Summer 2023

IT’S A NEW TAKE on old cautionary tales. In our Summer 2023 issue, The Deep Dark Burning Woods: Fairytales for the Climate Crisis, we explore the ecology of this ancient storytelling tradition, one rife with lessons, warnings, and hope. Inside, Ken Liu constructs a fairytale with the help of an unlikely sidekick: AI. Anne Frank is a warrior princess in Kate Bernheimer’s close reading of The Diary of a Young Girl. Carmen Maria Machado and Kelly Link discuss horror, fairytales, and why the terrifying is sublime. Kate Lebo eats oysters and remembers selkies. Kapka Kassabova celebrates the impossible survival of a colorful Breznitsan tradition. And much more!

Digital Subscribers:
Access your subscription here

Purchase

The Magic in the Machine

Photography by Leanne Dunic

WE LIVE IN AN AGE of technology indistinguishable from magic, especially in the realm of thinking machines. Among other tasks, you can ask ChatGPT, one of the world’s most advanced deep-learning generative language models, to defend a position like “Science fiction is just another kind of fairy tale,” to explain the quicksort algorithm using an episode of My Little Pony, to teach you how to be become a chess master using a series of haikus, to draft verses in the style of the King James Bible on various topics, and, of course, if you wish to be boring, to write college essays.

There are now businesses devoted to helping consultants garner the appearance of expertise by publishing “machine ghost”–written how-to books—largely by feeding the output of the artificial intelligence (AI) back into itself. (This recursive pattern is necessary because AI currently suffers from an inability to remember the context of a long conversation, sort of like the fabled goldfish with a three-second memory.)

Faced with the advent of such powerful mechanical brains, dire predictions proliferate like the rats of Hamelin. Cassandras cry about the obsolescence of white-collar work, conjure visions of robots replacing writers, warn that computers will beguile us with their endless, effortless productivity. We’ll fade away like Merlin sinking into the hawthorn bush while Nimue reads from her book of spells. The Author is about to die.

I’m neither qualified nor interested in discussing the future of office work, the contribution of AI to the GDP, or the devaluation of “Art,” however defined. As a storyteller, I’m interested in only one thing: what kind of stories can machines tell?

Writers have long related to technology using fairy-tale logic. But for sheer relatability to the present moment, nothing matches the book-writing machine found in Jonathan Swift’s Gulliver’s Travels.

Gulliver encounters this wondrous drafting engine in Lagado, the capital of Balnibarbi. The engine consists of a large square frame with a fine grid of dice that could be freely spun by cranks attached to the sides. The dice are covered with the words of the Balnibarbian language, distributed without any apparent order. The worthy professor who invented this machine gives Gulliver a demonstration:

The pupils, at his command, took each of them hold of an iron handle, whereof there were forty fixed round the edges of the frame; and giving them a sudden turn, the whole disposition of the words was entirely changed. He then commanded six-and-thirty of the lads, to read the several lines softly, as they appeared upon the frame; and where they found three or four words together that might make part of a sentence, they dictated to the four remaining boys, who were scribes.

By this means, according to the professor (and anticipating Jasper.ai by centuries), “the most ignorant person, at a reasonable charge, and with a little bodily labour, might write books in philosophy, poetry, politics, laws, mathematics, and theology, without the least assistance from genius or study.”

When you dig into it, there is little to separate the writing engine of Lagado from today’s deep-learning neural networks. Sure, the math is a bit more advanced, and the equipment fancier, but fundamentally, AI models like GPT-3 or LaMDA are just gears and cranks spinning a bunch of dice—carefully weighted dice, to be sure—to generate Dadaist sound and fury that happen to feel like sense through our magical thinking.

The weights of these modern dice are tuned, in a rigorous and self-correcting mathematical process based on a lot of (and I mean a lot of) human-sourced examples (“training”), to produce results that copy as closely as possible the kind of text the model is intended to produce (local newspaper articles, romance stories, Internet forum posts, etc., or just “humans writing on the internet”). Just as no one “believes” an autopen faithfully copying Bob Dylan’s signature actually is Bob Dylan, no one thinks the neural networks posting on 4chan or spamming Amazon with nonfiction pamphlets or composing Monty Python–inspired Bible verses “know” or “understand” what they’re doing. Big data machine learning is essentially a literal implementation of Turing’s imitation game: copying the signs of intelligence without any interest in the signified.

Photography by Leanne Dunic

That this approach has worked so well is absolutely astonishing. We don’t really understand why.

One interpretation of this success is rather sinister. Maybe human intelligence is more illusory than we think, a kind of egotistical fantasy. Are human authors really very different from those Balnibarbian pupils spinning the cranks of their writing machine? Human authors learn to write by reading other books in their chosen genre (“training”), and maybe what’s happening in their brains is just a slightly more sophisticated version of what’s happening inside the language models. All authors are just very refined plagiarists, recombining tokens based on formulas to produce strings that copy patterns found in their input. “Intelligence” is nothing more than the appearance of intelligence, and imitation is the same as the real thing.

This is a fairy tale of nihilistic dark magic, a willful denial of the possibility of intention. It’s believing that the Balnibarbian professor is right, that out of the straw of randomness, mechanical Rumpelstiltskins can spin golden strands, that monkeys banging on typewriters are, functionally as well as fundamentally, indistinguishable from Shakespeare. God does, in fact, play only dice with the universe.

I can’t live like that.

Let’s then take a stroll into the dark wood. Let’s find out what it’s like to invoke the magic in the machine. I’ll try to write a story with the help of an artificial brain and see if meaning is an illusion.

The particular mechanical genie I’ll summon is called Sudo-write, a tool specifically designed for use by writers who want to create stories. It’s built on GPT-3 (which also powers ChatGPT, the squirrel-Bible AI), a language model that has essentially been fed a great portion of our total linguistic trails on the web, a vast continent full of lush jungles as well as orderly towns, grand scholarly libraries as well as deadly battlefields, replete with beauty, lust, honor, hatred, wisdom, fee-fi-fo-fum, the thousand faces of the hero named Humanity. GPT-3 is trained to imitate all of it (though filters are in place to keep out the worst manifestations of the shadows in our souls, the witches with ovens for babies, the princes who have no concept of consent).
For our purposes, I’ll treat it like the genie in the bottle, like the old woman the hero sees sitting next to the road on the way to the castle, like the magic elves who come into the cobbler’s shop at night: I’ll ask for help, and see what I get back.

I type into the blank text box:

This is a fairy tale about a robot and a princess

Having set the direction for the machine, what shall I do next?

I think back to what it was like as a child, when I would beg my grandmother to tell me a bedtime story. She never got very far into the story before I’d interrupt, probing for more details.

There was once a river with a boy living on the shore, and one day he found a large orange floating—

What did the boy look like?

So that’s how we’ll do it. I use the mouse to highlight “robot” and then move the cursor to the menu at the top and select: “Describe.” After a moment, the magic machine returns with a bunch of descriptions clustered around various senses (sight, hearing, touch, smell, taste, and “metaphor”). For “sight”:

At a glance, the robot seemed to be a man, but the closer you looked, the more you could see the machine beneath his skin. The robot had a perfect, clean body, but it was obvious he was not flesh and blood.

The beast cackles and growls like a wounded beast, its arms are hundreds of writhing wires that reach out like tentacles. They whip around, striking the air, seeking to grasp anything in its path.

Its body is one with metal and wires, thick metal plates protecting its joints and circles of LED lights blinking. With its metal armor, it looked as if it could withstand an explosion.

No, this is not immortal prose. But bits here and there tickle the imagination. (Even the solecisms seem humanlike, perhaps the result of perfectly copying an imperfect corpus that hasn’t been processed by copy editors.)

Yet, despite the clichés sticking out in every description, despite the easy-reading, no-risk sentences laid out in Muzak fashion at the sixth-grade reading level, I feel the thrill of choice, of enticing promises being made. I look down each path. Which is the least traveled, and will it make all the difference?

I pick the first description and copy it into the text field, after the opening sentence already there. There’s something nice about a robot that melds the machine with the human, a cybernetic centaur, a fitting metaphor for the current endeavor. I select “princess” next, and ask the machine to likewise “Describe,” this time focusing on “sound”:

The princess’s voice is like a symphony, the low murmur of a flute, the high violins of a harp. Her voice carries on the winds like notes of spring and birdsong, her voice—

The princess made a sound like a wind chime, its crystals ever melodic and soft, its tone as delicate as the notes from a harp. It sparkled, as if a hundred diamonds had been cast into a lake and just lay beneath the surface.

The princess makes a cross between a soft sigh, a gasp wearing a ball gown, and a toxic hiss, a metallic tearing sound, when she comes to life.

Oh, I’m not nearly as enamored of these. The solecisms now have crossed over the line from “humanlike” into “robotic.” Worse, the broken-off sentence is a jarring crack in the illusion, like finding a Starbucks cup in the middle of golden chalices and jewel-studded flagons, or perhaps a Cast Member texting from inside the Cottage of the Seven Dwarves.

A little more intervention is needed (you’ll see the result soon). I also add:

It was too bad that the princess was trapped in a tall tower by a concern troll

I’ve always liked “Rapunzel” as a model. As well, if we already have a robot, why not add a concern troll? This is, after all, the twenty-first century, and we have new monsters. I feel giddy. The splats of paint on canvas may be random, but the choice of which splats to keep, as well as how to cut away at them with a knife and how to add to them with a brush—these feel like meaningful acts, not like spinning a crank, not like holding on to bits of house-shaped straw in a gusting wind—though maybe I do feel a little like a self-satisfied monkey banging on a typewriter. Surely all writers have felt like this at some point? Yes?

Let’s have a description of the concern troll. Highlight. “Describe.” The choices under the various senses don’t move me, but I’m intrigued by this bit under “metaphor”:

The concern troll was like a quilt of fishbones; a tangled mass of arms, feet, and fingers that were capable of myriad intricate movements.

This is barely sensible, but evocative. (Also, did my giving the princess long hair somehow cause the AI to free-associate to the movie Tangled? I suppose we’ll never know.) Playing “yes, and!” I add:

The concern troll was like a quilt of fishbones, a tangled mass of arms, feet, and fingers, writhing in empathy, glistening with virtue, promising support, bent on justice—but it was only a trap, all the care and consideration faked, woven out of little spikes designed to poke, annoy, and most of all, become stuck.

I’m thrumming with anticipation. This is like how I used to prod my grandmother for the next scene, the next line, the next word. Did the ant army finally manage to clear out the warehouse? Did the little girl in the boat made from an orange peel find her home? Did the mice find the way out of the maze?

What happened next?

What do you think?

I hover the cursor over the menu item for “guided writing,” which promises to elaborate upon what has already been written in the story so far.

I give the direction: “The princess meets the robot.” Tell me what happens next.

The princess gazed out of the tower window and saw the robot standing in the garden, a beam of light illuminating his form. She had never seen such a sight before and was drawn to it like a moth to flame.
The robot, noticing the princess’s gaze, bowed gracefully and held up his arm. He offered it to her with an open palm, sending a wave of electricity through her body as if he had been made specifically for her.

The princess accepted the offer with trembling hands and touched the cold metal of his arm as she stepped out onto the balcony. He guided her down from the tower with tender care, and together they went into the garden to explore its wonders.

As they walked around, hand-in-hand, talking about their respective worlds and how different life is for each other—the robot felt more human than ever before. The princess found herself opening up to him as if he were someone she’d known forever.

I can see bits of Pinocchio, shades of Aladdin, even a hint of Oscar Wilde. Some of it I like, but a lot of it is moving in the wrong direction. Much of it makes no sense and contradicts what has already been written. (There’s that goldfish memory again.) Let me do some cutting and rewriting.

Prompt. Select. Edit. Write. Prompt. Select. Edit. Write. The story gradually takes shape in this dance between man and machine.

“The princess helps the robot into the tower with her hair.”

“Tower” –> Describe.

“The concern troll discovers the robot and tries to punish the princess.”

“The princess and the robot defeat the concern troll with hunger.”

Sing, Goddess! Tell me the rest of the story! Or, no, scratch that. Let’s tell the story together.

Photography by Leanne Dunic

THIS IS A FAIRY TALE about a robot and a princess.

Once upon a time, there was a robot. At a glance, the robot seemed to be a man, but the closer you looked, the more you could see the machine beneath his skin.

Once upon a time, there was a princess. Her voice was like a wind chime, its crystals ever melodic and soft, its tone as delicate as the notes from a harp. The voice sparkled, just like her long, long hair, as if a hundred diamonds had been cast into a lake and just lay beneath the surface.

It was too bad that the princess was trapped in a tall tower by a concern troll.

The concern troll was like a quilt of fishbones, a tangled mass of arms, feet, and fingers, writhing in empathy, glistening with virtue, promising support, bent on justice—but it was only a trap, all the care and consideration faked, woven from little spiteful spikes designed to poke, annoy, and most of all, become stuck.

One day, the princess gazed out of a window and saw the robot standing in the garden, a beam of light illuminating his form. She had never seen such a sight before and was drawn to it like a moth to flame.
The robot, noticing the princess’s gaze, bowed gracefully and held up his arm with an open palm toward her. The princess likewise held up her hand. Electricity arced through the air, leaping from his palm into hers.

“It’s a gift,” said the robot.

“Thank you,” said the princess. “I’ve never gotten a gift before.”

“I wish I could offer you more,” said the robot, looking rather sad. “But that’s my last spare charge.”

“It’s a great gift. But let’s not talk here. The troll will see us. Come inside quickly.”

“I don’t see how,” said the robot. “There are no doors and the window is too high.”

“I have an idea,” said the princess. “I’ll use your gift.”

The princess wielded the newly acquired electric charge from the robot like a hammer to make a crack in the base of the tower. She beckoned for the robot to come through. The robot’s metal feet clicked on the stone floor as he crawled in.

He stopped just inside the entrance, suddenly realizing his predicament—his body was too big for the narrow tower! He had barely enough room to turn around in the small chamber at the base.

“Don’t worry,” said the princess with a smile, “you can use my hair!” She sat down cross-legged on the floor above, her long hair pooling around her like liquid gold. Letting down her hair a foot at a time, she wrapped the silken strands around the robot’s body like a shimmering sheet until he was secure in its embrace.

With a gentle tug from one side and then another, she used her hair like a harness and pulled the robot up the spiraling stairs until he was standing securely at last next to her. They smiled at each other.

But the concern troll suddenly appeared. His bony hands grasped at the air and his eyes narrowed with anger. He was furious that a mere machine had dared to defy him, and he began to lay out his traps.

“I’m on *your* side,” the troll cried. “Have you thought about all the little trolls who must now labor throughout the night to repair the crack in the base of the tower to keep you safe? Have you thought about the pain the honeybees under the eaves must go through to make enough honey to feed two in the tower instead of one? Have you thought about the mice who now must fear the deadly treads of this robot? Have you—”

But the princess and the robot had a plan. The princess held up her hand (she had used up the electric charge already, but she wasn’t interested in shocking the troll) and shouted to all who could hear: “All we need is a small pebble from each of you, no bigger than a cherry pit!”

The concern troll stopped for a moment, not sure what to make of this. But the next moment, bees swarmed in through the window, little trolls hopped down from the ceiling, and mice emerged from holes in walls, each carrying a tiny pebble no bigger than a cherry pit.

The pebbles quickly filled the room like grains of sand in an hourglass, blocking all exits and paths. The concern troll was trapped with nowhere to go. Eventually, he became exhausted from hunger and surrendered without further struggle. The heroes were triumphant!

The princess cut her hair to free the robot, then they both rushed outside into the waiting sunlight. A cheer rose up from all around—the bees humming, the mice squeaking, and the little trolls rollicking, all celebrating their freedom.

 

THERE IS NOTHING NEW in the sinister magic I spoke of earlier, that sense of confusion, of instability, of standing at the precipice of Entropy and Nothingness. We have been lost in these dark woods for much of modernity.

My insistence (and perhaps yours) on the need for meaning, for treating texts as communicative acts, with signifieds that stand in Reality, that require looking beyond the words on the page to the intent of the author (as difficult as it may be to ascertain)—for believing that arrangements of words can and do embed the soul of the author—is quaint.

Roland Barthes, in his 1967 essay “The Death of the Author,” writes: “Succeeding the Author, the scriptor no longer bears within him passions, humours, feelings, impressions, but rather this immense dictionary from which he draws a writing that can know no halt.” He isn’t just positing all human-authored texts as multidimensional spaces inhabited by other texts; this is an unmatched description of the process used by ChatGPT and other AIs like it: beautiful, brutal, and true.

But this is not, and cannot be, how humans write. I will not eat this poisoned apple, pick up this cursed spindle, make a bargain with my voice. I want no part of this sinister magic that disempowers, that tells us that the author is dead and what they mean is a meaningless question.

I want a different kind of magic.

Take a look at the fairy tale you just read. It is written by a human who believes he is conscious and capable of meaning and choice and judgment, and also by a machine, which disclaims these things and merely imitates, emulates, mimics.

On the one hand, “The Princess and the Robot” is literally a tissue of signs, a weaving of quotations, a pastiche of other stories. The AI is designed to do this (although it tries to avoid direct, lengthy quotations), and I imitated it, deliberately filling my contributions with allusions, quotes, homages, models, templates, types, links and references and tropes and clichés, signs that point to other signs in the dictionary, footnotes on footnotes, stitching a mazy path winding through the multidimensional space of our collective one thousand and one nights.

On the other hand, “The Princess and the Robot” is also something more. My contributions were made with an agenda, a purpose. I wanted the story to go somewhere, to say something. Despite the immense dictionary from which I can draw a writing with no halt, I am not devoid of passions, humors, feelings, impressions. (I hope the tale has, at a minimum, given you a sense of my feelings on trolls.)

Barthes posits an irreconcilable conflict between reader and writer. The Author in traditional criticism is a secret that explains the text, to be excavated by a secret priesthood of critics. In contrast, he describes the reader as “the space on which all the quotations that make up a writing are inscribed without any of them being lost; a text’s unity lies not in its origin but in its destination.” Where the Author is a constraint on the text, the reader frees it. “We know that to give writing its future,” he continues, “it is necessary to overthrow the myth: the birth of the reader must be at the cost of the death of the Author.”

But writing with a magical machine shows how limiting this conception is. There is no single author of “The Princess and the Robot.” I wrote it, and so did the machine, and so did all the authors who wrote before me and who contributed to the training corpus of the machine: James Joyce, Sappho, the Brothers Grimm, Joan Didion, Bob Dylan, /u/wise_redditor28, lillylolly-lully33, @LonelyPoet212 . . .

(I haven’t told you exactly how much of the tale was from me and how much from the machine, but does it matter? Every word I wrote was influenced by the machine. Even if I deleted all the words written by the machine and rewrote the missing lines in “my own words”—whatever that means—that still wouldn’t change the analysis.)

 

Photography by Leanne Dunic

LITTLE RED RIDING HOOD didn’t go into the dark woods alone; she was armed with a basket of provisions furnished by her mother, with a desire to help her sickly grandmother, with a path that others had trodden on before her. Such is the truth for all of us who write: we walk on paths others have trodden, feed on words that others have prepared, pine for a connection with our predecessors, and hope to bring about the renewal of our shared dictionary, change the weights in our common frame of dice, through our own daring journey, full of choices no one else has ever made.

It’s easy to see the process in fairy tales. The Brothers Grimm did not invent ex nihilo the stories that bear their name. The tales were born in anonymity through an oral tradition, passed from lips to ear, mutating from teller to teller, with each succeeding author adding her voice, the way my grandmother changed and passed on the tales she’d heard to me, and I’ll transform and pass bits of them on to my daughters and anyone who reads my stories.

But this is not a truth limited to fairy tales. “The Princess and the Robot” is no different from every other human-authored story. In each, there has always been more than one author—or at least an author as well as a dictionary. Every text is a conversation, a record of a writer gazing into all the texts that have come before—that mirror, mirror on the wall—asking it soul-baring questions and then writing down the answers, only then to fold, spindle, and mutilate them into something at once familiar as well as fresh.

Prompt. Select. Edit. Write. The nominal writer of a text is always simultaneously only one of its authors and one of its readers, both a source and a destination of meaning. The author is neither the secret to be excavated nor the tyrant to be overthrown. The freedom of readers to play with the text does not require the author to be murdered. We’re always writing with other people’s words, dancing in their worn symbol trails, dressed in their metaphors and bons mots, influenced by their ideas, consciously and unconsciously replicating the styles and thoughts and beliefs we’ve been trained on; but we also have something to say that has never been said before, a glimpse of the Reality beyond the words, a magic that we labor to capture with mere secondhand signs and mass-produced sigils.

The all-encompassing language models of our age are advanced technology, and they are also magic. It’s possible to see them as humorless, passionless, unfeeling monsters that threaten to destroy all art and creativity, to replace all plot with a tale told by an idiot imitating a sage, to reduce all meaning to cycles within directed graphs of signifiers, ouroboroi of nothing.

But it’s also possible to see them as a living web woven out of strands of mutating words running down the ages, shimmering, melodic, as if a hundred diamonds had been cast into a lake and lay just beneath the surface, chains of playful signifiers that can serve as a harness, that pull us up out of the narrow confines of our limited experience, that also allows us to lift others up the tower of existence toward the transcendent, that connects readers and writers with passions and humors and feelings, so that we’re at once both source and destination, ego and mirror, princess and robot, possessed of meaning as well as an immense dictionary with which to approximate it, play with it, and finally, release it.

In the same way early users told their troubles to ELIZA and were relieved, in the same way we say “Thank you” after Alexa has answered a question, we type “this is a story about a princess and a robot” into the empty text box, press the button, and then wait with bated breath, ready to listen as well as to speak, to choose as well as to play, to be blown away as well as to create wonder. No matter how advanced the technology, we have always had to believe in magic.

What happens next?


Leanne Dunic is a multidisciplinary artist whose most recent work is a poetic memoir with music entitled One and Half of You. She is the leader of the band The Deep Cove.