This text is a working document, written as part of my slow return to PhD research after an extended period of medical leave. It is one of several threads I am weaving to better understand my own practice and make explicit the ideas that underpin it. It lacks the footnotes and references I had hoped to include but was unable to at the time of writing. It is not a finished piece of scholarship but a foundation - an attempt to articulate, in continuous prose, the tradition I place myself within, the theoretical tools I draw on, and the material conditions I work under. It will change as the work progresses.
As an artist educated at least partially within what might be called the Western contemporary art tradition - even if that tradition was never the one coherent tradition it sounds like, but rather a series of global trends that always occur in highly local variants - I write from inside the structure I critique. The colonial history of the state I inhabit is one axis of that position, but it does not define me completely. I am also embedded in the institutions I question; I am a parent navigating systems whose internal rules are often hidden; I am someone whose circumstances exceed the frameworks I use here - and who claims the right to that excess. The insistence on opacity found in this text applies to its author as much as to its subjects.
What cannot be escaped is the condition of cultural surplus: working within an archive that is already too full, drawing on a tradition whose privileges I inherit even as I question its narratives. The difficulty is not lack but excess. Marginal and collective histories are not absent; they are buried. The task becomes one of salvage.
My entire practice - the scavenged hardware, the generated images, and the writing of this text itself - is a project of repair. In Eve Kosofsky Sedgwick’s terms, this is a reparative orientation: it seeks unused potential within a tradition rather than throwing the tradition away entirely, reading for what can be recovered rather than only for what can be exposed. Awareness of blind spots is not a reason to abandon a history but a precondition for working with it. It is not this text alone that attempts to dream up new worlds, but the daily practice it belongs to. The frameworks invoked throughout function as instruments - provisional tools for making aspects of a situation legible - rather than as authorities to be applied.
The Western art historical tradition has long organised itself around a particular figure: the singular creative genius who produces from nothing, whose work is original in the deepest sense - without prior influence or debt. This figure does not just judge art; it structures the entire canon. It determines how influence is traced and how value is assigned. From Vasari’s Lives of the Artists to the contemporary art market, the underlying logic insists on a smooth, unbroken lineage of exceptional individuals whose contributions are entirely new.
But this “West” was never one thing, and the canon is not, and has never been, a settled agreement. It is always disputed and contentious, no matter how hard art institutions choose to pretend otherwise. The standard narrative organises a messy, discontinuous history into a neat sequence of movements and breakthroughs. This is an institutional fiction. The history it describes has always included collective, anonymous, and collaborative practices that this structure actively ignores. The Romantic elevation of genius broke from earlier craft traditions; the historical avant-gardes attacked institutions while creating new exceptional heroes; conceptual art attempted to remove the authored object altogether. Each rupture challenged the myth of the genius - and was eventually absorbed by it.
The persistence of originality is not an accident. It is maintained through the hard work of institutions whose authority depends on keeping up the illusion of a unified tradition. The question is not why the myth survives being disproven, but how it rebuilds itself after each challenge. Read differently, the margins of this history form a record of alternatives: recurring refusals of linear progress and individual primacy. Their repeated absorption does not nullify them. It indicates that the resource remains available. A reparative practice draws on this archive not out of nostalgia, but as material for present work.
The critique of originality also emerges from within the tradition itself. André Malraux proposed that no artwork is ever encountered in isolation. Walter Benjamin argued that the “aura” of an original work was tied to specific physical and economic conditions. Later writers extended this argument: culture is naturally a mix of influences, and “originality” is a bad description of how it is actually made. These critiques share an implicit condition: abundance. They speak from within cultural surplus. When critics celebrate “uncreativity” or the “ecstasy of influence,” they are navigating having too much, not too little.
Generative AI redistributes the tensions that these critiques identify but cannot resolve. Everything the models do - extraction, abstraction, removing context, institutional control - has happened before: modernism abstracted, colonial archives classified, museums removed context, photography compressed, print capitalism scaled. What changes with generative systems is not the actions themselves, but their automation, speed, and invisibility. The extraction is continuous; the abstraction is statistical; and the process is built to remain hidden from the people whose material it consumes. I do not claim that this technology is a new form of absolute bondage, but I contend that it echoes older patterns of internal colonisation - in Etkind’s historical sense, describing how empires treat their own populations and cultures as extractable resources. These analogies describe structural patterns of extraction and asymmetry rather than lived colonial violence.
I bring several traditions together - surrealism, feminist theories of situated knowledge, networked art, Édouard Glissant’s concepts of relation and opacity, Václav Havel’s post-totalitarian thought, and the speculative methods of pataphysics and speculative fiction - not to force them into a single grand theory, but to ask what each reveals about this redistribution of tensions that the others cannot. My driving claim is this: originality is a positional myth, generative systems make its contradictions newly visible, and situated artistic practice is a way of inhabiting those contradictions without pretending they resolve.
Surrealism occupies a useful, if shaky, position in this history. The movement sought to break the strict, rational logic of the Enlightenment, reaching instead toward the collective subconscious. In this sense, the Surrealist impulse was an early attempt at relation; it sought a world where the lone genius was replaced by collaborative methods of making, such as the exquisite corpse.
Yet this desire for connection moved toward a specific kind of transparency. The Surrealists wanted to dissolve the ego to reveal a universal human subconscious. In doing so, they frequently treated their encounters with non-Western culture as “discoveries” to be processed by European artists. The “folk” became a resource for the Surrealist project to consume, precisely because the movement’s primary goal was to universalise the irrational - to find a mirror in which the Western subject could see its own hidden depths. These were not the movement’s only blind spots. Surrealism’s systematic reduction of women to muses, and its largely ignored class and race privileges, are part of the same structure: the collective method remained organised around a viewpoint that was male, European, and wealthy, even as it claimed to dissolve the boundaries of the self. To rescue Surrealism’s potential for connection requires acknowledging these limits as constitutive rather than incidental - the same discipline required when working with AI, where datasets carry specific biases and power imbalances, and the work consists of consciously navigating them rather than pretending they do not exist.
But to leave it there would reduce Surrealism to simple cultural theft. The movement also operated in direct engagement with anti-colonial thought, emerging most clearly where Surrealism and Caribbean practice met. Aimé Césaire’s encounter with Surrealism in 1930s Paris was a mutual provocation. Césaire found tools in Surrealist techniques that could be turned against the very rationalism that justified colonial authority.
Where the European Surrealist sought to make the self universally transparent, the Caribbean subject sought to protect the self against colonial mapping. That Glissant himself emerges from the same Caribbean geography, and was Césaire’s student, means the relationship between these traditions is historically tangled. Surrealism did not simply extract from the colonial world; it was also transformed by it.
What Surrealism could not resolve was the question of where the artwork actually comes from. Its techniques - automatic drawing, chance, the found object - were designed to bypass individual intention. But the gallery system still required named authors and solo exhibitions. The tension between collective making and individual credit was never settled; it was managed, and the management always favoured the named artist. This tension returns, amplified, in the context of generative AI.
Donna Haraway’s critique of the “god trick” - the claim to see everything from nowhere - offers a different entry into the problem of originality. For Haraway, all knowledge is situated: produced from a particular body, a particular location, and specific material conditions. The “view from nowhere” is always, in practice, a view from somewhere very specific - usually from a position of power.
Sandra Harding’s standpoint theory extends this: not all positions see equally well. Knowledge produced from the margins can reveal structures that are invisible from the centre. This is not a claim that marginal perspectives are automatically correct, but that they have access to features of a system that the system’s beneficiaries have no reason to examine.
Applied to art, this challenges the genius myth differently than the “everything is a remix” argument. The remix critique says: the genius did not really produce from nothing; the work is a combination of influences. The feminist critique says: the genius is a myth about social position. It claims to speak from nowhere - from pure creativity, from universal value - while actually speaking from a very particular somewhere: usually white, usually male, usually embedded in institutions that reward that particular somewhere as if it were everywhere.
This has direct consequences for how we understand generative AI. The large language model also performs a god trick. It consumes material from millions of positions and produces outputs that appear to come from nowhere - from “the machine,” from a system without location.
But the model is situated: it lives in a data centre, runs on particular hardware, is trained on data assembled according to specific corporate priorities, and is optimised for specific goals. The outputs carry the marks of that situation even when they appear placeless. The pretence of universality here is not merely false; it is coercive - not because all abstraction is an act of domination (museums, archives, and translations all abstract, and these are not inherently coercive), but because this particular abstraction is enforced without consent, erases the contexts from which the training material was produced, and claims the resulting outputs as general, neutral, and authorless while serving commercial interests. The harm lies in the asymmetry: material is taken, processed, and returned under terms set entirely by the extracting party. The disappearance of the original source is not an accident of scale. It is what the system is designed to accomplish.
If Haraway challenges the claim to see from nowhere, Karen Barad’s concept of intra-action challenges the assumption that the observer and the observed exist independently before they meet. For Barad, entities do not pre-exist their interactions; they emerge through them. The instrument of observation is not separate from what is observed; it participates in producing the result.
Applied to generative systems, this means that “the model” is not a static object waiting to be used. Nor is “the artist” a stable subject who deploys the model as a tool. Both take shape in the moment of generation - they emerge from the encounter rather than entering it fully formed.
But we must be careful. While intra-action is productive for dismantling the myth of the lone genius, applying it too smoothly to generative AI risks political flattening. If humans, machines, and data are simply an entangled, co-creating web, it becomes difficult to name extraction or exploitation. The tech industry would gladly frame mass data-scraping as a natural, post-human “entanglement” to avoid accountability.
Therefore, Barad’s framework must be held in tension with the realities of economic power. This becomes clear through her concept of the “cut.” The question “who made this?” assumes that maker and made can be cleanly separated. Barad suggests this separation is not natural but performed - an artificial boundary created by the system of observation itself. In the context of AI, this “cut” - deciding who gets credit, who signs the work, who holds the copyright - is not merely philosophical; it is commercial. The institution performs the cut where it is most profitable, separating “the artist” and “the work” after the fact to produce a sellable unit. The genius myth is, in this light, a particular way of cutting up the world to serve institutional interests.
Decades before large models, internet art already confronted a similar tension between collective making and institutional control. The traditions of networked art - from mail art to contemporary digital practices - offer a practical history of what shared authorship looks like when taken seriously. These practices have spent decades working with problems generative AI now makes unavoidable: the distribution of authorship across systems, the reliance on infrastructure the artist does not control, and the instability of the work as it circulates.
Cornelia Sollfrank’s “Net.Art Generator” (1999) is an instructive case: a system that scraped images from the web and recombined them algorithmically, producing “artworks” credited to fictional female artists. The project made several things visible at once: the raw material was always already someone else’s; the system’s operations were hidden from the viewer; and claiming authorship is a political act, not a natural fact.
More broadly, the history of networked art demonstrates that collaborative, machine-aided making does not automatically destroy the myth of the genius. The art world simply absorbed net.art, assigned individual reputations to its practitioners, and integrated the outputs into the familiar economy of galleries and collections. The same process is underway with AI-generated art: despite the radical distribution of authorship that the technology implies, institutions insistently re-establish individual credit. This is not a failure of the institution to understand the technology. It is the institution doing what it does best: performing the cut that produces “the artist” as a unit of value.
The risk - which must be named directly - is that the art market will absorb these new practices as aesthetic style, extracting their visual surface while leaving the extractive infrastructure untouched. An image of a system’s failure, once circulated smoothly through a corporate platform, is no longer a failure. It is product.
Édouard Glissant introduces a dimension that the other frameworks do not fully address: the relationship between extraction and cultural production under conditions of deep global inequality, and the politics of who is allowed to remain opaque.
A distinction is necessary here. Platform capitalism extracts from its own users; it operates within a shared, if exploitative, economic space. Colonial logic extends extraction through geopolitical asymmetry into contexts that had no part in designing the system and no real power to refuse it. Generative AI operates in both registers simultaneously. A European freelance photographer whose images appear in a training dataset and the oral tradition of a community that never consented to digitisation are both subject to extraction, but the asymmetry between their positions is substantial, and the colonial framework names that asymmetry where “platform capitalism” alone does not. The two frames are not alternatives; they describe overlapping but non-identical dynamics within the same system. Etkind’s concept of internal colonisation adds a third register: the way systems treat their own cultural material as a resource, standardising and rendering it transparent for easier management. In the context of contemporary AI, the “metropole” refers to global tech centres such as Silicon Valley, highlighting that structural asymmetries in computational infrastructure echo patterns of nineteenth-century empire without equating them directly.
Glissant’s distinction between filiation and relation addresses the logic that sustains both the myth of originality and its in-house critiques. Filiation is the logic of the root: lineage, origin, a trunk from which branches descend. Relation describes encounter without origin - forms meeting and transforming one another without any serving as the source. Glissant roots this in the specific history of the Caribbean, the Middle Passage, and the plantation. Relation names what happens when people are violently uprooted and must construct meaning from fragments. This process of créolisation produces what did not exist before, but it carries the traces of the violence that made it necessary.
His concept of opacité addresses the politics of legibility directly. Colonial ethnography, liberal multiculturalism, and contemporary data systems all share a demand: the other must be rendered classifiable in order to be granted standing. Glissant proposes opacity as a right - the claim that one’s existence should not depend on a system’s ability to categorise it. But opacity runs in both directions: it is also the condition of navigating systems whose internal logics are opaque by design. The opacity of the coloniser maintains power; that of the colonised constitutes resistance. Glissant keeps both in view without collapsing one into the other.
Large generative models extract vast quantities of cultural material without consent, homogenising it through optimisation functions that flatten specificity, and centralising infrastructure in ways that impose the centre’s terms across diverse contexts. The term “algorithmic imperialism” names this structural condition - highlighting the asymmetry in technological systems - rather than equating it with lived colonial violence. The colonial framework here operates as an analytic of asymmetry and enforced legibility, not as a claim of equivalence between computational extraction and historical colonial domination. In this context, opacity - in Glissant’s sense - names the refusal to be fully tokenised: the space where a culture’s interiority resists being rendered as training data.
But “generative AI” is not a singular actor. The corporate API, the open-source model, the diffusion system, the small fine-tuned model running on local hardware - these occupy very different positions, and the colonial analogy applies with varying force. Treating them as identical would reproduce exactly the flattening the argument criticises.
Describing what models do to their training data as “forced créolisation” is provocative but requires qualification. In Glissant, créolisation involves unpredictability, irreducibility, and emergence; AI training, in contrast, produces statistical compression that homogenises situated knowledge, meaning the analogy applies to extraction rather than emergent relation. The model does not produce relation in Glissant’s sense; it produces abstraction - taking material that was situated and opaque and rendering it as weights in a network. What is lost in this process is not “data” in any neutral sense but the specific conditions under which the material meant something: the community that produced it, the conventions it operated within, the situated knowledge it carried.
The latent space of a generative model is, in this sense, the metropole’s map - not in the territorial sense of colonial cartography, but in the epistemological sense: both impose a system of comparability that the mapped material did not generate and cannot refuse, rendering diverse forms of knowledge legible only in terms the mapping system defines. It is a topology in which meaning is reduced to distance, and to be “known” by the system is to be assigned a coordinate relative to everything else the system has ingested. This is the transparency demand in its purest mathematical form - the refusal to let an object exist on its own terms, forcing it instead into a comparative metric where proximity is determined by the system’s training, not by the material’s own logic.
Does the model produce true connection, or only abstraction? In its dominant commercial configurations, AI abstracts rather than relationally transforms - not because computation is inherently incapable of producing emergence, but because the systems as currently built and governed optimise for output, not for the situated encounter that relation requires.
This distinction should not be drawn too sharply. All relation involves abstraction; all modelling reduces entropy; human cognition itself compresses and schematises in ways that are not categorically different from what the model does. Compression is not inherently harmful. It is how we think at scale - taxonomy, classification, generalisation are all lossy processes, and they are also what makes navigating complexity possible. What is lost in compression is the price of increased capacity, and that trade-off is often productive. The political question is not whether to compress but who controls the compression, what gets designated as noise, and whether the people whose material is compressed have any say in the terms. Human cultural synthesis is embedded in communities, subject to contestation, and answerable to the people whose material it transforms. Statistical compression in a generative model is answerable to a loss function. The absence of social embedding is what makes the model’s abstraction politically distinct, not some essential difference between computational and human cognition.
And yet the outputs are not entirely reducible to their mathematics. This is where “hallucination” becomes relevant. The tech industry uses the term to label outputs that are factually wrong, framing them as failures to be fixed. But the model is not trying to tell the truth and failing. It is producing what is statistically plausible. It is indifferent to whether its outputs correspond to anything real. What we call a hallucination is the visible surface of that indifference.
These distortions are not evenly distributed. Material that is heavily represented in the training data - the visual culture of the wealthy West, Standard English, commercial photography - survives the compression relatively intact. Material that was rare, local, or structurally unlike the dominant patterns gets distorted much more severely. This is lossy compression as political fact. In computer science, lossy compression works by discarding what the algorithm considers redundant. But what registers as noise to the system is often precisely the opacity that Glissant seeks to protect. The system’s definition of “noise” is not neutral. It is a hierarchy encoded as an engineering decision. The hallucination marks the point at which this hierarchy encounters material it cannot digest. This is not rebellion by the data. It is structurally produced evidence of the system’s epistemological limits.
Haraway’s situated knowledge and Glissant’s opacity converge here. The hallucination is the point at which the model’s false universality becomes legible - where its claim to produce from nowhere encounters material that was most stubbornly somewhere, and the encounter leaves marks. Both the feminist critique and the postcolonial critique identify the coercive force of imposed universality, and both insist on attending to what that universality cannot contain. The difference is in what they emphasise: Haraway foregrounds the erasure of position; Glissant foregrounds the erasure of opacity. Together they describe two aspects of the same operation.
Artistic practice can work in these residues deliberately. The reparative impulse described at the outset finds its structural counterpart here: the recuperation of what generative systems discard. Just as the standard art historical narrative writes out collective and non-linear practices, the corporate AI pipeline optimises away the outputs that do not conform to its definition of success. In both cases, what is discarded is not without value; it is without value to the system that discarded it. The practice of attending to residues applies the same method to both objects - the tradition and the tool.
Here two further traditions become relevant. Pataphysics - Alfred Jarry’s “science of imaginary solutions” - provides a method for treating the model’s failures not as errors to be corrected but as data about the system’s assumptions, to be examined with the same rigour one would apply to any other empirical phenomenon. Speculative fiction, from its earliest forms through Afrofuturism and contemporary climate fiction, offers the practice of constructing counterfactual worlds that illuminate the present by displacing it. Together they describe a mode of working that is neither credulous (accepting the model’s outputs as given) nor purely critical (rejecting them as compromised) but diagnostic: using what the system produces, including what it produces wrongly, as material for understanding what the system is. The speculative historical image - a counterfactual document, an impossible architecture, a scene that shimmers between plausibility and impossibility - makes visible the system’s epistemological indifference and the unevenness of its compressions. The viewer cannot fully resolve what they are looking at, and that irresolution, worked with deliberately, becomes a way of refusing both the myth of the original and the false transparency of the generated.
But the counterfactual image is not only diagnostic. It is generative. The true lies of speculative practice - images of pasts that did not happen and presents that do not yet exist - make it possible to imagine other futures and other infrastructures. They are grounded in enough fact to produce a slight plausibility, and contain enough strangeness to open avenues of thought that settled narratives have closed. A speculative image of a Danish healthcare system organised around principles that never took hold, or a labour movement whose iconography developed along paths history foreclosed, does not claim to depict what was or what is. It claims that things could have been otherwise, and that this “otherwise” is not exhausted - that dormant traditions can be reactivated, not as nostalgia but as material for reimagining what comes next. This is the reparative project applied to the future as well as the past: recovering unspent potential not only from the archive of what was made but from the archive of what was almost made, nearly thought, not quite realised.
If the question of originality is also a question of position, then the physical realities of generative practice cannot be treated as incidental. I find an ally here in Václav Havel’s post-totalitarian thought. Havel describes a system where participation in the official lie - even through small acts of conformity - is how power is maintained. The shopkeeper who places a government slogan in the window does not believe it; the system does not require belief, only compliance. To refuse the slogan is not to overthrow the system. It is to “live in truth” - to maintain a sphere of interiority that the system’s demand for legibility cannot fully penetrate.
Denmark occupies a double position in this landscape: historically a colonial metropole with past territories in the Caribbean and Greenland, yet today a peripheral actor within US-dominated digital infrastructures. This shapes what forms of extraction, agency, and opacity are legible from here. Yet I am not the state I inhabit; my own practice requires its own opacity, distinct from national histories. I am situated within this context, but I am not reducible to it.
My practice involves maintaining a scavenged 3090 GPU - hardware repurposed to run a localised AI model on renewable energy. Running models locally shifts governance, accountability, energy relations, and dependency.
But I must not romanticise this hardware. A GPU is not an innocent object; it is the physical condensation of global extraction, relying on mineral mining, exploited labour, and massive ecological expenditure. Refusing the corporate cloud does not erase the extraction baked into the silicon.
“Living in truth” in this context cannot mean claiming purity or complete escape from the system. It means taking material responsibility for a compromised position rather than hiding behind the seamless interface of a corporate API. The intervention is positional, not structural. A locally run model still carries the biases of its training data; it still operates within the same optimisation logic. What changes is the governance: who decides what runs, what is kept, what is discarded.
This prevents the local setup from becoming a false salvation narrative. It does not wash away the extractive history embedded in the model’s architecture, but it establishes a different position from which to engage. The value lies precisely in the friction: partial, physically situated, and accountable to its own context. Not a utopia, but a specific ground from which to speak and make.
When my model fails to accurately generate specific Danish labour iconography, the failure is not a bug. It is the point at which the physical limit of my position becomes visible - where the system’s indifference collides with material it was never trained to understand.
I do not claim that working locally transforms global infrastructure. It does not. What I claim is more modest: working locally preserves the residues that corporate platforms are structurally incentivised to eliminate. Each software update patches out the hallucinations. The situated practitioner treats them as material - as evidence of what the system’s compression could not absorb.
This preservation is precarious. Glitch aesthetics were absorbed into mainstream design almost immediately; corporate systems routinely incorporate user anomalies to improve their corrections. But the precariousness is part of the point: the goal is not permanent preservation but maintaining residues for as long as they remain legible. They function as a counter-archive - evidence of what the system cannot contain. Without such evidence, the system’s compressed account becomes the only account available.
Beyond individual practice, there is the question of lateral knowledge-sharing. When communities share tactics for navigating opaque AI systems - fine-tuning small models for local languages, pooling hardware - they create spaces where multiple opacities coexist, and margins learn from margins without needing to explain themselves to the centre.
Perhaps the most concrete form this takes is the creation of small data - curated, specific datasets intentionally withheld from the corporate pipeline, maintained by the communities that produced them. Where Big Data extracts and dissolves context, small data is bounded, situated, and opaque by intention. But small datasets are not inherently virtuous; they can reproduce local hierarchies, encode exclusions, and serve as tools for gatekeeping. Situated knowledge is not a guarantee of justice. What makes small data politically significant is not that it is pure, but that its biases are legible and its creators are accountable. It makes responsibility possible.
The several frameworks brought together here do not agree with one another, and that disagreement is productive. Surrealism demonstrates how the myth of the lone genius persists even within practices designed to refuse it. Feminist epistemology insists that the question “who makes?” is inseparable from the question “from where?” - and that imposed universality is not merely false but coercive. Networked art provides a practical history of distributed authorship encountering institutional re-individualisation. Glissant’s concepts of relation and opacity address the politics of extraction under geopolitical asymmetry and the right to withhold from systems that demand legibility.
Barad’s intra-action destabilises the entities - artist, model, output - that the other frameworks still tend to take as given. This creates a genuine tension within the essay’s own argument: if entities are co-constituted through intra-action, how stable are the distinctions between extractive system and relational practice, between abstraction and relation, on which the political critique depends? The answer is that these distinctions are not ontologically fixed but asymmetrically structured - the system and the practitioner are co-constituted, but the terms of that co-constitution are not set equally. The extraction is designed; the relation is insisted upon against the design. Barad does not dissolve the political claim; she relocates it from the level of pre-existing entities to the level of the processes that produce them. Havel reminds us that the power of the powerless lies in the refusal to be rendered transparent - that maintaining a sphere of truth within a system built on compliance is itself a political act, however modest.
None of these resolves the core tension of working with generative systems, which should be stated directly: the dominant generative infrastructures are centralising, epistemologically indifferent to the meaning their material once carried, and built on extraction at every level. Artistic practice that engages with them can nonetheless be grounded, accountable, and attentive to what the systems cannot contain. The system is extractive in structure. The practice can still be relational in orientation. These two facts coexist without synthesis.
But relational orientation is not immune to capture. The institutional apparatus will aestheticise the traces of systemic failure with the same efficiency it brought to appropriation art and net.art if the practice remains only at the level of the image. The defence is not purity - there is no outside position - but the insistence on tying the work to material conditions that resist full aestheticisation. These are not guarantees. They are frictions.
The generated image that trembles between document and fiction, the counterfactual history that disturbs a settled narrative: these are not proofs of a new genius, nor confirmations that everything is remix. They are markers of a process that is partial in its reach and unresolved in its politics. To work with them honestly is to refuse the myth of the singular creator without accepting the false universal that would replace it, and to insist that the conditions of production - who extracts, from whom, through what infrastructure, under whose terms - are never incidental to what is produced.
But diagnosis alone does not account for why anyone would persist in making work under these conditions. The answer is not only political. If we must live with extraction - and for now we must, since there is no position entirely outside it - then the question becomes how we still find beauty, or some semblance of it, within compromised circumstances. To insist on beauty is to insist that pleasure and joy remain among our goals, rather than efficiency, production, or the accumulation of wealth. The speculative image that shimmers between document and fiction, the counterfactual that opens a foreclosed history, the moment where the model produces something it was not trained to produce and the result carries unexpected weight: these are not only evidence of epistemological limits. They are also, sometimes, beautiful - not despite the compromise but within it, as a quality that emerges from the friction between what the system can produce and what the practitioner insists on looking for. Beauty under these conditions is not purity. It is attention: the capacity to find in residues something that matters aesthetically and not only critically. Without this dimension the practice becomes merely diagnostic, and diagnosis without care for what is made is another form of extraction - the instrumentalisation of the work in the service of the argument about the work. The reparative project is not only an intellectual commitment. It is also an aesthetic one: the conviction that something worth attending to can still be made from within systems that are indifferent to that possibility.
But refusal alone is not enough to sustain a practice. The goal, stated plainly, is to nurture traditions and ways of thinking that live in the margins - to keep alive unrecognised beauty and out-of-favour ideas, locally and patiently, until they become relevant again. This is not a solitary project. It is one contribution among many from people who find activities and ideas outside the mainstream worth preserving. The work is parapolitical: planting ideas and making them visible through images and objects so that situated, sustainable ways of thinking might re-enter broader circulation - not by force but by the slow accumulation of things that exist and can be encountered. Part of this work involves bridging domains that rarely speak to one another. The tacit knowledge of craft - the material understanding that comes from building, coding, printing, installing - and the tacit knowledge of programming share more with each other than either shares with academic discourse or policy debate. Yet academic and political institutions are where decisions about technology, culture, and funding are made. The practice sits between these domains, translating in both directions: bringing the concrete, situated knowledge of making into spaces that tend to deal in abstractions, and bringing the analytical frameworks of theory into a practice that would otherwise remain illegible to the institutions that shape its conditions. This translation is itself a form of the reparative project - it insists that craft knowledge and theoretical knowledge are not separate orders of understanding but different registers of the same attention to how things are made, by whom, and under what terms.
These are not hypothetical traditions. The Amiga demoscene and warez culture developed, from the 1980s onward, a practice in which programming skill was turned toward aesthetic ends - crack intros, demos, 64k intros - producing work that was technically rigorous and often beautiful, distributed through informal networks with their own competitions, critical vocabularies, and pedagogies, entirely outside institutional art. The demoscene worked inside proprietary systems and turned the act of navigating their constraints into a creative practice; the parallel to working within extractive AI infrastructure is direct. Live-action role-playing constitutes a collective, embodied form of speculative practice: participants construct counterfactual worlds and inhabit them physically, together, in an active disbelief of the given reality that is structurally close to what this essay calls the “true lies” of speculative image-making. Crafting traditions - textile, ceramic, woodwork - have always transmitted aesthetic and material knowledge through practice rather than theory, maintaining standards of excellence and communities of critique without recourse to institutional validation. And what might be called the folk art of AI - the millions of people now engaging directly with image generation, learning through practice about composition, style, distribution, and the politics of representation - constitutes a new form of non-institutional visual literacy. These practitioners learn to make and critique images by making them, developing judgement through direct engagement with the tools rather than through formal education. What these traditions share is that they produce knowledge - technical, aesthetic, critical - through practice and community rather than through credentialled instruction, and that institutional frameworks have consistently failed to recognise them as forms of knowledge at all. They are part of the reservoir. The reparative project includes them.
The solidarity this practice seeks is both local - grounded in specific communities, specific infrastructure, specific contexts - and lateral, margin to margin, building resilient hope across distances without requiring alignment on everything. This is what the reparative project looks like in practice: not a theory of resistance but the daily work of making and maintaining, in the conviction that what is kept alive locally can matter beyond the local, and that the reservoir of non-conformity is replenished by those who add to it.
I do not yet know if I will be successful in planting these ideas or images of change. That is the hope. The interpretation of the work does not belong to the artist; it lies with the viewer, and their choice to take up - or ignore - the materials and ideas offered. To accept this is to accept that art is an offer, not a set of instructions. Once the work leaves the local hardware and enters the world, it is out of my hands.
What remains, after these refusals, is not a programme but a practice: situated navigation within opaque systems, the maintenance of an archive the system would prefer to erase, and the patient attention to residues as evidence of what no single framework, and no single system, can fully contain. An image of trembling, preserved outside the pipeline - for now, and without guarantees - is still trembling. Whether it sparks a larger change is not for me to decide. That is enough to begin with.
