Mario Carpo

Publication History:

“Out of Order.  Mario Carpo on Mannerism, the Canon, and Generative AI.” Artforum 63, 2 (February 2025): 102-108

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

Mannerists can never catch a break. It doesn’t help that the man who started it all, Giorgio Vasari (1511–1574), was a terrible writer and an uninspired artist. Yet in retrospect one cannot help but concede that he may have been on to something. When he wrote Lives of the Artists (released twice, in 1550 and 1568, and often hailed as the first history of art), he divided artists into separate groups, or sets, based on common features in spite of the artists’ individual differences, and he called those sets “manners”—thus anticipating today’s notion of artistic styles. Vasari’s general story line was that, starting with Giotto and Cimabue at the end of the thirteenth century, painters rediscovered that their primary task was the imitation of nature, and once that course was set, they never stopped improving their imitative skills. Painters first learned to draw nature as it is—and it took a couple of centuries for Italian, mostly Tuscan, painters to master that trade. Then came perfection in the art of mimesis, which, following the anecdote of Zeuxis’s five Crotonian maidens, is achieved by picking the best out of many models, thus rendering nature not as it is, but as it should be. This, Vasari argues, is the bella maniera, the good style, which painters of his time, from Leonardo to Raphael to the best of all, Michelangelo, had finally mastered in full.

The problem with teleological models of history is that once the final station is reached, the story is over, and nobody knows where to go next. The consensus among art historians is that late-Renaissance painters, having concluded that everyone had already learned how to imitate nature, started imitating one another—in particular, imitating one another’s style, regardless of content. This is the modern meaning of Mannerism, which, as of the nineteenth century, acquired distinctly “sinister” connotations. [1] Mannerists came to be seen as painters who had lost their footing, forsaken Renaissance naturalism, and traded that worthy mandate for a futile game of endless self-referentiality. For the same rea- sons, some in the twentieth century started to see the Mannerists as heroes, not as villains.

None of the above applies, or ever applied, to architecture. Vasari deals with painting, sculpture, and architecture in similar terms (and he actually pioneered the notion of these three arts as closely related “arts of drawing”). But architecture, unlike painting and sculpture, does not imitate nature—at least, not directly. The mimetic principles already outlined in Vitruvius (primarily the idea that the classical orders imitate the structure of a primitive wooden construction) were, from the start, fig leaves, and widely understood as such. Renaissance architects were not meant to copy nature; their real business was to imitate ancient architecture. In that regard, Renaissance architects were much closer in spirit to Renaissance writers than to Renaissance painters: Renaissance writers, too, tried to revive a classical language—mainly Cicero’s Latin—by imitating a corpus of extant ancient sources.

Not surprisingly, Renaissance architects and writers ended up contriving very similar imitative strategies. Architects and writers alike had to compile catalogues of reliable models, then find ways to imitate them without replicating them ver- batim. One of the best methods they found consisted of looking for common features, or regularities, in the models they chose, and extrapolating from those regularities a set of usable rules. For writers, these were the rules of grammar and syntax; for architects, these were the rules of the classical orders, published in handbook form as of 1537. Writers and architects, working in different media, ended up writing, respectively, rules for imitating the style of ancient writers and rules for imitating the style of ancient builders. Insofar as Renaissance writers and architects were imitating manners, or styles, and not nature, all Renaissance writers and architects were, in a sense, Mannerist from the start—without knowing it.

In fact, architects did know it. From the beginning, alongside the rules for building in the style of the ancients, Renaissance architects set forth rules for transgressing them. The idea was that once one knows the rules, one may also break them, but only if the rules that are being broken are still visible or intelligible, thus implying that showing the awareness of a rule, and at the same time its deliberate infringement, could by itself be a message and convey meaning. The Italian architect Sebastiano Serlio was the first to theorize various ways of breaking design rules—the very same rules he had so carefully spelled out in his own books; he called these transgressions “licenses,” using a term he borrowed from coeval moral theology, meaning something venial you do knowing you shouldn’t. Giulio Romano is often seen today as the master of Mannerist licentiousness, but it was Vasari, again, in 1568, who offered the best formula: License is what is outside of the rule, but can be subsumed under the rule without creating confusion or breaking the overall order. [2] More than three centuries later, Ferdinand de Saussure could not have said it better.

Early in the twentieth century, Saussure famously posited that all languages evolve through a dynamic interplay between established norms and use, convention and invention, code and creativity (langue and parole); in the 1960s and ’70s, Roland Barthes and Umberto Eco, among others, popularized an artsy version of Saussure’s linguistics, asserting, almost as a scientific principle, that breaking some rules of a language would make most messages more meaningful (or, to be precise: would increase the amount of information being conveyed). [3]

This version of modernist semiotics was hugely popular among avant-garde artists and designers of that period, and this is how, improbably, the classical theory of Mannerist licentiousness and the structuralist theory of language imperceptibly merged, as they both led to the same plan of action, which was as follows: First, choose a language; next, learn its rules—its code; then, break or tweak some of those rules, but not all, and not in full, so the language will keep functioning, and you can avail yourself of all of its subtleties and nuances to express your difference, detachment, distance, or disagreement. As a consequence, courtesy of semiotics, twentieth-century Mannerism became an explicit critique of precedent, which classical Mannerism had always been—without saying so.

And sure enough, as soon as the social and political priorities of architectural modernism abated, and designers could again express their interest in form, Mannerism was primed for prime time. Robert Venturi’s seminal Complexity and Contradiction in Architecture, first published in 1966, was reportedly meant to be titled Mannerism in Architecture; the title was changed at the last minute, but the picture on its cover page remained that of Michelangelo’s Porta Pia gates in Rome. [4] Venturi kept reiterating his interest in Mannerism throughout his long and influential career, but what he and Denise Scott Brown meant by Mannerism changed over time, shifting from a direct reference to historical Mannerism to a general idea of eclecticism and multiplicity of sources. It is, instead, with the work of the so-called New York Five (Peter Eisenman, Michael Graves, Charles Gwathmey, John Hejduk, and Richard Meier) that the general principle of what I would call critical Mannerism was solidly established: The New York Five chose architectural modernism, not classicism, as their precedent of reference; and they creatively reinterpreted, critiqued, and transcended the formal rules of their language of choice. [5] In other words, they did to early Le Corbusier what Serlio had done to Bramante.

Though the spirit of this Mannerist game was eminently clear in most work of the New York Five, the protagonists did not directly invoke Mannerism itself—not even Colin Rowe (who prefaced the original publication of Five Architects in 1972) or Manfredo Tafuri (who prefaced a related Italian publication), both great experts on all things Mannerist. Yet, as Joan Ockman has recently pointed out, both Rowe and Venturi had precociously embraced “the inevitability of historical precedent as a source of formal invention,” [6] and from this springboard came the leitmotif of Peter Eisenman’s lifelong endeavors: Historical precedent is key to all formal invention, because without precedent there is no language; but so is the critical disturbance of precedent, because without that disturbance there is no innovation, hence no meaning. License is what turns copy into creative imitation. If this idea of Mannerism has been variously embedded in architectural theory for the past fifty years, there are some reasons for its recent revival. As I pointed out in a 2023 essay in Artforum, the technical logic of generative artificial intel- ligence oddly vindicates some core principles of the classical tradition in the visual arts. [7] Generative adversarial networks (gans) and more recent, GPT-based cross- modal generators, like Dall-E and Midjourney, create new images derived from a dataset of existing images used to “train” the system. In art-historical terms, AI-based image generators imitate the datasets they have been fed. But this is imitation in the classical, not in the modernist, sense of the term: While for mod- ernists all imitation was about copying, replication, and plagiarism, in the classi- cal tradition, imitation was about working from models, seen as a source of inspiration to be creatively transfigured and transcended. Likewise, today’s gen- erative AI produces new images that are meant to be recognizably similar, but never identical, to the exemplars from which they derive—never mind if the orig- inal dataset is custom-made (as in the old gan systems) or ready-made and generic (as in today’s GPT systems).

As everyone who has tinkered with generative AI knows, the computational process of imitation (i.e., of interpretive transformation) of any chosen set of models happens within the “black box” of a machine; the only way to change the output is to tweak the input, changing the datasets (or, in more recent systems, the verbal prompts, which in turn tweak the subsets of data being brought into play). But regardless of the technical tricks one needs to master to play the game well, the general spirit of that game is always brutally transparent: When generative AI creates images, everything is generated out of something that already exists—something that is already out there: The traditions, codes, conventions, or languages we use and refer to are the precedent that makes our creation possible and our message understandable. Thus, by the way it works, generative AI also reminds us that, regardless of technology, all artistic expression starts with the awareness, acknowledgment, invocation, and selection of our precedent of reference—and of our reference to precedent. Precedent is that canon into which we inscribe ourselves, and which gives meaning to our voice. And we know full well today that every canon is based on preference, and that preference is often a proxy for prejudice. Datasets are by definition exclusionary: When we make a dataset, or even when we just use one, we put something in to kick someone out. It is as simple as that.

And let’s not beat around the bush: Precedent, or a dataset, in architecture and design means tradition and history. And every time that architects in particular invoke history, we have reasons to be alarmed. Architectural traditions have often served as signposts and standard-bearers of nations, cultures, and civilizations; and of all canons and traditions, the classical language of architecture has more often been used or invoked as a signe identitaire (identity marker) of a certain idea of Europe—and as such it was in some cases even mandated by law, in some Western countries, as recently as a few years ago. Fascists, Stalinists, racial supremacists, and bigots of all times and places typically endorse architectural canons—that one, or any other—for similar discriminatory purposes.

But this is the uncomfortable truth: In spite of all the above, we still need canons, and always will, because without canons—without languages—we can neither speak nor be heard. We must use the languages we know, because we know no other. Yet when we do so, there is one more thing we can still and always do: We can at least try to make it clear that we are conscious of the rules of the languages we are using. We can critique a language we did not choose. We can use our parole inside and against a langue we do not like. Generative AI cannot do that—not now, based on the current state of the art, or I would suggest, for some time to come. But if generative AI, which is a technology, reminds us of the inevitability of precedent, Mannerism, which is an idea, reminds us that we can deal with precedent and still do the right thing.

Classicists endorse canons; Mannerists critique canons. And if these two terms I use here for ease of reference, classicism and Mannerism, belong to the European tradition, the idea of canon now applies to every tradition. Generative AI is a global technology, and it works the same everywhere, regardless of what we train it on. At any rate, even without reference to that technology, many designers of our time have already ended up being Mannerist, sometimes deliberately; a recent book by the architect and educator Francisco Gonzáles de Canales brilliantly surveyed many of them, from Office Kersten Geers David Van Severen to Lütjens Padmanabhan Architekten—a list to which I would also add the great master of late-modernist rule-breaking, Valerio Olgiati. [8] In the West, generative AI may beget even more totalitarian classicism—either of the pre-Mannerist or of the post-Mannerist ilk. But in between these two extremes—between Bramante and Bernini, so to speak—lies the narrow path of reason; and it is a notoriously difficult path. The sad destiny of Mannerists of all times, of all who choose a critique of precedent instead of endorsement or excess, is that many will see you as a nobody, and many will see you as a turncoat. But if anyone has better ideas for these nefarious times, please tell me.

NOTES

1. The term is from Gaetano Milanesi, the author of the first critical edition of Vasari’s Lives, published in nine volumes between 1878 and 1885. Giorgio Vasari, Le vite de’ più eccellenti pittori, scultori ed architettori, with annotation and comments by Gaetano Milanesi (Florence: Sansoni, 1879), 4:8.

2. Translation mine, from Vasari, Le Vite, 4:9.

3. Ferdinand de Saussure’s Cours de linguistique générale was written between 1906 and 1911 and published post- humously in 1916.

4. See Francisco Gonzáles de Canales, The Mannerist Mind (Barcelona and New York: Actar, 2023), 41, and footnote 47.

5. Five Architects: Eisenman, Graves, Gwathmey, Hejduk, Meier (New York: Oxford University Press, 1975), first published by Wittenborn in 1972, following a meeting held at the Museum of Modern Art, New York, in 1969.

6. Joan Ockman, “Form Without Utopia,” Journal of the Society of Architectural Historians 57, no. 4 (1998): 448–54.

7. “Imitation Games. Mario Carpo on the New Humanism,” Artforum, Summer 2023, 184–88.

8. Canales.