Automated and machine-learning systems can now produce images, texts, and sounds with little or no direct human intervention, reviving with new force a question Michael Betancourt posed in 2008 about art made through automated technology: what role must the artist have for a machine-produced work to be acceptable as art? The candidate problem is to give a principled, non-ad hoc account of the human-agency condition under which a machine-produced object — even one exhibited without its generating device — is admissible as an artwork in virtue of the artist's instantiated conventions, together with a treatment of how presentation and attribution modulate that admissibility. The philosophical, legal, and empirical literature leaves this unresolved: standing definitions of art locate art-status in the artist's role yet remain contested, courts have fixed a human-authorship rule for copyright purposes only, and empirical studies document that agency cues shift appreciation without supplying a normative criterion.

Background

Artists have produced works through automated devices since at least the early twentieth century: Mary Hallock-Greenewalt's Sarabet, an electro-mechanical instrument for playing colored light, is a canonical precursor, and from the 1960s onward conceptual art made the intellectual specification of the work — Sol LeWitt's dictum that the idea becomes a machine that makes the art — rather than manual execution carry the artwork. The readymade — a mass-produced object, such as Marcel Duchamp's Fountain (1917), presented as art purely by the artist's act of selection and framing — had already made art-status depend on choice and presentation rather than craftsmanship. Photography, film, and electronic music further accustomed audiences to works whose final form is produced by an apparatus the artist configures rather than by hand. Contemporary generative AI (systems trained on large corpora that produce novel images or texts from a prompt or from seed inputs) pushes this trajectory to the limit: the machine's output may involve minimal, indirect, or contestable human contribution, and it is routinely exhibited detached from any generating device.

Twentieth-century aesthetics explains why this creates a real problem rather than a merely rhetorical one. On an institutional view, being art depends on the discursive and institutional setting — the 'artworld' — that confers the status; on Jerrold Levinson's intentional-historical definition, something is art when its maker intends it for regard in the ways prior artworks were correctly regarded, so the artist's role, not the material or technique, carries art-status. Betancourt's 2008 account of art via automated technology worked in this second, role-based register: what makes a machine-produced object a candidate artwork is the artist's intellectual process and the conventions the artist instantiates and thereby makes effective in the work. 'Conventions' here means the rule-sets an artist designs and instantiates — the explicit, project-specific constraints that govern how the object is produced and how it is to be regarded — including the historically accumulated norms of an art form that the artist activates in doing so.

Betancourt's own account, however, stops short of a criterion. The article's abstract frames the interpretive problem as one of situatedness — 'where should the artwork be situated — in the objects the machine produces, in the machine itself, or in the design for the machine?' — identifies human agency in the creation of art as central to it, and reports that the works examined 'suggest a framework that recognizes the artist's role as the designer of the art, rather than as necessarily the fabricator of those works': a role-based sketch of an answer, not a standard precise enough to decide cases. What remains open, in the surrounding literature, is whether any principled, widely accepted criterion exists for when a machine-produced object — especially one shown without its generating device — should be admitted as an artwork on the basis of the artist's instantiated conventions; 'conventions' has not been made precise enough to apply; and the sociocultural dynamics of presentation — how exhibiting the machine together with its products changes attribution and critical judgment — are decisive but uninvestigated, even though the source's own framing question makes the situation of the work central. The problem thus sits at the intersection of the philosophy of art, the young field of the aesthetics of computer and AI art, empirical reception studies, and the law of authorship, and it can be stated without any specialized formalism.

Problem Statement

Give a principled, non-ad hoc answer to Betancourt's open question — what role must the artist have for a machine-produced work to be acceptable as art? — stated at the source's natural generality: under what conditions is a machine-produced object admissible as an artwork in virtue of the artist's intellectual process and instantiated conventions, including the limiting case where the object is exhibited without its generating device? At that generality the question has three parts. (1) What is the criterion — the minimal role the artist must occupy (specifying conventions, selecting and configuring the system, intending regard-as-art, or some other specified relation), in necessary-and-sufficient, graded, or relational form — with 'instantiated conventions' made precise enough to apply to concrete cases rather than left metaphorical, and discriminating — it must withhold art-status from at least some machine-produced objects and must not be satisfied by the mere production, selection, or presentation of an object? (2) How does the criterion stand against the recognized challenge classes: the conceptual-art and readymade precedents in which the artist's intellectual specification rather than manual execution carries the work, the historical automations (photography, film, electro-mechanical instruments) that became accepted art without a new theory, and the outputs of largely autonomous generative systems? (3) How do presentation variables — especially exhibiting the generating machine alongside its products, and the attribution of the work to a human artist — bear on admissibility, so that the criterion covers rather than ignores the same-object-different-framing cases?

References

  1. Michael Betancourt (2008). 'Intellectual process, visceral result: human agency and the production of artworks via automated technology.' Journal of Visual Art Practice 7(1): 11–18. DOI: 10.1386/jvap.7.1.11_1. https://doi.org/10.1386/jvap.7.1.11_1
  2. Jerrold Levinson (1979). 'Defining Art Historically.' The British Journal of Aesthetics 19(3): 232–250. DOI: 10.1093/bjaesthetics/19.3.232. https://doi.org/10.1093/bjaesthetics/19.3.232
  3. Dominic Lopes (2009). A Philosophy of Computer Art. Routledge. DOI: 10.4324/9780203872345. https://doi.org/10.4324/9780203872345
  4. Mark Coeckelbergh (2017). 'Can Machines Create Art?' Philosophy & Technology 30(3): 285–303 (first published online 2016). DOI: 10.1007/s13347-016-0231-5. https://doi.org/10.1007/s13347-016-0231-5
  5. Rebecca Chamberlain, Caitlin Mullin, Bram Scheerlinck, and Johan Wagemans (2018). 'Putting the art in artificial: Aesthetic responses to computer-generated art.' Psychology of Aesthetics, Creativity, and the Arts 12(2): 177–192. DOI: 10.1037/aca0000136. https://doi.org/10.1037/aca0000136
  6. Jean-Marc Deltorn and Franck Macrez (2021). 'Authorship in the Age of Machine Learning and Artificial Intelligence.' In: The Oxford Handbook of Music Law and Policy. Oxford University Press. DOI: 10.1093/oxfordhb/9780190872243.013.3. https://doi.org/10.1093/oxfordhb/9780190872243.013.3
  7. Ziv Epstein, Aaron Hertzmann, the Investigators of Human Creativity, Memo Akten, Hany Farid, Jessica Fjeld, Morgan R. Frank, Matthew Groh, Laura Herman, Neil Leach, Robert Mahari, Alex 'Sandy' Pentland, Olga Russakovsky, Hope Schroeder, and Amy Smith (2023). 'Art and the science of generative AI.' Science 380(6650): 1110–1111. DOI: 10.1126/science.adh4451. https://doi.org/10.1126/science.adh4451
  8. Thaler v. Perlmutter, 687 F. Supp. 3d 140 (D.D.C. 2023), No. 1:22-cv-01564, decided August 18, 2023. https://www.courtlistener.com/opinion/9421250/thaler-v-perlmutter/
  9. Stephen Thaler v. Shira Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025), No. 23-5233, decided March 18, 2025, affirming 687 F. Supp. 3d 140. Official opinion: https://media.cadc.uscourts.gov/opinions/docs/2025-03-23-5233.pdf; mirror: https://www.courtlistener.com/opinion/10357586/stephen-thaler-v-shira-perlmutter/
  10. Melissa J. Dolese (2026). 'Beyond ethics and aesthetics: perceived lay theory violations and agency disruption in AI-generated art.' Frontiers in Psychology 17. DOI: 10.3389/fpsyg.2026.1716814. https://doi.org/10.3389/fpsyg.2026.1716814
  11. Ivan Z. Stojilović (2026). 'Art, authorship, and AI: Predicting authorship biases from personality traits and attitudes.' Psychology of Aesthetics, Creativity, and the Arts. DOI: 10.1037/aca0000885. https://doi.org/10.1037/aca0000885
  12. Catrin Misselhorn (2025). 'Artificial Intelligence, Authorship and Aesthetic Responsibility in Art.' In: Ästhetik, Digitalisierung und Künstliche Intelligenz, pp. 17–36. Brill. DOI: 10.30965/9783969753477_003. https://doi.org/10.30965/9783969753477_003