A significant part of the image production made by artificial intelligence presents elements that we associate with the surreal, especially when the prompts are open and not directed towards photorealism or a naturalistic representation. This effect must be read as a cultural symptom. Yves Citton (2014), in L’économie de l’attention, recalls that attention functions as a collective ecology: it is modulated, cultivated and fought for in environments saturated with stimuli. From this perspective, the aesthetics of the strange gains visibility because it occupies the center of this ecology, capturing views, redistributing perceptual energy and ensuring circulation on networks. Machinic surrealism thus emerges both as a result of machinic functioning and of cultural logic that privileges the unusual.
The journey begins in the surreal, where juxtaposition intensifies collisions and suspends causality; moves towards the strange (weird), which introduces an active outside, a force that reconfigures expectations; enters the disquieting (uncanny), when the familiar returns with a minimal deviation that disturbs; and crosses the valley of the unsettling (uncanny valley), a zone in which the almost-human intensifies repulsion as the resemblance grows without full congruence. This grid guides the reading of AI images and offers a grammar for the circulation of the strange.
André Breton's surrealism (2016), proposed in texts such as the Manifestos of Surrealism (1924–1946), presented psychic automatism as a path to liberation: writing or spontaneous creation without the mediation of reason, in search of a direct connection to the unconscious. This practice had a political dimension, as it intended to subvert structures of rational and bourgeois control, establishing an insurgent imagination capable of dissolving hierarchies. Psychic automatism emerged as an aesthetic technique and strategy for subjective and collective emancipation.
When looking at AI, the temptation arises to draw an immediate parallelism: the algorithm also produces automatically, combining signs without conscious intentionality. The analogy, however, soon reveals its fracture. Surrealist automatism was a deliberate gesture of liberation of the subject; machinic automatism emerges from a technical exteriority: it recombines visual residues from statistics. The affinity between the two is not in the origin (psychic vs. machinic), but in the structure: in both cases, creation emerges from non-linear processes and juxtapositions that escape discursive rationality. AI simulates automatism without interiority; Still, human perception tends to read the result as if it were the product of a deep imagination.
Much of the imagery produced by AI tends towards surrealism because the operating logic of these systems comes close, in a structural way, to what surrealism sought: the unexpected combination, the collision of signs, the encounter between elements that should not coexist and which, precisely for this reason, reveal a poetic and unsettling intensity.
Walter Benjamin (2012) observed how mechanical reproduction dissolved the “aura” of the original. AI, by multiplying and recombining images, creates a regime in which the aura is diluted, leaving the combinatorial logic. This process comes close to what Breton called “convulsive beauty”: the beauty that emerges from shock and montage, from the juxtaposition of elements without context.
From a technical point of view, the surrealism of AI images arises from the difficulty these machines have in sustaining coherence: duplicated faces, excess limbs, distorted spaces. What would be a failure in realism, in the surrealist sense becomes style. Deleuze and Guattari (2003), in A Thousand Plateaus, speak of the “war machine” as a deterritorialization device; AI acts like this, deterritorializing established visual codes and recombining them in an anarchic, almost automatic way, in a kind of machinic unconscious.
There is also a cultural dimension. In the contemporary digital ecosystem, the strange, the grotesque and the uncomfortable are valued. Surrealism appears as the aesthetic form that best circulates on the networks: capable of causing laughter, shock and viral sharing. AI does not invent surrealism; converts it into a continuous production regime, adjusted to the attention economy.
Machinic surrealism gains intensity because it is part of an economy that privileges attention as a scarce resource. Shoshana Zuboff (2020), in The Age of Surveillance Capitalism, shows how platforms colonize individual experience, capturing data and transforming every gesture into a predictive commodity. In this regime, the AI images that circulate most are not those that are coherent and predictable, but those that interrupt, shock or cause perceptual hesitation.
Citton (2014) emphasizes that attention is a collective process of manufacturing and redistribution. The strange, the bizarre and the disturbing participate in this shared fabrication, guiding sensibilities and ways of seeing on a social scale. Images produced by AI not only capture individual attention, but shape a collective regime of perception, in which the surreal operates as a mechanism of cohesion and cultural dispute.
I call the machinic unconscious a heuristic on three planes: in the technical one, generative models recombine patterns in latent spaces and produce correlations; in the aesthetic, the gaze interprets these correlations as signs of depth and attributes meaning to the collisions; in the political, historical residues, class, gender, race, coloniality, permeate the data and guide affections. The articulation of these plans allows each image to be read as a synthesis of functioning, perception and governance.
While historical surrealism sought to free the gaze and thought from a repressive rationality, machinic surrealism functions as a mechanism of capture and monetization. They share a formal structure (automatism) and diverge in political horizons: emancipation in the first case, instrumentalization in the second.
Mark Fisher (2025), in The Weird and the Uncanny, suggested that the strange and the unsettling reside not only in the unreal, but in the fracture of the real. AI images are precisely that: visual fractures, echoes of an impossible reality. Surrealism stops being just an artistic movement and becomes the spontaneous aesthetics of the machine—a reflection of a technical process that simulates the dreamlike without the dreamer.
The question that arises is whether these machines could be programmed to avoid such combinations. In theory, yes; in practice, with clear costs and losses. Diffusion models do not “know” what is surreal or realistic: they learn statistical patterns and generate recombinations. If the objective was just photographic realism, it would be necessary to train filtered datasets and apply restrictions on anatomical consistency, perspective and physical logic. This is feasible in areas such as medicine or product photography, with reduced freedom of imagery.
Historical surrealism cultivated dislocations and exoticisms that fed an archive of strangeness. The recent technical image adds its own economy: the dissolution of the aura identified by Benjamin finds a second life in the “poor image” of Hito Steyerl (2012); software grammars establish manipulation routines (Manovich, 2001); and the new tends to establish itself as a repetition program (Groys, 2022). AI condenses these lines: it recombines inherited exoticisms, operates through codified grammars and transforms novelty into calculable routine — a machinic surrealism with platform materiality.
But the reason why surrealism is not avoided is twofold. On the technical side, forcing absolute coherence requires computational effort, algorithmic corrections and highly controlled datasets. On the cultural side, audiences tend to look for the strange, the unexpected, the “fascinating mistake.” Surrealist aesthetics gain value, so elimination doesn’t matter. Citton (2014) proposes thinking about attention as a shared ecology. From this perspective, the strange and the disturbing modulate this ecology: they absorb perceptual energy and redistribute it according to the logic of the platforms, saturating the collective space with dissonance and shock. The machine could exclude the surreal; the culture that uses it programs the device to produce what best circulates. Federico Campagna (2018), in Technic and Magic, goes further: technique, more than executing instructions, establishes a regime of reality. AI's tendency towards the surreal expresses this world regime in which the real merges with the artificial and the machinic imagination becomes naturalized.
Attention is organized through metrics and continuous trials: ranking by click probability and viewing time, cold starts with reinforcement of novelty, recommendation cycles that promote content of high perceptual interruption. The stranger acts as a logistics operator: it prolongs dwell time, triggers comments, activates sharing. In test cycles, retention at 3 and 10 seconds tends to rise with perceptual interruption; the system forwards similar pieces, reinforcing the unusual editorial line.
Walter Benjamin (2012) warned that the technique fetishizes: it transforms historical processes into the appearance of neutrality. Vilém Flusser (1998), in Essay on Photography, adds that devices hide programs and force players to play according to invisible rules. Image production by AI operates in this regime: images seem natural and inevitable, while material conditions, datasets, invisible work, energy consumption, remain opaque, covered by the fascinating surface of the strange.
When talking about AI surrealism, it is not enough to think about error or technical failure. What emerges from these images is the feeling that the machine dreams. It is a dream without a dreamer, a non-human unconscious, even though it is composed of human waste. Donna J. Haraway (2022), in A Cyborg Manifesto | The Companion Species Manifestoproposes the figure of the cyborg to think about these crossings: neither natural nor artificial, but hybrid in which the distinction loses meaning. AI generates images that appear to come from a machinic collective unconscious because it recombines fragments of visual memory, faces, gestures, objects, styles, without a linear narrative.
Aesthetics gain density when they agree with documentation and accountability. Datasheets for Datasets (Gebru et al., 2018) and Model Cards (Mitchell et al., 2019) clarify provenances, gaps and risks; cartographies such as the Nooscope Manifesto (Pasquinelli & Joler, 2020) expose material chains, invisible work and energy dependencies. From here, curators take on diversity and fairness metrics, regular audits and socio-technical impact reports, reconfiguring the experience of the stranger.
Federico Campagna (2018) distinguishes between technical realism, where reality is a calculable and repeatable system, and magic, where language creates worlds. AI is located on this threshold: technically it has no interiority, but its productions are interpreted as if they came from within. The images seem like messages from a non-human unconscious, a radical “other” that speaks to us through strangeness.
Deleuze & Guattari (2004), in Anti-Oedipus, introduce the idea of desiring machines: systems that produce flows, cuts, incessant recombinations. AI does not dream because it wants to; dreams because its production logic is homologous to the flow of desire, associative, fragmentary and disturbing. Machinic surrealism visually translates this process.
There is also a political dimension: this machinic unconscious appears saturated with prejudices, colonial imaginaries, gender and racial stereotypes present in the datasets. Each strange image expresses a historical collective unconscious filtered by algorithms. AI surrealism is a social symptom.
If the machinic unconscious emerges as a collage of cultural residues, it is important to see which residues and how they carry political history. Achille Mbembe (2017; 2019) shows how modern power is exercised through the unequal management of life and death. datasets reproduce inequalities: racially discriminated, feminine or dissident bodies appear distorted, hypersexualized or marginalized; normative bodies stabilize as standard. Sara Ahmed (2014) argues that emotions circulate as forces that consolidate social boundaries. The machinic unconscious absorbs fear, desire and repulsion and returns them in images that reinforce stereotypes. The politics of the machinic unconscious is produced as a synthesis of past and present colonialisms, hierarchical genders and socially distributed affects. The surrealism of AI thus also becomes a symptom of a digital necropolitics.
In a first scenario, a hand proliferates fingers and produces perceptual hesitation; comments soar, shares multiply and the watch time rises with the attempt at deciphering. The technical feature points to guidance elevated and heterogeneous hand data, a combination conducive to the anatomical glide that fascinates. In another scenario, a pipeline for radiology or product works with controlnets, anatomical guides and strictly curated dataset; the central metric migrates to precision and functional error, fantasy recedes and operative confidence emerges. To read cases, a short protocol is useful: (1) objective (circulation vs. precision); (2) technical restrictions applied; (3) success metric (shares/retention vs. functional error).
Machinic surrealism designates the statistical recombination that the eye reads as depth; machinic automatism formally approaches psychic automatism and operates without interiority; the ecology of attention describes the fabric of capture and redistribution that gives traction to the unusual; the cultural residues mark the data and guide affections; and the necropolitical synthesis appears when aesthetics and differential governance of lives converge in the circuit of images. This lexicon, used in flow, stabilizes the reading and supports the critical proposal.
Historical surrealism presented itself as a gesture of emancipation: an attempt to free the imagination from the conventions of reason and morality. Machinic surrealism emerges in a regime of capture: its images are part of an ecology of attention shaped by digital platforms, where the strange functions as an engine of circulation. The machinic unconscious exposes the grammar of capture of our time and opens up a field of aesthetic and political intervention. Transparent curations, diversity metrics and auditing devices transform the stranger from a blind motor of circulation into a public laboratory of shared imagination.
References
Ahmed, S. (2014). The cultural politics of emotion (2nd ed.). Edinburgh University Press.
Benjamin, W. (2012). About art, technique, language and politics (M. A. Cruz, M. L. Moita, & M. Alberto, Trans.). Water Clock. (Original work published in 1936)
Breton, A. (2016). Manifestos of surrealism. Free Letter. (Original works published 1924–1946)
Campagna, F. (2018). Technic and magic: The reconstruction of reality. Bloomsbury Academic.
Citton, Y. (2014). L’économie de l’attention: New horizon du capitalisme? La Découverte.
Deleuze, G., & Guattari, F. (2003). A thousand plateaus: Capitalism and schizophrenia 2. Assírio & Alvim. (Original work published in 1980)
Deleuze, G., & Guattari, F. (2004). The anti-Oedipus: Capitalism and schizophrenia. Assírio & Alvim. (Original work published in 1972)
Fisher, M. (2025). The weird and the unsettling (L. C. Nunes, Trans.). vs. Editor. (Original work published in 2016)
Flusser, V. (1998). Essay on photography: Towards a philosophy of technique. Water Clock. (Original work published in 1983)
Gebru, T., Morgenstern, J., Vecchione, B., Vaughan, J. W., Wallach, H., Daumé III, H., & Crawford, K. (2018). Datasheets for datasets. arXiv.
Groys, B. (2022). Art in flow (P. E. Duarte, Trans.). Black Orpheus. (Original work published in 2016)
Haraway, D. J. (2022). A cyborg manifesto | The Companion Species Manifesto (A. M. Chaves, Trans.). Black Orpheus. (Original works published in 1985 and 2003)
Manovich, L. (2001). The language of new media. MIT Press.
Mbembe, A. (2017). Critique of black reason (M. Lance, Trans.). Antigone. (Original work published in 2013)
Mbembe, A. (2019). Necropolitics (S. Corcoran, Trans.). DukeUniversity Press. (Original article published in 2003)
Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., Raji, I. D., & Gebru, T. (2019). Model cards for model reporting. In Proceedings of the Conference on Fairness, Accountability, and Transparency (FAccT) (pp. 220–229). ACM.
Pasquinelli, M., & Joler, V. (2020). The Nooscope manifesto: A cartography of the limits of AI. (White paper).
Steyerl, H. (2012). The wretched of the screen. Sternberg Press.
Zuboff, S. (2020). The era of surveillance capitalism (L. F. Silva & M. S. Pereira, Trans.). Water Clock. (Original work published in 2019)