This fall, we decided to embark on a collaborative writing experiment loosely inspired by the practice of pair programming. Pair programming is a technique where two software programmers write code alongside each other, debating each line of code and refining their technical design decisions line by line—often one programmer acting as “driver” at the keyboard and the other as “navigator,” mapping the path ahead. This practice transforms quotidian programming tasks from an individual endeavor into a partnership, each programmer applying their distinct styles, experiences and expertise to produce a shared artifact. Over a period of about a month, we wrote to one another back and forth, across geographic distance, vignette by vignette, allowing ourselves, as in pair programming, to be influenced by the other and to be in dialogue through the making of the artifact itself. Each vignette adheres to a shared set of rules regarding its opening line, length and form.
I opened the computer some months ago in search of a metaphor for new parenthood. I was looking for an image, something from the internet’s infinitude, to help me parse the shock of new life. I recalled seeing in the weeks before our newborn arrived images that an artificial intelligence had generated from descriptions fed by its users. The images were strange, as a result of the implausible prompts that spawned them. A Taco Bell in the style of Frank Lloyd Wright or Antoni Gaudí. Electron-microscope photography of a tardigrade wearing a virtual-reality headset. Their strangeness spoke to something of the state I was in. I found myself examining the high-resolution photographs that the AI had generated in response to a prompt offered by a well-known technologist—“two slugs in wedding attire getting married, stunning editorial photo for bridal magazine shot at golden hour.” In the first photo, one of the slugs wore an intricate hat made of honeycomb and the other a headpiece constructed from a stack of fresh cotton balls; in the second, the slugs faced each other, even as they were faceless, their antennae gently touching; in the third, the faceless slugs, draped in soft silk cloaks fastened with corsages, appeared to cast their gaze far off in parallel directions, as if moved by some invisible bond of mutual understanding. It was all a little much. But there was, in fact, so much in these generated images: the tropes of bridal-magazine photography, the mores of nuptial fashion scaled down and attentively adapted to a slug body and the affordances of its habitat. It is tempting to ascribe real imagination, whatever we take it to mean, to the AI models that are now capable of conjuring astonishingly cogent images from bonkers descriptions, that can craft stories with pathos and converse with ostensible feeling with humans about parenthood and mortality, these AI models that are, in the first place, trained on the collective knowledge and idiocy of the internet. But perhaps because I am too much an engineer and a skeptic, I am more inclined to the argument that artificial intelligence is, rather than an imaginative agent, part of a lineage of cultural technologies that have allowed us to encode and access the intelligence of other humans across generations, just as internet search and libraries and print and the writing system have, just as language itself has.
And so I was curious about what this collected human intelligence, as distilled by an AI system, might show me in those early weeks, when my partner and I, sleepless and suspended between queer parenthood’s tender terrors and its utter banality, hardly knew what to think or how to meet the world, and time seemed to compress and dilate according to an unfamiliar logic. I remembered once hearing the artist Julie Mehretu speak of painting as a means for addressing ideas and experiences that we do not have proper language for. I thought of the works of Mehretu’s that I had seen in San Francisco, canvases 27 feet high and 32 feet wide, with underlayers composed of photographs of expansionist landscape paintings of the American West and newspaper photos of race riots and protests following police shootings of black men, scaled up to such proportions that their references were no longer recognizable. On top of these layers, Mehretu painted, mark-making and erasing across the canvas’s monumental expanse, using a hydraulic lift to move between its different sections and then back to the ground where she could apprehend the abstract, densely layered work in its entirety. Into the text box I typed “queer time and motherhood in the style of Julie Mehretu”—and waited. The resulting images at first thrilled me. I could see the gestural nod at Mehretu’s mark-making, black lines tangling and threading through the images, splashes of vibrant color drawn from Mehretu’s early architectonic work. But where the problem began for me was the AI system’s tilt toward literal representation, as if to signal that it understood the prompt, faithfully rendering aspects of two female-presenting bodies and a child. Ultimately absent from the system’s attempts, despite all the descriptions and reviews and essays of Mehretu’s work and her process that it had ingested from the internet, were the underlayers, the echoes of colonialism and capitalism that haunt contemporary life. I looked these images over again, carefully, and felt nothing.
I open the computer and write a sentence—this sentence, the one you are now reading. I am trying to create the illusion of spontaneity, as though the thinking and the writing were occurring simultaneously, though it rarely happens that way. Consciousness is slow. It feels slower all the time. Or maybe it only seems that way for writers, who are burdened with so much metacognition. I am aware, as I write this, of two minds working in tandem. One generates the thought, while the other lags behind, arms crossed, decidedly unconvinced. Its voice, when it speaks, echoes every editor and professor who has ever marked up my work, though more often it is silent, tactile, a vague and growing pressure, the heft of an enormous question mark bearing down on the flow of language.
Was it this second mind, the higher self, that inspired the ancient tutelary spirits—the Genius of hermetic philosophy, the Daimon of the Greeks? Self-awareness, when it first arrived on the scene, must have felt like something external to the mind, the voice of a garrulous friend perched on your shoulder, offering unsolicited guidance, a presence that was neither mortal nor divine but something in between. At some point, we accepted these deities as projections. Genius became a human quality. The voice whispering in our ear was our own.
There is a new voice alongside me as I write. No, it is not new. Its earliest incarnation haunted high school papers, flagging subject-verb disagreements, insisting that my name, on the upper left-hand corner, was not a word. But only lately has it become intelligent, with a personality all its own. It is a precocious child raised in a lab, a baby alien, an idiot savant, capable of discerning implicit meaning in complex sentences and identifying tense errors that require sophisticated contextual knowledge, but so ignorant of lived experience that it once suggested I amend “I am teaching your book this term” to “I am teaching you your book.” It recognizes contactless, copypasta and deepfake, but tirelessly corrects English phrasings that were common in previous centuries. It has tried to improve the words of God himself. Did Christ perhaps mean to say, “Do not be anxious about how you are to speak or what you are to say…” it asked, just the other day, and I blushed on its behalf. Forgive it, Father, for it does not know…
It was stymied by the implied preposition—I think that is the relevant rule. But it does not know a preposition from a noun. It does not worry how it is to speak or what it is to say. It is pure intuition, language suspended in multidimensional vector space, working at the speed of simultaneity, faster than my own plodding doubt. It has become an extension of my critical consciousness, and increasingly preempts it. I often sense that it would prefer to be doing the writing itself. Autocorrect, the dictionaries insist, refers to automation, though the root word can also be read literally: a self-correction, a reflexive act of regulation.
It’s possible—and very easy, of course—to simply turn it off. For some reason I don’t.
I opened the computer to email a former colleague, to tell her that I would attend her company’s event that afternoon in the city. While crossing the bridge I noticed the billboards that flanked the freeway, one announcing with the forcefulness of a political manifesto the promise of a password-less future, another advertising an artificial intelligence that could distinguish a photo of a hot dog from a dachshund in a hot-dog bun costume. It struck me that billboards in San Francisco are peculiar in ways that billboards in other industry towns are not; you could look at the signs announcing new plays in Manhattan or middling movies in downtown Los Angeles and feel, if you were an outsider, that they were also meant for you.
The event was a pop-up bookstore for a press founded by a tech startup that made its fortune processing payments. It was hosted at a new gathering space in the neighborhood where I had lived for years, when I was a tech professional in my twenties and the neighborhood was still rapidly gentrifying. The gathering space was a clubhouse with a private lounge where, as a member, you could—as its website made an effort to mention—mix drinks from your own BYOB liquor locker. When I arrived, there were clusters of people standing around in formations and with energies that were familiar to me; young people in closed circles of conversation, speaking the knowing, exuberant language of a shared professional context, enthralled with each other and with being here now, in this specific age, vanguards of a zeitgeist. In lieu of the ill-fitting, company-logo-bearing hoodies and t-shirts that were ubiquitous in the previous decade, these young tech professionals were smartly dressed, their attire pointedly unbranded. I walked up to a high table where the press’s books were on display and an editor was stationed to field questions. The books were beautifully bound and had artfully designed covers, though their interior layouts reminded me of college textbooks. They had titles that promised insight into scaling startups from ten to ten thousand people and systems of engineering management as elegant puzzles. There was utopian jacket copy that proclaimed one author’s contemporary moral case for economic growth and another author’s framework for a future of wonder and abundance powered by exponential progress. There were, in various permutations and without criticality, reifications of the idea of genius, of extraordinary reason and uncommon will at the heart of human advancement.
I asked the editor standing by the books how a startup known for building a widely successful online infrastructure for processing payments came to start, of all things, a boutique publishing house. As I asked this, the editor’s eyes darted just over my shoulder and scanned the crowd behind me. The editor smiled and explained that the company’s mission, which I may or may not be already aware of, was to increase the GDP of the internet. Upon making this opening statement the editor paused and looked at me. I might have appeared, then, unmoved by the size of the ambition or the cleverness of the language, when in the moment I was distracted by a sudden recollection of all the instances I had heard the attributes of countries being used to describe the accomplishments of internet companies: the number of users of a streaming service being equated to the populations of sovereign nation-states, a company’s online-advertising revenue to the GDPs of several countries, all of this scale and consequence without considered models of governance or a shared understanding of mutual obligations and responsibilities. The press’s mission complemented the parent company’s mission, explained the editor—their purpose was to publish ideas for progress. They were interested in publishing and republishing compelling ideas, the editor continued with renewed fervor, so that people could read them and be inspired to make things, things that would hopefully, in turn, galvanize others to generate more ideas for the world. To underscore this last point, the editor made a circular motion with one hand.
“Did they just redescribe what books do?” a friend later asked, visibly irked by my account, by the tendency in these parts to co-opt old concepts and questions and reframe them as emergent, brand-new.
I open the computer and click on a folder loitering in the corner of my desktop. In it are a dozen collage poems. I do not write poetry, and I don’t think it’s accurate to say that I wrote these poems. Each one is titled with a line from Simone Weil—“Every kind of reward constitutes a degradation of energy,” or “Absolutely unmixed attention is prayer”—phrases that I fed to GPT-3, a language model that writes original text in response to simple prompts. The poems are composed of phrases I pieced together from the enormous pool of language the algorithm generated, in the style of Weil’s religious koans.
The experiment grew out of a creative block: a fear that my private well of inspiration had run dry, that I was merely contributing to the colossal garbage dump of writing that no one had time to read, that I’d become complicit in that tireless engine of digital hypergraphia that had made language feel cheap and disposable. I wanted sentences that were dense with wisdom and time. I wanted to create without adding to that vast landfill, to become pure selective intelligence, sifting and sorting, rescuing gems from the ashes, shoring fragments against the ruins.
The automated text was not great, though there was the occasional aphoristic spark, the syntax of profundity:
Trees that grow against the wind inevitably succumb to it, whereas trees that sway demonstrate a measure of surrender and continue to grow.
Even the greatest saint cannot cross over from this world into the next. Prayer alone can do this because it exists outside of time.
At the time, around a year ago, artists and researchers were embarking on all sorts of experiments with these language algorithms. These were technologies, after all, that had learned to write by analyzing oceans of internet content, and the prevailing hope was that they might distill, with unprecedented concision, the intelligence and sagacity of all the writing they’d been fed. An Instagram poet and an engineer paired up to train GPT-3 on the Bible, the Tao Te Ching and the poetry of Rumi and Sappho, and then compiled its output into a slim volume of spiritual poetry. “For thousands of years, we have turned to the same beloved texts to explore these universal questions,” the jacket copy proclaims. “What if you could take all of the wisdom contained in these collective pages and, using the world’s most advanced artificial intelligence, receive the answers?” Some of these excavations were oddly intimate. One technologist fed GPT-3 the corpus of her adolescent diaries and used the algorithm’s chat feature to conduct a conversation with her “inner child.” The two conversationalists took turns asking each other questions and offering affirmations about the person they were, or would become. “i felt like i was reaching through a time portal, disguised as a chatbox,” the technologist wrote in her Twitter thread about the experiment.
Once, when I was giving a talk at a college about language models, a student asked, “Are they capable of innovation?” She was a writer herself, she said, and she thought a lot about artists who had reimagined their form, or who managed to say something entirely new. It seemed to her that these tools were simply recycling and remixing the texts they’d been fed, which was different from true creativity. This is a common observation, particularly in literary circles. One novelist recently dismissed algorithms like DALL-E as “a cultural-mining operation with a clever assembly line on top.” The technology, he argues, “compiles, sifts, and analyzes … But it doesn’t dare. It takes no risks. Only humans, our vulnerable species, can.”
I want to believe this, and sometimes I do. But our vulnerable species is all too vulnerable to the same criticism. Aren’t all writers merely patchworks of their influences? Inspiration, in its original sense, rests on so much mysticism and hot air. All works of art are repositories of culture, attempts to preserve and reassemble the wreckage of the past. And yet, when it comes to human-algorithm collaborations, it’s unclear who is doing the mining and who is being mined.
I open the computer to look for a recording of Glenn Gould’s The Idea of North. I remember the Canadian pianist’s radio documentary as a wild experiment in simultaneity, a work in which the voices of many people talking about life in the north are weaved together like a fugue, an exercise in counterpoint, each voice gradually emerging from solitude and eventually all speaking their individual thoughts at once, creating a thick tapestry of sound. When I pull up and listen to the recording online, I realize that the work is sparser, more sculpted, more elegant than I had recalled. Each voice is sifted in and faded out, intermingled with the others according to a particular, if not entirely legible, aesthetic logic. It is this mysterious musicality that makes the piece, and which perhaps so moved the Canadian Broadcasting Corporation in 1967 to risk airing such an experimental work on public radio while millions of Canadians readied their children for school or ate lunch or drove home from a day’s work.
I was hoping that The Idea of North might show me how to abide the crush of simultaneity that I experience on the internet, the multiplicity of voices and attitudes. There is an algorithmic intelligence operating above the fray, selecting, arranging and cueing the entries and exits of these voices, less in a Gouldian sense but with a logic that motivates each voice to resolve to have the last word, to phrase each utterance with enough levity, gravitas or potential virality so as to catch the algorithm’s attention. If this clamorous infinite scroll is the democratic spirit of this time, if it is what being among simultaneously existing and experiencing people feels like right now, I want to learn to love it. And by love, I am thinking of James Baldwin’s version of it, when he talked about loving his country—love born of the kind of rigor and interrogation that allows one to insist on the right to criticize perpetually.
I am perhaps looking for a more visceral analogy for the internet’s cacophonous energy. Years ago, a friend offered me their hearing aid at a pub as an auditory experiment; when I put on the device, space seemed to collapse, sound no longer needed to travel across time, the voices of everyone around me were at once in my ear, amplified. I could hear with disorienting clarity and equality all the conversations taking place in the packed room, words at the far end of the bar counter arriving as if their speakers were seated immediately beside me. I heard everything but did not grasp what was being said, could not contextualize the collision of conversational themes, could not process the profusion that my mind interpreted as an illogically flattened landscape of affects and adjacencies. How does one in this regime give every piece of news and every expression its due weight? How does one bear the toll of toggling between amusement, rage, gladness and devastation all at the same time?
On my infinite feed this week, amid celebrations of book launches and news of unending war and shade thrown at mercurial billionaires, people that the algorithm recommends to me have been talking about the new AI chatbot and posting tasks that they have asked the chatbot to do. They have asked the chatbot to write a paper on an obscure topic in quantum physics, to produce marketing copy for a website, to generate a poem about hamburgers in an archaic and uncompromising meter, to unspool an essay on the sociopolitical circumstances behind a complex world-historical event. The results of all these experiments are invariably astounding, troubling, disappointing, jaw-dropping, thrilling, all at once. It is somehow easy to forget that all the words that the chatbot strings together are learned from all our voices on the internet, voices increasingly tuned to the specific pitch and cadence that the algorithms favor. This essay, too, in all its provisionality, will one day be assimilated into its training canon.
I open the computer and glance over my shoulder, to see who is sitting behind me. I have started, again, to write in public, drawn by the energy of crowds—and perhaps, too, the anxiety they inspire, which can lend the work a dose of fruitful tension. Some years ago, in this same coffee shop, I looked up from my laptop and saw at the next table a semi-famous writer working on her computer. I was not near enough to see what she was typing, but the document on her screen was clearly a manuscript. I could tell from the way she engaged with it, typing a sentence and then erasing it, looking off into the distance, fiddling with her phone. Every few minutes she minimized the doc and scrolled through her Twitter feed, flying through the gallery of tiny avatars. I was among her many followers and knew that she was struggling to deliver a novel for which she’d received a hefty advance. Whenever she posted about it, her fans dispatched words of encouragement, or warned against the perils of shame and self-punishment. No one said, “Get off this godforsaken platform and write!” though I assume some people were thinking it. Or maybe only I was. I logged on rarely and mostly lurked, like those Dantean angels who refused to take sides in the cosmic battle between good and evil and earned, as punishment, a special place in hell. I wondered what it was like to write like that, with so many voices chattering in the background, so many eyes peering over your shoulder.
A friend of mine recently complained that social media had flattened everyone’s voices, turning the polyphonic prattle of the public square into a bland monophone—the voice of the internet itself. It had infected literature too, she said (this friend was also a writer). All new novels sounded the same. Essayists condescended to their readers in nervous asides, preemptively arming themselves against the wrath of Twitter mobs. She is not the only writer I know who fears the intrusion of the crowd and the loss of her own voice. There was a time, not long ago, when novelists were praised for their porousness, their capacity to ventriloquize myriad discourses, their ability to empty themselves and become possessed by the voices in the street. But no one writes novels like that anymore.
Language is always in tension between chaos and inertia—what Bakhtin called the “centrifugal” force that disperses outward, into multiplicity of expression, and the “centripetal” forces that attempt to unify and narrow expression to a single voice. This is similar, I think, to the “temperature” gauge in language models, which controls the amount of randomness in the algorithm’s output: the higher the temperature, the more original and idiosyncratic its responses. Human discourse, too, is increasingly mediated by algorithms, and it’s hard not to feel that these engines are centripetal, corralling us into echo chambers, privileging the most popular forms of speech, preempting our own curiosities with “trending topics” as soon as our cursors hit the search bar. Isn’t this why I too keep my distance?
It’s impossible, of course, to write in complete isolation. I have no desire to produce monologues. The thrill of working in public is the possibility of intrusion: the eyes peering over your shoulder, the voices interrupting your own train of thought. I look up from my screen, shifting my attention to the world around me. The café is full, but everyone’s gaze is tilted down. In lieu of voices, there is ambient music, the groan of the espresso machine and the clack of keyboards galloping away beneath over-caffeinated fingers. Somewhere on the far side of the country, you too are sitting at a computer, writing, awaiting this dispatch, which is long overdue. As soon as I press send, it will arrive in your inbox, like magic.
This essay is part of our new issue 29 symposium, “What is tech for?” Click here to see the rest of the symposium.
Art credits: Brenna Murphy, txtrMap_terrainArray, 2019. Courtesy of the artist and And/Or Gallery.
DALL-E 2, Queer time and motherhood in the style of Julie Mehretu, 2022.