Utsav Gupta is a Stanford Master of Liberal Arts candidate researching human purpose in the age of AI and co-lead of the Stanford HAI Cognitive Security Task Force. He is founder and co-CEO of Filarion, building AI and spatial-computing systems across education, law and design, and serves as Commissioner on the Palo Alto Utilities Advisory Commission. Views are his own.
At Stanford, the debate over AI art is no longer hypothetical. A January Daily survey found that 48 of 50 surveyed undergraduates regularly use ChatGPT. Across campus, Stanford scholars are building tools to help visual artists steer AI systems more precisely. In theater and performance studies, Michael Rau has staged “Hamlet.AI,” a production developed with playwright Michael Yates Crowley that uses image generation and AI-written lines streamed to actors mid-scene.
AI is already part of art-making at Stanford. What remains open is how we judge it: Are we willing to discuss AI art with the seriousness we give to other art forms? And can we agree on a standard for when AI-assisted work earns that name?
We should. AI art is art whenever the use of AI does not erase human intention. Whether any given output is good or original or ethical is a separate question. Art is defined by choice, framing, judgment, revision and meaning as much as by the artist’s hand touching the material. In the autumn 2024 course “Art Meets AI: Algorithmic Bodies in East Asia,” undergraduates used ChatGPT-4o to test whether a generative model could capture the wabi-sabi aesthetic of a Japanese tea house. Their work, exhibited at Stanford Libraries during Love Data Week 2025, was not the absence of artistic decision-making. The students chose a tradition, a question, a reference set, a set of outputs to reject and a final image to defend. Those choices are the work.
By that measure, “AI art” covers a range of practices rather than a single category, and Stanford needs a working standard to tell them apart. The most useful one is a set of questions, which I will call the “five questions,” that we can apply to any AI-assisted work: Where is the human contribution? What was disclosed? What was transformed? Who was copied? Who benefits? Those questions, not the slogan “AI is just a tool,” should govern how we judge what students, faculty and The Daily itself produce.
This does not mean that typing one vague prompt and accepting the first result is the same as painting for weeks in McMurtry. But not all art requires the same kind of labor. A photographer does not carve the landscape into being, and a DJ does not invent every sound wave from scratch. We still understand these practices as art because artistry often lies in mediation: what the artist selects, arranges, cuts, repeats, emphasizes or refuses.
Art history gives us a useful warning against technological panic. Painters declared photography the end of painting in 1839; photojournalists later worried Photoshop would erase photographic truth. Both panics were wrong about the death and right about the change. The Metropolitan Museum of Art describes photography, from its birth, as both an artistic medium and a scientific tool. That dual identity, far from diminishing photography’s artistry, helped make it modern. Britannica explains that Duchamp’s use of mass-produced objects helped redefine art as an intellectual, not merely material, process. A urinal became “Fountain” because selection and idea mattered.
The analogy only goes so far. A camera does not ingest the archives of every photographer who came before it. A readymade does not generate a thousand variations on someone else’s catalog overnight. Generative models do both, at scale, and that scale is one component of the difficulty in answering these questions. AI art belongs in that lineage, but its scale raises the bar.
Stanford has a particular reason to take that lineage seriously. The Whitney Museum traces Harold Cohen’s AARON, one of the earliest AI artmaking systems, through Stanford’s Artificial Intelligence Lab from 1973 to 1975. Cohen used code to capture what artists know and how they work. Decades before today’s text-to-image tools, Stanford was already asking whether machines could be part of art-making.
That history should make us more ambitious than the usual pro-AI and anti-AI clichés. The strongest argument for AI art has nothing to do with making creativity effortless. In fact, the best AI art often does the opposite. It forces the artist to wrestle with a powerful system that is frequently wrong. It requires iteration and taste. A model might produce an image in seconds, but meaningful work still depends on the human capacity to know what the image is for.
On the other side, the anti-AI critique is at its strongest right here. Stanford’s Ge Wang calls prompt-based generative art “the least imaginative use of AI imaginable,” likening companies that promise effortless creative output to operators selling helicopter tickets to mountaineers: removing the difficulty also removes what made the work meaningful. But rather than rejecting AI art, that critique should push us to demand AI art that uses the technology to ask more, not less, of the artist.
The harder objections are economic and legal. Artists have reason to worry about training data, attribution, style imitation and market displacement. Stanford’s Samuel Goldberg and UCLA’s H. Tai Lam found that when one online marketplace allowed AI-generated images, total sales rose 39 percent and variety improved, while the number of human-generated images and active human producers fell sharply. The U.S. Copyright Office’s Part 3 report on training data, released in May 2025, is more cautious than the courts have been so far: federal judges in Bartz v. Anthropic and Kadrey v. Meta both found AI training to be fair use in June 2025, while the Office concluded that training can be transformative but that “making commercial use of vast troves of copyrighted works to produce expressive content that competes with them in existing markets” falls outside established fair use. Part 2 had already concluded that the outputs themselves can be copyrighted only where a human author contributes sufficient expressive elements; that is, prompts alone are not enough.
The five questions I propose are evaluative, asking us to look harder at the work rather than label it. A studio class should judge an AI-assisted piece by the gap between the brief and the final image. A TAPS production using AI projections should be reviewed for whether the AI does dramaturgical work or only spectacle. The Daily’s graphics section, which still draws every illustration by hand, should edit AI-assisted illustration by the same standard it applies to its own: whether the image earns its space.
Ge Wang’s warning lands on the graphics desk before it lands anywhere else in the paper. The least imaginative use of AI imaginable would be using it to replace a hand-drawn illustration with a faster machine-drawn one. The most imaginative would be using it in circumstances that no illustrator could meet alone: a series across a dozen variations, a real-time graphic tied to live data, a piece whose final form is the artist’s choice among options the model surfaced. The five questions tell the difference. A graphics policy that asks for the second use and refuses the first would set the bar.
The unproductive response would be to pretend that the tool is either a miracle or a poison. AI does not turn every user into an artist. It also does not make every user a fraud. It is a collaborator, a constraint, a source of error and, sometimes, a mirror. It can flatten creativity into polished sameness. It can also help one visualize an idea they could not otherwise express.
We are a university where code and culture constantly meet: at the Center for Computer Research in Music and Acoustics (CCRMA), in TAPS, at the Stanford Institute for Human-Centered AI, in McMurtry studios and in dorm rooms. Whether AI art is “real” matters less than whether we can build the vocabulary to distinguish art from automation and experimentation from shortcuts.
AI art is art. The machine does not erase the artist. Look at the work itself.