Why AI Will Never Rival Human Creativity

Predictive mechanisms preclude the originality needed for true art.

 Study for Les Demoiselles d’Avignon, Pablo Picasso, 1907. (Photo by Luiz Souza/NurPhoto via Getty Images.)

AI might put artists out of business. It will not, however, replace them. It will not—cannot—make good art, great art: true art. Which is to say, original art. This is, I know, a dangerous prediction (“dangerous prediction”: a redundancy). But unlike the techies and pundits, in their glorious ignorant smugness, I have some sense of what art is and how it is created.

AI operates by making high-probability choices: the most likely next word, in the case of written texts. Artists—painters and sculptors, novelists and poets, filmmakers, composers, choreographers—do the opposite. They make low-probability choices. They make choices that are unexpected, strange, that look like mistakes. Sometimes they are mistakes, recognized, in retrospect, as happy accidents. That is what originality is, by definition: a low-probability choice, a choice that has never been made.

The African masks in Picasso’s Les Demoiselles d’Avignon, to take one of a million examples, were a low-probability choice. So were the footnotes in David Foster Wallace’s Infinite Jest. So was the 40-second chord at the end of The Beatles’ “A Day in the Life.” So is every new metaphor. Elizabeth Hardwick, who wrote criticism at the pitch of art, was famous for her adjectives: “the clamorous serenity of [Frost’s] old age,” Plath’s “ambitious rage,” the “aggressive simplicity“ of the old New York aristocracy. None of these were probable. There are words, in art, for that which is: derivative, stale, clichéd. Boring. 

Low-probability choices are leaps: lateral and unpredictable, associative and idiosyncratic. Where do they come from? Inspiration, we say, a word that explains by not explaining. Inspiration is mysterious (not the same as mystical, though some would say it’s that, as well). Its nature is obscure. It is neither conscious nor unconscious but instead involves a delicate and frequently elusive interplay between the two. It is serendipitous—like standing in a thunderstorm, said Randall Jarrell, and hoping to be struck by lightning. That is why successful works cannot be replicated even by the artists who create them. Every new one is a voyage of discovery, its destination unforeseeable—the very opposite of creating, as the AIs do, to a set of specifications. “The main thing in beginning a novel,” wrote Virginia Woolf, “is to feel, not that you can write it, but that it exists on the far side of a gulf, which words can’t cross: that it’s to be pulled through only in a breathless anguish.” Quality in art is an emergent property: it arises in the doing, in a dialogic dance between the artist and the work. As the work takes shape, it shows the artist what it wants to be.

Joan Didion, wrote Joyce Carol Oates, “began Play It as It Lays with no notion of character or plot or even ‘incident.’ She had only two pictures in her mind: one of empty white space; the other of a minor Hollywood actress being paged in the casino at the Riviera Hotel in Las Vegas.” Saul Bellow found the clue to The Adventures of Augie March, his exuberant early masterpiece, in the sight of water flowing down a street. Why not write a novel, he suddenly thought, that would have “as much freedom of movement as the running water”? Chatbots, in creating text, have only other texts to draw on. Artists draw on the totality of their experience. Both Salman Rushdie, a novelist, and Martin Scorsese, a filmmaker, have talked of being influenced by the music of the Rolling Stones. How do you program a chatbot, or a filmbot, for that—program it to draw on information that belongs to an entirely different medium? How do you program it for “influence”—that non-logical process of absorption, digestion, suggestion—at all?

And Rushdie and Scorsese were not only influenced by music. Like every artist, they were influenced—were shaped, were made—by everything they’d seen and heard and smelled and touched, everything they’d thought and felt and done. Experience: not just the source of art but its very substance—what it’s made from, what it refers to, what it is tested against. It might be possible to code for low-probability choices, I guess, but how would the computer know if the results are worth a damn? Art is good insofar as we recognize it as true, as corresponding to our experience of the world, both inner and outer. But for AIs, there is neither experience nor world. No sights or sounds, no joys or pains, and also no awareness—no idea whether what they make is true, and therefore whether it is good.

In art, what’s more, the true and the good are unstable. Every great breakthrough in art is rejected at first. All truly original art—Cubism, bebop, the dance of Merce Cunningham—is reviled by the standards of its day. Often it is judged to not be art at all. It is noise, or nonsense, or something that your five-year-old can do. AIs create, perforce, according to existing standards. If the true and the good are beyond their ken, all the more so are tomorrow’s true and good, the ones that don’t exist yet, the ones great art brings into being.

We argue whether artificial intelligence is truly intelligent, but even if it is, intelligence and creativity are very different things. Part of the confusion in discussions of AI and art undoubtedly arises from the degraded conception of creativity that has taken hold, in recent years, in tech. Nothing is original, techno-pundits like to say; “everything is a remix.” This is a banality that grew up to become a stupidity. That new creations build upon existing ones has long been a cliché, but the techies have stretched it to mean that nothing is ever original: that creativity involves, and only involves, the rearrangement of existing parts. Which makes you wonder how we ever managed to progress from the first painting in the first cave. Assisting these arguments is the concept of the meme, the idea that elements of culture propagate themselves from mind to mind, just as genes do from body to body. But the meme hypothesis (and it is only a hypothesis) fails to recognize that minds are capable of altering their contents. We don’t just passively transmit ideas and images, nor do we simply recombine them. Somehow, we manage to generate new ones: manage to create—through processes we do not understand and, I do not think, will ever replicate outside the human brain—the elements of culture to begin with. Or, at least, some of us do.

None of this, however, is to say that AI art—artificially generated songs, novels, visual images, even films—will not supplant the human kind. In the age of mass production, people have shown an unending willingness to accept cheap crap in place of costlier quality: in food, in consumer goods, and, more recently, thanks to the internet, in culture. Indeed, having turned art into “content”—limitless, interchangeable, disposable—the internet has already eroded taste to such an extent that fewer and fewer people are capable of distinguishing between crap and quality in the first place. Or bother to. As for artists, those rarities who bring the new to birth, good luck.

William Deresiewicz is an essayist and critic. He is the author of five books including Excellent Sheep, The Death of the Artist, and The End of Solitude: Selected Essays on Culture and Society.

---

Follow Persuasion on Twitter, LinkedIn, and YouTube to keep up with our latest articles, podcasts, and events, as well as updates from excellent writers across our network.

And, to receive pieces like this in your inbox and support our work, subscribe below: