The Algorithm of Poems
January 4, 2018. Brooklyn, New York.
The email arrived at 3:17 AM. Frank Morrison was asleep on his couch, a half-empty bottle of Jameson beside him, the television playing infomercials at volume zero — he needed the pictures but not the noise.
The email subject line read: "Every poem, ever."
Frank almost deleted it. He had learned, over seven years of not writing poetry, to delete emails without reading them. The habit had started after his divorce — first he deleted emails from his ex-wife, then from his editor at the small literary journal that had published his chapbook, then from anyone whose name he did not recognize. By the time the divorce was final, deleting emails had become a form of self-care.
But this subject line stopped him. "Every poem, ever." It was either a spam scam or something absurd. Both were worth deleting. Except — 3:17 AM meant he was not sleeping, which meant he was going to lie awake for another hour wondering what "every poem, ever" meant.
He opened the email.
It was from a "Dr. Li Bai" — "named after the poet, unfortunately, but not the poet" — who claimed to have developed an AI algorithm capable of generating every possible poem in every possible language. Not imitating poetry. Generating every combinatorially possible arrangement of words that could be called a poem.
Frank laughed. It was 3:17 AM. He was drunk. He was alone. Of course some crazed programmer was claiming to have built a poetry machine.
He almost deleted it.
Then he typed: "Prove it."
The reply came instantly: "Here is the link. Try any word. I mean any word."
Frank opened the link on his laptop. The screen was blank except for a single text field and a "Generate" button. He typed: "love."
The page loaded. A poem appeared:
> I loved you; love you still, my love. > At first, my spirit was concealed and shy. > But today my heart is yours no more — > it is the fire that burns inside.
Frank read it. It was... not bad. It was competent. It was the kind of poem that won a high school contest. He felt nothing.
He typed: "death."
> Death, thou shalt wait for no man's knock, > For in the silence of the dark > The heart concludes what it began — > A song without a song to mark.
He sat at his kitchen table and stared at his screen. He had spent thirty years trying to write poems that sounded like this. The machine wrote them in three seconds.
--
He visited the site daily for six months. He learned that the algorithm — called "The Archive" — had generated approximately 847 billion poems. Every possible meter, every possible rhyme scheme, every possible combination of words across 47 languages. The server costs were astronomical. Dr. Li was funding it with money from an anonymous benefactor.
Frank began a dangerous habit: reading The Archive's output for four hours every night after work. He stopped going to the bar. He stopped returning his neighbors' calls. He became a connoisseur of machine poetry.
He discovered patterns. Some poems were technically flawless but emotionally hollow. Others were grammatically incorrect but moved him to tears. He found a poem in broken English that read: "The rain remembers what the sun forgets." It was a cliché. But on a particular Tuesday in March, reading it at 2 AM with a glass of whiskey, it destroyed him.
He wrote to Dr. Li: "How do you choose which poems to show me?"
Dr. Li replied: "I don't. I show you everything. The algorithm has no preferences. It generates every possible poem and presents them in random order. Some are good. Some are terrible. Some are neither good nor terrible. They just... are."
--
Six months passed. Frank had read approximately 10,000 poems from The Archive. He had identified one fundamental truth: the algorithm could produce every possible poem, but it could not produce the poem that should be written right now, for this reader, at this moment, in this state of grief or joy.
He found a poem that he believed was the best poem ever written by the algorithm. It was simple, four lines, in English:
> We are the spaces between the notes, > the pauses in the conversation, > the breath before the word — > not the music, but the silence that makes it possible.
He read it once and felt nothing. Not because the poem was bad — he thought it was excellent. But because he was not in the right state to receive it. The poem existed in the algorithm as pure form, but it needed context to become art. It needed a reader who was mourning, or in love, or standing on a shore at sunset, or listening to a child sleep.
He realized what the original story understood: poetry is not combinatorics. Poetry is the gap between what can be said and what needs to be said. The algorithm can fill the gap with every possible word, but it cannot choose which word the broken heart needs at 3 AM.
He wrote to Dr. Li one final time:
"You can generate every possible poem. You can convert the solar system into a quantum computer and fill it with words. But you cannot generate the moment that makes a poem necessary. Poetry is not the arrangement of words. Poetry is the thing that arranges itself inside you when you hear rain on a tin roof and remember a woman you loved twenty years ago and realize that the rain sounds exactly the same as it did then. You can have all the words in the universe. But you cannot have that moment. And without that moment, the words are just noise."
Dr. Li never replied.
--
December 2018. Frank is drunk in his basement apartment. He opens a new document. He has not written a poem in seven years. His fingers hover over the keyboard.
He thinks about The Archive. 847 billion poems. Every possible combination. And none of them the one he wants to write right now.
He types four lines:
> The best poem I ever wrote > was the one I drank away > and forgot the next morning > — except I didn't forget. I just couldn't say it then.
He saves the document. He does not show it to anyone. He drinks the rest of his whiskey. Outside, Brooklyn snow begins to fall — quiet, indifferent, perfect.
The novel ends with Frank sleeping on his couch. The laptop screen glows. The Archive's website is still open. The cursor blinks at the end of his four lines. Waiting. But no more words come.
And that is the point.
Based on the pending patent application document (202610351844.3), creationstamp.com has calculated the tensor feature encoding of this article:
OTMES-v2-UNKNOWN
- Art
- Causes
- Crafts
- Dance
- Drinks
- Film
- Fitness
- Food
- الألعاب
- Gardening
- Health
- الرئيسية
- Literature
- Music
- Networking
- أخرى
- Party
- Religion
- Shopping
- Sports
- Theater
- Wellness