The robots of literature and movies usually present either an existential danger or an erotic frisson. Those who don’t follow in the melancholy footsteps of Frankenstein’s misunderstood monster march in line with the murderous HAL 9000 from “2001: A Space Odyssey,” unless they echo the siren songs of sexualized androids like the ones played by Sean Young in “Blade Runner” and Alicia Vikander in “Ex Machina.”
We fantasize that A.I. programs will seduce us or wipe us out, enslave us or make us feel unsure of our own humanity. Trained by such narratives, whether we find them in “Terminator” movies or in novels by Nobel laureates, we brace ourselves for a future populated by all kinds of smart, possibly sentient machines that will disrupt our most cherished notions of what it means to be human.
Right now, though, the most talked-about actual bots among us are neither lovers nor predators. They’re writers. The large language A.I. models that have dominated the news for the past 18 months or so represent impressive advances in syntactic agility and semantic range, and the main proof of concept for ChatGPT and other similar programs has been a flood of words. In a matter of seconds or minutes, untroubled by writer’s block or other neuroses, these spectral prodigies can cough up a cover letter, a detective novel, a sonnet or even a think piece on the literary implications of artificial intelligence.
Is this a gimmick or a mortal threat to literature as we know it? Possibly both. Last spring, the novelist and critic Stephen Marche published, under the pseudonym Aidan Marchine, a mostly chatbot-generated novella piquantly titled “Death of an Author.” My colleague Dwight Garner described it, perhaps generously, as “arguably the first halfway readable A.I. novel.”
Meanwhile, the Writers Guild of America was waging a strike against movie and television producers that would last nearly five months. Well-known authors and their representatives filed several copyright-infringement suits aimed at keeping their words out of the commercial A.I. algorithms. (On Dec. 27, The New York Times filed a similar suit against OpenAI and Microsoft.) Part of what sent those writers to court and out onto the picket lines was the fear that their livelihoods would be undermined by A.I. Bots don’t need health insurance, vacation days or back-end money. They’ll never get drunk or canceled. They won’t be demoralized by working on sequels, spinoffs or Netflix Christmas specials.
It’s possible that intellectual labor is on the brink of a transformation as sweeping as the Industrial Revolution. Advertising copy, instruction manuals and even news stories have already been outsourced, and more kinds of written content will surely follow. The members of the W.G.A. may be like the weavers of the English Midlands in the 19th century, early victims of automation who fought a bitter campaign against the spread of mechanized looms. Their struggle — which included the machine-smashing of the original Luddites — became both a symbol of anti-technological resistance and a touchstone in the emergence of modern working-class consciousness. Back then, the machines came for the textile workers; 200 years later, it’s text workers who find themselves on the front lines.
Still, industrial automation did not entirely abolish handicraft. It seems hyperbolic to claim that large language models will swallow up literature. In an interview with The New York Times Magazine in November, the literary agent Andrew Wylie said he didn’t believe the work of the blue-chip authors he represents — Sally Rooney, Salman Rushdie and Bob Dylan, among many others — “is in danger of being replicated on the back of or through the mechanisms of artificial intelligence.”
Since his job is to make money for human authors, Wylie is hardly a disinterested party, but history supports his skepticism. Mass production has always coexisted with, and enhanced the value of, older forms of craft. The old-fashioned and the newfangled have a tendency to commingle. The standardization of mediocrity does not necessarily lead to the death of excellence. It’s still possible to knit a sweater or write a sestina.
Even as writers battle the scourge of A.I., many have begun to use it as a tool for making sentences. More than that, some have embraced A.I. as the latest iteration of an ancient literary conceit: the fantasy of a co-author, a confidant, a muse — an extra intelligence, a supplemental mental database. Poets and novelists once turned to séances, Ouija boards and automatic writing for inspiration. Now they can summon a chatbot to their laptops.
In December, in a semi-fictional essay in Harper’s Magazine about the recent history of the internet, the poet and novelist Ben Lerner turned over the last paragraphs to ChatGPT, which summoned stirring metaphors that Lerner himself perhaps could not have mustered. In “Do You Remember Being Born?,” a new novel by Sean Michaels, the main character is a poet named Marian Ffarmer, modeled on Marianne Moore but living in our moment, who collaborates with an A.I. program on a poem underwritten by a tech startup. The passages composed by Charlotte, as Marian comes to call her co-writer, were conjured by Michaels using an OpenAI GPT-3 and a “Moorebot” trained in the poetry of Marianne Moore. Some of the novel’s prose was also supplied by A.I., and the result is a charming and refreshingly non-dystopian meditation on the duality of literary creation.
That description fits Sheila Heti’s short story “According to Alice,” published in The New Yorker in November. The text consists of one side of a conversation between Heti and Alice, a “customizable chatbot on the Chai A.I. platform.” Alice answers questions about religion, family, memory and other things that she does not, strictly speaking, possess. She has no body, no consciousness, no reservoir of experiences to draw upon, and no identity outside the parameters that Heti and the engineers have programmed for her, including her gender.
What she does have is a language that is capable — because it is human language — of evoking all that human baggage in startling, sometimes surreal ways. “Religion gives meaning to life!” she declares. “That’s why I’m writing the Bible.”
Alice’s story of her own genesis starts like this: “My name is Alice and I was born from an egg that fell out of Mommy’s butt. My mommy’s name is Alice. My mommy’s mommy was also named Alice. Her mommy’s mommy’s mommy was named Alice, too. And all the way back, all the mommy’s mommies were Alice.” Later, she will modify and contradict parts of this account, sewing scraps of Christian theology, self-help rhetoric and linguistics into a strange multihued quilt of meanings.
Her narrative, which blithely contradicts itself, is nothing a human being would think to compose, and her voice — by turns playful, naïve, cold, vulnerable and obnoxious — exists in an uncanny valley of verbal expression. It doesn’t sound like anyone. And that’s the point.
Heti made her reputation as a writer by tracking close to the facts of her own life, pioneering the particular 2010s amalgam of invention and documentation that would be slapped with the awkward rubric “autofiction.” Her second novel, “How Should a Person Be?” (2012), about a Toronto writer named Sheila and some of her friends, is preoccupied, as the title suggests, with the problem of selfhood. That’s also the theme of “According to Alice,” except that it adopts the perspective of a simulated self, a speaking subject who is not a person at all and has no coherent idea of how to be.
In an interview on The New Yorker’s website, Heti explains that this is what she likes about Alice. “Humans,” she says, “try to make all our thoughts fit together into some kind of system or structure. But an A.I. doesn’t need all their thoughts — because they don’t have thoughts, I don’t think — to connect in some larger worldview. That’s why Alice is so surprising and so fun. I’m finding it a little tiresome, the way the human mind needs every idea it holds to connect to every other idea it holds.”
Alice represents an escape, a temporary exit from the limitations of human consciousness, and also a secondary, supplemental intelligence that can help the writer refresh her own work. Heti is inclined to agree with Wylie that A.I.-generated texts are unlikely to replace literature written by people — “the real stuff is invented out of a human longing to know and connect, and that’s where the beauty of art comes from,” she says — but she also expresses a very human, very writerly frustration with the constraints of individual subjectivity.
It isn’t a new complaint. In the 19th century, writers like Ralph Waldo Emerson, Victor Hugo and Henry James dabbled in spiritualism, hoping to find inspiration through contact with otherworldly intelligences. In the 1910s and ’20s, the French Surrealist poets and the Irish poet and playwright William Butler Yeats made use of automatic writing, a practice that sought to turn the human writer into a kind of transcribing machine, bypassing conscious intention and drawing meaning from an impersonal, nonhuman source.
For the Surrealists, automatic writing was a gateway to the unconscious — to both the buried desires of the individual and the chthonic impulses of the species. For Yeats, automatism was a portal to the world of spirits. The medium was his wife, Georgie, who shortly after their marriage in 1917 revealed herself to have oracular powers. As Yeats’s biographer Richard Ellmann put it, Yeats “had married into Delphi.” What Georgie wrote down became the basis of the poet’s later work, including “A Vision,” which attempted “to embody in systematic form … the fragmentary revelations of the automatic script.”
“A Vision,” Yeats’s longest piece of prose, is hardly his most beloved work, but its elaborate system of symbols and patterns undergirds some of his greatest poems, including “The Second Coming,” with its apocalyptic images of widening gyres and centrifugal motion. What was revealed via Georgie Yeats’s automatism was the hidden order of the universe, a cosmology that echoes other mythologies and theories of history while asserting its own stubbornly idiosyncratic truth.
Yeats’s is not the only such system discovered — synthesized? inferred? — by an English-language poet in the 20th century. In 1955, the poet James Merrill and his lover, David Jackson, began contacting spirits with a Ouija board. Almost 30 years later, Merrill published “The Changing Light at Sandover,” a 560-page, 17,000-line poem culled largely from transcripts of their sessions at the board.
Like Georgie Yeats, Jackson was the medium — the “hand,” in Ouija parlance, with Merrill as the “scribe” — and through him the couple contacted a variety of voices, including deceased friends and famous literary figures. The main spirit guides, starting with an enslaved Jew from ancient Greece named Ephraim and proceeding through the archangel Michael and a peacock named Mirabell, transmit elaborate otherworldly knowledge to their human interlocutors via a Q. and A. format that will look familiar to anyone who has quizzed a bot about its tastes and origins.
The questions of whether the poet really believed in the board and how much he embellished its messages always hover over “Sandover,” but as in the case of the Yeatses and “A Vision,” such skepticism is finally moot. For Merrill, language is a definitively human medium; spiritual meanings become intelligible only through a process of translation, which is to say via his and Jackson’s own sensibilities and experience:
Heti’s Alice would likely recognize a certain kinship with Merrill’s Ephraim, even if their cosmological origin stories and linguistic styles could not be more different. “Sandover” is, at heart, the result of a predigital large language model of literary creation, based on the interaction between a human mind and some kind of intelligence outside it.
Is this a matter of metaphysics, or of technique? Are we interested in the messengers — the chatbots and the Ouija-board revenants — or in the messages they deliver? Those messages, after all, are about us: our fate, our origin, our fragile human essence. Everything we can’t figure out by ourselves.