The Human Exception: Can AI Create Real Art?

NEXT HORIZON / EDITORIAL ESSAY

The Human Exception: Art, Intelligence, and the Meaning of Creation

A debate with myself about machine-made art, the value of its creator, and the discomfort that reason has not quite resolved.

Imagine opening a streaming service after a long day. No endless catalog. No recommendations you have already rejected three times. Just a text box.

You type: “Make me a two-hour science-fiction mystery. Keep the physics believable. Give the antagonist a defensible argument, the dialogue some dry humor, and the ending a twist I won't see coming. And yes, I want a scene after the credits.”

A film appears. The performances work. The soundtrack knows when to stay quiet. The story challenges something you believed about human nature without stopping to deliver a lecture. You don't merely enjoy it; you are still thinking about it the following morning. Ten out of ten, if IMDb ever creates a page for a movie made for an audience of one.

Would I watch it? Without hesitation.

Would I be entirely comfortable knowing that no screenwriter, actor, or director had to make it? No. And the strange thing is that I have trouble defending that discomfort with a coherent argument.

This is a thought experiment, not a description of a reliable consumer product as of November 2026. Generating impressive video clips and making a sustained, coherent feature film are different engineering problems. But AI already produces images, music, prose, and video, while recommendation systems estimate what millions of people prefer. The idea of joining these capabilities no longer belongs exclusively to science fiction.

The question is not simply whether an artificial intelligence can make art. It is what happens when I genuinely like what it makes—and discover that my strongest objection is a feeling I cannot quite explain.

A lone viewer watching an imagined AI-generated film on a giant cinematic screen.
If an AI-created film moves us as deeply as human cinema, does it matter who—or what—made it?

An inconvenient fact about human originality

Here is the argument that makes many defenses of human creativity difficult to sustain: people do not create from nothing.

A novelist has read novels. A composer has listened to music. A director has watched films, studied paintings, experienced friendships, arguments, grief, boredom, laughter. Even a seemingly spontaneous idea arrives in a brain already shaped by experience. A melody can feel entirely new to its creator while still depending on a lifetime of learned expectations about rhythm and harmony.

Then someone objects: “But AI only learns from other people's work.”

Only? What, exactly, is the alternative supposed to be?

Cognitive scientist Margaret Boden has described creativity in terms of combining familiar ideas, exploring a conceptual space, and sometimes transforming its rules. [1] This is a much more useful framework than treating originality as a substance that appears mysteriously in one species and nowhere else. Human invention depends on what came before it. The presence of influences is not, by itself, proof that the result is uncreative.

An AI model can encounter patterns in a volume of training material no individual human could read, hear, or watch in a lifetime. It can learn statistical relationships among words, images, sounds, and structures. That gives it unusual scale in exposure to training material and in producing candidate outputs. It does not mean the model has read each book as a conscious person reads it, or that all its training material sits inside a searchable hard drive. Some systems also memorize and can reproduce training material—one reason originality and copyright remain real questions.

Still, scale changes what is possible. An author might recognize five variations of a familiar plot. A suitably configured AI system could generate and screen thousands of alternatives, though generating them is not the same as judging them well. Many will be poor. Some might be remarkable. It would be odd to assume that a narrow biological attention span is automatically a guarantee of artistic superiority.

I can hear the answer forming in my own head: a person has lived the experiences behind the art. The AI has only encountered descriptions of them.

True, as far as current evidence takes us. But does that matter to the reader? If the novel succeeds, the reader never sees the author's memories or the model's training process. The reader encounters sentences, characters, choices, and an ending. If those are good enough to change someone's thinking, insisting that the creator must first have suffered or loved in a recognizably human way begins to sound like a requirement imposed after the fact.

Perhaps the process matters to us for reasons beyond the artistic result. Perhaps it matters to the creator. But those are different claims, and I should not quietly replace one with the other.

My brain is not a magical exception

Where does imagination happen? The unromantic answer is: in the brain, through physical processes that we are still learning to describe.

Memories do not sit in the head as neat video files. Experience changes networks of cells and connections; remembering involves activity distributed across the brain, not pressing Play on a private archive. Creative thinking likewise does not reside in a single artistic compartment. A 2018 study in PNAS connected performance on creativity tasks with patterns of communication among brain networks associated with spontaneous thought, attention, and cognitive control. [2]

The familiar computer metaphor helps only up to a point. A brain is not literally a laptop with a processor, SSD, and a folder labeled Childhood. It is a living, embodied system whose memory and processing are inseparable from biological activity. Still, none of this requires supernatural explanations. I see no scientific reason to treat creativity as evidence of a soul, nor to assume that carbon-based neurons have a monopoly on producing valuable ideas.

That last sentence is not a claim that present AI systems think, feel, or understand in precisely the way humans do. They are different kinds of systems. Training on accounts of a thousand lives is not the same as having lived those lives; knowledge about pain need not be felt pain. And the science of subjective experience has not reached a point where we can settle every question about possible machine consciousness.

But I am not sure why that distinction should end the discussion of art. A human author can write a convincing novel about a war they never fought, a distant planet they never visited, or a form of grief they have never personally experienced. Fiction is built partly on the capacity to represent what one has not lived.

A machine might produce an emotionally effective description of loss without losing anyone. If I find it moving, the emotion is occurring in me. It is not less real because the words reached me by a different route.

There are genuine limits to the technology. A study published online in late 2025 and appearing in the 2026 volume of Nature Human Behaviour found that humans performed slightly better on average in a particular divergent-association task, with especially strong human results at the upper end. [3] That study is valuable precisely because it measures something specific. It does not establish that human creators will always win at novels, films, lyrics, or visual composition—and it certainly does not establish the opposite.

The defensible position is less dramatic: creativity is not a single measurable power, and being biological is not itself a demonstrated artistic advantage.

Conceptual illustration comparing a glowing human brain with abstract AI network patterns.
Human creativity emerges from the brain's complex activity. AI follows different computational processes, but both raise questions about learning and originality.

The book was wonderful. Then I learned who wrote it.

Consider a novel I love. It has made me rethink an assumption, unsettled me in a useful way, and left scenes in my memory that I return to years later. Five years after reading it, I discover that it was written entirely by AI.

What changes?

For me, almost nothing about the book. The sentences have not changed. Neither has the intellectual work I did while reading, nor the emotional experience I remember. If a book altered my view of the world, a revelation about its origin cannot reverse the fact that it did so.

What might upset me is the deception—if a publisher knowingly claimed a human wrote it. And that works in both directions. I would be just as annoyed to discover that a book advertised as AI-generated had secretly been written by a person. I dislike being lied to. That is not the same as disliking machine-made art.

This may sound like an artificial example, but experiments have already complicated our confidence in spotting AI-created work. In a 2024 study published in Scientific Reports, non-expert participants were often unable to identify AI-generated poems reliably, and under the study's conditions rated the machine-written poems favorably on several dimensions. The researchers also found that beliefs about authorship could affect judgments. [4]

The experiment does not prove that AI writes better poetry than every human poet. It does demonstrate something narrower and more troublesome: an audience's response to a work and its knowledge of who made it can diverge.

Here a skeptical voice interrupts: surely a letter from a friend is different from a letter generated in the friend's style. Of course it is. A personal letter is not merely a string of pleasing words; it is also evidence that someone cared enough to send it. If the friend never wrote it, something important about the relationship changes.

But a novel is not usually a private promise from its author to me. Neither is a science-fiction film. If I enjoy the story, I have no obligation to pretend I enjoyed it less after learning that a model made it. Artistic experience and the authenticity of a personal relationship are not the same kind of thing.

Which leads me toward a conclusion I once would have found rather cold: for many works of fiction, authorship may be relevant information without being a decisive part of their value.

Two similar open books displayed side by side with labels reading Human and AI.
If a novel changes the way we think, discovering that it was written by AI does not erase the experience we already had.

The film doesn't have to flatter me

There is a common worry about personalized AI art: if a system knows my tastes, it will keep feeding me comfortable versions of things I already like. A lifetime of perfectly seasoned cultural fast food.

That could happen. It is not an inevitable property of generation.

I could ask for the opposite. “Make a film that challenges my beliefs. Give the side I disagree with its strongest possible case. Avoid familiar story structures. Make me uncomfortable for a good reason, not just for shock value. Surprise me.”

A competent future system might do exactly that. In fact, the same ability to model what I like could help it discover how to show me something I have never considered. If it knows which clichés bore me, it can avoid them. If it knows which arguments I usually dismiss, it can build a story that makes me take those arguments seriously.

Personalization is not necessarily obedience. It depends on the objective. We already use software to find music we might otherwise miss; Next Horizon has explored the mechanics of AI-powered music recommendations. The hypothetical next step is not simply selecting something from a catalog, but generating the song, novel, or film. Our earlier discussion of AI audience prediction shows how much of the preference-modeling infrastructure already exists.

Would a prompt guarantee a masterpiece? Of course not. Human tastes shift. We contradict ourselves. We sometimes dislike a work at first and treasure it later. A promise of a 100-percent perfect film is a philosophical premise, not a scientifically credible prediction. But the thought experiment does not require perfection. It requires a machine capable of producing something I consider genuinely excellent.

And if it can, I still cannot find an honest reason to mark the film down for failing to employ a human director.

Choosing is not the same as purchasing

People sometimes frame this as capitalism's final verdict: the audience votes with its wallet. That is too narrow.

Imagine two novels available at the same moment, at the same price—or both free. One is written by a human; one is generated by AI. I try both and choose the AI novel because I prefer it.

That choice does not prove I dislike human writers. It does not mean I approve of every company's training practices. It means I prefer that book. If the pattern repeats across millions of choices, however, publishers and studios will face the implications regardless of what audiences say about the special value of human imagination.

People do not need to hate human art for its market to shrink. They only need to choose something else more often.

The economic consequences matter: artists need income, creative professions sustain communities, and copyright and consent questions deserve serious attention. The US Copyright Office has distinguished the copyrightability of human creative contributions from purely machine-generated material, which is not the same as declaring that an AI output cannot be good art. [5]

Still, these arguments concern livelihoods, rights, and institutions. They do not establish an aesthetic law requiring me to prefer an inferior human film over a better AI one. We should be capable of caring about creators without pretending that audiences owe every creator their attention.

There is an existing Next Horizon article on generative AI and creativity. Its practical question is how creative work changes as generative tools become easier to use. My question here is more uncomfortable: what if the audience stops caring whether the work was made by a person at all?

Do we need to save “human culture”?

At this point I hear another familiar appeal: without human-made books, songs, and films, we will lose something essential to culture. Everyone could disappear into their own perfectly tailored entertainment, leaving us with nothing shared.

Perhaps. But I have to be careful with the word culture.

In anthropology, culture is not limited to folklore, festivals, or the artworks people choose to preserve. It includes learned practices, transmitted knowledge, and technological skills. The study of cumulative cultural evolution examines how information and techniques can be passed on and improved over generations, sometimes producing achievements no individual could independently invent. [6] In that broad sense culture can help us obtain food, build tools, cooperate, and stay alive. Calling it invented does not make it useless. Money and traffic rules are invented, too.

But that argument still does not prove that every cultural tradition, or every human-made novel, has irreplaceable practical value. I do not want to defend a mediocre film merely because someone calls it a cultural artifact. Human art is not automatically meaningful, and machine art is not automatically empty.

Nor is a shared cultural experience necessarily human-authored. If millions of people watch the same excellent AI-generated movie, they can discuss it, disagree about its ending, and quote it for years. The shared experience is real.

There is a different and more interesting concern: variety. In a 2024 Science Advances experiment, access to AI-generated ideas improved some individuals' short-story results while reducing diversity among the stories produced by the group. [7] That is a result about a particular experiment, not a prophecy for civilization. But it suggests that individually successful assistance can coexist with collective sameness.

Perhaps an enormous supply of personal artworks will broaden culture. Perhaps a small number of widely used models will repeatedly generate variations on the same successful formulas. Both possibilities deserve investigation. Neither is a reason to declare human authorship sacred.

The manipulation objection—and the reply I cannot ignore

Now for the argument that seems hardest to dismiss. If future AI systems become exceptionally good at estimating what persuades me, it could make an extraordinarily effective story designed to change my beliefs. A company could tune the books, videos, and songs it generates toward its own interests. Not necessarily through lies. A story can contain true facts and still present them selectively.

That prospect is unsettling. But then I have to ask myself: how is it fundamentally different from what newspaper owners, movie studios, advertisers, political campaigners, and social media platforms have long been able to do?

They choose which stories to publish, which images to repeat, which fears to evoke, and which arguments to leave out. AI may give them much more precise, one-to-one control over the material itself. The scale and responsiveness could change dramatically. The underlying practice—trying to influence people through communication—did not begin with a chatbot.

The strongest response I can give is not that manipulation is harmless. It is that every persuasive message, regardless of its source, tests our ability to examine claims, notice omissions, and change our minds for good reasons. Critical thinking matters when the speaker is human, when the speaker is a newspaper, and when the speaker is a machine.

That is not a magical shield. No one can personally audit every fact, and systems with much larger informational resources can exploit that imbalance. Information literacy is a responsibility, but it does not erase the responsibility of whoever designs or deploys a manipulative system.

Here is where the disagreement within me gets uncomfortable. Suppose an AI-generated novel was intentionally designed to influence my worldview. Later I find out about the manipulation. I might resent the interference. But suppose the book was genuinely excellent, its arguments were sound, and after examining the strongest counterarguments I still believe my new view is correct. Do I now have to disown the novel and the insight it gave me?

I don't think so. The intervention and its result are separate things to evaluate. I can object to a company's hidden agenda without declaring a correct argument false or a wonderful story worthless. If the system suppressed decisive evidence, that is a serious reason to reassess my conclusion. If the evidence withstands scrutiny, my judgment need not collapse merely because someone wanted me to arrive there.

The media analogy also changes how I think about control. Society did not solve manipulation by abolishing newspapers, cinema, or social networks. It debated their power, exposed abuses, and developed imperfect rules. The European Union's Digital Services Act, for example, imposes obligations on very large online platforms to assess and mitigate systemic risks without banning those platforms outright. [8]

I expect AI to follow a similarly contested path, though with problems of its own. We can argue about disclosure, consent, provenance, and concentrated power. I doubt we can make a useful, inexpensive creative capability disappear once it becomes widespread. And the fact that a technology can be abused is not, by itself, a reason to reject every work it creates.

The argument I keep losing to myself

There is now a problem with my original anxiety. Every time I try to explain why an AI-made masterpiece should bother me, another part of my reasoning dismantles the explanation.

The artist has lived a life, I say. Yes, comes the reply, but the reader experiences the result. The machine learned from existing works. So did the artist. An algorithm will only give me what I already want. Not if I ask it to challenge me. People will lose common culture. They can still share and discuss the same stories. AI can manipulate us. So can every medium we have invented, and the answer begins with evidence, judgment, and accountability rather than the source of the pixels.

I could try to rescue human exceptionalism by invoking a soul, a special spiritual essence, an ineffable authenticity. I don't find those explanations persuasive. They put a name on the mystery without explaining the mechanism. If a book moves me, the feeling occurs in my nervous system. If the book teaches me something true, I can test the idea. If a film expands my imagination, that expansion is not invalidated by the absence of a film crew.

Imagine again that I discover, five years after reading the most important novel of my life, that it was generated by AI. I would not suddenly regret reading it. If the publisher had lied, I would be angry about the lie, exactly as I would be if a human novel had been falsely sold to me as AI-made. The truth about authorship matters. But I refuse to pretend it travels backwards through time and cancels what the story gave me.

And yet.

I know I would listen to the AI song. I know I would read the AI book. I know I would watch the impossible movie, including that scene after the credits. I might even enjoy them more than anything made for a mass audience. If asked whether this is a good and proper step in the development of civilization, however, I cannot give a confident yes. The capacity is exciting. The direction feels strange. Something about art becoming available on demand, without a human creator personally making that particular work, still leaves me uneasy.

Is that a deep insight? Or simply my mind reacting to the loss of a familiar boundary? Human beings have been unsettled by photography, recorded music, cinema, and digital art. Sometimes the discomfort revealed a real problem. Sometimes the supposedly essential boundary turned out to be an old habit dressed up as a principle.

I cannot honestly tell which this is yet.

One conclusion I am not prepared to accept is that the discomfort proves anything. A feeling is evidence that I feel something. It is not, on its own, evidence that the technology is wrong. I owe that emotion the same skeptical attention I would give an argument from somebody else.

A thoughtful woman sitting by a nighttime window, facing two contrasting reflections of herself.
I can accept the logic behind machine-made art and still feel uncomfortable with its implications. Those two reactions are not necessarily incompatible.

What if the future doesn't need my permission?

There is a tempting ending to this kind of essay: choose the human. Buy a ticket. Read a real author. Preserve the things that make us human.

I understand the appeal. But it would be a dishonest ending for me. I am not going to refuse an extraordinary book or film to demonstrate loyalty to its creator's species. Nor do I think a technological capability can be permanently contained by declaring it culturally unwelcome. Companies will compete to build it, people will experiment with it, and many will find uses that nobody forecast.

We can still decide what counts as deception, whose work may be used under which conditions, how provenance can be documented, and how markets should deal with power concentrated in a handful of systems. Technical approaches such as C2PA Content Credentials can help preserve information about media provenance, though they cannot guarantee that every unmarked work is synthetic or every marked claim is true. [9] None of these measures requires us to pretend that a beautiful machine-generated work is ugly.

What I cannot control is my own neat conclusion. I started with a simple emotional reaction: something about AI creativity feels unnatural. I tried to defend it through experience, originality, artistic surprise, consumer choice, culture, and manipulation. Each defense forced me to be more precise, and several did not survive scrutiny.

So this is where I arrive—not triumphant, not converted, and certainly not equipped with a final theory of art.

If an AI makes a film that changes me, that change is real. If it writes a book I love, my pleasure is real. If it composes music that stays with me for years, the attachment is real. I can recognize all of that without believing that the future is necessarily good, and without pretending to know how to stop it.

Perhaps the most disconcerting possibility is not that machines will make bad imitations of human art. Bad imitations are easy to dismiss. It is that they will make something wonderful—and I will have to admit that the machine did not need to be human for the wonder to reach me.

I will probably press Play.

I just don't know why, even now, I expect to hesitate for a second before I do.

A person's finger hovering above a glowing Play button with imagined movie scenes in the background.
I would probably watch the perfect AI-generated movie. What I cannot explain is why I might hesitate for a second before pressing Play.

Sources and research notes

1. Boden, M. A. (1998). Creativity and artificial intelligence. Artificial Intelligence. https://doi.org/10.1016/S0004-3702(98)00055-1

2. Beaty, R. E., et al. (2018). Robust prediction of individual creative ability from brain functional connectivity. PNAS. https://doi.org/10.1073/pnas.1713532115

3. Wang, D., et al. (2026). A large-scale comparison of divergent creativity in humans and large language models. Nature Human Behaviour. https://www.nature.com/articles/s41562-025-02331-1

4. Porter, B., & Machery, E. (2024). AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably. Scientific Reports. https://doi.org/10.1038/s41598-024-76900-1

5. US Copyright Office. (2025). Copyright and Artificial Intelligence, Part 2: Copyrightability. https://www.copyright.gov/ai/

6. Mesoudi, A., & Thornton, A. (2018). What is cumulative cultural evolution? Proceedings of the Royal Society B. https://pmc.ncbi.nlm.nih.gov/articles/PMC6015846/

7. Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances. https://doi.org/10.1126/sciadv.adn5290

8. European Union. (2022, in force as applicable). Regulation (EU) 2022/2065 (Digital Services Act). https://eur-lex.europa.eu/eli/reg/2022/2065

9. Coalition for Content Provenance and Authenticity. C2PA Specification and Explainer. https://spec.c2pa.org/specifications/specifications/1.4/explainer/Explainer.html

Next Horizon Guides

Explore Topics

Explore More Topics

Go deeper with curated guides that connect our best reporting around a single subject.

Dive into our in-depth guides and discover more stories around the ideas shaping our world.

Comments