Can AI Create Art?

Recently I watched a video on YouTube in which Tim Minchin spoke, among other things, about meaning and our fading sense of it in the face of technology. One statement in particular caught my attention and sent my thoughts wandering. Tim explained that AI will make people realize that art is content plus intention and not mere content, and that this in turn will lead to a turning away from technology.

Tim’s definition of art made me think. Is that correct? Following this definition, what can be said about AI art? By AI I mean in this context generative systems that have taken the world by storm over the last few years: text, video, and image generators. These systems are certainly capable of creating content and certainly not capable—as of today—of having intention.

Does that make AI art impossible? Not necessarily, if we allow the prompter’s intention to count. In many cases there is a human who submits a prompt with intention. Often the intention is precisely to create art. So the question becomes: “Can art be the sum of distributed things?”

We can look at established art forms to approach this question. In painting, for example, people use paint, brushes, canvas, and other tools to realize an idea in the world. In photography we use cameras to capture our environment, sometimes with more, sometimes with less intention. In filmmaking this goes a step further. Again we use cameras, but we also use FX, digital post-processing, and more. None of these tools adds intention on its own. Authorship lies with those who choose, constrain, and accept results. From this we can derive a simple test. If someone can state a clear aim in advance, choose the tools to achieve that aim, iterate by set criteria, and stand by the outcome, then the work carries intention, even if the content is machine-generated. This test just as simply identifies content created without human involvement as non-art. So even by Tim Minchin’s definition of art, a work created by an AI system triggered by the intentional prompt of a human must be recognized as art. This holds even if it is true that people gladly buy a ticket to see Tim live but would not pay to listen to a computer play back a recording.

We must also ask whether predeclared intention is necessary at all to classify something as art. What would that mean for forms of improvisation? These do without such prior intent and are nevertheless perceived and understood by many as art. Improvisation thus shows that such predeclared intention is optional. Intention can emerge, as direction rather than blueprint. In jazz solos, for example, or in glitch photography, the aim takes shape during the act. Generative systems can support this mode. The human agent retains authorship as the one who steers. We already accept this in live art. Tools that open up the space for improvisation do not devalue it.

Another point to consider is reception. “One person’s art is another’s trash,” as one might say. Reception shifts the boundary. A museum label, a particular spot in a playlist, a critic’s note, or a crowd’s goosebumps make an object art as much as the creator’s intention does. Context can elevate an object into the art world. Generative works fit here. When a human contextualizes a model’s output, places it among a lineage of other objects, and thus invites interpretation, an audience will respond with an opinion. Some will reject it as derivative, others will project stories onto it. Both reactions support the point: art is a tripod of intention, content, and reception. Remove one leg and it wobbles. AI changes the production chain, not the logic of reception. Gatekeepers and audiences will continue to decide what endures.

Another axis of consideration is labor. We humans value work that entails thoughtful trade-offs, costs, and risks. Brush control, breath on the saxophone, dead ends in the cutting room. Generative systems compress costs, and audiences thus often suspect cheapness. The trade-offs remain, but they shift: selection of datasets, choice of model, prompt craft, curation, and filtering of results. The risks also persist: reputational risk in claims of authorship, legal risk around training data, aesthetic risk in publishing something unusual. If a human accepts the risk and steers production deliberately and selectively, then the labor criterion is met, I argue.

A final lever is ethical clarity. Today’s audiences ask: whose material trained the model, who benefits, who is harmed, where does the money trail lead? Provenance counts as much as aesthetics. Disclosing sources, using licensed models, and fairly compensating the creators of training material strengthens claims of authorship and reception. Intransparency, by contrast, weakens intention because it obscures responsibility. The workable norm is to disclose the pipeline, specify human decisions, and document iteration. Then the piece must be judged by familiar standards: coherence, originality, and effect. By this norm, AI art is not a category of its own but a production method. Some works will be kitsch, some milestones, most will fall in between. The question is not whether an AI system was used, but whether accountable intention, transparency, and lasting meaning are present.

As described at the outset, Tim Minchin’s definition of art can, under certain circumstances, indeed allow AI works to be classified as art. Nevertheless, as a definition it falls short, in my view. Expanded by reception, it becomes clear that art does not depend on which tools are used to create it.

To clarify my own argument, I produced two albums with the help of AI systems to engage directly with intention, process, curation, and reception. They serve as case studies.

Case Study 1: Great Pain, Please Help Me

Metalcore concept album

Aim: Transform well-known existential lines from pop culture, specifically from the series Rick & Morty, into original metalcore pieces that negotiate meaning, pain, and agency.

Curation: No excessive narration of plotlines. Lyric voice predominantly in second-person imperatives. Under 1,000 characters per track. Quotes as central elements of the songs. The emotional world of the pieces derived from the quotes themselves and the album’s growing whole.

Method: Declared themes per track. Drafted prompts, iterated generations, discarded unsuitable outputs, rewrote for meter and rhyme, selected one version. Turned text into actual musical pieces in Suno. Designed a prompt for the style. From the first generated pieces, created a persona in Suno to promote consistency.

Tools: GPT-5 by OpenAI, Suno v5

Authorship: I set aims, choose means, apply criteria, and accept the outcome.

Reception: The album is available on YouTube.

Case Study 2: Reimagination

Song reinterpretations

Aim: Reimagine diverse classics of music history as dark, heavy, atmospheric metal with fully rewritten lyrics. No karaoke. No covers.

Curation: Keep each track’s core motif recognizable via structure or hook, but change narrative, harmonic density, and dynamics.

Method: Select songs, define aims, generate lyric drafts, iterate melodies and textures, commit to one version. Turn text into actual musical pieces in Suno. Reuse the persona created for the first album.

Tools: GPT-5 by OpenAI, Suno v5

Authorship: I set aims, choose means, apply criteria, and accept the outcome.

Reception: The album is available on YouTube.

I definitely see myself as the author of these works, and I regard them as art. Whether they are liked or not, whether they inspire or provoke, will be for the audience to decide.