In one of those canonical episodes of the Rod Serling-era Twilight Zone now preceded by its own cultural influence, a bookworm finds himself alone on Earth after a nuclear apocalypse. Based on a short story by Lynn Venable, the much-parodied, much-referenced “Time Enough At Last” depicts a bank teller eternally frustrated in his efforts to find time for reading. Serling narrates, as usual, unstintingly:
Witness Mr. Henry Bemis, a charter member in the fraternity of dreamers. A bookish little man whose passion is the printed page, but who is conspired against by a bank president and a wife and a world full of tongue-cluckers and the unrelenting hands of a clock. But in just a moment, Mr. Bemis will enter a world without bank presidents or wives or clocks or anything else. He'll have a world all to himself... without anyone.
One day, while Bemis is shirking in the bank vault—doing guess what—a hydrogen bomb destroys his city, killing everyone but him, and by the time he exits the vault nuclear war has destroyed all life on the planet. Bemis is confounded. He looks for his boss, his wife. They’re all dead. He finds years’ worth of canned food at a destroyed grocery store. He lights a candle, eats. But he is utterly alone. Bare life is not enough. He finds a handgun and resolves to commit suicide, but before he can pull the trigger, he glimpses a blown-apart public library in the distance and discovers that it is surrounded by scattered, unscorched books. “All the books I’ll ever want! All the books!” He is transported from despair to elation.
The cultural impact of the episode seems to revolve around the piece of supposed cosmic irony that ends the tale: once the bookworm finally has time to catch up on his beloved reading, he breaks his glasses and—there being, we assume, no other unbroken glass left on the planet—finds himself eternally unable to read, just as all of time has unfolded before him.
Is this really ironic, or is it just unfortunate? What’s ironic about bad eyesight? Maybe if he were to trip over one of the giant piles of books he constructs in the final scene, in anticipation of a lifelong bout of blissful reading, then the desire to read would have at least have brought about its own frustration. That would have been sort of ironic, but still a cheesy, contingent irony, not, like, the irony of Antigone. Or is the irony that the professor has devoted himself to useless literature rather than some practical skill like lens grinding, such that it is precisely his myopia (developed, as my optometrist likes to tell me, over a lifetime of squinting at small print) that has now ironically sentenced him to a bookless life?
No, the thing that I suspect actually resonates with viewers, maybe not entirely consciously, and that has made the episode so enduring in a collective mind shaped by pop culture, is another irony: literature in the absence of readers.
The readerless body of literature frightens and unsettles. Stacks of books written to be read, with no one to read them, or—somehow worse—with one person to read them, but no one else to talk or write with, after, or about them. A literary community of one. Like the traveler in Borges’ Library of Babel, a solo reader in the stacks of mute civilizations is vertiginous. It feels morally wrong, futile, and incomplete. There’s something likewise horrifying about a solipsistic and pseudo-Buddhist idea I’ve sometimes encountered in arguments for radical empathy, which posits that there is only one being who is infinitely reincarnated as every creature on earth, and that everything you do or say, you are doing or saying to yourself in a different life. One shudders to think of this world, a world without, or with only fictional, difference. A final solution to all conflict: atomic narcissism.
The goosebumps that rise at the phrase “readerless literature” come from the same uncanny feeling. It’s the feeling I imagine also follows you if you ever encounter an untranslated or long-dead language: the persistent appeal of Egyptian or Mayan glyphs, or other, still-untranslated scratches: Linear A, Voynichese. All these words no one can speak or ever will. Not to mention the indecipherable, eternally mysterious cave and rock art whose functions and meanings we can only guess at. I have met one or two rock art archaeologists and can say they are the most self-effacing, unsure, unsteady academics in the world. Everything they venture is the shadow of a guess. The shadow passes across 10,000 years to try and touch another human mind, but the mind is locked inside a bank vault and fire rains down everywhere.
“It writes you” “It writes through you” “It is written”
Here’s an idea with staying power across cultures and eons. The idea goes that artists are just conduit, that we are medium, that we are the water and not the wave or current, the tinder but not the light or heat. This pervasive idea, seemingly universal in human societies, tells us that it—the written, the sounded, the spoken, the drawn, the sculpted—moves through us but is not from us. Every artist has felt it, has credited muses, gods, spirits, or drugs, the furor poeticus, afflatus, or spirit writing, has looked upon the finished work and been unable to recall any conscious act of creation. We are all charter members in the fraternity of dreamers.
Some writers feel compelled to look into or through the conduit of dreams. Is it organic or artificial? Formalist devices can help illuminate the machinery of inspiration. Oulipo-style tools, which come in two basic forms, constraints and generative “story-making machines,” help us uncover the stochastic, the determined, and the potentialities of a literature uncompressed by the ramrod of intention. Like the one where a poet drafts a poem, then replaces every seventh word with the one directly following it in the dictionary of one’s choosing. Or the one where you roll dice to find out which of your characters will live or die. (To my knowledge, no one has yet written the definitive essay that connects Calvino, Perec, and Mathews with Gary Gygax, architect of the most popular Oulippean generator in history, Dungeons & Dragons.)
Over the past few months I’ve been checking out the Twitter account https://twitter.com/images_ai. The account collects computer-generated images created through a machine learning process that takes as input a string of words. A user puts in a short piece of text, then pieces of software transform that phrase into an image using machine learning technologies and one or several giant online image datasets. I even tried making my own; one result ends the section above.
At first, I didn’t really understand how it all worked: how the software produces such variety and how it seems, at times, even to contain wit. How is a machine able to decide “style”? Like this mid-century (Twilight Zonish?) abstraction it made for “the emptiness” or the de Chirico-esque, bombed-out landscape of "With indifferent geometry and a long Sharp shadow cast by the low morning sun…”?
Kelsey [Festival Trap Remix] @writerremixed@unltd_dream_co @images_ai "With indifferent geometry and a long Sharp shadow cast by the low morning sun..."
Googling around, I came across a beginner-friendly explanation of how the program works. A partial answer is that there are more human-determined initial variables than I’d originally thought, which can help guide the mechanical hands: pick a data set, pick a starter or a target image, select areas where you want objects to form.
But the work being done is still non-human. The machine still learns from itself and self-corrects. If you play around with it enough, you’ll start to get a feel for what “decisions” come out of the choices you make at the outset and which are the uncontrollable actions of ghosts in the machine.
(Another Twitter account, https://twitter.com/ai_curio , illustrates this point nicely by posting experiments similar to @images_AI, except they mostly use starter images from which the algorithm must draw inspiration. In the tweet thread, the account posts the source “image prompts” to demonstrate how the software has interpreted its inputs. Click through the below image to see an example.)
By contrast, many of the images that @images_ai publishes are less restrained or controlled by presets, less guided, and therefore more chaotic, and curiously ugly. Here’s “A Landscape in Bolivia”:
Here’s “A View From the Hill”:
Here’s “Skipping through the windmills of your mind”:
This one reminds me of a Marc Chagall painting, at least metaphysically: the floating human figure, folk-art landscape, something magically disturbed about the color grading…Chagall often painted dreams, and these feel like dreams from the machine.
Are they like dreams? Or do I only want to say they are like dreams so as to place them in a comfortable human context? Are they like my dreams? Are they like machines? When I was in college, studying cognitive science and philosophy of mind, projects like this were mostly dreamed-of. Those that existed were more rudimentary. I remember for one class assignment I built a basic machine-learning Sudoku player, followed by a machine-learning sentence generator that, much like today’s conservative pundits, was easily confused by pronouns. “Creative writing” machines were interesting failures from what I could then tell, and there was no vast image database out of which an “artist” machine like this one could even be imagined. We students debated among ourselves: Was machine learning a kind of thought? Would there ever be thinking objects?
There seems little question now. Whatever human hands went into evolving the generative process, look at these images, and watch videos of their development over tens of thousands of iterations. It would be odd to say there’s no thinking there.
There is an essay in the current issue of The Believer called “Ghosts” by Vauhini Vara which gives new form to the conduit philosophy of creation.
As a little disclaimer, Vara is my former editor at The New Yorker. I’ve also written about her work before as a friend/fan, in an early newsletter segment on underread essays, of which this entry could I guess be thought of as a continuation. Her Harper’s story on Indian-American spelling bee champions is a favorite of mine, and I’ve taught it in workshops to illustrate a generous and breezy approach to feature magazine writing.
“Ghosts” is something else: a neo-Oulippean experiment in death meditation. It moved me more than any recent essay I can remember. No praise felt sufficient in the moment for the rarity of the experience of reading it. As soon as I finished, I started over again. The next morning, I read it a third time.
“Ghosts” opens with an introduction that describes the process of its composition. Vara worked with “an artificial intelligence model that was being trained to write human-like text.” The model’s name is GPT-3, or Generative Pre-Trained Transformer 3, and it is has energized the world of AI research since its debut last year. “If you fed it a bit of text,” she explains, “it could complete a piece of writing, by predicting the words that should come next.”
What follows are eight attempts (or iterations or experiments) in which Vara collaborates with the machine on a short essay about her older sister’s death from Ewing sarcoma when both girls were in high school. With each successive attempt, Vara feeds a few more sentences or paragraphs of the same text she has written—an understated, laconic eulogy—into the machine, which then proceeds to “finish” the essay for her. Vara claims that she and her editor intervened rarely and insubstantially. (“My and my editor’s sole alterations to the AI-generated text were adding paragraph breaks in some instances and shortening the length of a few of the stories; because it has not been edited beyond this, inconsistencies and untruths appear.”)
In the first few entries, the machine’s contributions are “wrong,” not only in the sense of false, but in the sense that they are wild, off-topic, and strange-sounding, drawing too much out of certain words of Vara’s, locking onto the wrong clues, and failing to grasp the appropriate elegiac tone for the essay’s subject. Or so a human might think on reading it. The first attempt takes as its input a single sentence of Vara’s: “My sister was diagnosed with Ewing sarcoma when I was in my freshman year of high school and she was in her junior year.” The machine’s result is ridiculous—a story of quick recovery and, bizarrely, lacrosse. Where did lacrosse come from? The machine has understood the sister’s diagnosis but leans too hard on what the phrases “high school” and “junior year” might signify for the rest of the story.
The second result, following five sentences of Vara’s, is somehow worse, or “worse.” The sister vanishes from the essay altogether, and the machine produces a strange—if compelling—story about jogging, which ends with this scene:
It was a pretty good run. It took me about thirty-three minutes. I was pretty proud of myself when I finished. I sat down on the wall of the Battery and had a drink of cold water. I was just wiping the sweat off my face when I looked up and there was a guy standing there. He was a big guy, at least two hundred and fifty pounds, six foot four or so. He had a beard and he was wearing a tank top. He looked like an ex-Marine. He must have been in his late twenties. He had a six-pack of beer in his hand. He said, “Boy, you don’t look like much.”
The specifics are weirdly arresting: the run that takes “about thirty-three minutes,” the “wall of the Battery,” the tank top and six-pack of beer. The sentences have a fairly natural rhythm, too. We’re compelled to believe what’s happening despite the dream logic, despite knowing the conditions of the essay’s construction. Such is the alchemy of narrative—we want to believe, and to make meaning.
The more text Vara gives the machine, the “better” it performs, which is to say, the more closely it follows the tone and content of her own writing, and the more “realistic” the essay becomes—meaning, in some sense, conventional, although never uninteresting. Each attempt works in its own way, setting off sparks in the reader’s mind.
I don’t often describe writing as brave because I don’t feel that most writing requires bravery, unless there’s a direct threat of arrest or violence involved. Like “violence,” “bravery” is one of those easily abused words. But I found myself feeling that “Ghosts” was brave in its writerly risk-taking, not only in the usual sense of disclosing private pain but in the courage required to hand over your expressiveness to another flawed and fumbling entity. Vara surrendered own gift, her contact with the muses, over to a musebot to fumble with (and fumble it did), and the object of attention she chose to forfeit was such a rare and precious one: her sister’s death. It’s hard to imagine many writers taking this leap. We are possessive of our art. We don’t tend to quote other writers when we are in pursuit of the quiddity of experience because we are jealous more than anything of our talents—our eyes and minds—but here Vara hands hers over not to a Shakespeare but to a mediocre and arguably soulless collaborator, a glorified spellchecker, a machine.
One way to read the essay is by understanding Vara and this machine as being in conversation, speaking to each other. It occurred to me that the surrender of one’s own control over a narrative is also what happens whenever we try to communicate with another person, and that “Ghosts” models the Wittgensteinian frustrations of turning private thoughts into public statements. Hearing your own thoughts come “back at you” in alien, transformed shape, with stories that aren’t true and sentiments you don’t share—why, that’s exactly what happens when you try to communicate the impossible—to communicate anything!—through the human kludge of morpheme, syllable, sentence! Our imperfect, antiquated code. Yet something was communicated: the machine returned stories of grief and loss, and the more Vara surrendered to it, the more it seemed to grasp her mindset. Communication was hard and imperfect, but not impossible.
This thought led me to more and more readings of “Ghosts,” an exploding narrative fractal: more ways in which the essay’s formal experiment, for all its simplicity, seemed to model so much of what literature tries to do, what humans do when they talk to or past one another, and even what love and grief do inside the loving, grieving mind. The cyclical nature of the exercise—Vara writing the same words over and over, with spiraling, sometimes nonsensical results coming back at her across the corpus callosum—also depicts the mental intrusiveness and circular thinking of the overpowering moments in life: love, trauma, grief. The bizarre fantasies, the lies and omissions, the repetitiveness, the unplanned epiphanies. It felt like watching raw consciousness grieve in bicameral counterpoint. The machine’s text resembled, sometimes with spooky verisimilitude, the way a distressed mind can get caught in patterns of thought or flights of fancy: the way thinking about your dead sister might prompt a strange fantasy about meeting a creative writing (!) professor at college who says he loves you, and that you love him. Or a fantasia about traveling in space.
Some of the machine’s responses are almost…Biblical? Vedic? They approach some kind of ancient pattern poetry:
I will tell you how it felt for me. I felt I had lost half of myself. I felt I had lost my right arm. I felt I had lost my left leg. I felt I had lost my tongue. I felt I had lost my heart. I felt I had lost my mind. I felt I had lost my eyes. I felt I had lost my ears. I felt I had lost my breath. I felt I had lost my voice. I felt I had lost my smile. I felt I had lost my laugh. I felt I had lost my tears. I felt I had lost my future. I felt I had lost my past. I felt I had lost my parents, as well. I felt I had lost everything. I felt I had lost everything.
The strangest moment in the whole essay feels like a piece of early 2000s alt lit or a copy/paste accident:
I’ve turned the space station into a spaceship. I’m a ghost, and I’m in a spaceship, and I’m hurtling through the universe, and I’m traveling forward, and I’m traveling backward, and I’m traveling sideways, and I’m traveling nowhere. I’m hurtling through the universe, and I’m a ghost, and I’m in a spaceship, and I’m hurtling through the universe, and I’m a ghost, and I’m in a spaceship, and I’m hurtling through the universe, and I’m a ghost, and I’m in a spaceship, and I’m hurtling through the universe, and I’m a ghost, and I’m in a spaceship, and I’m hurtling through the universe, and I’m a ghost, and I’m in a spaceship, and I’m hurtling through the universe, and I’m a ghost, and I’m in a spaceship, and I’m hurtling through the universe, and I’m a ghost, and I’m in a spaceship, and I’m hurtling
Yet another unfolding of “Ghosts” is the collective grief it secretly contains. Hidden beneath the story of Vara’s sister death, reflected in the machine’s eight attempts at completion, are the raw data points on which the machine drew: millions or billions of pieces of text, sourced online. How many of them are accounts of other loved ones gone? How many departed wives, fathers, and in particular how many sisters—the grief over which a hundred thousand siblings typed into a hundred thousand text boxes—became spectral nodes in the network that produced these writings? How much of them is remembered in any of Vara’s echoes? A word? A whole phrase?
As close as these collective griefs may seem to us, they’re unreachable. We don’t have the “raw data” and could never hope to process it if we did. The machine thinks unlike us—it is so unlike us we might consider its thoughts unthought. Its memories are unmemories. There is only process here, no Cartesian theater, not even the warming fiction of a homunculus inside. Why does the result feel so human?
As Serling narrates Bemis’s tour through a nuclear wasteland, he describes a devastated landscape full of American cliché that sounds pulled from one of Vara’s fragmentary machine-tooled replies:
Seconds, minutes, hours—they crawl by on hands and knees for Mr. Henry Bemis, who looks for a spark in the ashes of a dead world. A telephone connected to nothingness. A neighborhood bar, a movie, a baseball diamond, a hardware store, the mailbox at what was once his house and is now a rubble. They lie at his feet as battered monuments to what was but is no more. Mr. Henry Bemis, on an eight-hour tour of a graveyard.
A telephone connected to nothingness: a one-way message with no possibility of receipt or response. How else should we define grief? What else but the echo of your own voice, a voice that is at the same time all the voices you have ever heard, all the voices that have ever been, assembled uniquely into your own? A voice that is calling out to someone both no longer there and already contained within the cry?
The other, other, other part of “Ghosts” is that nothing we grieve is ever lost; the accumulated information of the machine, the babble that writes for us and with us, will keep writing and accumulating, we can presume, even after the last human shatters his glasses and dies. This digital body of memory now contains, forevermore, the text of “Ghosts,” an experimental essay about death that is somehow immediate, unfussy, and even fun. An essay that allows its grief to be transformed in flows of reading and writing beyond anyone’s control.
I came across a beautiful line in a beautiful novel last week: Dreams and Stones by the Polish writer Magdalena Tulli. It’s a book that seemed to touch on all my thoughts at once as I was reading it, in sculpted, inimitable (if anything remains inimitable in the era of GPT-3) paragraphs, a novel that is also a prose poem or essay or fever dream about an imagined city or cities. The line I marked could as easily have appeared in my last newsletter as in this one, so maybe my concerns are closing in on each other:
But since nothing in the world can be completely and finally destroyed it is clear that the letters written on the water still exist somewhere and will continue to do so forever along with the city abounding in fragile teacups and flammable furniture, the city of the grotesque, safe, entirely free of disasters and unsusceptible to pathos. Whoever recalls its misfortunes has to laugh: Its sorrow has a false bottom in which merriment is concealed.
Last month I was interviewed about my reporting on Xinjiang for a newsletter project called The Postscript. I am not a natural interview subject, as my interviewer, New Zealand-based writer Brian Ng, gently pointed out to me…a few times. But I really liked the way this one came out, much to his credit. This month, Reeducated screened at the Venice International Film Festival, and will screen at the Human Rights Film Festival in Berlin—opening tonight! If you’re in Berlin and want to see the film, feel free to reach out. You can watch Reeducated in the VR section, inside the "Ton 3" hall at the Berlin Union Film Ateliers where the festival is taking place.
Also: This is my 20th newsletter! It’s gone through a few changes in subjects and tone since I started it in 2018, but I’ve enjoyed playing with ideas, drafts, and discarded material here. I’m just going to keep writing down the stuff I’m thinking about in this space, once a month, in a format pretty much like what you’ve become used to. Unless I change it. I’d love to hear from you if you like it or don’t! Thanks for reading so far.