Stochastic Histories

stochastic histories /stəˈkæstɪk ˈhɪstəriz/ n. pl. [from Gk. stokhastikos “capable of guessing,” from stokhazesthai “to aim at, guess at” + L. historia “narrative, account”] 1. Fabricated historical narratives generated through iterative probabilistic sampling with artificial-intelligence systems, producing documentation that appears authentic through the accumulation of plausible details. 2. A mode of artistic practice that exploits AI’s pattern-matching capacities to reveal how all historical knowledge operates through stochastic processes of inference and reconstruction. 3. Speculative archives that occupy the space between memory and invention, demonstrating that the distinction between genuine and generated documentation has become computationally irresoluble. First usage: K. Varnelis, 2025. See also: pattern-making, hallucination, probabilistic inference.

This is a personal essay, a position statement on my recent work with AI. Ultimately, as with all of my work, it is for myself. It a retrospective piece that makes sense of my work with AI in a way that I had not done before. I am not sure I will continue working in this vein or head in another direction in photography or technology, leaving specific investigations of AI. It still, however, may prove useful to readers interested in the topic of AI and human creativity.

Since 2021, I’ve called my AI-generated pieces Critical AI Art. It’s a terrible name—misleading and increasingly intolerable as the work has evolved. The time has come to rename it Stochastic Histories, a title that reflects what the work actually does. Critical AI Art misaligns my practice with two opposing but ultimately identical failures: the prosumer kitsch of Midjourney prompt jockeys producing fantasy waifus and hyperreal landscapes, and the institutional theater of “critical” artists performing algorithmic audits for progressive credibility. Despite surface differences, both serve the same void. One packages it as technical novelty, the other as moral virtue. Neither changes or reveals anything. Both exist to be consumed and forgotten.

The Midjourney jockeys string together technical specifications they don’t understand—“hyper-detailed, ArtStation, Octane render, volumetric lighting, 8K”—cargo cult incantations for the aesthetically degenerate. They’ve created an arms race where every pore must be visible, every hair strand individually rendered, as if resolution could substitute for vision. Their subject matter never varies, the same collision of sci-fi, anime, and softcore imagery, endlessly recombined in a simulation of the “what’s new” feed on DeviantArt. As I detailed in my essay “California Forever or, the Aesthetics of AI Images,” these are prosumers in Alvin Toffler’s sense, merging producer and consumer roles to create kitsch marked by mass appeal and derivative aesthetics. Such work triggers the same neurological response as slot machines: just novel enough to release dopamine, just familiar enough to require no thought. Their work circulates as novelty, accumulating likes and shares for images of Emma Watson, Cannabis Goddess of Mars, bathed in golden-hour lighting, a modern-day version of Andres Serrano’s Piss Christ, while contributing nothing to genuine artistic inquiry.

Meanwhile, self-proclaimed “critical” AI artists operate from the same fundamental emptiness, though with institutional representation secured through their positions at American universities or funded by grants from the Creative Europe Programme. Whether the work involves AI or not is irrelevant: walk into any contemporary art institution and you’ll find the aesthetics of the corporate diversity training module transplanted to white-cube walls. Every exhibition includes positionality statements, as if confessing privilege or claiming intersectionality absolved them of participating in the very systems they purport to critique, together with declarations listing the tons of carbon expended to exhibit the work. They claim to “center” the “voices” of the “marginalized” while speaking to the elite, producing “disruptions” that don’t disrupt, “interrogations” that never wait for an answer. Rest assured: surveillance systems everyone already knows exist will be exposed, and algorithmic biases that have been common knowledge for years will be uncovered. Curatorial heads will not roll; on the contrary, they will nod, and appropriate boxes will be checked. This moralizing approach reduces art to forensic reports on systemic oppression, as if the purpose of art were to prove what we already know about power rather than to discover what we don’t yet understand. They perform the rituals of exposure while knowing these gestures have become empty theater, that their “interventions” will be absorbed into the academic-industrial complex as another line on a CV, another grant secured, another exhibition lets everyone feel they’ve confronted power while leaving all structures intact. Now that this crowd has come to AI, their work continues in the same vein, offering the usual litany of complaints from yesterday’s war.

As Peter Sloterdijk diagnosed in his Critique of Cynical Reason, we live in an age of enlightened false consciousness: people know very well what they are doing, but they do it anyway. The critical intellectual who exposes ideological operations pretends their audience is naïve, but the audience, like the intellectual, is already cynical—both know the game is rigged and continue playing it. Critiques persist even though such gestures change nothing. are made even though such gestures change nothing. Institutions, too, have learned to incorporate their critique as immunization—funding the very work that claims to expose them, thereby proving they can take it, neutralizing any actual threat. The art world pays for its own flagellation, emerging stronger and unchanged. Both camps—the Midjourney jockeys and the critical artists—exhibit what Sloterdijk calls the unhappy consciousness of modern cynicism, aware their work is empty but unable to stop producing it. The real revelation is not that technology reflects the biases of its training data. It is that all cultural production has always been a stochastic process, and the panic about technological “harms” is itself a symptom of our refusal to acknowledge that creativity was never the province of individual genius but has always emerged from the vast, impersonal machinery of cultural recycling.

But all this hardly matters. We are in the post-AI era. Adobe Photoshop now includes generative AI as just another tool. Microsoft Word now includes an AI Assistant. It is no longer a novelty worth celebration or critique. Doing anything interesting with it requires treating AI not just as a medium but as a distinct mode of production. Unlike mechanical reproduction—copies of originals—or mass media—singular messages broadcast to many—or even digital computation, which simulated reality through deterministic processes, AI operates through probabilistic extraction and recombination. It ingests the totality of digitized human expression, identifies patterns across this vast archive, and generates new instances that never existed but feel as if they should have. Its hallucinations—those moments when pattern matching produces impossible responses—aren’t failures but the source of its creative potential. Crucially, AI reveals what was always true but hidden: all cultural production operates through similar stochastic processes of pattern recognition, recombination, and variation.

This term stochastic histories deliberately plays off Emily Bender’s dismissive “stochastic parrots.” “Stochastic” means probabilistic; “parrot” means imitation. Together, her phrase describes how large language models merely repeat patterns of language learned from vast datasets, generating fluent but potentially meaningless or biased output through probabilistic sampling. But unlike Bender, I was trained as a historian of cultural production. Where she sees mindless repetition, I see an unintended insight into how culture itself operates. Stochastic histories are fabricated narratives produced through probabilistic sampling with AI systems. Each project requires months of work—thousands of prompts refined and recombined until coherent worlds crystallize from the collision of probabilities. I do not select the resulting texts and images merely for technical quality but also for their capacity to provoke a kind of productive uncertainty. The term also recognizes that all historical and speculative production—not just that produced by AI—operates through similar processes of pattern-matching and plausible inference, constructing narratives from fragments and filling gaps with what must have been there or what might yet be. The “parrot” was never the insult critics imagined; parrots are remarkable pattern-matching machines, and so are we.¹

My stochastic histories lie within a specific lineage, both in broader movements and in the circumstances of my own biography as an artist and theorist. For me, the key departure point is David Wilson’s Museum of Jurassic Technology, which I first encountered in 1996 and have returned to repeatedly (the Museum is next door to the Center for Land Use Interpretation, with which I began collaborating with the late 1990s). The Museum presents detailed exhibits on obscure scientists, bizarre natural phenomena, and peculiar inventions—some real, some fabricated, most existing in ambiguous spaces between documentation and invention. Wilson, who had a previous career in Hollywood special effects and understands how reality is constructed frame by frame, does not critique truth so much as create a space where epistemic certainty dissolves. I once had the rare opportunity to see the backstage of the Museum, where the exhibition rooms were revealed as theatrical flats within a vast interior space, the vitrines accessible from hidden corridors for maintenance—the museum itself a stage set masquerading as architecture, which in turn throws into question the relationship between display and servant spaces in traditional museums. The desire to know whether something is “real” becomes the content, but simultaneously, standing in the Museum’s quiet, darkened galleries, gazing at exhibits while hearing distant sounds—a fox screaming, crickets chirping, a slideshow narrator explaining a preposterous history of the museum, the haunting voice of an opera singer, bells drifting from other rooms—one becomes acutely aware of one’s own body, achieving what the minimalists of the 1960s sought but far more effectively. The experience reveals something crucial: unlike the critic who maintains distance to perform exposure, the Museum makes visitors aware that the very act of evaluating makes them constituent parts of the work.

In Mr. Wilson’s Cabinet of Wonder, Lawrence Weschler connects the Museum to the pre-Enlightenment Wunderkammern—Renaissance collections where natural specimens, artistic creations, and technological marvels existed before disciplinary boundaries solidified. Cabinets of curiosities emerged from the era of natural philosophy, when investigating nature meant contemplating the world in all its marvelous variety, when the boundaries between the miraculous and the mechanical, the divine and the empirical remained productively unclear. The Wunderkammer took the stage alongside another technology for managing early modernity’s information explosion: the book index. Both arose in the sixteenth century as print culture and global exploration overwhelmed existing systems of organization, creating too many books to read linearly and too many objects to fit familiar categories. Natural philosophers assembled these collections not as data sets but as chambers for contemplation, where a unicorn horn (actually a narwhal tusk) might reveal divine providence as surely as it demonstrated nature’s ingenuity. Like the index that allows readers to access text through alphabetical entry points, the cabinet’s drawers and shelves permitted nonlinear exploration. Just as index entries create connections across distant pages, meaning emerged through juxtaposition rather than sequence—a narwhal tusk might sit beside a mechanical duck, an ostrich egg beside a dodecahedron carved from ivory, each object a node in a network of correspondences between microcosm and macrocosm. The wonder of the Wunderkammer did not so much fix meaning as create a productive uncertainty, a cognitive state of openness. These collections revealed that knowledge itself requires an architecture—literal shelves and cases, but also conceptual frameworks that determine what belongs with what, what counts as evidence, what deserves preservation.²

Wilson’s Museum extends this logic into contemporary practice, but with a crucial difference: where the Wunderkammern cultivated wonder through genuine rarities and (usually) honest misidentifications, Wilson achieves this through what the Russian Formalists called ostranenie—defamiliarization in the form of a meticulous institutional fiction. At his Museum, the vitrine functions as a technology of belief: placing something behind glass with a label does not simply display knowledge; it produces it, transforming objects into evidence, the everyday into the remarkable. His Museum operates as a three-dimensional index, where each exhibit points to others through conceptual correspondence rather than alphabetical order, making visible what the book index conceals: that all systems of organization are arbitrary impositions on the universe of things and ideas. The Museum reveals that our frameworks for organizing knowledge are themselves fictions, that the difference between the Leiden University anatomical collection and the Museum of Jurassic Technology is one of degree, not kind. By faithfully reproducing every convention of museum display—the authoritative labels, the scholarly apparatus, the hushed galleries—while filling them with impossible content, Wilson makes the familiar institutional frame suddenly alien. We begin to notice the arbitrary confidence of wall text, the coercive authority of the vitrine, and the way dim lighting and reverent silence produce belief regardless of what they frame. Through careful presentation, the plausible feels impossible and the impossible feels documented, but more importantly, the very distinction between these categories becomes unstable.

At this point, it is also worth citing Joseph Cornell’s boxes, which function as both precursors and miniature, portable counterparts—private Wunderkammern assembled from the detritus of consumer culture. Cornell explicitly referenced the tradition, but where Renaissance collectors displayed unicorn horns and ostrich eggs as trophies of colonial reach, Cornell assembled his cosmologies from hotel advertisements, celestial maps, and scraps found in New York’s junk shops. His boxes did not fabricate evidence like Wilson’s Museum; they transformed genuine ephemera into hermetic worlds through careful juxtaposition, creating wonder from the worthless.

It seems beyond coincidence, then, that in Count Zero, cyberpunk author and pundit William Gibson centered his novel around an AI that creates Cornell boxes. In a 1989 interview, Gibson explained that Cornell’s assemblage method was “the key to the whole fucking thing, how the books are put together and everything”—describing his own creative process as “grabbing little hunks of kipple, and fitting them together.” The Boxmaker wasn’t just a character; it was Gibson’s explicit metaphor for how novels get made, for how all cultural production operates through stochastic assemblage from existing fragments. Today, any image generator can produce convincing Cornell-style boxes in seconds, complete with nostalgic patina and surrealist juxtapositions. Gibson himself now recoils from AI art as “weird as hell.” This is not mere contradiction but reveals something structural: Gibson’s blindness to the implication of his own insight is not a failure of reasoning but a constitutive necessity—the concept of “human creativity” requires that the machinery remain invisible. The moment AI makes pattern-matching and probabilistic recombination explicit and operational, the entire framework collapses. Gibson’s position is literally impossible to maintain coherently, which is precisely what is “weird as hell.” What appears as personal inconsistency exposes how meaning, authorship, and artistic value have always depended on a foundational blindness to their own mechanical operations. We cannot simultaneously acknowledge that creativity operates through stochastic probabilistic processes and preserve the human/machine distinction—yet we must try, because abandoning either claim threatens the entire edifice of how we understand cultural production.³

Cornell dedicated two works to another precursor of stochastic histories, Jorge Luis Borges—”For Jorge Luis Borges,” a collage, and “The Puzzle of the Reward (for Jorge Luis Borges)”—creating a direct link between the assembler of found objects and the architect of impossible libraries. Where Cornell transformed commercial detritus into hermetic worlds through physical juxtaposition, Borges achieved similar effects through textual accumulation. His Tlön, Uqbar, Orbis Tertius presents an entire fictional world that gradually replaces reality through sheer documentary weight—encyclopedic entries, scholarly footnotes, and discovered objects bearing inscriptions in imaginary languages. The story isn’t just about a fictional world but about how fiction becomes real through the apparatus of scholarship, how plausibility emerges from the proper citation format and cross-reference.

Pierre Menard, Borges’s character who rewrites Don Quixote word for word three centuries after Cervantes, reveals that identical repetition produces difference—the same words mean something entirely different when produced in a different era, by a different hand. This isn’t plagiarism but a theory of cultural production: all writing is rewriting, all creation is recombination. The Library of Babel, containing every possible book, suggests that originality is merely a matter of selection from a pre-existing field of possibilities. Borges understood that culture constructs itself through pattern and variation, that authenticity and fabrication are categories we impose rather than qualities inherent in things.

This tradition of productive fabrication extends through various methods and media. Contemporaneous with Wilson, the photographer Joan Fontcuberta constructs entire scientific archives with obsessive thoroughness. His Fauna presents complete taxonomies of impossible animals—the Solenoglypha Polipodida, a snake with legs, documented through skeletal diagrams, field notes, habitat studies, and photographic “evidence.” His Sputnik project (I purchased a copy at the Museum of Jurassic Technology) resurrects Ivan Istochnikov, the lost Soviet cosmonaut, playing directly with the persistent conspiracy theory that cosmonauts died in space before Gagarin, erased from official history. Fontcuberta provides everything—mission patches, crew photographs where Istochnikov has been carefully removed, leaving ghostly gaps, technical documents, commemorative stamps, even childhood photographs. The project gains power precisely because it builds on existing suspicions about Soviet secrecy, making viewers question whether this is revelation or fabrication. Many refuse to believe it’s fiction even after being told, insisting that Fontcuberta must be exposing actual suppressed history. Every element mimics the authoritative grammar of scientific documentation. Fontcuberta doesn’t critique truth so much as demonstrate how truth effects are produced through proper formatting, comprehensive citation, institutional presentation—and how they become even more convincing when they confirm what we already suspect.

Another artist whose work was essential for the world-building nature of my stochastic histories is Christoph Büchel, whose exhibit at Los Angeles Contemporary Exhibitions(LACE) in the fall of 2003 affected me deeply. It seemed, upon approaching the gallery, that some kind of horrific event had happened at its Hollywood Boulevard space—unsettling in the post 9/11 years—the windows were boarded up, but a representative appeared and had us fill out an unsettling disclaimer absolving the gallery of liability. We entered into a darkened room that clearly once was the gallery, but now was strewn with refuse and rubble. Light came from a door that, when opened, led into a disturbingly small, well-lit space with a locker at the end, an institutional space, perhaps a school or factory or military installation. The place appeared to have seen better days: some tiles in the floor had broken away and an animal hole led off somewhere through the dirt. Eventually, we discerned that we could enter into the locker, crouch down into a hole, and emerge under a desk in another, much larger room (again all this is happening in what we expected was the white cube of the gallery we had visited many times). This room was connected to a large complex of rooms filled with what appeared to be missiles, military equipment, and institutional furniture, an unsettling space with the air of Cold War menace, an excavation into another world unknown to civilians. This wasn’t a representation but rather a total environment, removing all reference to the gallery. Where Wilson’s Museum maintains the institutional frame while making it uncanny, Büchel eliminates the frame entirely.

There is architectural precedent for stochastic histories as well. In particular, I have found the work of the Italian radical architecture group Superstudio compelling. Even though modernists had worked with media from the start, the wide accessibility of color printing by the late 1960s meant that saturated, glossy color had become the norm. Superstudio took the medium itself as a site, understanding that mainstream architectural publications such as Domus and Casabella would be as important as the gallery. The glossy magazine page became a space where fictional architectures could exist with the same presence as documented reality. Carefully selecting documentary photographs to build upon, Superstudio understood that photomontage carries different authority than drawing—the photographic substrate insists on its own truth even when obviously manipulated. Their method was explicitly critical, showing how modernist rationality taken to its logical extreme would produce a world of total homogenization. Yet, the critique worked through seduction rather than revulsion, through images so compelling they made dystopia feel inevitable, even desirable.

The most perverse precedent is that of architect Jean-Jacques Lequeu, whose drawings I had the privilege to see at the Morgan Library a week before the pandemic lockdown. Lequeu’s drawings combined Neoclassical rigor with fantastical, often erotic and transgressive imagery that remained unknown until their rediscovery in the 1930s by Emil Kaufmann, who encountered them in the Bibliothèque nationale de France in Paris. His work includes visionary designs for temples, grottos, and civic buildings that blend archaeological precision with theatrical excess, featuring allegorical figures, exotic references, and occasionally explicit sexual content. But in the only book-length study on Lequeu, French architecture historian Philippe Duboy claims that Marcel Duchamp, while working at the Bibliothèque, fabricated Lequeu’s drawings, attributing them to this third-rate figure from the revolutionary era. According to Duboy, Duchamp invented this 18th-century visionary architect complete with theoretical texts and fantastical building designs. Many scholars dismiss this as conspiracy theory, but their rebuttals seem to rest on little more than the audacity of Duboy’s claim. The vertigo here is absolute: either we’re looking at authentic 18th-century visions that anticipate surrealism by a century, or we’re looking at Duchamp’s forgeries that have been absorbed into architectural history as fact. The authentication machinery—carbon dating, stylistic analysis, archival provenance—cannot resolve this; it can only accumulate probabilities that feel like certainties. This conundrum shows how modernism might construct its own prehistory through forgery, how we recognize certain patterns as “visionary” because they match what we’ve been trained to expect from an outsider genius.⁴

Where these predecessors employed physical assemblage, institutional mimicry, photomontage, or architectural intervention to produce their fictions, I work with AI’s probabilistic sampling—a method that doesn’t construct or manipulate but rather summons plausible realities from latent space, making the process of authentication itself the medium. Stochastic histories emerge from this tradition, but also respond to our own time, when AI has become inextricably entwined with cultural production. This work, however, is not merely an interrogation of AI. Stochastic histories set out to produce a particular vertigo—not simply uncertainty about what is real, but the recognition that this is exactly how the real gets constructed. The AI hasn’t learned to fake history; it has learned history’s own methods of self-authentication. When we cannot distinguish the generated from the genuine, we’re forced to confront how all archives operate through probabilistic accumulation of plausible details. The dataset is Benjamin’s archive in its raw form, stripped of curatorial narrative—every photograph that has ever been digitized, every document scanned, every architectural drawing uploaded, all flattened into latent space. We find ourselves staring directly at culture’s unconscious, and in that moment we glimpse something like the Lacanian Real—the point where representation collapses and the machinery of meaning-making becomes visible.

My process embraces both chance and curation. Each image demands hundreds of iterations, each prompt a negotiation with probability space. Although I aim to produce works that appear plausibly real, the stochastic element of this production nevertheless  reveals itself most powerfully in glitches—the hallucinations where the system generates mistakes—a three-story-tall streetlight, a misshapen human figure, a façade in the process of digital decay. These glitches show the moments where the probability field tears and something genuinely alien emerges. A building that shouldn’t exist but feels absolutely inevitable. A document that describes events that never occurred but should have. These hallucinations are terrifying precisely because they feel more plausible than fact—they reveal how thin the membrane is between history and its probabilistic shadow.

One reader of the latest revision of the Canals of Vilnius essay was disappointed by the quotes about the fire by Alessandro Cilli, a Jesuit priest (note that these are not in the current version, but will appear next week). She knew that I wrote the fictional history with help from an AI, and she felt the AI’s writing was unconvincing. And yet, Cilli’s quotes are all real: they seem unreal, too strange to be believed.

Stochastic histories make visible that the past is continuously reconstructed through pattern and probability, that authenticity is a learned aesthetic, that culture reproduces itself through endless stochastic variation. Cilli’s quotes seem unreal while Lequeu’s drawings are considered authentic by most scholars because the possibility that Cilli’s quotes are real and Lequeu is a forgery by Duchamp is too unsettling. We are all stochastic parrots. Every historical narrative, every archival discovery, every recovered memory operates through the same probabilistic logic—filling gaps with what should have been there, constructing continuity from fragments. Meaning has always emerged from vast, impersonal processes of pattern matching and recombination. Thus, the old saw that the dataset is biased misses the point entirely—it is literally culture’s biases made operational, our collective patterns of recognition transformed into probability distributions. When I work with these systems, I’m not using a tool but inhabiting a space—a vast probability field containing every possible history that could feel real. 

The recursive nature of this practice reveals itself in the “7 Fables of Accelerationism,” where AI systems trained on archives containing Superstudio’s imagery from the 1960s generate new speculative futures that cite those radical visions. Here, I project stochastic histories into both the future and the past. Just as Superstudio’s photomontages pointed back to the constructivists and early modernists whose visual language they appropriated, my fables emerge from probability spaces where the Continuous Monument—as well as the images that Superstudio trained themselves on—exist as patterns to be recombined as well as conditions they could not have anticipated. The dataset doesn’t just enable the work; it determines its visual genealogy, making every generated image a stochastic descendant of its training data.

Stochastic histories operate in the space between memory and invention, between what was and what should have been, between human intention and machine hallucination. They’re not critiques or spectacles but investigations into how the past gets constructed, how the future gets imagined, how culture reproduces itself through endless variation. Turning Gibson on his head, I very much hope that these works are “weird as hell.” In revealing how probability shapes our expectations and how the line between history and hallucination dissolves at the level of pattern-matching, they make visible that culture has always been a stochastic process, and that being a stochastic parrot might well be the only kind of consciousness there is.

¹ See The Generative Turn: On AIs as Stochastic Parrots and Art for a fuller discussion of how structured repetition enables rather than constrains creativity. ↩

² Lawrence Weschler, Mr. Wilson’s Cabinet of Wonder (New York: Pantheon Books, 1995). ↩

³ Genevieve von Petzinger. The First Signs: Unlocking the Mysteries of the World’s Oldest Symbols. (New York, NY: Atria Books, 2016). ↩

⁴ Phillipe Duboy, Lequeu: An Architectural Enigma (Cambridge: The MIT Press, 1987). ↩