Surrealism began with a problem that now sounds unexpectedly contemporary: how can an artist make something without allowing conscious intention to control every decision? André Breton’s 1924 Manifesto of Surrealism proposed psychic automatism as one answer. André Masson let his pen move across paper without beginning from a predetermined composition. Surrealist writers used automatic writing. Artists experimented with chance, collage, decalcomania and photographic processes that allowed unexpected forms to interrupt deliberate planning.
A century later, generative artificial intelligence introduces a different kind of surrender. An artist can enter an instruction, provide images or source material, set parameters and receive results whose individual visual details were not manually drawn. That sounds superficially close to automatism. Yet the mechanism is radically different.
Stockholm’s Moderna Museet has created an unusually productive place to examine that difference. The Subterranean Sky – Surrealism in the Moderna Museet Collection opened on October 26, 2024, marking the centenary of Breton’s manifesto, and is currently scheduled to remain on view through January 17, 2027. The exhibition combines nearly 200 works from Moderna Museet’s collection with around 30 loans, moving from Claude Cahun, Wifredo Lam, René Magritte, Meret Oppenheim, Man Ray and Toyen to contemporary artists including Agnieszka Polska, Tarik Kiswanson, Thale Vangen and Fatima Moallim.
Most provocative for the AI debate, the museum explicitly includes Polska’s AI-generated video works beside Moallim’s automatistic line drawings. Moderna’s The Subterranean Sky exhibition therefore makes a question visible that Surrealism itself could never have anticipated: if automatism once meant loosening the artist’s conscious control, what happens when part of that control is handed to an algorithm?
Breton’s Automatism Was About The Human Mind
It is tempting to describe AI generation as the newest form of automatic art.
The comparison needs limits.
Breton’s definition of Surrealism centered on psychic automatism. His project was concerned with thought, dreams, free association and the unconscious. Surrealists wanted to weaken rational censorship so material that normally remained suppressed, accidental or psychologically inaccessible could enter artistic production.
André Masson’s Automatic Drawing (1924) provides one of the clearest visual examples. According to MoMA’s account of Masson’s automatic drawing, he began without a preconceived subject or composition, moved his pen quickly and allowed fragments of bodies and objects to emerge from an initially abstract network of lines.
Masson did not become absent from the process.
That distinction matters.
He recognized forms as they appeared. He occasionally developed them. He decided when an image was interesting enough to continue. Automatism reduced one kind of intentional control while leaving many other forms of artistic judgment intact.
The Surrealists were not trying to create an artwork with no human authorship.
They were trying to discover parts of human mental activity that deliberate reason might suppress.
Generative AI does something else.
A model does not reveal an artist’s unconscious merely because the artist cannot predict every pixel it will produce. Its output emerges from computational operations shaped by training data, model architecture, parameters, prompting and subsequent human selection.
Unpredictability is shared.
The source of that unpredictability is not.
The Subterranean Sky Treats Surrealism As A Method, Not A Look
Moderna Museet’s curatorial framing helps avoid another common mistake: defining Surrealism by bizarre imagery alone.
The museum describes Surrealism as a revolutionary literary and artistic movement rather than a particular style. Dreams, desire, chance, free association, occultism and psychoanalysis appear throughout the exhibition, yet the visual results vary enormously.
That definition is useful for AI art.
A generated image containing floating clocks, hybrid creatures and impossible architecture is not Surrealist simply because it resembles familiar Surrealist imagery. Such an image may imitate the surface appearance associated with the movement without participating in its intellectual project.
The more interesting question concerns method.
Does the artist use a system to disturb habitual thought?
Does chance create a result that forces the artist to respond differently?
Does the process reveal associations the artist did not deliberately plan?
Does the final work question rational assumptions about reality, identity, desire or perception?
These are more demanding tests than asking whether an AI image looks dreamlike.
Moderna’s exhibition makes this distinction through historical breadth. Classical Surrealists appear beside Fluxus, conceptual art, film, photography and contemporary practices. Curator Lena Essling and assistant curator Corina Oprea have constructed the exhibition around Surrealism’s continuing mutations rather than a fixed visual formula. The museum’s exhibition press material identifies contemporary AI video, automatic drawing, hypnotic sculpture and organic objects as different ways that Surrealist questions persist.
AI enters the exhibition as one branch of that longer history.
It does not become the destination of Surrealism.
Agnieszka Polska Uses AI To Build Impossible Histories
Agnieszka Polska provides a much stronger test case than generic text-to-image generation.
Born in Lublin, Poland, in 1985, Polska works primarily with film, animation and digital image-making. Her projects examine technology, information, social responsibility, ecological relationships and the strange emotional effects produced by mediated images.
One of her important recent works is The Book of Flowers (2023).
The film uses artificial-intelligence tools together with archival flower imagery. The Adam Mickiewicz Institute describes Polska’s process as beginning with 1950s stop-motion footage of flowers, which she transformed using AI-based tools. The resulting film imagines an alternative history in which flowering plants and human beings have existed in unusually close symbiosis for thousands of years. The institute’s Agnieszka Polska exhibition overview explains how the work combines technological manipulation with speculative storytelling.
The result is visually strange, but strangeness is not its most Surrealist quality.
Polska begins with recognizable historical material.
The algorithm destabilizes it.
Flowers develop forms that can seem animal, mechanical or biologically impossible. Documentary-looking imagery drifts into speculative fiction. Something that resembles archival evidence begins describing a history that never happened.
That operation has a real affinity with Surrealist strategies.
Max Ernst assembled familiar 19th-century engravings into impossible collage narratives. Man Ray let ordinary objects become ambiguous shadows in rayographs. Surrealist photographers altered scale, orientation and context until reality stopped behaving conventionally.
Polska’s technical process is different, but the structural move is recognizable: existing visual information is passed through a system that breaks its original meaning.
Readers interested in an earlier technological version of that disturbance can compare Polska’s process with thesurrealists.org’s examination of Man Ray’s photographic innovations. Rayographs and generative models belong to radically different technological periods, yet both raise questions about how much an artist must directly control for an unexpected image to become artistically useful.
Fatima Moallim Shows What Human Automatism Still Looks Like
The most revealing comparison inside The Subterranean Sky may not be between Polska and Breton.
It may be between Polska and Fatima Moallim.
Moallim, born in Moscow in 1992 and based in Stockholm, works across drawing, performance, video and site-specific installation. One of her recurring methods involves drawing directly onto walls or large surfaces while allowing the line to develop in response to the immediate surroundings.
Moderna Museet describes Family Album (2021) as an example of her performative automatic drawing. Figures and delicate lines spread across gray linoleum, forming an alternative to a conventional family album. The work connects with memories of migration and family history, including her parents’ flight from Mogadishu to Moscow and later Sweden.
The museum has described her drawing process as one in which a line gradually forms a narrative while the work is being made. Moderna Museet’s material on Fatima Moallim documents how her practice joins spontaneous drawing with migration, alienation and memory.
This is much closer to classical automatism.
The body acts.
The line appears.
The artist reacts.
Memory and form become entangled during the performance.
Place Polska’s AI-assisted moving images beside Moallim’s physical line and an important distinction emerges.
Both processes can generate unforeseen forms.
Only one depends on a generative computational model.
That does not automatically make one more artistic than the other. It does mean that the word automatic now describes processes with very different relationships to human agency.
AI Is Not An Electronic Unconscious
The most seductive argument around AI and Surrealism may be the weakest one: that a generative model functions like a technological unconscious.
The metaphor works visually.
Generative systems can produce bizarre associations. Objects merge. Anatomical structures mutate. Architectural spaces collapse into one another. A prompt can return combinations the user did not anticipate.
Yet resemblance does not establish equivalence.
Surrealist automatism was tied to theories of human thought. Breton and his circle believed that reducing rational control could bring normally inaccessible mental associations closer to the surface.
There is no comparable basis for saying that a generative model possesses a subconscious in the human psychological sense.
What it does possess is an ability to generate new outputs from patterns learned during training.
This difference becomes especially important when AI systems produce imagery that feels uncannily familiar. A dream can recombine memories from a person’s experience. A generative model can recombine statistical relationships derived from vast quantities of training material.
The mechanisms should not be confused simply because both can produce surprise.
The better Surrealist question is not whether the machine dreams.
It is what the human artist does with the machine’s unpredictability.
Chance Alone Never Made Something Surrealist
Surrealism has always contained a complicated relationship with chance.
Automatic writing loosened conscious planning. Decalcomania allowed unpredictable textures to appear through transferred paint. The exquisite corpse prevented individual participants from seeing the whole drawing while adding their contributions.
Yet the Surrealists still selected, interpreted, published, exhibited and contextualized results.
Chance was a method.
It was not a substitute for artistic thought.
That distinction becomes decisive with generative AI.
Typing a prompt and accepting the first result may involve unpredictability, but unpredictability alone does not create a Surrealist artwork. A shuffled playlist is unpredictable. So is a software error.
Artistic significance develops through framing, selection, modification and context.
Polska’s The Book of Flowers provides a useful example. The AI-generated transformations belong to a larger constructed narrative about humans, plants, ecology and imagined history. The technology operates inside an authored film rather than replacing the need for authorship.
The U.S. Copyright Office reached a related, though specifically legal, distinction in its January 29, 2025 report on AI and copyrightability. The Office stated that using AI as an assistive creative tool does not prevent copyright protection for human-authored aspects of a work, but that prompts alone generally do not establish sufficient human control over expressive elements. Its AI copyrightability guidance is not a definition of art or Surrealism, yet it highlights the same practical divide between using a system and determining the expressive form of the resulting work.
The artistic question remains broader than copyright.
Still, human decisions matter.
Training Data Creates A Problem Automatism Never Had
There is another major difference between a pen moving automatically and a generative model producing images.
The pen does not contain millions of earlier artworks.
Generative models are trained on large datasets. That means discussions of AI-assisted Surrealism cannot stop at chance or unconscious-looking imagery. They must include questions about source material, licensing, attribution and creative labor.
This is where romantic comparisons between AI and automatism can become misleading.
When Masson drew automatically, the ambiguity came through his own physical action and psychological associations.
When a generative model produces an unusual image, part of the result depends on relationships learned from its training data.
That fact does not dictate whether a particular AI-assisted artwork succeeds aesthetically. It changes the ethical and authorship framework surrounding the process.
The U.S. Copyright Office’s continuing artificial intelligence study separates copyrightability from the distinct question of generative-AI training. Its 2025 reports make clear that these remain active policy issues rather than settled artistic permissions.
For contemporary Surrealism, this creates a boundary Breton never had to confront.
Breaking rational control is one question.
Whose material helps the machine generate that break is another.
The Exhibition’s Older Works Stop AI From Looking Historically Unique
One of the strengths of The Subterranean Sky is that AI does not appear in a technological vacuum.
The exhibition contains nearly 200 collection works and around 30 loans spanning Surrealism’s historical development and later artistic consequences. Claude Cahun, Wifredo Lam, René Magritte, Meret Oppenheim, Francis Picabia, Man Ray and Toyen appear beside later figures including Leonora Carrington, Maya Deren, Graciela Iturbide, Robert Rauschenberg, David Alfaro Siqueiros and Dorothea Tanning.
Conceptual art and Fluxus extend the story further.
Then come Polska, Kiswanson, Vangen and Moallim.
Seen in that sequence, AI becomes less unprecedented.
Artists have repeatedly built systems for disrupting intention.
Cameras produce optical information the hand cannot duplicate.
Darkroom experiments alter photographs.
Collage imports images the artist did not create.
Film editing makes separate moments collide.
Performance introduces bodily contingency.
Fluxus scores establish rules whose results depend on whoever performs them.
AI is different from all of these.
It is not completely disconnected from them.
The history of modern art is filled with artists designing situations in which they cannot fully predict the result.
Can AI Art Actually Be Surrealist?
The answer depends on what “Surrealist” means.
If it means imagery that looks strange, dreamlike or impossible, AI can produce that appearance almost instantly. That is the least interesting answer.
If Surrealism means the historical movement organized around Breton and his contemporaries, a generative model obviously cannot retroactively join it.
The useful meaning lies between those extremes.
AI-assisted art can engage Surrealist methods when a human artist uses computational unpredictability to disrupt habitual thinking, test associations, transform source material or construct a meaningful tension between intention and chance.
That does not mean the algorithm itself becomes Surrealist.
It means an artist can place the algorithm inside a Surrealist problem.
Polska’s work makes this distinction visible. She uses AI within films that ask larger questions about ecology, technology, memory and imagined histories. Moderna Museet does not present AI as the replacement for automatism. It places AI-generated video beside Moallim’s physical automatic lines and a century of works built through dreams, chance, desire and free association.
That juxtaposition is the argument.
Automatism After Algorithms Still Needs An Artist
The Subterranean Sky remains on view at Moderna Museet in Stockholm through January 17, 2027, with curator Lena Essling continuing to lead programs around the exhibition during its final months. It is an unusually useful exhibition for an AI era precisely because it refuses to treat Surrealism as an antique visual style.
The movement’s deeper legacy lies in methods for disturbing control.
That is why AI belongs in the conversation.
It is equally why AI cannot settle the conversation.
A generative model can surprise its user. It can transform archival material. It can produce visual mutations too complicated for an artist to anticipate individually.
It cannot make those surprises meaningful on its own.
Surrealist automatism was never simply a celebration of whatever happened accidentally. Its importance came from what artists believed those accidents could reveal and from the cultural, psychological and political questions they built around them.
A century later, the technological system has changed.
The artistic problem has not disappeared.
How much control can an artist surrender before authorship becomes unclear? How much unpredictability is productive? When does a generated accident become a chosen artwork? What does a human recognize in an image that the system producing it does not need to comprehend?
Moderna Museet’s answer is deliberately unresolved.
Fatima Moallim lets a line become a story through bodily movement.
Agnieszka Polska lets AI transform archival flowers into speculative organisms.
Between them lies a century of artists trying to escape the limits of deliberate reason.
AI can participate in that history.
The algorithm does not need to dream.
The artist still has to decide why the strange image matters.