The Mutilated World
AI-generated images proliferate online, spreading through diverse political ecosystems and flourishing in the darkest corners of the internet. But, as A.V. Marraccini writes, what would it mean to create the plan of a real city from AI images, and how could we think through its politics?
I see one of the first trees blossom. It is mid-February; it is too early for that. I think it is an ornamental pear with the same level of certainty I am struggling to apply to Aristotle on form versus matter. My laptop is making a high-RAM-use-buzzing-sound like a pollinator looking at the tree. It is churning out machine generated images. They are probably matter, in the sense that they are electrified atoms passing through nameless binary gates. They also do have a form, which is to say the form of an image, which is dictated to some degree by a prompt that vectors out into latent n-space like a branch set in ikebana. Is a digital form a non-natural one by definition?
Heidegger says yes, but I am always suspicious of his thought. The type of randomness that makes images fuzzy and unfuzzy again in the training of these algorithms is called Gaussian noise. It all goes back to movement, to if you drop a random set of marbles into evenly distributed pegs, they will form the normal curve, to the fact that e and pi are these always-already-inscribed constants of nature that just keep cropping up everywhere. The normal curve is shaped like a bell; this bell is the distribution pattern of the noise, the way you can imagine the white specks are sprinkled across the expanse of a black square like cut-signal television, the way they’re always ring slant-rhyming oranges and lemons. The natural constants are called natural because they are pervasive in the world in ways we can’t always completely understand—when we divide circles, when we float out in the edges of the universe until its heat death in the plane of our Euclidean space, they persist and persist, the same despite everything, a kind of stubborn patterning. For instance, consider the fractal geometry of the pear tree’s branches splitting, each one into two and so on.
Anyway, from the noise, the semi-almost-random-constant-of-the-universe noise, the algorithms make images, because they are trained backwards and forwards in it: add noise to the known, de-noise the unknown, and so on. The data point for the unknown that is a pear tree probably lies close to the apple, the quince, the lemon… or so you would think, but technically this is an ornamental pear that doesn’t fruit, so it could also live close to roses or wisteria. The data point is a string of n-numbers in matrix brackets. Maybe if you could see n-space and squint at it would look like a tree, growing, splitting up branches stuffed with life and movement and phusis.
I know the ornamental pear not out of some knowledge of trees, but because in avoiding reading Aristotle I pulled an all-nighter playing a city building computer game. One of its digital assets for a city park was a small rendered ornamental pear tree, its tiny branches dotted with white pixels. But the images of cities that the AI generator builds are not like game cities – like Sim City, the famous one, made by Will Wright, and now held even in the collections of MoMA. I love these games; their implicit ordering. But the pictures of cities I make using the software we mostly call AI grow: they aren’t planned like sewer lines and railways. They are taken out of the Gaussian noise in little partial derivative steps, that turn into columns and cupolas.
[book-strip index="1" style="buy"]
In the Categories, Aristotle explains that an arm is no longer an arm if it doesn’t function as arm – say if you cut it off. But what if it has too many fingers? This is the current problem with AI images. The algorithm has seen the pattern in images of hands, they go finger-finger-finger-finger, but it doesn’t know when to stop adding fingers, or how they bend. (Human artists, incidentally, sometimes have problems with hands, because hands are difficult). Midjourney, the software I am using, in the variants of neoclassical cities it imagines, treats columns like fingers too. There are a lot of them, in vast rows, growing uncannily in to the distance. The images have too many columns, and too many cupolas. They sprout out of nowhere in the air, as if to say there is some minimum number that makes a city looks like a city and you just sprinkle them on without the buildings beneath. The don’t hold up pediments, mostly, or spring from adjacent towers. They have form in that they are generated from many other images of buildings, but their function is at a remove. They are not quite architecture, in the way the severed hand is not quite a hand because it can no longer do the acts of hands.
There are problems with entrances and egresses too. The doors are obscured. There are always too many stairs. Sometimes the water of the imaginary Mediterranean looks more like a swimming pool in Vegas. This makes sense, actually. I am making these images of neoclassical cities because of a project called the Praxis Society. They want to build a real one, with the assistance of AI, somewhere in the Mediterranean. They use images like this on their website and Twitter to imagine their utopia.
Praxis is a Network State, an idea based on a hack work of political philosophy – more glorified blog post than a real book – by a venture capitalist named Balaji Srinivasan, associated with the techno-libertarian VC firm Andreessen Horowitz. Although published in 2022, the ideas in Srinivasan’s book, The Network State, have been floating around the far right and right-libertarian sectors of Silicon Valley since at least 2016. In the book, the network state is ideological, united at first online and, theoretically at least, powered by the Blockchain – although so questionable is the necessity of this factor in its implementation that it more seems like an excuse to play Philosopher King for those nominally limited to the tech sector by association with crypto. The Praxis Society is one such state, a technocratic autocracy funded by billionaire venture capitalist and darling of the Silicon Valley right, Peter Thiel, among others.
[book-strip index="2" style="buy"]
Praxis Society is also unique for the way in which it joins Silicon Valley money and the New York City literary scene as legitimating intellectual garnish. In the Praxis Society’s own online journal, the names of buzzy debut novelist Jordan Castro, and an assistant editor at Paris Review, sit adjacent to disjointed musings from ex-VC investors who think they’re the new Socratic class because they can regurgitate a Wikipedia article. They collectively want the Mediterranean and its neoclassical turns because they think they are the inheritors of this tradition; one they refer to in explicit terms of Nietzschean heroics and the new frontier. They make screeds and videos along these lines, flirt with fascism as if it’s edgy and new. They add the images they generate from the noise periodically to facet them. Sometimes they look like Vegas, sometimes more like Speer.
A quote from Srinivasan’s book sticks with me:
I look out of the window at the pear tree, set into the streets of London – once Londonium, built on the Polybian grid of a Roman army camp. I finger the spine of Braudel’s two volume history of Mediterranean, thick with textual non-emptiness. There is no world without historical constraint. There is no world without e or pi, and in the conditions of our known universe at least, Gaussian noise still always results in that normal distribution, that sloping curve implied and inscribed into everything. I look at the generated images, I count the too many stairs with my eyes. The cartilage in my knees is weak. The nerves in my right arm were aggravated by the hypermobile joints in my hands, wrists, and elbows. I carry a gene for this; a disease with a name. The Praxis website and founder’s documents talk about the crucial need for “health and vitality”. They invite an expert on natalism – that is, the need for certain members of society to give birth to pass on their superior genes – on their nascent state podcast.
This network state, this product of crypto-blockchain technology with which I am intimate, mathematically and philosophically, these images of dream cities that I can summon with a click—they don’t want me there, in their city, no more than Heidegger would have in his Black Forest hut when he refused to recant the politics of the Nazi party to Herbert Marcuse after the war. The brokenness of my body, it’s queerness, has an incantory power to summon the history that the founders of these alt right crypto states want to erase and forget. The Italian writer and director Pier Paolo Pasolini, in making the infamous degradation film Salo, reminds us that fascism is ultimately expressed too, in power over bodies, their life throbbing into them, the sheer force of phusis that ends when you cut something off, a state, a history, a genital, a limb.
Pasolini’s poems in the Italian sub-dialect of Friulian are what made me think about pears in the first place, to notice the ornamental tree. The first and only early extant Friulian poem is form about 1380. I can’t find an English translation, but I read one in my broken Latin with some Italian. To a beloved:
Pasolini was possibly assassinated on a beach in Ostia for political reasons, but anyway. There is the ornamental pear, breaking the old gridlines of London’s city, and here is Aristotle, his Categories perhaps somehow less perplexing in the tangle of the Greek. Aristotle’s Athenian Constitution is a much easier text. When I look at the Praxis Society website I also think: write my name on all the ostraka, send me into exile from whatever this is ahead of time, though the means of its conception as just as much mine by intellectual right. I am internet-native, digital-native, crypto-native and an alpha and beta tester of the software used to make the AI images the city builder use. I want nothing to do with them: I choose history and not the story of the blank slate, where Augustine never stole pears, where the City of God wasn’t righteous and angry and wrong about all the same things centuries ago anyway. I look at the red spine of the copy of Ovid on my shelf behind the desk. Send me to Tomis! I’d rather be there.
Ah, but the images. The little cities, one after the other accumulating on my screen. Should they belong to them, the Network State and its odious far right politics? Can I take some of them with me? Gaussian noise is everywhere, like e or pi, after all. There are some constants. When I see the mathematics of it, the way the image generation algorithms are trained, I feel brave. Gaussian Noise too, is movement, is phusis the sense that Aristotle uses it in the Physics, that means a kind of animate force, and not simply the techne or handicraft of Heidegger.
The media theorist Byung Hyul-Chan makes a compelling case for the sterility of many digital images and things. I think of this when I see early attempts at the metaverse. He rightly notices that they are opposite to the baroque sensibility of ornament, seen as the purest expression of God. But the cities made my machine learning, by Midjourney and other software like it, aren’t devoid of ornament at all. The too many columns, like the overgrown fingers, are evidence of this. The columns themselves are of some strange, overproliferate order, with acanthus and ribbing gone wild and to seed. There is hope there, in the strangeness that seeps in. For Chan, digital things are Non-things, data replacing sensibility. But these cities, these images, to me are almost-things, there Almostness rooted in Aristotelian movement as the animate basis for the natural even as they are manifestly un-natural. They aren’t quite constructed though; more found by working de-noising programs backwards on pure noise to generate an image, like backwards-finding the root of the tree from the last splitting twig, with a bud and white flower on the end.
Almostness defies the ends for which these cities, their vistas, might be productively misused, might escape the ends of their creators. They can cast you out of their state, enact an absolute and self-aggrandizing power, but they can’t defy the gravitational laws of physics and the second law of thermodynamics any more than I can. The prospective exiles of the Network State stand in equal aesthetic contention for the means of machine generated images of cities as the presumably hale and ubermensch-like city dwellers. The software doesn’t check your polis. It also, because it is trained on the entire ossified product of human artistic labor – over five billion text-image pairs in the LAION dataset alone – cannot escape historical constraint. Every single image, every word, is bound up in and by it, espaliered to its fence as if in some ornamental garden. It is not a surprise that I have noticed this and that people like the Praxis Society fail to see the contradiction. Exiles from real and imagined city-states, after all, are often particularly attendant to history, and particularly aware of the pitfalls of claiming to remake, renew, or erase it, whether rhetorically or with images alone.
And yet, the dataset has inborn algorithmic biases. The software does in a sense check your polis. If you use “ideal city” or “ideal future city” as a prompt with the current Midjourney default trained styleset, a smattering of mid-century New York or some very waterlogged isle with a touch of Schinkel appears. The images look like stylised video game covers or those of pulp novels. They tend toward the palette of teal and a warm orange. The are almost highbrow and almost lowbrow at once. The algorithm uses visual variations on the English word ideal to poster-title the images: Iolal, Llsal, Dilata, Dilaly. Some of them take stranger turns than others, into nineteenth century orientalism and Magritte-like hovering spheres. These are idealised cities yes, but maybe the ones Marco Polo tells the Khan about in Italo Calvino’s novel: Almost-Venices.
The best of the almost-cities, to me, are the ones with strange errors, where the imperfections of algorithm breathe a paradoxical life-phusis into their being. One city has a fairly typical idealised body of water, but it is furrowed with what look like roots or cracks in the ice. The buildings seem strangely Gothic instead of neoclassical. The edge of the city seems to seep into ruin, crumbling into the sea. This is not the kind of city the neat autocracies of Praxis select as a sample image from their AI outputs. It is implicitly historical: ruins imply time. It is imperfect, messy, splitting at the roots. Sometimes the things that even a trained algorithm that is inclined to prefer wholeness emerge from the noise in latent space to something that resists itself. Adam Zagajewksi, a Soviet poet of exile, describes the spirit of these images best in a line demanding that we “praise the mutilated world.” Despite itself, the medium of diffusion generated images can give us just world to praise.
[book-strip index="3" style="buy"]
The mutilated world is full of almost-things. Another favourite image from my data set shows what is supposed to be a neoclassical façade, but it is more like some demented garden folly. The columns are irregular, even if topped by neat concrete (you can imagine here a deeply frustrated Mussolini). There is no building per se, just a colonnade with such uneven rhythms that it is accidentally postmodern. It glows in a golden light in front of a futuristic cityscape but one has to wonder: is it an almost-city, full of almost-buildings like this one too? The light is the same cast as that in Pasolini’s Boccaccio. On the cusp of evening, so many illicit things can come into being. The many fingered algorithmic hand draws differently. It is a strange body, too. Fascists hate strange bodies, hate things outside the norms of hearth and home, blood and soil. Praxis likes to talk about their city sitting on a new heroic frontier. The anti-heroism of this building is that it does nothing. It is an ornament to the world. It refuses the idea of a new manifest destiny, because it is an almost-monument or an almost-future-ruin, but cannot resolve into either. The clean-slate aesthetics of the Network state can’t accommodate the phusis that comes out the noise particles.
Joy is evasive and an almostness. For my last set of images, I use the same set of prompts but add the stipulation of a pear tree to them. Some of the images show just that: a pear tree growing in an idealised future neoclassical city. But they get strange quickly. The weight on the word pear causes the algorithm to mis-size the fruits, making them looming and gigantic. Trees grow into columns into walls, instantiating the metaphor of the AI generated almost-thing as bound to phusis as growth intractably. Sometimes, the buildings nest in a giant pear or a giant tree like playhouses. This set also produces an oddity: a futuristic, clean-lined building that grows out of fleshly-pear appendages, a Pasolini orgy of breast-buttocks fruit.
Perhaps this is where you live if the terrifying pseudo-perfections of the AI generated Network State exile you. Perhaps this evasiveness is a tool and not a punishment. Implicit in almostness, in the mode of functioning that relies on noise-signal continuums, is also the potential for strangeness, for resistance. The advocates of Network States like Praxis use AI generated images because they are obsessed with the technocratic flourish. But a flourish, like an acanthus leaf, is an ornament, and ornaments are dangerously, biologically inflected things. These generated images of architecture offer solace in their inseparability from the noise that generates them, from phusis and strange growths. They can make us blushing and brave in their failure to normalize. This is what we say to tyrants, as we sit in our boat bound for the rocky shores of Tomis: I wave my six-fingered hand. I count the too-many hypothetical columns. I laugh. I praise the mutilated world.
A.V. Marraccini is an essayist, critic, and art historian. Her first book, We The Parasites, was recently published by Sublunary Editions. She will be the Critic in Residence at the Integrated Design and Media program at NYU starting this coming fall.