By Robin Lee
AI-generated art is often met with skepticism. Critics question not only its artistic legitimacy but also its ties to copyright infringement and the appropriation of human labor. Yet at the same time, AI art represents one of the most compelling intersections of science and art today. Rooted in algorithms, trained on massive datasets, and executed through computational systems, it is a clear product of technological innovation. Still, its outcomes often resemble human creativity—surprising, visually striking, and emotionally engaging. In a world shaped by rapid technological change, AI art reflects deeper shifts in how we create, perceive, and define art itself. This paper explores those shifts through a thought experiment. Focusing on generative AI systems used in visual art, such as AICAN, DALL·E, GANs, and embodied robots like Ai-Da, I examine how these technologies challenge traditional ideas of creativity. Can a scientific machine truly create art? And what do our answers reveal about the evolving boundaries between artistic expression and scientific design?
To assess whether AI can create art, we must first consider what it means to be creative. One influential framework comes from Margaret Boden, who defines creative work as that which is new, surprising, and valuable.1 This product-based view sets a relatively inclusive standard, and by that measure, many AI-generated artworks already qualify. Scholars like Claire Anscomb note that AICAN’s paintings are stylistically novel and not direct copies of its training data.2 Similarly, Lindsay Brainard contends DALL·E 2’s ability to generate coherent and unexpected images based on text prompts.3 Even if these results are unintentional, they are often found valuable by human viewers. These examples suggest that generative AI systems are not just capable of producing creative artifacts—they are engineered to do so. Systems like GANs and diffusion models operate through recombination and probabilistic variation, producing outputs that depart from known patterns. In this sense, novelty and value are not the side effects but the intended function of the generative AI systems. Still, novelty and value alone may not be enough. Berys Gaut argues that creativity also requires agency.4 In his view, only agents—beings capable of intention and skill—can be creative. A pearl formed by an oyster may be beautiful, but it is not creative. Gaut believes that praiseworthy creativity must involve intentional acts guided by knowledge-how. For example, if an artist accidentally creates a beautiful painting by spilling paint, we do not credit them with creativity because the outcome was not intentionally realized.5 Claire Anscomb applies this reasoning to AI, arguing that systems like AICAN do not possess beliefs, intentions, or evaluative awareness.6 They optimize based on code written by humans and lack autonomous judgment. Thus, while generative AI systems’ outputs may appear creative, these scholars argue that the processes behind them lack the intentionality and understanding required by traditional definitions of artistic agency, and therefore AI-generated art cannot be considered truly creative or artistic.
While traditional theories of creativity emphasize agency, it is worth asking whether this assumption should go unchallenged. Not all creative acts stem from conscious intention or skillful control. Marcel Duchamp’s Fountain and Andy Warhol’s Brillo Boxes, for instance, are widely considered creative despite lacking traditional technical effort or aesthetic intention in their production. As Smith and Cook observe, these mass-produced objects became art only when reframed within a cultural and conceptual context. Their creativity lies not in how they were made, but in how they were positioned and interpreted.7 If such works can be recognized as creative despite their origins, then why should AI-generated images be excluded, especially when they too are subject to cultural framing and aesthetic reception? Of course, some may argue that Fountain or Brillo Boxes were creative only because the human artist made the conscious decision to reframe them as art. AI, by contrast, lacks human intention, emotion, and awareness. But this view overlooks the possibility that creativity exists in degrees, and that systems without full human agency might still participate meaningfully in creative processes. Generative AI models like Dall-E are built on deep learning architectures, specifically transformer-based neural networks, that analyze massive datasets of text and images to learn probabilistic associations between visual forms and linguistic prompts. Their outputs are not random but the result of thousands of training iterations, where the model gradually refines its ability to produce novel, coherent imagery based on statistical patterns. Ai-Da, an embodied drawing robot, integrates camera vision, AI algorithms, and a robotic arm. She uses computer vision to interpret her environment and convert it into gestural outputs through motor control, effectively learning to draw by mapping visual input to mechanical response. In both cases, the system’s “style” is shaped by architecture, data exposure, and iterative training—just as a human artist’s expression is shaped by context and experience. If we are willing to call microbial evolution creative for generating novel, adaptive traits through interaction with environmental pressures, then it becomes plausible to see AI’s outputs as a form of emergent, machine-driven creativity.
This perspective encourages us to move beyond rigid comparisons and consider how creative agency might emerge differently in machines. Lopes proposes that agency is not an all-or-nothing quality, but a set of capacities that can be gradually assembled—such as embodiment, perceptual sensitivity, and the ability to make aesthetic judgments.8 Some of these features are already visible, however faintly, in systems like Ai-Da and GANs. From this view, the question is not whether AI is already an artist, but what kinds of artists we want machines to become. Still, I do not offer this as a defense of AI at all costs. While AI art invites fascinating possibilities, it must not become a commercial tool that undermines or replaces human creators. Rather than dismissing AI outright or embracing it uncritically, we should remain attentive to how it is designed, used, and valued; so that it expands, rather than diminishes, the future of art.
1 Margaret Boden, “Creativity,” in The Routledge Companion to Aesthetics, 2nd ed., ed. Berys Gaut and Dominic McIver Lopes (London and New York: Routledge, 2005), 477–88.
2 Claire Anscomb, “Creating Art with AI,” Odradek: Studies in Philosophy of Literature, Aesthetics, and New Media Theories 8, no. 1 (2022): 13–51, https://odradek.cfs.unipi.it/index.php/odradek/article/view/168.
3 Lindsay Brainard, “The Curious Case of Uncurious Creation,” Inquiry (2023): 1–31, https://doi.org/10.1080/0020174X.2023.2261503.
4 Berys Gaut, “The Value of Creativity,” in Creativity and Philosophy, ed. Berys Gaut and Matthew Kieran (New York: Routledge, 2018), 124–39, https://doi.org/10.4324/9781351199797-8, 130.
5 Ibid.
6 Anscomb, “Creating Art with AI.”
7 Adam Smith and Michael Cook, “AI-Generated Imagery: A New Era for the Readymade,” paper presented at the 2023 ACM Designing Interactive Systems Conference (DIS ’23), Pittsburgh, PA, July 2023, https://doi.org/10.1145/3610591.3616432.
8 Dominic McIver Lopes, AI Art and Artists: What They Are, What They Could Be, What They Should Be (unpublished manuscript, shared in PHIL 339, University of British Columbia, 2024).