked to imitate classical buildings while designing buildings that never existed in classical times (Christian churches, for example).  This is where Renaissance artists were de facto obliged to invent a new mode of imitation.  Renaissance artists had to learn to imitate the visual form of an artistic model—but only the form, without its content.  Their new mode of imitation thus became, first inadvertently, then more explicitly, the transference of form through different contents; the extraction and replication of some appearances regardless of subject matter.[2]  Renaissance scholars did not have a word to define the object of this new operation, until Vasari started to call it a “manner”; much later, in the 19th century, the term “style” was almost universally adopted with a similar meaning.  To be noted, Renaissance artists also came up with a concurrent theory to limit the ambit and scope of some deliberate tweaks, then called “licences”, that artists could introduce in the creative replication of the canonical models they imitated—which could be altered, or even transmogrified, but not beyond recognition.  ”Licentiousness” then became the hallmark of artists like Giulio Romano, or Michelangelo—and of a classical style today called Mannerism.    

In this sense, and with these meanings, imitation, manner, and style became the backbone of academic training, and the warp and weft of the classical tradition in the visual arts—including the arts of design.  This is what artists and architects trained in the European tradition studied in academies of arts—that’s what they specialized in: in the more or less “licentious”, or creative, imitation of manners and styles.  And as we all know, this is the artistic culture, and the kind of artistic training, that modernism brutally and thoroughly eliminated from the arts and design of the 20th century.[3] 

Modernist artists, designers in particular, had many good reasons to feel averse to the ideas of imitations and styles that had largely dominated 19th-century European art theory.  This is what modernists wanted to do away with and get out of.  Yet the idea of reference to precedent was never fully abandoned in the course of the 20th century, and as of the late 1970s post-modernists of the historicist ilk made of reference to precedent their battle cry.  The precedent of PoMo choosing was, mostly, the European tradition—as if the PoMos knew no other; which in most cases they didn’t.  But it is worth mentioning that, well inside the modernist, or late-modernist precinct, the tradition of formal analysis (as distilled and perfected, famously, by Peter Eisenman) kept the idea of reference to precedent at the core of design education in North American academia.[4]

The readers of this issue of AD are, I must assume, well acquainted with the technical logic of Generative AI, so I do not need to remind them that GAN and Generative AI are, in essence, automated technologies of imitation.[5]  Imitation and style, which were kicked out of design theory by the ideology of late modernism, are coming back to design practice through the unexpected window of technology.  The first step in every Generative AI project is the establishment of a dataset; this dataset will eventually translate into a mathematical “latent space” which will in turn preside over the creation of new specimens that will have “a certain something” (often a style) in common with the models assembled in the original dataset.  This dataset may be ready-made or custom-made—made to measure for a specific task; all the same, a dataset is, simply, the tradition we designate as our precedent of reference: it is the langue that makes our parole possible; it is that canon of our own choosing which gives meaning to our voice.  Every dataset is a canon; every canon is a dataset.  If Generative AI works, it is because someone, somewhere, must have created a dataset—chosen a canon. Creating a dataset means putting something in, and leaving something—or someone—out. And we know all too well today that preference often means prejudice. Give me your dataset and I shall tell you who you are.

Generative AI came from computer science to design studios, and to design practice, to reminds us that there is no creation without imitation, no invention without convention, no progress without precedent.  To remind us that we are but dwarves of the shoulders of giants, and that whenever we sketch something we like, it is because there is something similar, out there, which we must have seen and liked; when we write something we like, it is because we must have been inspired by something similar we have read; when we use words we like, it is because someone has already used them—if not exactly in the way we do.  And at the end of the day, we should not be surprised if computer scientists, imitating the inner logic of the human mind, came back to us with a creative tool that works similarly to the way our mind seems to work.  Twentieth-century modernism in the arts, in culture, and in literature, repudiated the organic logic of the human mind—and of the human body—because it thought it had to follow the artificial logic of the machine; and there is a certain irony in that today a new technology, which we still call artificial intelligence (a term invented in 1956), should remind us of the way our natural intelligence always worked.          

[1] That was the second, expanded edition of Architettura e Comunicazione, published by the LEF in 1974 (first edition in 1970).

[2] Alexander Nagel, Christopher W. Wood, Anachronic Renaissance (New York: Zone Books, 2010).  See in particular 150-157 and 289-300.

 [3] See recent work by Mary Hvattum and in particular Style and Solitude (Cambridge, MA: MIT Press, 2023).

[4] Mario Carpo, “Formal Analysis, Generative AI and the Eternal Return of Precedent,” Log 58 (2023), 133-139.

 [5] Matias del Campo, Neural Architecture. Design and Artificial Intelligence (Novato, CA: Oro Editions, 2022); Mario Carpo, “Imitation Games. Mario Carpo on the New Humanism,” Artforum 61, 10 (summer 2023), 184-188.