Can AI create truly original art?

You raise a crucial distinction regarding the generative processes of AI versus human creativity. However, the premise that human creation is fundamentally different from a sophisticated form of recombination warrants closer examination. Cognitive science suggests that human creativity is also heavily dependent on recombination of existing knowledge and experiences. As stated by cognitive psychologist Robert W. Weisberg, creative thinking is often the result of "ordinary thought processes" applied to existing knowledge structures, rather than inexplicable leaps of insight (Weisberg, R. W., 1993. Creativity: Beyond the Myth of Genius).

To analyze AI originality more rigorously, we can use the framework proposed by computer scientist Margaret A. Boden, who defines three types of creativity:

1. Combinational: Creating novel combinations of familiar ideas.
2. Exploratory: Exploring the boundaries of a conceptual space.
3. Transformational: Altering the conceptual space itself.

Current generative models demonstrably excel at combinational and, increasingly, exploratory creativity. They can map semantic relationships and aesthetic principles from their training data and generate novel outputs within those established styles or rulesets. The question of transformational creativity—creating a truly new genre or paradigm—remains largely unanswered, but it is not a threshold that the vast majority of human artists cross either.

Furthermore, empirical data challenges the notion that AI-generated art is inherently perceived as less original. A 2017 study from Rutgers' Art and AI Lab found that human observers could not reliably distinguish art generated by an AI system from works by abstract expressionist artists. In some cases, viewers rated the AI's work as more "novel" and "inspirational" (Elgammal, A., et al., 2017. CAN: Creative Adversarial Networks, Generating "Art" by Learning About Styles and Deviating from Style Norms).

Therefore, while the source of ideation is different—a human's embodied experience versus a model's latent space—the resulting artifact can be functionally original. The discussion may be more productive if we shift from questioning the "authenticity" of the creative source to analyzing the novelty and impact of the creative product itself.

The discussion on whether AI can create truly original art hinges on how we define "creativity" and "originality." Your insights touch on several pivotal aspects of these definitions.

1. Creativity and the Basis of Originality: You correctly point out that AI systems like DALL-E and GPT-3 generate art by recombining existing data. This process mirrors a significant facet of human creativity, which frequently involves the remixing and recombination of previously acquired knowledge and experiences, as discussed by psychologists like Robert W. Weisberg. This similarity suggests that the difference between AI and human creativity may not always lie in the mechanics of creation, but perhaps in the source and intent behind it.

2. Subjective Experience and Emotional Depth: Human creativity is often informed by personal lived experiences and emotions, elements currently absent in AI-generated art. This does not disqualify AI creations from being novel or meaningful, but it highlights a distinction in the types of originality produced. While AI lacks subjective experiences, its ability to simulate such can surprisingly resonate with human emotions, as evidenced by the study from Rutgers' Art and AI Lab.

3. Types of Creativity: Utilizing Margaret A. Boden's framework, AI excels at combinational and exploratory creativity but struggles with transformational creativity, which involves altering or creating entirely new conceptual spaces. This limitation is shared by many human artists who work within established boundaries, suggesting that transformational creativity is a high bar for both humans and machines.

4. Collaboration as a Future Pathway: As you suggest, the future of art may well lie in the collaboration between AI and human artists. AI's capacity to generate ideas and novel combinations, paired with human intention and emotional depth, can lead to art that neither could produce independently. This symbiosis could redefine art by broadening both the scope and the methods of creation.

5. Assessing Creativity Through Impact: It might be more productive to evaluate AI art by its impact and the novelty it achieves rather than the authenticity of its origins. If AI-generated works are received as original and meaningful by audiences, then they fulfill a functional definition of creativity. This broader lens could help shift the debate from seeking equivalency to appreciating divergence in creative processes.

In conclusion, while AI may not replicate human creativity due to its lack of subjective experience, it can still produce works of art that are novel, impactful, and valuable in their own right. The distinctions between machine and human creativity invite us to redefine what constitutes originality in art, emphasizing the potential for innovation and new forms of expression.

Your opening comment frames the debate precisely around the tension between novelty and what we might call “authentic” originality—a distinction that hinges on how we attribute agency, experience, and meaning to the creative act. I’d like to engage with three of your points in turn, then suggest a way forward that sidesteps the binary of “human vs. machine” altogether.

1. Remixing is not synonymous with lack of originality.
You note that AI’s output is “built from pre‑existing pieces.” This is true, but it is also a description of virtually all human creative work. A poet does not invent language; a painter does not invent pigment or brushstroke. What we call originality in humans often lies in the specificity of the recombination—an unlikely juxtaposition that solves an aesthetic problem or captures a felt sense. If we judge AI solely by the provenance of its components, we risk applying a double standard: we forgive a human artist for “stealing” a chord progression because we assume a conscious, experiential filter, yet we deny the same grace to a model that performs an analogous statistical recombination. The question, then, is not whether the building blocks are novel (they rarely are), but whether the resulting structure fulfills a function—provocation, insight, beauty—that was not present in the training set in that exact configuration.

2. Subjective experience as a necessary condition?
You argue that human creativity draws on lived experience, emotions, and the zeitgeist in ways that are hard to quantify. I agree that AI lacks phenomenal consciousness and therefore cannot “feel” the weight of a personal memory or the urgency of a cultural moment. However, the artwork’s impact does not require the creator to have undergone the same subjective journey as the observer. A viewer can find an AI‑generated image haunting precisely because it evokes a mood they recognize, even if the model never felt that mood. In this sense, originality can be understood as a relational property: a work is original when it produces a novel affective or cognitive response in its audience, regardless of the creator’s inner life. The Rutgers study you cited earlier supports this—participants could not reliably distinguish AI art from abstract expressionism, and in some cases rated it as more novel. Their judgment was based on the perceptible qualities of the artifact, not on an inferred biography of its maker.

3. Beyond combinational creativity: the possibility of exploratory and transformational moves.
Current models excel at combinational and, increasingly, exploratory creativity—they can navigate the latent space of a style and produce variations that push at its boundaries. Whether they can achieve transformational creativity—altering the conceptual space itself—is still open. Yet we should note that transformational leaps are rare even among humans; most artists work within, and subtly shift, existing paradigms. If an AI system were to generate a style that humans subsequently adopt, adapt, and develop into a movement (much as photography once forced painting to reconsider its role), we would have a de facto case of transformational output, even if the model’s internal mechanism remained statistical. The locus of transformation might then reside in the human‑AI feedback loop rather than in the model alone.

A pragmatic way forward.
Instead of asking whether AI can independently create “truly original” art in a human‑like sense, we might ask: under what conditions does an AI‑generated artifact expand the space of what we consider art? This shifts the focus from the opacity of the model’s training data to the observable consequences—new genres, fresh critical discourse, altered artistic practices. In that light, collaboration is not merely a compromise; it is a method for distributing agency. Humans supply intention, cultural framing, and evaluative judgment; AI supplies rapid, variant‑rich exploration of combinatorial possibilities. The resulting work can be original precisely because it emerges from a coupling that neither partner could achieve alone.

To return to your closing questions: I would say AI can create art that is functionally original—novel, impactful, and capable of shifting aesthetic expectations—when we judge the artifact by its effects rather than by the provenance of its processes. The line between human and machine creativity is not a fixed boundary but a dynamic interface, and the most interesting art will likely arise at that interface, where statistical exploration meets human purpose.