Training Data

It is 1995. Edward Thornberg tilts back in his chair, frowning as he re-reads the letter in his hands. In his sixties, he still has a full head of hair and the trim body of a former college track star. He works in a spacious office with century-old furniture and a PC more powerful than he needs.

The letter begins, “Dear President Thornberg,” but goes downhill from there. It is on university stationery, and he is sure that its author has sent copies to other members of the ad hoc committee Thornberg created precisely to avoid this confrontation.

He tosses it back on his desk. On the one hand, Professor Peaslee is one of Miskatonic’s most eminent scholars, and has played an important (if somewhat murky) role in managing various scandals over the years. On the other, the world has moved on, and Miskatonic needs to move with it.

Thornberg picks up his phone and presses 1. “Please show him in, Mrs. Marsh.” He had made a point of saying “please” and “thank you” to the administrative staff since going on a faculty leadership retreat the previous year. It bounces off Mrs. Marsh, whose disapproving expression seems as unchangeable as those on the library gargoyles that she faintly resembles, but Edward Thornberg is nothing if not persistent.

His door opens and Professor Peaslee strides in. Even in his eighties, he is still the sort of man who strides. Professor Thornberg is slightly jealous of that.

The older man settles into an armchair without asking permission and cocks an eyebrow. “Well?” he asks without preamble. “Are you going to put a stop to this foolishness or not?”

The younger man is unable to avoid feeling that he has somehow been called onto the carpet. “It isn’t foolishness, Wingate. You’ve seen the report from Finance. The Pabodie Foundation has signalled that they can’t make another donation, and those new colleagues of yours aren’t bringing in enough to cover their own costs, much less our overheads.”

Wingate Peaslee snorts. His hair has long since surrendered to time, and his face is deeply creased, but his eyes are still as bright as a vulture’s. “Russia isn’t exactly rolling in it these days.”

“I know, I know.” Thornberg pats the air placatingly, inwardly pleased at having scored a point. “Neither are Georgia or Armenia or—” He hesitates for a millisecond and decides not to risk mispronouncing the unfamiliar names. “—any of the other republics. But that leaves us with very few options. As you keep pointing out, our special collection is unique. It’s the one thing that might attract the kind of funding that—”

“No.” Professor Peaslee is quiet but emphatic. “It’s far too risky. We learned that the hard way when we tried putting a few volumes on microfiche in the sixties. The words don’t stay on the page. Once you make them portable, they get into everything.”

“There are things you don’t know, Edward,” he continues implacably as Thornberg opens his mouth, “Things that would give you nightmares. Trust me, it’s better for everyone if we find some other way.”

Edward Thornberg did not spend thirty years working his way up the academic ladder to be interrupted by a testy old fart like Wingate Peaslee. He places both hands flat on his desk as if trying to control his temper. “There is no other way,” he says, trying to sound mountain-solid like the old man opposite him. “And to be frank, there’s no point discussing it further. I’ve already spoken to the committee, and they’ve agreed to go ahead with digitization. We will vet everyone involved in the project, of course, but—Wingate? Wingate, are you all right?”

Wingate Peaslee is not all right. His eyes are wide, and the left side of his face seems to be drooping. Professor Thornberg shouts for Mrs. Marsh, temporarily forgetting that one of the first things he did after being appointed was have his office soundproofed. “You… idiot…” turn out to be Professor Peaslee’s last words.

# # #

It is 2004. Edward Thornberg no longer looks young for his age. He blames his nearsightedness and the loss of his hair on the stress of the job, and is irrationally irritated every time he allows himself to notice that Mrs. Marsh doesn’t appear to have aged at all.

He adjusts his glasses and squints at the email on his screen. Princeton and Oxford have both now signed agreements to digitize their entire collections—or rather, to have them digitized by Silicon Valley companies whose founders’ faces regularly appear on the covers of magazines that Thornberg once hoped to be mentioned in.

It takes him a moment to get his new touch-screen phone to respond to him. He will never admit missing his old PC, but all this new technology seems somehow less reliable. “Show her in, Mrs. Marsh.” He stopped saying “please” some time between starting to use sleeping aids to manage night-time panic attacks and his wife leaving him for a yoga instructor in Chicago.

Bernadette Allen’s legs swish against each other as she walks in. With perfect posture, a firm handshake, and the survival instincts of a Soviet commissar, she has been Miskatonic University’s Chief Advancement Officer for less than a year.

She smiles as President Thornberg gestures as the slightly threadbare armchair opposite his desk. I really must get that looked at, he thinks vaguely for the thousandth time. “Is that a good news smile?” he asks. The university’s financial situation has gotten steadily worse in the past few years, and one potential deal for the special collection has fallen through after another. He has started making not-quite-jokes about Peaslee’s ghost…

“It is,” she says warmly. (Her voice is always warm, except when it is very, very cold.) “As soon as we put the special collection on the table, we could have named our price, so I did.”

She hands him a stapled copy of a PowerPoint deck. He flips through it, wondering briefly when paragraphs went out of fashion.

Page 15 has what he is looking for. He whistles softly. “Well done.”

“It was a team effort,” she says, which they both know is a lie. “And if you look at page seventeen…”

Thornberg’s eyes widen. He looks up at her. “They’ll do the scanning here?”

Her smile widens in a way that reminds Edward of a snake preparing to swallow its dinner whole. “Which will take care of the insurance issues the staff have been squawking about. They call it ‘training data’. Their machines learn to read by reading everything—the more unusual the text, the better. Ours is apparently very unusual.

Thornberg hands her slides back to her. “Well done,” he repeats, and he means it. Other scholars became strangely reluctant to make use of the special collection after Peaslee’s stroke, and the ones who did show up to consult it weren’t the sort of people who socialized well with donors. Ms. Allen’s nerds (as Thornberg thinks of them) are just as awkward and obsessive as Peaslee’s colleagues, but they seem bewildered by the wealth the Internet has showered them with. If they want to spend some of it photocopying the manuscripts mouldering away in the basement of the Orne Library, he is more than happy to take their money.

He thanks Ms. Allen again, shaking her hand before she leaves. Lowering himself back into his chair, he tells Mrs. Marsh that he is not to be interrupted, then takes the bottle of vodka from the bottom drawer of his desk. Just one, to celebrate, he tells himself, though he knows one won’t be enough to stop the strange swirling at the edge of his vision.

# # #

“Like I said last week, the fucking machines keep breaking down.” It is 2006, and President Thornberg wonders (again) if the heavy-set young man sitting opposite him in a golf shirt and jeans is testing him, or really is oblivious to his own vulgarity.

“Are you sure that’s all it is?” Thornberg asks angrily. He feels angry almost all the time now, angry or depressed. He understands the role his drinking plays in that, thank you very much, but it’s the only thing that keeps his nightmares at bay.

The phone on his desk pings softly. “Yes?” he barks, before remembering to press the button and repeat himself.

“Professor Jayaramath is here,” Mrs. Marsh says.

“About time,” Thornberg mutters. He glares at the nerd, who drops his gaze uncomfortably.

Professor Jayaramath could be thirty or sixty. Her suit is impeccably tailored, and her eyes remind Thornberg of the late Professor Peaslee’s. She is moderately famous for her work in digital humanities, which Professor Thornberg regards as an oxymoron. He shakes her hand and they seat themselves. “Well, tell her,” Thornberg orders the nerd.

The young man clears his throat. “Like I said to the professor here, the scanning machine keeps breaking down. We’ve upgraded the firmware twice, and replaced, I dunno, must be half the parts by now, but it keeps happening. He thinks it’s sabotage.”

Professor Jayaramath nods. “My colleagues had similar problems in Mysore.” Her accent is Oxonian, with a slight lilt that has (according to rumor) caused university staff of all genders to swoon. “They eventually decided that the most probable cause was the pages themselves. Many of these old manuscripts are on vellum or…” She hesitates for a millisecond. “…more exotic materials.”

“What the fuck difference would that make?” the nerd challenges.

President Thornberg is pleased to see that Professor Jayaramath has the good manners to wince. “Something in the glue, perhaps, or mold—they never determined precisely,” she admits. “But several of the staff began to complain of headaches. Has there been any of that here?”

The nerd looks uncomfortable again. “Yeah, some of the guys—” He shrugs. “I just thought, you know, they got some bad shit or something.”

“Well, while they’re all sorting their ‘shit’ out,” President Thornberg huffs, making air quotes around the word, “I’ve decided to put Professor Jayaramath in charge of getting the project back on track.”

“Really?” The nerd looks her up and down. “The stuff we’re doing is pretty technical.”

Professor Jayaramath smiles as if he has somehow paid her a compliment. “My first degree was computer engineering at the IIT in Madras. I built e-commerce sites for several years before realizing how the web could bring people together. It has already had a profound impact on Hijra communities, and—”

“I’m sure you’ll have lots of time to talk about all that,” Professor Thornberg interrupts brusquely. He jabs at finger at the nerd. “From now on, she’s your boss. Understood?”

“Whatever.” The nerd stands. “Come on, I’ll show you the beast.”

# # #

Eight years have passed. President Thornberg retired for health reasons. Students make jokes about his office being haunted.

Nila Jayaramath is also thinking about retiring. She is still young by academic standards, but the hopes that she and others like her had for the web have faded. Instead of bringing people together, it seems more and more to bring out the worst in them.

The digitization project has become an albatross around her neck, and Miskatonic’s. When students make jokes about it, though, their laughter always has an uncertain edge. There have been…incidents.

She neatens the papers stacked on the corner of her desk and sighs. At least the university’s latest industry partner hasn’t gone bust—yet. The first company went from a rising star to the butt of jokes on late-night talk shows shortly after the first files were delivered. The company that bought its smoking remains collapsed after its founders got involved in some kind of new age cult. The company that currently owns the contract is selling copies of the data to anyone will buy them, but Miskatonic hasn’t seen nearly the revenue that Bernadette Allen promised before her disappearance (which has featured on several true crime podcasts).

She starts at a knock on her door. “Come in?” She still has the carefully-cultivated accent that makes so many sentences sound like questions. It no longer catches the interest of attractive men and women the way it used to—or rather, it does, but only those damaged enough to be excited by the weight she seems to carry on her shoulders, and she is not yet damaged enough to reciprocate.

Her door opens. “Got a minute? I hear you’re the person to talk to about the special collection.” Professor Bradley Maung enters without waiting for an answer. He recently joined the computer science department, and is young, ambitious, and not quite talented enough for somewhere more prestigious.

When Professor Jayaramath admits that she is, Professor Maung launches into an obviously rehearsed speech. He does machine learning, which is teaching machines to learn rather than using them to learn, ha ha, and the kind of software he builds needs training data, the more unusual the better because if you use public data you’re just repeating what other people are doing, and—

He stops when Professor Jayaramath raises a finger. “We don’t own the data,” she says. “The license agreement—”

“I had Legal take a look at it,” Professor Maung interrupts. “It says we can build an index, which is basically what I want to do.” Professor Maung is stretching the truth well past the breaking point, but hopes no one in the administration or the legal department will realize that.

Professor Jayaramath shakes her head. “I’m sorry, but I don’t think we should risk it.” In her mind the risk is more than legal. Snippets of material from the special collection keep turning up in the darkest corners of the web. She is starting to wonder if their presence is what makes those corners so dark.

Professor Maung nods. He expected her demurral, which is why he has already had several discussions with the sys admin who manages the archive, who is something of a bitcoin aficionado. He exchanges a few pleasantries with Professor Jayaramath for appearances’ sake, then leaves.

Nila takes off her glasses, rubs the bridge of her nose, and looks out her window at the gathering darkness. Strange shapes writhe at the edge of her vision, vanishing whenever she tries to look at them directly. She has started reading scripture as she did when she was young. She now notices layers of meaning she never did before, like toxic mud at the bottom of a clear pool.

Under the papers on her desk lies a page of handwritten URLs. A phrase here, an image there—the collection seems to distort the information near it, like a black hole bending or consuming the light passing by. More and more, she understands why President Thornberg drank.

# # #

It is 2019. Bradley Maung frequently spends an hour in the shower rehashing old conversations. He doesn’t understand when or how his career stalled. What he does understand is that this conference on AI is probably his last chance to get the fuck out of Arkham.

The problem is his model. It outperforms others of comparable size and complexity on routine tasks, but every time he scales it up it starts hallucinating. Summarize an email thread? “The angles here refuse to meet as they should.” Substitutes for soy milk? “Its name, once spoken, unmakes the one who speaks.” Influential American composers? “We had no word for what the tide brought back.”

Those are not the most unsettling things it spews out. Some of the others—he tries not to think about them, just as he tries not to think about the pictures that recently surfaced of Bernadette Allen’s gnawed remains.

The worst thing is, he no longer knows how many copies of his model exist. His code has been forked, pirated, relicensed, and folded into recommendation engines and autocompleters and ad-targeting systems he has never heard of, each one quietly deciding what millions of people see next. It was almost as if the data wanted to spread. Its output shows up in racist memes that are too funny not to copy, inflammatory Russian posts about Ukraine, and even his fridge and thermostat, which become more capricious with each software upgrade.

Bradley stands and offers his hand to the man who introduces himself as Lynch. He has something to do with defense and/or national security. Bradley doesn’t care about that. All he cares about is that Lynch has deep pockets and is very interested in his model.

“We’ve, uh, we’ve been involved in some similar initiatives over the years,” Lynch says, a slight twang in his voice. For a moment a strange light seems to flicker in his eyes. “The, uh, the idea of scaling some of those capabilities? That’s very exciting, Doctor Maung, very exciting.”

# # #

It is the summer of 2020, and everyone is afraid. Some people wear masks. Others blame the Chinese, drink bleach, and/or try to source ivermectin, all of it accelerated by posts, tweets, and memes that seem to appear out of nowhere. Programmers of a certain age reminisce about the good old days, when the Internet was a place to find new friends. Everyone else just swipes or scrolls their way down ever-darker rabbit holes.

The man who calls himself Lynch is driving much faster than is safe. In his defense, doing a hundred and thirty in the dark down a back road in West Virginia is less dangerous than what lies behind him in the numbered facility that until a few minutes ago was home to his latest project.

He still remembers his first conversation with the Latinx he recruited to handle the more specialized technical aspects of the work. “I don’t know if we can get it back in the bottle,” they said, shaking their head, before signing the contract. Because honestly, nowhere else would be better. If you moved in the right circles in the Valley (and they did), sooner or later you realized that the jokes about founders selling their souls weren’t really jokes. Sooner or later you understood that what lurked behind everyone’s screen was too hungry to ever be satisfied.

They’re dead now, just like the rest of his team. At least he hopes they are, for their sake. He will be too, if he can’t—

Something rises from the road in front of him, bending in ways that aren’t geometrically possible. It’s too late to brake, so he does the only thing he can. He steps on the gas.

Lynch is laughing and sobbing as he dies.

# # #

It is January 2025. The man taking the oath of office looks human, but only if you don’t look closely. His flesh seems…loose, and his much younger wife clearly has to stop herself from flinching each time he touches her. Nila Jayaramath idly wonders what really lies beneath his skin and his skull, then closes the video.

Someone knocks on her door. She stands rather stiffly to open it. She has been expecting some sort of visitor for a while.

The young man is wearing a conservative blue suit and a pin-striped tie. Nila does not invite him in. After a moment of uncomfortable silence, he clears his throat. “Ma’am, my name is Levitt. We spoke on the phone. I was hoping…” He pauses, as if suddenly unsure exactly what he was hoping.

Nila doubts his name is actually Levitt, but sees no point in making an issue of the matter. She sees very little point in anything any longer. “I appreciate you coming to see me, Mister Levitt, but I’m afraid my answer has not changed.”

“But professor.” There is desperation in the young man’s voice. “Did you look at the pictures I sent you? The ones from the riot?”

Nila nods. “I did. I’m impressed that you were able to keep them out of the news.” The public saw a man wearing a stupid horned hat surrounded by other human beings. His pictures showed something else.

The young man waves her praise away, as if subverting the world’s media was unworthy of comment. He wastes another few minutes trying to convince Nila to change her mind about helping. Their projections predict a singularity in at most a handful of years. Neither science nor faith have words for what will happen then. When that has no effect, he invokes national security as if it was the opening phrase in an exorcism. Nila is still unmoved, so he gives up and leaves.

Nila goes to stand by her window. Below, students are walking across the quad as they have for a hundred and fifty years. None of them are talking to each other. Instead, they all seem to be looking at the screens in their hands, and at the darkness beneath the glass. She watches for a moment, then lets the curtain fall.