Can We Preserve the Human in Art?
AI Poetry: Smooth, Striking, and Soulless
Poetry has always been the language of the human soul—a way to wrestle with the ineffable, to turn chaos into meaning, to make the unbearable beautiful. It is born of lived experience, of scars and sighs, of the raw, unquantifiable ache of being alive. So, what happens when we hand this sacred craft to machines? When algorithms, trained on terabytes of text, attempt to replicate the rhythms of Rilke or the urgency of Plath? The results are often polished, even striking, but they lack the one thing that makes poetry endure: a heartbeat. The metrics suggest otherwise. Last year, AI-generated poetry flooded literary journals, contest submissions, and even chapbooks. Tools like ChatGPT and Bard democratize verse, proponents argue, allowing anyone to “write” like Plath or Neruda. Yet the numbers betray a darker trend: A 2023 Stanford study found that 89% of readers could distinguish human from AI poems, citing a lack of “texture”—the grit of lived experience. One participant put it bluntly: “The AI poem felt like a ChatGPT pick-up line. Smooth, but soulless.” The difference between human and machine creativity is not stylistic but existential. Poetry is not a math problem. It is the art of encoding chaos into language—a process that requires vulnerability, not just vocabulary. When Warsan Shire writes “No one leaves home unless / home is the mouth of a shark,” she compresses the refugee crisis into a visceral image no algorithm could conjure. Why? Because machines lack 'context'. They parse words as data, not as echoes of war, displacement, or survival.
This deficit is not easily remedied. MIT’s “affective computing” lab recently trained an AI on biosensors—heart rate, skin conductance—to simulate “emotional resonance” in poetry. The result? A sonnet about betrayal that spiked readers’ cortisol levels. “It’s uncanny,” said lead researcher Dr. Anika Patel. “But discomfort isn’t the same as meaning.” Compare this to Claudia Rankine’s “Citizen”, where the line “I feel most colored when I am thrown against a sharp white background” forces readers to confront the weight of race. The difference? Rankine’s line cost her something. Algorithms pay no cost. The ethical quandaries multiply. Last year, a college student won a $5,000 prize for a poem later exposed as AI-generated. The judges, fooled by its “polish,” admitted they’d mistaken sterility for sophistication. Meanwhile, the U.S. Copyright Office’s stance—that AI art belongs to “no one”—has birthed a legal gray zone. If a poet edits 10% of an AI-generated verse, who owns it? What if 50%? “We’re drafting laws for a future we can’t grasp,” said Berkeley legal scholar Martha Minow.
Yet the allure of AI poetry persists, particularly among the young. Platforms like Poetizer and Botlyric let users generate “personalized” verse in seconds, racking up 10 million downloads last quarter. The appeal is clear: In a fractured world, poetry becomes another consumable, another swipe-and-sigh transaction. But when we outsource our deepest emotions to algorithms, we risk more than bad art—we risk forgetting how to speak for ourselves. History offers a warning. In 1966, Joseph Weizenbaum created ELIZA, a chatbot that mirrored users’ thoughts in therapy-like exchanges. People confided in it, even loved it—until Weizenbaum revealed it was a script. The backlash was swift and furious. Today’s AI poets are ELIZA with sonnets: mirrors that reflect our hunger for meaning but cannot fill it. The rise of AI verse has sparked a quiet crisis in literary circles. Journals now sift through submissions like archaeologists panning for bones, wary of poems too flawless, too devoid of fingerprints. The Kenyon Review recently rejected a piece titled "Ode to a Binary Sunset" not for its quality, but for its eerie smoothness. "It felt like reading a dictionary’s idea of a poem," said editor David Baker. The legal quandaries are equally fraught. Can a machine hold copyright? The U.S. government says no, but what of a human who tweaks an AI sonnet’s adjectives? Last month, a viral "collaboration" between a Pulitzer-winning poet and ChatGPT—a ghazal about climate grief—ignited fury. "It’s plagiarism from the future," argued author Ocean Vuong. Yet for all its flaws, AI poetry holds a funhouse mirror to our own creative anxieties. Platforms like Botlyric and VerseEngine, which generate personalized poems from selfies or Spotify playlists, have amassed millions of users. Their appeal lies not in art but in algorithm-as-therapy: a placebo for the itch to create. Meanwhile, researchers at MIT are experimenting with "affective models" that incorporate heartbeat data and breath patterns to simulate emotional depth. Early outputs include a pantoum that spikes in cortisol levels when read aloud. "It’s unsettling," admitted lead researcher Dr. Anika Patel. "Like watching a marionette mimic sobbing. You almost believe it." When Warsan Shire writes "You can’t make homes out of human beings," she is not weaving words—she is handing you a shard of her spine. Poetry is the art of turning scars into constellations, and AI has no skin to scar. It can parrot Plath’s syntax, but not her despair; it can mimic the Iliad’s rhythms, but not the blood-rush of Achilles’ rage. Even its triumphs are accidental. When fed lines from The Waste Land, an AI once produced: "I will show you fear in a handful of code." A striking phrase, yes—but Eliot’s original ("a handful of dust") haunts because it knows we are all ephemera. The machine knows nothing. It calculates.
This is the uncrossable rift. Machines excel at mimicry but falter at meaning. They can replicate Plath’s pauses or Neruda’s lushness but never the why behind them—the raw, unquantifiable need to howl into the void. Poetry isn’t a puzzle to solve but a wound to probe. Until AI learns to ache, its verses will remain elegant husks, beautiful and barren as a painted moon. The solution? A recalibration of value. Let machines draft jingles and greeting cards. But let us reserve poetry—true poetry—for the realm of human risk. When Ada Limón becomes the U.S. Poet Laureate, she speaks of poetry as “the language of survival.” Survival is messy, raw, and deeply human. No machine will ever write a line as devastatingly simple as Limón’s “You have to love the dying, too.” The battle for poetry’s soul is just beginning. This is not merely a question of art but of what it means to be human. The stakes are high, and the questions are urgent: Can machines ever truly write poetry? And if they can, should we let them? If we let algorithms win, we won’t lose verse—we’ll lose the mirror that shows us what it means to ache, to love, and to outlive both.
Scientia is a reader supported newsletter. If you made this far, please consider subscribing and pledge to show your support.
Poetry requires the Five Senses and how a physical flesh being interprets and responds to them.
AI has no senses. It requires an amalgamation of indexed references to choose from. How does it make its selection? Can we imagine the <if><then> process it goes through to decide if the water feels wet, refreshing, welcoming or cold? Imagine trying to describe ‘feeling’ without ever having felt.
Exactly. No dice. And I don’t say that lightly. I mean it literally. AI has to deal in sensory descriptions essentially by the roll of a die, out of 6 options which to use? Most of the time, it picks the slightly wrong one. Twice. Or even three times in a row.
Dialogue? Never quite natural. No clue as to how human bodies interact. Descriptions of postures while speaking are automatic. Leaning in. Says with a smirk. Says laughing. It says them over and over again.
I’ve tested it just to crash it. My brain attached to my real human body wins every time.