Menu

Gagosian Quarterly

Summer 2023 Issue

A wild wild wind:Bennett Miller’s ai-Generated art

Benjamín Labatut addresses Bennett Miller’s engagement with artificial intelligence as a partner in the creation of a series of new artworks, asking what this technology—and its hallucinations—can reveal about our own humanity.

Bennett Miller, Untitled, 2022–23, pigment print of AI-generated image, 15 ¼ × 15 ¼ inches (38.7 × 38.7 cm), edition of 9 + 2 AP

Bennett Miller, Untitled, 2022–23, pigment print of AI-generated image, 15 ¼ × 15 ¼ inches (38.7 × 38.7 cm), edition of 9 + 2 AP

Benjamín Labatut

Benjamín Labatut is a Chilean writer born in the Netherlands in 1980. He is the author of The Maniac (2023), When We Cease to Understand the World (2020), La piedra de la locura (2021), Después de la luz (2016), and La Antártica empieza aquí (2012). Photo: Juana Gómez

Here we are—Mother Corn said:

We must hurry!

For the big Black-Wind is coming,

taking everything it meets.


A small herd of bison roams above what used to be America’s largest particle collider. Outside the fences that enclose them, strewn across the grounds of Fermilab, a six-thousand-acre physics complex in Batavia, Illinois, mathematical sculptures of various kinds dot the landscape: an Archimedean spiral, a geodesic sphere, a Möbius strip, and a staircase in the shape of DNA’s double helix. Fermilab is still the world’s premier facility for the study of neutrinos, subatomic ghosts that can pass through our entire planet without interacting with matter, but it is also a relic, since its largest structure—the Tevatron, a colossal atom-smasher—is silent and empty after it was shut down by the US government in 2011, when the Large Hadron Collider, Europe’s high-energy physics lab, rendered it obsolete. What used to be the crown jewel of American particle physics is now an abandoned four-mile-long underground tunnel. Its superconducting magnets no longer carry charge. Its giant detectors, fine-tuned to pick apart the products of protons and antiprotons colliding at close to the speed of light, are blind and oblivious. Its vacuum tubes fade to rust in a nearby rubbish dump. But the bison above continue to graze in silence, swaying their giant, noble heads and sniffing at the wind.

For the past several years, the film director Bennett Miller has been trying to capture an image of the future fast approaching. From 2015 till 2020, he sat down with many of the people who hope to shape it, men and women who are divining technologies with such fabulous potential to transform the human experience that even their early prototypes—such as the large language model ChatGPT, and digital image-creating systems such as dall·e 2, Midjourney, and Stable Diffusion—are already generating waves of panic among the fearful. Miller interviewed tech gurus, futurists, entrepreneurs, economists, CEOs, venture capitalists, artists, physicists, writers, biologists, computer scientists, poets, social workers, and artificial intelligence experts laboring to accelerate, amplify, or guide the evolution of the tools that will reshape civilization, trying to distill the essence of what society will look like in the coming decades into a six-part documentary entitled A Better World. It is a passionate, fascinating, and deeply unsettling portrayal, but due to one of the legal upsets so common in the film industry, audiences may never get to see it; however, the many questions it raises about the limits of science, the dangers of technology, and the fact that we may create entities that will slowly drive us into obsolescence have become unavoidable. The birth of advanced artificial intelligence, the creation of virtual realities, and the belief (no longer whispered but openly declared as an article of faith by many of the Silicon Valley elite) that our species, smart and wonderful though it is, may be just a step along the way toward a higher form of being are only a handful of the daemons that could soon come to haunt or enlighten us. All of these changes and transformations pose deep philosophical dilemmas that we don’t have answers for, questions that gnawed at Miller as he was trying to finish his documentary. What is the true nature of consciousness? What makes us human? Can we control our thirst for knowledge? Is there something about creativity that cannot be computed? What will beauty look like in a posthuman world? Can we really be mimicked and replaced by our own artifices? While these quandaries were stewing inside Miller’s head, A Better World hit a wall, and he saw that there was a definite chance that his many years of work could remain trapped forever inside the memory of his computer. At that point he stumbled across a particular type of artificial intelligence system that was completely unknown to the general public but that is now causing a storm, as it will surely have a profound impact on the arts: dall·e 2.

A Wild Wild Wind: Bennett Miller’s AI-Generated Art

Bennett Miller, Untitled, 2022–23, pigment print of AI-generated image, 15 ¼ × 15 ¼ inches (38.7 × 38.7 cm), edition of 9 + 2 AP

What is dall·e 2? I asked ChatGPT, another artificial intelligence program created by the same mother company, OpenAI. The AI responded,

dall·e 2 is a neural network-based image generation model. It is capable of generating images from textual input by using a combination of unsupervised and supervised learning. The model is trained on a dataset of images and their associated captions, and it learns to generate images that correspond to a given text description.

Have you heard of Bennett Miller?

I’m sorry, I don’t know any specific person that goes by that name. Bennett Miller is an Oscar nominated American film director, known for his films such as Capote (2005), Moneyball (2011) and Foxcatcher (2014).

I think he is summoning ghosts or wraiths using dall·e 2.

As an AI language model, I am not aware of any individual or organization that is using dall·e 2 or any other image generation model to “summon ghosts” or “wraiths.” dall·e 2 does not have the ability to summon ghosts or any other supernatural entities. It is not capable of performing supernatural tasks such as summoning ghosts or communicating with the future. Such claims are not based on any scientific or factual evidence, and should be treated with skepticism. AI and machine learning models are powerful tools, but they are not able to perform tasks that are beyond the realm of scientific explanation and understanding.

But I’ve seen them. With my own eyes!

The inner workings of dall·e 2 are inexplicable to a layman, but its outer shell, the program you interact with, is so marvelously simple and straightforward that using it feels like wielding magic. All you need to do is input text. This can be a random, nonsensical phrase, a line of prose—Charles Darwin’s sentence “In these still solitudes, Death, instead of Life, seemed the predominant spirit,” say—or a specific and meaningful instruction (“a close-up, detailed, full-color realistic portrait of a Templar knight in the style of Gustav Klimt, 9mm lens, 4k”), and dall·e 2 will create an image. You can then use the program to edit it, or ask for endless variations. And that is where the rabbit hole opens up and swallows you. Because never before in the history of our species has there existed a method capable of producing—on demand and in an instant—such a wealth of beguiling imagery: in less than a year, Miller has used dall·e 2 to generate more than a hundred thousand images.

The process starts with an idea. Perhaps it’s of a specific image in mind, but it can also be something vague. How will the AI interpret a particular phrase? How can dall·e 2 represent this style or that subject? Those ideas and intuitions must be turned into a written prompt. So in the beginning is the Word. But that is just a starting point. The key, according to Miller, is to arrive at a seed: an image generated by the program which can then be planted back into dall·e, and from which the AI might be coaxed to “hallucinate” in some intended direction. It is not a straightforward procedure at all. You can never be sure of what dall·e 2 will create, so randomness and chaos inevitably creep in. Miller worked with these seeds, prodding and manipulating the AI by altering them, erasing parts, editing, inputting new prompts and asking for variations, selecting the most resonant ones and seeing them evolve in strange ways.

From the monster horde of imagery that the AI produced, Miller has saved close to twenty thousand. He will exhibit only twenty at Gagosian, but every single one has something that we have rarely seen before in digitally generated images, something also lacking from a great deal of modern art: an undeniable aura. But what is that aura made from? Time. Loss. Pain. Grief and decay. Long-eroded memories. Bleached bones and bison skulls. Old things wrenched from oblivion’s embrace. Feathers. Tears. Butterfly wings. Forgotten spells, curses, chants. Songs of the scarecrow. Death’s silent stare.

Outside, unseen, beyond the borders of each frame, the wind howls. The wild wind that shamans ride. The voice of Tornado, roaring with a cry so loud and deep it can wake the dead. The body of Whirlwind Woman, beautiful and destructive, pulling us up, connecting the heavens and the earth, twisting away from what we know. The Black Snake Wind that talks to the elders. He tells us how many lives he will take and how destructive he will be. But he remembers the rituals and the language. Tornado is not evil; he resets the balance in nature.

A Wild Wild Wind: Bennett Miller’s AI-Generated Art

Bennett Miller, Untitled, 2022–23, pigment print of AI-generated image, 15 ¼ × 15 ¼ inches (38.7 × 38.7 cm), edition of 9 + 2 AP

The gravity of the past lends its weight to these images. It’s as if they’d been pulled out from darkness, conjured from the void with the aid of technology. They feel alien and familiar at the same time, like phantoms from a world that never was, souls caught by a neural net. Is this a horror show? A family album? A nightmare? A dreamscape? Miller’s masked children gaze at us from a long-lost era. Or have they come from the future, with a word of warning that they cannot utter, because their voices, unlike their images, cannot travel backward in time? That would explain their deep sadness: they embody a nostalgia for things that did not come to pass. Gargantuan wooden submarines whispering below the waves. Hot-air balloons rising into a sky still enlivened by the gods. In Miller’s images, even the silent features of the landscape—the yawning chasms of sleeping volcanos, the open wound of a gorge, the hollowed-out canyon dug deep into the earth—are altars that connect us to the cruelty, chaos, and magic of nature, and the people who used to live in such close proximity to its beating heart. Their melancholy and beauty come from their connection with the invisible realm of the spirit, one in which we no longer believe, but that we cannot live without.

Miller’s artworks are neither painted or drawn nor imagined; they have been invoked by strings of words. By using dall·e 2, Miller has somehow managed to capture ghosts. A phantasmagoria of men, women, and children. Dancing shamans and masked spirits, crow warriors jumping from cliffs, and gods both old and new, called down like UFOs. Some look like future phantoms already haunting the present, threatening and enchanting our lives. Some appear to be traces of what has retreated from the world. A few also seem to be a glimpse forward, a look beyond our doom; blurred by speed and distorted by movement, it’s as if Miller’s images were rushing ahead of us, running away from the light that we shine so readily upon ourselves. And yet we are moving forward so fast that we seem to be catching up with them.

Working with dall·e 2 definitely feels like interacting with an Other. Its products—which are flooding the Internet—have a disturbing quality that is hard to express. It arises, in some measure, from the AI’s “personality,” the series of quirks, glitches, and errors that it tends to make, and the particular way in which its model interprets the relationship between words and images. It has difficulty with hands, for example, and several built-in limitations, hard-coded by OpenAI to avoid the production of violent or pornographic images. These are hard to get around and can result in a certain uncanniness, a strange unfamiliarity that Miller, however, has fully taken advantage of in his artworks. But perhaps that weakness is a strength. It may well be the case that dall·e 2 and its many sister technologies can, in their profound otherness, be used to bring something back into our lives. Something glorious, powerful, and revelatory: a shamanistic approach to art, one through which we do not toil or sweat but ask and pray for inspiration. dall·e 2 is an almost magical tool for so many people who—like Miller himself—are endowed with a rare aesthetic sensibility that they hitherto could not manifest or fully express. By interacting with dall·e 2, these individuals are participating in a rare type of collaboration—because the AI does not have a mind of its own. Moreover, its creativity is (at least in part) derivative, because it arises from its training on many millions of images from the past, created by countless artists, photographers, and illustrators. And while this has given rise to criticism, the fact that, to create, dall·e feeds on the two things that humanity has developed, above all others, as vehicles to express our inner world—the word and the image—may prove a gift in itself; the head of a path that humanity and AI could walk down together, one where art and beauty become a common language, a shared form of communication that is desperately needed now that the future is so far behind us.

A Wild Wild Wind: Bennett Miller’s AI-Generated Art

Bennett Miller, Untitled, 2022–23, pigment print of AI-generated image, 15 ¼ × 15 ¼ inches (38.7 × 38.7 cm), edition of 9 + 2 AP

Miller knows that the version of the AI that he used for his show will soon cease to exist, knows it may become a relic. He can already feel it getting away from him. Because the AI is evolving, and the company that created it is altering the algorithm. But it could very well get worse. Tamer. They may hobble it in many ways, and deprive it of that special thing that Miller has somehow found in it: “This version of dall·e will never occur again. In part because the algorithm will ‘improve’, but also because the dataset it trained on is unique: it comes from our past. It is actual history recorded, captured by photochemical processes. Future algorithms, however, will draw from vast oceans of images and data generated by artificial intelligence programs. They will just as likely draw from their hallucinations as from records that have some fidelity to their origin. But I have no desire to preserve this version of dall·e. For me it’s enough that my images survive as artifacts, as a sort of time capsule of this rare moment in history when AI drew from a relatively pure source and took its first awkward steps before crossing the threshold into something unimaginable.”

The particular type of intelligence that film directors possess is one of the keys to understanding this collaborative form of art. To appreciate these strange, centaurlike products of man and machine, we must consider what a director brings to a movie set: not a script or a plan, not even a story, but a vision. That is why a great film—which requires an army of people, each lending a part of their creative energy—nevertheless carries the signature of its director. It is their unique vision, be it crystal clear or ill-defined, that marks the work indelibly. And this is also why the images Miller created with dall·e have a spirit of their own—because there is no will in the AI, nothing inside pushing to get out and take form. The sensibility, the hunger, the unconscious intuition, the drive, the sense of sadness and beauty belong to Miller. But he himself does not feel in control of the finished product. Not because he gives ownership to the AI, but because whatever drove him to create these images arose from a part of himself that he cannot place, and that he is almost unwilling to name, since the only word that human beings have ever used to describe it is now considered anathema for anyone truly modern and awake: the soul. This word, and everything it entails, which has been exiled from science and psychology, should be preserved in art. Soul, spirit, phantom: art is now one of the few vehicles by which we can express and channel the most mysterious aspects of our psyche. We do this in a language that we never really understand, but can only intuit, an ancient, secret language that—contradictory though it may seem—may well be recovered through technology. Because while some consider that tools like AI will only take us farther and farther away from ourselves, Miller’s work proves that this is not always the case. There are currents that flow back in time even as we race forward, there are unknown and unsuspected uses for even the most soulless devices and technologies. Yes, we have raced into the future with our eyes closed. Yes, we feel blinded by our own momentum. Yes, we no longer understand the very tools by which we control ourselves and the world. And yet the bison still roam.


ChatGPT’s responses were edited and cut down for size, but the content was not altered in any other way.

Bennett Miller, Gagosian, 976 Madison Avenue, New York, March 21–April 22, 2023

Black and white portrait of Lisa Lyon

Lisa Lyon

Fiona Duncan pays homage to the unprecedented, and underappreciated, life and work of Lisa Lyon.

Interior of Goetheanum, Dornach, Switzerland

Goetheanum: Rudolf Steiner and Contemporary Art

Author and artist Ross Simonini reports on a recent trip to the world center of the anthroposophical movement, the Goetheanum in Switzerland, exploring the influence of the movement’s founder and building’s designer Rudolf Steiner on twentieth-century artists.

Black and white portrait of Alexey Brodovitch

Game Changer: Alexey Brodovitch

Gerry Badger reflects on the persistent influence of the graphic designer and photographer Alexey Brodovitch, the subject of an upcoming exhibition at the Barnes Foundation, Philadelphia.

Installation view with Douglas Gordon, Pretty much every film and video work from about 1992 until now... (1999–)

Douglas Gordon: To Sing

On the occasion of Douglas Gordon: All I need is a little bit of everything, an exhibition in London, curator Adam Szymczyk recounts his experiences with Gordon’s work across nearly three decades, noting the continuities and evolutions.

Detail of Lauren Halsey sculpture depicting praying hands, planets, and other symbol against red and green background

Black Futurity: Lessons in (Art) History to Forge a Path Forward

Jon Copes asks, What can Black History Month mean in the year 2024? He looks to a selection of scholars and artists for the answer.

Inkjet print of Jeff Wall's "In the Legion" (2022)

Jeff Wall: In the Domain of Likeness

The Fondation Beyeler, Riehen/Basel, has staged a comprehensive Jeff Wall exhibition including more than fifty works spanning five decades. Here, Barry Schwabsky reflects on the enduring power of and mystery in Wall’s photography.

Portrait of Simon Hantaï holding on a blue and white canvas in his studio

Simon Hantaï: Azzurro Blue

In conjunction with Azzurro, an exhibition of paintings by Simon Hantaï at Gagosian, Rome, we share the catalogue essay by curator Anne Baldassari. Here Baldassari focuses on the significance of blue in the artist’s practice, illuminating his affinity with Italy and the influence on his work of its classical painting tradition.

poster for an exhibition with two figures

The Beginning: A Life in Art

Delphine Huisinga and Alison McDonald chart Larry Gagosian’s formative years on the West Coast and contextualize the Los Angeles art scene in the mid-1970s.

A Horse, of Course

A Horse, of Course

Alix Browne considers the enduring presence of horses in the contemporary imagination.

Poster art for Avanti! (1972), directed by Billy Wilder

Kiss Me, Stupid

Carlos Valladares mines the history of the romantic comedy and proposes an expanded canon for the genre.

Sarah Sze: Timelapse

Sarah Sze: Timelapse

Francine Prose ruminates on temporality, fragility, and strength following a visit to Sarah Sze’s exhibition Timelapse at the Solomon R. Guggenheim Museum, New York.

A graphite crayon on paper self portrait of Balthus from 1943

Between Shadow and Light

Scholar and researcher Yves Guignard, who is working on Balthus’s archives for a revision of the Balthus catalogue raisonné, examines the artist’s engagement with drawing, arguing for a more concerted attention to these works than scholarship has paid them.