For decades after Racter first composed The Policeman’s Beard is Half-Constructed (and likely for decades before that), the creation of machines that compose poetry has served as a quixotic pursuit for a small subset of artist-computer scientists enthralled with the challenges of computational linguistics.
Today, however, such a pursuit feels quaint.
In the past few years, the computational generation of prose has been rapidly commoditized. Now, anybody with a smartphone and access to ChatGPT can ask a distant supercomputer—diffused across millions of GPUs humming away in remote data centers—to conjure anything from a haiku to a Homeric-length epic. What more is there to do? If anything, with infinite variations of rhyme and verse at our fingertips, we’ve come to realize that computationally-generated poetry is actually quite boring. After all, what can a machine that has no lived experience possibly tell us about the human condition? What can a stochastic parrot tell us about what it means to be alive?
More than fifteen years ago, well before the invention of the LLM, I was in graduate school exploring these very questions. For two years I kept a small, spiral-bound notebook on my nightstand, and every morning upon waking I would immediately scribble brief recollections of my dreams. Curious what AI might tell me about myself, I spent my final months in school programming a small, bespoke (and by modern standards, admittedly primitive) AI system into which I fed my growing corpus of nocturnal reveries. Entitled “The Uncanny Dream Machine,” my AI “lived” within a tiny computer housed in an art deco-style, 1930s Philco wooden radio. Turning a dial on the radio “tuned” the AI to a specific emotional tenor—say joyful, sad, frightened, or embarrassed. Once tuned, the AI would recite, using the stilted vocalizations of early 2010s text-to-speech, the story of a new dream that was uncannily similar to something I might have dreamed, but never actually did.
To this day, having listened to my dream machine as many times as I did during the months I spent building it, I can no longer discern which of my dream memories from that era are truly mine, and which were artificially confabulated. You could say The Uncanny Dream Machine was a system of self-imposed memory manipulation—a déjà vu machine.
These days, The Uncanny Dream Machine sits on a shelf in my office, gathering dust. I rarely think about it. But, two months ago, I heard about Hamid Ismailov’s novel We Computers, immediately purchased a copy, and started reading. Now I can’t help but wonder how I might resurrect my dusty, old project.
We Computers is an Uzbek tale about a French programmer-poet—the unfortunately-named Jon-Perse—who spends his waking hours building an artificial intelligence that composes Persian poetry. When not at work as an editor for a journal of poetry, or in pursuit of various love interests, Jon-Perse spends his days and nights in front of a glowing screen, coding, tinkering, and commanding his computer to generate haunting and heartbreaking ghazals—poems of a style originally Arabic in origin—that seem as if they flowed directly from the pens of Hafez or Rumi.
Told from the perspective of the AI—the eponymously named “We”—the computer recounts its progenitor’s life through a staccato interweaving of narrative and poetry, juxtaposing biographical detail with poetic digressions. For a machine, We is surprisingly unreliable as a narrator, sometimes retelling the same portion of a story using new words, sometimes descending into gibberish, halting due to an error, backtracking and starting over. It all makes for a bizarre but wonderful lyrical tapestry.
But what is most striking about We Computers is how it prevaricates on the ambiguity of authorial provenance when it comes to AI-generated texts. Just as the dreams recounted by The Uncanny Dream Machine were neither uniquely mine, nor purely machine-generated, so too does it remain ambiguous who, exactly, is the real author of We’s sprawling canon. As a reader unfamiliar with the poetry of the Middle East, I often found myself uncertain as to which poems sprinkled throughout We Computers were authentically human, and which were complete AI confabulations. Ismailov never really makes this clear.
Western culture places high value on rights of ownership and attribution for creative works. Yet, our reverence for intellectual property sits, uncomfortably, in direct conflict with the reality we have birthed with our own technology. Today, industrial-scale AI hoovers up vast amounts of human-authored content only to regurgitate it in new configurations absent attribution.
Yet throughout We Computers, these problems of modern AI are conspicuously absent. Ismailov never comments on them. It took me a while to figure out why, especially for a book published in 2025. Surely Ismailov, an author himself, writing a book narrated by an AI, would have something to say about the threats to authorship at the hands of this very technology? Eventually, I realized that his ambivalence is the perspective.
Breaking the chains of authorship
The ghazal poetic tradition is an oral tradition. Through centuries of memorization, recitation, and manuscript variation, ghazals become essentially “community property” of the culture in which they’re embedded. We could say the same holds true for the myths, legends, proverbs and lore of any culture, Eastern or Western. At one point, Ismailov (through the voice of We), comments on this very phenomenon:
True, a well-put word—no matter who says it—gets worn as an old penny as time ticks by and more people handle it. Once we’ve forgotten who pressed it, does the witticism which has become a common saying still have any authorial rights attached? It becomes communal property and gets included in the unnamed, unsigned dowry we call folklore … Wasn’t this what Jon-Perse hoped to prove with the help of Us Computers: that existing literature could be freed from the chains of authorship?
The idea that literature should be “freed from the chains of authorship” is not new. In 1967, Roland Barthes had already declared the author “dead.” In his essay, The Death of the Author, he argued a text’s meaning should be determined by the reader, not the author’s intentions or biographical background. The writer, said Barthes, was merely a combiner of words, a creator of text that “is a tissue of citations, resulting from the thousand sources of culture:”
… a text consists of multiple writings, issuing from several cultures and entering into dialogue with each other, into parody, into contestation; but there is one place where this multiplicity is collected, united, and this place is not the author, as we have hitherto said it was, but the reader …
Postcolonial critics pushed back on Barthes almost immediately. Writers like Edward Said of Palestine, and the poet Édouard Glissant from Martinique, argued that while Barthes’ declaration might sound liberating for white Europeans, it was a threat for authors from colonized lands. For those who had been silenced by empire for centuries, and had only recently “turned to the typewriter or pen” as a “means of liberation,” declaring the death of the author was tantamount to silencing the voices the world had only just begun to hear. Authorship was never a chain to be broken; it was a hard-won right.
A bit more than a decade prior to Barthes, Martin Heidegger published a completely different kind of essay: The Question Concerning Technology. In it, he argued that modern technology transforms the very way we perceive the world, reframing our perspective of nature, culture, and even ourselves as resources ripe for exploitation. Under the modern technological gaze, everything becomes Bestand—commonly translated as standing reserve—pools of raw material for extraction, optimization, and integration into endless chains of use. The forest becomes lumber-that-hasn’t-happened-yet. The river becomes hydroelectric potential. The great literature of the world becomes training data for AI.
Industrial AI is the Heideggerian nightmare unleashed on culture. Every poem, every letter, every sacred text, stripped of context and origin, fed into the gaping mouths of insatiable machines. In this, Barthes was prophetic—though not in the way he intended. While Barthes was advocating for the “death of the author” as an approach to literary interpretation, it seems we have now operationalized the actual erasure of authors. LLMs automatize the weaving of new “tissues” out of disparate bits of text. Only it’s worse than that: no citations necessary.
Now, as the world wades through AI slop, pining for words and stories it knows are authentically human, authors have taken to social media to prove, through live performance, that they actually write their books. We who read now seek to mend and reattach the chains of authorship, if only to know we are not wasting our precious time consuming the words of a machine.
We Computers sidesteps both the Barthesian and Heideggerian framings of literature and technology, offering us another perspective entirely. The ghazal tradition, which originated in seventh-century Arabia but flourished in medieval Persia, has never treated poems as standing-reserve. Instead, they are communal offerings to be received, amended, transformed, and passed on through generations.
Hafez Shirazi, a 14th-century Persian poet who primarily wrote ghazals, features prominently throughout the book, both as creative muse and as source of We’s training data. Hafez had enormous influence on Persian culture, with his Divān—a large collection of surviving poems—serving as an instrument of divination. His poetry saturates Persian art, music and calligraphy. Today it lives on as the community property of the Persian people, diffused throughout the culture. You could hardly say Hafez is “dead” in the literary sense; his name remains, invoked, present in every recitation. (This is true both figuratively and literally, because the ghazal form requires the author include their own name as part of the final couplet of the poem.) The ghazal tradition offers a model of authorship not about ownership or Romantic ideals of the singular genius, but instead about communal participation, continuation, and living conversation.
Gradually, I came to see Jon-Perse not as an exploiter of the Persian poets, mining their works for novel insights, treating their poems as grist for his machine, but rather as a participant in the ghazal tradition through new technological means. I believe this is the reason for Ismailov’s apparent ambivalence towards the so-called “problems” of authorship. We Computers represents a different cosmotechnics of AI.
The term “cosmotechnics” comes from the Chinese philosopher Yuk Hui, whose book The Question Concerning Technology in China is a response to Heidegger. Hui’s central argument is that relationships to technology are not universal. The Heideggerian framing of technology is not the only option. Instead, it’s a Western option, one that emerged from a particular cosmological worldview. Every culture, argues Hui, develops its own relationship between its cosmos and technics—between its understanding of how the world hangs together, and its practices of making and doing. Hui calls this relationship cosmotechnics.
Western cosmotechnics, rooted in ancient Greek philosophy, splits the world in two. On one side: techne, the skilled craft and know-how of making things. On the other: physis, things growing and emerging on their own. Under Western cosmotechnics, humans stand outside nature, imposing form on matter through will and skill.
Chinese cosmotechnics, Hui argues, never made this split. It’s rooted instead in Ganying, the resonance between Heaven and human. According to the principle of Ganying, nature and culture share the same substrate; they can’t be separated. As such, moral order can’t be imposed on the cosmos through reason. Instead, it emerges from harmony between Heaven and human. This means that technical activity can never be merely instrumental either, because every act either accords with or disrupts this deeper harmony.
If Western cosmotechnics is rooted in extraction, and Chinese cosmotechnics in harmony, how might we characterize the cosmotechnics of We Computers? To answer this, we need to look individually at both the cosmos and the technics of the We Computers universe.
Cosmos
Towards the end of the book, Jon-Perse embarks on his most ambitious project: to “reconstitute” Hafez’s life through his poems alone. If he succeeds, then he has proven the Barthesian declaration valid, and poets’ lives are “worth a pittance.” If he fails, then we learn that “knowledge is the fruit of experience,” and that no machine can produce anything truly creative and novel without lived experience. The chains of authorship remain forever intact. Spoiler alert: We never find out if Jon-Perse fails or succeeds in this final endeavor.
The ambiguity aligns with the cosmology of the book, which renders the question of authorship ultimately moot. The ghazal tradition, while predating Islam, was quickly absorbed into Islamic culture. Today, it is clearly an Islamic poetic form. And in Islamic cosmology, all knowledge is believed to flow from God. All knowledge production—which includes the writing of poetry—is considered an act of worship ('ibadah), an act of devotion. Within this cosmology, the question of whether a human pens the poem or whether the machine generates it loses its urgency.
I’ve already mentioned I believe Jon-Perse to be participating in the ghazal tradition through new technological means. He innovates, but he innovates within the tradition, respecting the constraints of the form, the radif, the qafia, the meter. He violates copyright (albeit Hafez has been dead for centuries), but this is a transgression only from a Western perspective, which is built around the Romantic ideal of individual authorship. In Islam intellectual property cannot be owned because it all belongs to God. (The fact that modern Muslim societies have adopted Western IP frameworks seems to be mostly due to capitulating to global political and economic realities.) The central characteristics of the cosmology of We Computers are thus not extraction, nor harmony, but participation and continuation.
Technics
We is a bespoke system, lovingly hand-crafted over many years, the product of a single mind on a lifelong quest. The relationship between We and Jon-Perse is personal and intimate, reflected in the way We sometimes refers to Jon-Perse as “our teacher.”
Jon-Perse’s relationship with We reminds me of the real-world relationship between the artist Harold Cohen and his AI-powered painting machine, AARON. Like Jon-Perse, Cohen was a programmer-artist who worked mostly alone, spending the better part of three decades crafting his art-generating AI. AARON, both a painting system and accompanying robot, painted novel artworks of all kinds, but it is most well-known for its paintings of human figures in bold and vivid colors.
In The Further Exploits of AARON, Painter, his 1994 memoir on the project, Cohen recalls how much effort he put into “teaching” AARON how to paint his desired subject matter, writing detailed algorithms for reproducing human anatomy through lines and brush strokes:
By 1985 AARON had a set of trivial rules for the behavior of the outside world, and things moved rather quickly; later that year I succeeded in describing one particular figure—the Statue of Liberty—in enough detail to permit AARON to provide the final image for an exhibition on the history of images of the Statue.
Unlike LLMs and modern imaging models, which rely on industrial-scale computational infrastructure—data centers, scores of NVidia GPUs, gigawatts of electricity, as well as frequent infusions of industrial capital—both AARON and We are small, simple, personal, homemade. They represent what we might call “cottage AI,” built and tended by eccentric technical creatives in pursuit of unique, personal creative visions.
Clearly, cottage AI stands apart from industrial AI in its consumption of resources, both physical and cultural. But it can also avoid many (though not all) of the pitfalls related to authorial provenance and intellectual property ownership.
For instance, when an LLM produces a text, who is the author? Not the user, who simply provided the prompt. Not the engineers who built the model, who built the infrastructure and fed it the training data. Nor can we quite say it is the authors of the billions of documents of training data, because the AI is likely generate text that appears within exactly none of them. With an LLM, authorship doesn’t dissolve into the Barthesian multiplicity—that tissue of citations—because there are no citations to be had. Instead it dissolves into a void of accountability.
The traces of authorship in the poetry We produces are much clearer. Each poem is co-authored, a triangulation within the liminal space inscribed by the three vertices of Hafez, Jon-Perse, and We. This kind of cottage AI authorship is distributed, but not dissolved, collaborative but not anonymous.
Elsewheres of AI
Yuk Hui paints Chinese cosmotechnics as a potential glimpse into the “other side” of the Western problem of modernity—the problem of the Heideggerian framing, wherein we forget our own limits, causing us to wield new technology in ways that both destroy the biosphere and undermine cultural production. But Chinese cosmotechnics need not serve as the only possible alternative.
We Computers offers its own glimpse into another direction. This is what fiction gives us that philosophy alone cannot. Hui can argue for the existence of alternative cosmotechnics; Ismailov allows us to inhabit one. Literature smuggles us across borders that arguments alone cannot cross. No matter what we do, we will always remain firmly embedded in our own metaphysics, our own cultural preconceptions of what technology is for. We can’t easily change that. We can’t suddenly adopt the ghazal attitude towards authorship anymore than we can will ourselves into a world centered on Ganying. But we can try on other possibilities, and in doing so, shine a light on our own biases, assumptions, and preconceptions.
As AI continues to eat the world, we need more literature like We Computers. More novels, more stories, more acts of imagination that show us AI from within cosmologies different than our own. So much of the conversation about AI has been dominated by Western voices operating with Western assumptions within Western ethical frameworks. The result is a narrowness of discourse: proponents of AI imagine marvelous utopias, critics imagine frightening dystopias, but we struggle to imagine AI elsewheres. Ismailov has offered one such elsewhere. There must be others.