On 27 October 2026, Jason Kottke published an entry based on a text attributed to 713 Magazine on Instagram. One of the slides he reproduces contains an effective phrase: 'curation can look like creation'. The original Instagram post was not accessible during verification, so its date, specific authorship, and full text are not presented here as confirmed. Kottke allows one to trace the formulation; he does not certify its origin.
The phrase points to a confusion. Knowing which designers to cite, which music to listen to or which aesthetic to adopt produces an appearance of creativity because references also function as credentials. Selecting well requires attention, memory and judgement. It may form part of a creative practice and, under certain conditions, constitute it. But selection does not incorporate that status on its own; it may order external materials, confirm a consensus or exhibit a cultural position that is already available.
Generative artificial intelligence did not inaugurate this ambiguity. It has reduced the cost of obtaining culturally legible surfaces: a period atmosphere, a composition with an editorial appearance, a recognisable combination of traits. It has also industrialised an operation that platforms were already performing: converting taste into a sequence of measurable, recommendable and reproducible options. That is why the debate on AI does not begin with the prompt. It begins with an older question: what difference is there between recognising, choosing, producing and transforming.
Taste, judgement and production
In paragraphs 46 to 50 of the Critique of Judgement, Kant distinguishes the genius that produces from the taste that judges. He defines genius as 'the talent (natural endowment) which gives the rule to art'. The formula does not turn the artist into an irrational medium nor does it deny learning. It signals a limit: the rule of a beautiful work cannot precede it as a sufficient recipe. The product appears with an exemplary originality. It may guide others, but the creative procedure is not reduced to instructions capable of guaranteeing another result of the same order.
Paragraph 47 insists on this difference. Whoever produces by genius cannot fully explain how they found the idea nor transmit a method that ensures its repetition. A scientific demonstration can be reconstructed step by step. A work admits technique, correction, and deliberation, although these elements do not exhaust the appearance of its form.
Taste fulfils another function. It judges beauty, corrects the work, and renders communicable that which invention delivers without measure. In paragraph 50, Kant calls it 'the discipline of genius'. Invention may fall into the formless; taste cuts, orders, and adjusts. Kant goes so far as to grant it primacy when both enter into conflict. This disciplinary authority does not erase the difference between judging and producing. Without inventive matter, taste may choose, polish, or discard, but it does not generate by itself that which it judges.
Speaking of 'good taste' today as if Kant had certified a stable social hierarchy would be imprecise. His analysis of the judgement of taste does not offer a catalogue of correct preferences nor a test to identify refined individuals. The claim to validity of aesthetic judgement is not equivalent to the prestige of a specific community. The distinction of interest here is functional: the ability to recognise a valuable result is not sufficient to produce it.
Hume had described another figure in 1757: the competent judge. In 'Of the Standard of Taste', critical authority depends on delicacy, practice, comparison, the absence of prejudice, and good sense. He summarises it as 'strong sense, united to delicate sentiment'.
Practice matters as much to those who create as to those who judge. It allows for the perception of relationships that a casual glance does not distinguish and the comparison of one work with others. Hume does not require the critic to be the author of that which they evaluate. Knowing how to judge without knowing how to make is possible and legitimate. The error appears when that legitimacy is transformed into equivalence: a rigorous critique can explain why a work functions without becoming the work it explains.
The distinction also protects the autonomy of criticism. If only those who produce could judge, all evaluation would remain enclosed within the corresponding guild. Hume recognises a specific competence of the trained spectator. He does not confuse it with authorship.
Duchamp did not choose the object of best taste
The ready-made seems to refute that separation. Duchamp took manufactured objects, displaced them and presented them as art. The case is often invoked to assert that choosing is equivalent to creating. His own account complicates that conclusion.
In 'Apropos of “Readymades”', the text of a lecture delivered at the Museum of Modern Art in New York on 19 October 1961, Duchamp stated that the choice was not dictated by aesthetic delight. He spoke of 'a reaction of visual indifference' and of 'a total absence of good or bad taste'. The word with which he closed the idea was 'anaesthesia'.
Duchamp did not select the most beautiful object from a series nor compose an exquisite collection to demonstrate sensitivity. He attempted to suspend aesthetic preference. The gesture reorganised the relationship between object, author, title, institution and the act of designation. Material fabrication ceased to be the only place in which to seek artistic decision. It does not follow from this that any selection acquires creative power.
The operation is also not reduced to placing an object in a room. Choosing, naming, signing, dating, presenting and displacing form a sequence. The industrial object enters a structure of expectations and forces a revision of what counts as artistic work. The action is materially small and structurally extensive.
A choice becomes relevant for what it does in a field, not for the prestige of the one selecting nor for the aesthetic quality of what is selected. Fountain persistently altered the available questions regarding work, authorship and institution. The displacement was not a comment added to the urinal. It constituted the event that the object supported.
This prevents using Duchamp as a universal alibi. 'The artist chooses' describes a condition that is too weak. Those who buy a lamp, organise a folder of images or request a variant from a generator also choose. For the choice to acquire artistic depth, one must examine what relationship it establishes, what convention it alters, what dependency it reveals and what possibilities it leaves open afterwards.
The ready-made also does not eliminate craft. It displaces it. Craft may consist of producing an interpretive situation, knowing the field being intervened in and calculating what part of the object must remain intact for the operation to be legible. The absence of manual fabrication is not equivalent to an absence of work.
Curation became a sign of identity
David Balzer termed curationism the expansion of curatorial language beyond its professional sphere. 'Curating' came to denote the selection of products, images, music, clothing, travel or experiences. The term grants choice an authority that previously required the justification of relationships, exclusions and a stance towards the materials. The editorial description of his book summarises this inflation: 'Everyone, it seems, is now a curator'.
An exhibition may be a complex intellectual work. It constructs sequences, modifies the reading of familiar objects and renders visible relationships that no single piece contained in isolation. This possibility does not render every compilation a creation. Curatorship may denote a specific labour or function as a veneer for a list.
Platforms added another layer. Kyle Chayka argues in Filterworld that algorithmic recommendations tend to homogenise cultural consumption and production. Personalisation promises an individual relationship with the world; millions of users receive variations of similar repertoires. References cease to validate a unique journey when the same system distributes them, measures their acceptance and returns optimised versions.
Taste thus becomes an interface. Marking, saving, following, discarding and sharing fuels a computable representation of the individual. Subsequently, the system returns options compatible with that representation. The selection appears intimate whilst responding to a collective architecture of classification and promotion.
This circuit does not render all choices false. It alters their evidentiary value. Citing a little-known figure could indicate a singular investigation; when a platform recommends that same figure to millions of similar profiles, the reference ceases to demonstrate the journey it purports to show. It may remain fertile. It no longer suffices as a sign of originality.
Inequality also counts. Access to specific spaces, languages, publications, study time and professional networks facilitates the acquisition of a repertoire considered refined. Taste may present itself as individual merit whilst accumulating social conditions. Generative AI alters this distribution: it allows for the imitation of many of its signs without sharing the trajectory that previously rendered them scarce. It democratises an appearance and threatens a credential.
This threat explains part of the hostility towards generated results. Not all of it stems from a defence of labour, rights or the provenance of data. A portion protects the distinctive value of recognising certain codes before others. The machine does not merely compete with production. It competes with the social signal of having looked well.
The problem does not lie in sharing influences. Every practice works with precedents. As I examine in 'From Manga to Original Drawing', copying to learn, studying a solution and publishing an appropriation are distinct acts. The guide 'Using Visual References for Drawing' translates this distinction into daily work and copyright. A reference may nourish a work without determining it; it may also substitute for labour when treated as an identity ready for application.
The aesthetic distinction does not coincide exactly with the legal one. In Spain and the European Union, copyright protects original works and their concrete expressions; it does not, as a general rule, grant a monopoly over a style. This does not render any use of third-party material lawful: the reproduction of identifiable works, transformation, trademarks, the image of individuals and platform conditions may come into play. That an appearance is easy to request does not resolve who provided the materials or which uses are authorised.
AI makes an appearance of creation cheaper
Generative models allow for the description of styles, the combination of traits and the production of numerous variants in a short time. This capacity possesses instrumental value. It also cheapens an operation that previously socially distinguished those who mastered specific codes: recognising which composition, palette, framing or texture refers to a magazine, a school or an era.
What loses scarcity is not taste as a faculty of judgement. Its exhibition as a sign of authorship loses scarcity. An individual may obtain a culturally refined image without mastering the crafts recognised within it and without traversing the decisions that gave rise to those languages. The result may be effective and valuable. Its stylistic correctness still does not state what it reorganises or why it should count as creation.
The prompt adds a layer of selection. Choosing words, discarding results, combining images and repeating generations may imply sustained intention. The quantity of decisions does not resolve the status of the work. A thousand choices may pursue a foreseen appearance and a single decision may alter a field. Counting iterations or minutes of work would be as poor as counting brushstrokes.
Unpredictability is also insufficient. Models return results not entirely determined by the person writing the instruction, yet chance has been a component of art long before AI. The question is what the practice does with that variation: whether it uses it as supply, investigates its conditions, exposes the system, or limits itself to choosing the most convincing output.
Two studies allow one to separate image and attribution. Bellaiche et al. assigned labels of human or artificial authorship to AI-generated paintings and observed more favourable ratings when they were presented as human. Zhang et al. studied behavioural and neural responses in 2025 to identical works labelled as human or AI-generated; the AI labels received lower evaluations.
Experiments show that judgement changes when attribution changes, even if the image remains. They do not prove that authorship is irrelevant or that every rejection responds to an irrational prejudice. Information regarding provenance modifies what the viewer believes they are evaluating: effort, experience, intention, history, or responsibility.
Provenance also matters for reasons those experiments do not measure. Consent and the remuneration of those who contributed works to training sets, the displacement of labour, economic concentration, environmental costs, and the possibility of challenging a result are all significant. None of this is resolved by contemplating the image in isolation. Research into labels measures aesthetic evaluation or preference, not contractual justice, legality, or the distribution of power.
Criticism loses precision when it summarises all these issues into an abstract defence of 'the human'. Technological enthusiasm incurs the inverse error when it treats any formally competent result as creation. Both avoid identifying which relationship, decision, or transformation they consider relevant.
Institutional recognition does not resolve the problem either. An exhibition, an award, or a well-known signature influences the reading of an object. This context can update previously invisible possibilities; it can also turn a reputation into a substitute for examination. The institution participates in the historical status of art, but its acceptance is insufficient as an ontological criterion.
Craft offers an equally insufficient boundary. An arduous work may repeat a code without altering it. A materially minimal gesture may change a field. Measuring creation by hours would exclude conceptual practices; measuring it by capacity for choice would confuse decision and transformation.
The use of an AI tool does not decide the matter. It may be integrated into a practice that investigates its limits, makes its dependencies visible, or produces non-given relationships. It may be limited to generating a recognisable surface. Something similar occurs, with significant technical and legal differences, with a camera, an archive, or a found object.
Creation as reorganisation
In Art as Structural Surplus I propose another criterion. Structural surplus exists when a difference effectively, persistently, and non-trivially reorganises the field of possible configurations of a system without being absorbed by its primary instrumental function. Full artistic status further requires that this reorganisation be updated within a historical field of interpretation, exhibition, and response sedimented as art.
The criterion does not first ask who manufactured the object or how much effort was invested. It examines what difference it produced and whether that difference remains as a new capacity of the field. A list of references may generate surplus; a hand-painted work may not. An exhibition may reorganise relationships between works; an impeccable selection may confirm the available repertoire.
Persistence does not require indefinite material duration. An ephemeral action may stably modify subsequent possibilities of interpretation. Nor does every novelty produce surplus. An unexpected variant may be immediately absorbed as style, effect, or commodity without changing the rules that make it possible.
This framework allows for the distinction of three operations that are often conflated:
- Selection: choosing between possibilities already available.
- Combination: reorganising elements within an existing repertoire.
- Restructuring: modifying the field of possibilities and the relationships that determine what can be done or recognised thereafter.
The first two may participate in a creation. The third explains why some minimal choices acquire disproportionate consequences. Duchamp offers a demanding test because his material work was reduced. The displacement of Fountain opened persistent possibilities for interpreting authorship, the institution, and the ordinary object. The gesture is not exhausted by having recognised something interesting. It changed what the artistic system could do with a manufactured object.
AI may intervene in any of these operations. It may expand selection, accelerate combinations, or become the object of a practice that reorganises the field. It does not receive the status of author in advance, nor is it excluded by definition. The question lies in the effective relationship between system, artist, materials, institution, and reception.
This criterion avoids two shortcuts: reserving art for a human essence or calling every technical novelty art. It compels an examination of what reorganisation occurs, with what persistence, and within what historical field. It also avoids turning 'good taste' into a certificate of creation. A result may be perfectly attuned to the codes of the present and yet add no relevant possibility.
Kant separated the taste that judges from the genius that produces. Hume explained why judgement may possess authority without becoming authorship. Duchamp chose against taste and transformed the scope of choice. Artificial intelligence renders this sequence more visible because it allows one to request a visual culture without having traversed its history.
Possessing good taste merits recognition. Creating demands that something else occurs.
On the open conversation
This text separates verified facts, interpretation, and a personal philosophical thesis. If you work in aesthetics, creation, curatorship, or research into artificial intelligence and can provide a primary source or a precise objection, you may write to me.
The author’s relationship to the cases
I develop the criterion of structural surplus in the preprint Art as Structural Surplus, currently under peer review. I work with digital art and artificial intelligence tools, and I collect digital art. I disseminated the Kottke entry on Bluesky with the commentary from which this article originates. There is no economic or professional relationship with Kottke.org, 713 Magazine, or the other persons and entities cited.
Sources
- Jason Kottke, 'Have We Confused Having Good Taste With Being Creative?', 27 October 2026. The original publication attributed to 713 Magazine could not be verified directly.
- Immanuel Kant, Critique of Judgement, translation by Alejo García Moreno and Juan Ruvira, Madrid, 1876, §§ 46-50.
- Manuel Sánchez Rodríguez, 'The Rejection of the Concept of Good Taste in Kant’s Critique of the Power of Judgment', 2025.
- David Hume, 'Of the Standard of Taste', in La norma del gusto y otros escritos sobre estética, translation by María Teresa Beguiristáin, València, MuVIM, 2008, pp. 37-62.
- Marcel Duchamp, 'Apropos of “Readymades”', in The Essential Writings of Marcel Duchamp, 1975, p. 141.
- David Balzer, Curationism: How Curating Took Over the Art World and Everything Else, Coach House Books and Pluto Press, 2014.
- Kyle Chayka, Filterworld: How Algorithms Flattened Culture, 2024.
- Lucas Bellaiche et al., 'Humans versus AI: whether and why we prefer human-created compared to AI-created artwork', 2023.
- Wenyu Zhang et al., 'Neural correlates of evaluative bias against artificial intelligence-labeled versus human-labeled artworks', 2025.
- Juan A. Esteban Ruiz, Art as Structural Surplus: Toward a Relational Ontology Beyond Human Authorship.