Ishion Hutchinson's "D'Angelo in Starlight"

Background and Context

On September 28, 2026, The New Yorker published “D’Angelo in Starlight,” a short poem by Jamaican poet Ishion Hutchinson. Its most striking line—“Good God, the devil is up to his old tricks”—arrives as a seemingly offhand exclamation, yet in an era reshaped by generative algorithms it has acquired a symbolic weight far beyond the page. The phrase lands like a dissonant chord against the drumbeat of technological optimism, compelling a re-examination of those dimensions of creative work that resist datafication and imitation.

The line crystallizes a core paradox of AI-driven creative writing. Large language models, the engines behind today’s poetry generators, operate by statistical modeling of vast text corpora, predicting the next most probable word. When prompted to “write a modern poem about starlight,” they assemble high-frequency associations—night, solitude, the soul—into syntactically coherent, image-dense verse. This process is hyper-efficient but trapped in what Hutchinson’s devil calls “old tricks”: it can only recombine existing human expression, never distill the shudder behind “Good God” from lived experience. The poem’s power derives from an unrepeatable emotional current, fed by specific cultural memory, bodily perception, and historical consciousness—none of which are probability distributions.

Deep Analysis

Technically, the “old tricks” are pattern replication and probabilistic collage. State-of-the-art models fine-tuned on poetic corpora can mimic Emily Dickinson’s dash-laden cadence or Wisława Szymborska’s dry irony, and adversarial training can even produce “defamiliarized” imagery. Yet they lack a decisive dimension: intention. Intention implies that the author bears ethical and aesthetic responsibility for word choices; an AI’s “choice” is merely a mathematical output. Hutchinson’s line trembles with a particular speaker’s presence, a quality that statistical generation cannot fabricate because it has no self to tremble.

The title itself is a cultural cryptogram. “D’Angelo” gestures simultaneously toward the Italian for “angel” and toward the soul singer who reimagined Black musical tradition—a layered reference that constructs a meaning-space no current AI can autonomously design. Such density of allusion depends on a lived negotiation with history and identity, not on token prediction. The poem thus becomes a test case for what machine writing cannot touch: the intentional, situated act of making meaning that exceeds its training data.

Industry Impact

Commercially, AI writing tools have become efficiency levers for content. Since 2025, platforms offer assistants generating ad copy, web fiction, and personalized poetry from keywords. Self-publishing outlets host AI-generated poetry volumes, flooding the e-book market at near-zero cost and squeezing human poets. At The New Yorker, the deluge of algorithmic submissions has raised screening costs. Readers scrolling through machine-made “beautiful lines” are shifting aesthetic expectations: slow, chewy text is edged out by “poetic fast food.”

The competitive response is fractured. Condé Nast, The New Yorker’s parent, tests AI-assisted proofreading and translation to cut costs, while joining other journals to promote a “human creation certification” requiring authors to pledge no generative AI use. This contradiction mirrors industry anxiety: rejecting AI may forfeit efficiency, but embracing it could dissolve literary authority. Meanwhile, platforms emphasizing offline workshops, manuscript revisions, and embodied interaction try to restore the creative “aura.” Detection tools have progressed from perplexity analysis to stylistic-fingerprint tracing, but generative models keep closing the gap via reinforcement learning from human feedback. The arms race renegotiates the boundary of “creation.”

Outlook

Three signals merit attention. Legally, copyright rulings on AI-generated content are imminent. The U.S. Copyright Office has denied registration for purely AI works, but the standard for human contribution in AI-assisted works is murky.

If a model scrapes Hutchinson’s line, can he claim rights? This will test frameworks. Second, literary criticism is shifting toward “post-human poetics,” tracing how a text’s meaning mutates in networks, potentially eroding traditional authorship or creating new evaluative systems. Third, creative-writing pedagogy is diverging: some teach AI co-writing, others emphasize bodily awareness, fieldwork, and handwriting, treating imperfection as resistance to algorithmic smoothness. Hutchinson’s line may be the era’s footnote: as the devil works his data tricks, human creators must ask whether each word comes from a soul’s tremor or a forgotten ghost in the training set.