He came to a panting stop. And Greer, shy, ill-prepared Greer, suddenly said: “So you mean… like… a book is stored thoughts. But an ‘artificial intelligence’ is stored thinking?”

  And Marius said: “At last! Someone here fucking learn something.”

  He gave them their homework—with the instruction “Fucking read it this time or you don’t come to my class again, understand?” The students nodded attentively. Marius turned the screen off and screwed his face up with disgust.

  To the Universitatea din Bucureştì students in the room, he made a remark in Romanian starting with the word Americani that Zhen could only presume from their laughter must have been both filthy and insulting.

  Zhen waited until the students had left and Marius was trying to pack up his matchbox machine. She walked quietly down the stairs. He raised his head. There in the glance was everything. Decades of communication, of reaching out. A lifetime of learning—without knowing how you knew—who you could trust and who you couldn’t.

  “Fuck,” he said, smiling. He pulled Zhen into an embrace as warm as a furnace. “I hope you brought me some fucking disaster.”

  4. weaponized poetry

  martha

  In their anonymous hotel room near San Francisco, Martha watched Selah’s happymeal algorithm get to work. If you’d looked at the data for long enough—if you’d gotten used to the powerful back-end tools, pulling together information from dozens of different streams—you started to look straight through the numbers and see right into the users.

  There. In a tract house in Des Moines, Iowa, a forty-six-year-old woman finishes a show she’d started the night before. A nature thing, The Wonders of Africa. She likes that stuff; her kids used to watch with her, but they’re fifteen and seventeen now and everything she loves is stupid and boring. She’s only half watching the show as she cleans up breakfast. They all ate without a word of thanks, without even showing they knew there was a person there to thank. More conversation with their Anvil Chatterboxes than with her. She loads the dishes into the machine. Empties the coffeepot into the sink. Meanwhile a guy with a British accent is saying that these are the last eighteen wild elephants left in Africa, that they are guarded around the clock. She looks up. The old, wise elephant eyes look into hers. The unbearable hurt of suffering and no one noticing. The music swells. She’s crying, standing in her kitchen, holding a wet dishcloth that is dripping onto her moccasins.

  The show is over. The invitation to leave a comment flashes up on the screen. She types one quickly. Beautiful animals. We gotta do more to protect them.

  When she’s clicked away, and won’t see the comment change, Selah’s algorithm carefully flicks through all her previous comments to find the single word that will likely add the most weight to her thoughts. It is as meticulous as a poet.

  Happymeal reviews the criteria Selah has included. Selah has used the Anvil advertising code—the information Anvil uses to serve its users effective advertisements, tested a million times, iterated again and again. Short lists of adjectives work better than single adjectives—more unusual adjectives are more gripping but they must be easy to understand. Humans respond to animals that are like people. Without Selah having directly told it to, happymeal works through words pet owners use about their dogs and cats. What do you say if you love your dog, and you want others to love it too? Just one word. At last, it finds the word.

  It is a bead tumbling through a series of logic gates, noting each time what is marked as “success” using the reaction-emojis and what isn’t. It does not empathize with the woman in Des Moines or with the elephants or with the people who will read the comment. It is a procedure. Medlar and Anvil and Fantail have been curating and promoting comments based on this procedure for two decades now. The only difference is that this algorithm isn’t trying to sell anything and has been told not to make people angry.

  It is trying a word. If this doesn’t get the result it wants, it will try something else. It can do this several million times a second. The algorithm is weaponized poetry.

  On the screen, the comment becomes: Beautiful intelligent animals. We gotta do more to protect them.

  “That’s it?” said Martha.

  “That’s it times fourteen million, baby,” said Selah.

  Zimri Nommik had, after all, used these methods to direct human minds toward the products and services they were most likely to click on. These methods had made him the richest man on Earth.

  * * *

  In Ucklum, Sweden, a young man watches an American show about monster trucks. This is about the only part of his day he enjoys, the time before his parents come home. He hates living here. He wants to move to Gothenburg but he doesn’t have the money and his family can’t give it to him. He hates how the only social life is the fucking pizza place and none of the other kids in his band take it seriously. He hates how the rest of the kids at school are so happy and love the tiny town and talk all the time about winter sports. He hates the winter when the sun never rises. He hates the summer when it never sets. Through the window of the screen he reaches himself into the other world, pulls his head and shoulders through into the American world, where everything is big and angry and hot and loud. He feels the heat from the engines blasting his hair back from his face. If he had a monster truck, he’d paint it with a scene of hell and put all the kids from school in the picture.

  The show is over. The invitation to leave a comment flashes up on the screen. He stabs at the keys. Yaaaaaaaa, awesome! Monster trucks forever!

  Happymeal runs through each word. “Forever” is emotive, a suggestion of common cause. The algorithm does not know this, of course, not in the way humans know things—not by placing itself imaginatively in the mind of a reader. It is a bead selected from a box over a million million attempts. It has been taught to analyze the comments to see which are pro monster truck and to adjust them minutely to receive fewer hearts, fewer likes, fewer milliseconds of engagement. Happymeal runs a million different realities in one-tenth of a second. It picks the one it’s been told to favor.

  When the young man in Ucklum moves on to another show, on-screen his comment becomes: Yaaaaaaaa, awesome! Monster trucks.

  Selah said: “See this one? We’re predicting it’s gone from nine likes to two likes. It sounds a bit wet now. Like he couldn’t even be bothered?”

  Martha looked at it: “Some kid from fuckbum Sweden just lost seven little bits of reassurance.”

  Badger rolled their eyes: “Zimri Nommik does this anyway. My mom does this anyway.”

  Fantail did it as a matter of course: every user defaulted to seeing comments that the Fantail algorithms had been instructed to show them first. It was, Lenk had once told Martha, a full ninety minutes of work for any user to set up their own rules about what kind of information they wanted to see and turn off all the defaults and advertising filters and data extraction options for Fantail. That’s the ones you could turn off, if you knew what you were doing. Estimates were that an average user would need six hours to adequately research how to do this. A workday, just to stop Fantail from carefully showing you whatever it had figured out would make you spend the most time on the site. About 1.5 percent of users actually did it.

  But it had felt different when Lenk was doing it. Like the users, Martha could only control what Lenk did some of the time.

  This was them doing it, and watching it gave her a strange feeling.

  “Bloody hell, Anvil does this all the fucking time,” said Selah. “Stuff like this, but also fake comments, also not letting you see comments. Sketlish does too. I mean, you can pay for that, you know? That’s why we all feel like shit all day long. Want a lollipop?”

  They watched happymeal work all weekend. As countries woke and slept, as insomniacs posted sleep-deprived nonsense, as drivers posted from their moving vehicles, as people in Tokyo and Istanbul, N’Djamena and Cairo watched the same four shows. The comments on the nature shows became more finely honed advertising for the concept of animals and the natural world—its importance, its health benefits, its kinship with humanity. The comments on the monster truck shows became just a little duller, less sticky. The comments on the family drama were just a little kinder. Selah had said fuck it, why not try to make people be kind?

  The algorithms can’t do everything. But if they can make us more polarized, more angry, and more hateful, surely they can do the opposite of that. There is no “neutral” anymore. There is no leaving things as they would have been before the invention of the internet. Our minds have already learned how to interact with the algorithms and we are part of it.

  * * *

  In Des Moines, Iowa, a forty-six-year-old woman receives a notification that she has had twelve likes on her comment about elephants, which she does not for a moment feel uncertain that she wrote herself. This is more than her normal number. She feels mildly warmed. She clicks on another nature show as she gets the new day started. In Ucklum, Sweden, a young man has no reason to look again at his comment on monster trucks. He watches more videos anyway. The next time he’s invited to comment, he feels a minute fraction less bothered to do it.

  People learn quickly. Social approval is a powerful tool. We tend to put a great deal more emphasis on things that happened recently than things that happened long ago.

  “Fuck me,” said Selah on Sunday evening as she reviewed the data via the Medlar internal analysis GUI before erasing their tracks. “It’s already made a shift. Point zero three percent more positive comments on nature videos. Not just our shows—all shows. Fuck. That’s huge.”

  Martha reflected that Enoch had worked his whole life and hadn’t seen a 0.03 percent shift in anything he’d ever preached.

  “We can do it,” said Badger, and their eyes were wide and shining. “We can actually do it.”

  5. entente cordiale

  zhen

  “This is gift,” said Marius, later that afternoon.

  He spat a wad of tobacco into a Coke can on the kitchen table, leaving juicy brown flecks on his teeth. He had been a chain-smoker until he realized the smoke was bad for his electronics. Now he was a chain-chewer. He stuffed in another wad.

  Zhen and Marius had never met in person before. Zhen hadn’t known about the smell of tobacco that had seeped into the walls of Marius’s living space. Or that he lived across a floor of what had been an open-plan office, with wiry tufted carpets and graying window blinds. His bedroom was a corner office with a huge bed where, when they arrived back, his girlfriend Sarit would be naked, her curved ass only half-hidden by the blankets, awaking feelings in Zhen that had been dormant since she froze someone to death. She hadn’t known that Marius’s place was filled from floor to ceiling in every room but the bathroom with racks of electronic components and server units. Or that the fridge had a vigorous furry mold coating its interior back wall.

  “When I fought mold, it fought back. Spores. Growing. On my food. Outside fridge. Black… pieces everywhere. When I surrender, it stays in own territory. Now we live in harmony. Entente cordiale.”

  Marius laughed and took another swig of Coke. Unlike Zhen, he seemed to have no trouble keeping track of which cans were filled with Coke and which with tobacco and spit. Marius was, in no particular order, a hacker of some reputation, a visiting professor at Berkeley and Carnegie Mellon, an absolute asshole, and probably Zhen’s closest friend.

  They’d known each other for the best part of a decade. They were part of the Name The Day technology board ntd/tech, which was dedicated to using technology to prepare to survive the coming collapse of civilization, although Marius’s apocalypse “preparation” mostly consisted of gleeful acceptance of the inevitability of suffering, decay, despair, and eventual painful death. He was a legend on the board for the sheer range of his expertise in both technology and signs that humanity was on the verge of wiping itself out.

  Their friendship had begun with arguments and humiliation. Zhen was fairly well known on ntd/tech—there was a sub-forum where her videos were dissected, sometimes praised but mostly mocked. For a while, every time Zhen made a video with any computing software or hardware advice, Marius wrote a long post explaining why her take was “fucking amateur caca.” They’d argued, they’d called each other arrogant shitheads, they’d picked holes in each other’s points of view until suddenly Zhen realized that the wrestling had become a hug. When Marius had trouble getting his Israeli girlfriend into Malaysia, Zhen knew how to help. When some piece of technology smelled like bullshit, Marius was Zhen’s first call. Marius didn’t trust or like most people. So he had a lot of unused loyalty and kindness to give the few he could be bothered with.

  At the kitchen table, Marius wired Zhen’s phone up to a Romanian children’s toy computer keyboard, a set of motherboards sealed in plastic, and the screen from a smart washing machine. His hands were dirty but his equipment was clean. He worked quickly, humming to himself and occasionally offering gnomic thoughts.

  Like: “This is gift.”

  Zhen said: “The phone or… what, like… the whole situation?”

  “Someone save your life and want nothing back, that’s gift. She said she gave you gift.”

  He cracked open the back of Zhen’s phone so expertly he might have been shucking an oyster, removed component parts, and slotted them gently into his cobbled-together homebrew. It was like watching Zhen’s grandmother deal with every last grain of rice in the bowl, delicate and precise.

  The washing-machine screen pinged gently and brought up a rinse cycle symbol.

  “Is that something?” said Zhen.

  “Start of something.” Marius smiled. “Find out Sleeping Beauty secrets”—he stroked the top of Zhen’s phone with one index finger—“without wake her up. First, file list. Then, file contents.”

  Marius cracked open a can of creamed corn and scooped it into his mouth using vanilla ice cream wafers, with apparent enjoyment.

  “You want?”

  Zhen shook her head.

  “In the tunnels. She ask you for something, assassin lady? She say ‘Give me this or I kill you’? Or ‘I want to kill you because you did this’?”

  “Didn’t say anything to me. I’m pretty sure she was an Enochite.”

  “They don’t exist no more.”

  “Yeah so, they definitely exist enough to harass me online, OK? She had the key thing on her finger too.”

  Marius shook his head.

  “This don’t happen. Harass online is easy. Teenage boy sit in basement in Russia, write insults. No one can find him. Easy, safe. Government hate you, that turns into crazy person stab writer on stage. Religion hate you, yes, maybe.”

  “They are a religion? A newish religion. That’s what they are. I think they’ve been growing online, in private groups, for a while.”

  “Religious war against you? You didn’t do nothing.”

  “Tell that to that lady’s gun.”

  Marius chewed his corn thoughtfully. On one of the shelves next to the rows of motherboards there were three plastic-wrapped pallets of tinned creamed corn.

  “Maybe. Maybe you unluckiest person in the world. Maybe this is something else. You don’t got her phone? ID?”

  “She was a fucking… chunk of poison salt ice.”

  Marius gave several very quick nods of the head, scratching his beard like a student had given him an adequate and even intriguing answer to a computational problem.

  “You have two women. One want to kill you, don’t say why. One want to save your life, don’t say why. Like fucking fairy story. I think…” And he paused so long that Zhen thought he had potentially fallen asleep with his eyes open. “I think you not in so much danger.”

  “Oh yeah? Let me know how you feel about that after someone’s tried to shoot you in the head.”

  Marius picked a piece of corn out of his teeth with a dirty thumbnail. “The Enochites not so good they find you in Bucharest today, right? Don’t find you in Madrid. Don’t find you on plane. They are medium good. Find you when you not trying to hide, don’t find you when you hiding. Your schedule was on your fucking website. You not hard to find. Now you hard to find, no trouble.”

  “But she… she tracked me. Through the tunnels. Found me in the bin full of Valentine foxes. They had some kind of tracker on me.”

  “Maybe,” said Marius. “Maybe you made more noise than you thought. Moved in bin, foxes moved, maybe she saw you. Maybe put something on you at some event. Dot on clothing, tracer inside pocket. How many events you speak at every month? Ten? How often you throw out clothes after event? You never tried to make yourself hard to track.”

  “So what now—I have to do this forever?”

  Marius nodded. “Spend time making yourself invisible. Then be more careful than before forever.”

  “Thanks for the positivity.”

  Marius laughed. “You want positivity, you go America, white teeth big smiles. You want realism, you come former Soviet bloc, OK?”

  The rinse cycle symbol turned over twice. A list of files scrolled on the right side of the screen.

  “Hmm,” said Marius. “You had malware. Don’t know what yet.” He wrinkled his nose. “I can reconstruct maybe. If it works, might take two, three months. Have to try many different build models.”

  For now, all Marius could see was the date and time the malware had entered the system: 3:27 a.m. on the last night Zhen had slept in Martha’s suite.

  “You joined her Wi-Fi. Wi-Fi was compromised. There was special Wi-Fi for you, yes, not hotel, not normal,” said Marius, and it was barely a question.

  Zhen nodded.

  “Spiked,” Marius said sorrowfully, “middle of the night, uploads and restarts. You never know what happened.”

  “She said she gave me a gift. But she didn’t tell me what it was.”

  “And now she don’t talk to you?”