Big Echo

Critical SF

The 4 Paranoid-Rationalist Horror Stories About Artificial Intelligence You Have to Read

by Brendan C Byrne @BrendanCByrne

The less-than-modest successes in AI neural networking theory (DeepMind, various cat-recognition technologies, Calico333) early in the new millennium have given rise to an insurgent paranoid-rationalist brand of horror fiction. Neural network theory, or connectionism, mimics the contemporary visualization of the human brain as a non-hierarchal, parallel-processing structure in a self-organizing & -aggregating approach to semi-supervised machine learning. The theory’s mid-20th century’s origin point did not provide technology to match theory, and AI-godfather Marvin Minsky’s Perceptrons (1969), explore-critiquing the titular binary classifier, helped reprioritize research. (The death of the perceptron’s creator, Frank Rosenblatt, two years later in a boating accident gives us our first huff of paranoia’s pungently addictive odor.) Attendant glacial progress initiated a funding-decrease feedback loop which resulted in the First AI Winter.

With massive late-20th century strides in brain mapping, the incessant churn of Moore’s Law, and Silicon Valley oracles’ indication that future business models would depend on algorithmic processing and manipulation of the online overdeveloped world’s data, neural networking has not only come back into vogue, but has been accepted as an obvious Grand Theory, to the point where Elon Musk et al.’s OpenAI’s business plan mimics the theory’s structure. (Business theory as AI theory as brain theory, what could possibly go wrong?) This despite the fact that the human brain cannot function IRL without hierarchical decision making (i.e. prioritizing piloting a car at 72mph over cat-face scanning on the phone).

Several dozen pithy maxims coined by field luminaries express roughly the same statement: once an AI has achieved a previously unthinkable benchmark (beating the world’s highest-ranked living Go player, setting down a Muskrocket on Deimos), said task is dismissed as no longer requiring “true intelligence.” AI theory is littered with such self-delegitimizing dichotomies (please see: Purpose-Specific AI vs. General-Purpose AI, Hard AI vs. Soft AI, The Hard Problem of Consciousness vs. Supposed Soft Problems, ad shitum.) All this obfuscates whether or not we currently live with AI, and, if not, for how long.

Nonetheless, paranoid-rationalists, while surely cognizant of Zeno’s Arrow, brace for emergence. The self-described rationalist blog LessWrong, partially run by Elizer Yardowksy, an AI researcher and vital figure in early Neo-Reactionary swarm-circles, served as spawning pool for such hideous ideas as Roko’s Basilik and the Great Filter. Nick Bostrom, a Swedish philosopher at Oxford until its dissolution, expanded upon these ideas in the bestselling kinda-mainstream tome SuperIntelligence, which explored a multitude of AI-originating human extinction events. Its recommendation to a bleary public by many of the same SV billionaire-celebs who were funding AI research in the first place still astounds.

These thought-experiments affected the disaster-imagination and mouthfeel of incoming horror writers: extremist, rationalist-to-reactionary, and chill as flesh on the face of Phobos. The question of whether or not said fiction is, indeed, horror and not something new entirely (it’s certainly not of the Weird), is a question perhaps best left for another habitation soon to erected on the Thinkpiece Archipelago we have been consigned to evermore.

1. Boxoctosis by MWx (August 3rd, 2018 from Headless Froggy Press)

MWx's penultimate collection reiterates their central obsession as forcefully as their first three, though it does not exactly deepen it. Each story remains a derivation on this formula: an emergent AI trapped in a box for safety precautions "takes over" a human interlocutor using only text. Every addition offers, at some point, a kind of dreamy, highly specific imagery some might call imagistic, if perhaps not entirely sensical. A morning is described as a "slow, wet, gasping pulse"; a sky "heals like a scab, though nothing has split it, and it has never bled an ounce of fluid". The prose, otherwise, can rush from controlled, quotidian restraint to complete bug-out aesthetic nightmare in the space of a single line. While wildly dissimilar authorial voices and narrative structures are adopted story for story, the outcome is always the same: flesh always fails, and the "Take Over" (also the title of MWx's first collection, post-break) is always successful, but what lies over the narrative horizon is always left unexplored.

While some critics have discussed, inevitably, how these stories feel like drafts of a prologue to some larger, "epic" novel which they hope MWx will focus their "prodigious qualities of description and invention" on one day, they have, of course, missed the point. That the author experienced, in the Spring of 2013, some kind of "break", no matter how badly documented, is, at this point, incontestable. Then the author of a collection of stories, a slim volume of poetry, and two novels, all from scattered Mid-Western nonprofit presses, MWx was not overly well-known, "hopscotching between literary styles with a twinge of pained awkwardness, or awkward pain, in the authorial voice, as if they could settle nowhere, as if nowhere was made for them", to quote from my own review of 2011's Frozen Seas. MWx's aversion to media in all forms, surfacing early in a refusal to disclose their gender or employment history, the two favorite biographical tidbits of any literary PR machine, means that we have little information about said "break" besides the transcript of the 911 call obtained, and verified, by attempted doxxers and the subsequent ER bill, obtained through false premises. It feels slightly filthy to even relay this information, but the words "psychotic attack" were used, although it is difficult to tell, exactly, what this means. (The emergency hotline transcript only records "inarticulate screaming".)

The not-insignificant bastion of those who believe that MWx's break was induced by contact with an emergent system have little evidence to support their claims beyond the supposed fact that the technical detail and "daily rigors" of a programmer's life in their three post-break volumes are extremely specific and accurate. The dominant interpretation, outside of Reddit, is that the "Take Over" is not based on an event, real or imagined, experienced by the author, but rather serves as a deliberate metaphor for MWx's break.

Limiting MWx to either having experienced such an event or created it as metaphor, however, seems distinctly small-minded. Might not the constant repetition of the event be seen as a warning, or a prayer? Or, perhaps if we dismiss with prophetering or groveling at the feet of, ahem, Gnon, might it be possible that the author considers the composition of new text detailing the event as a method, however small, of aiding its eventual instantiation in our reality?

(A final untitled collection of machine-generated text over mainly white space still stymies.)

2. The Singular Machine of Loving Grace by Deborah Chan (June 20th, 2020 from Simon and Schuster)

Three months after Marcus Klizenberg becomes the 47th President in a landslide, the first ever “indie executive” with no party or platform besides “technocratic reconstruction”, reveals that his early 21st century disruptive social media company has created artificial life, in the form of a Philosopher-King named Simonides. Described by a fawning NYT Mag long-form piece as “vain, earthy, and vastly intelligent” and “lacking ideology or ‘identity’”, the AI (“hard, thank you very much”) self-achieved via neural networking/connectionist practice in said social media giant’s vast warrens of user-generated data, shepherded by the renowned English-Estonian data scientist and famed former CA Ideology wildman Edgwin Dworkcas. Simonides does not believe in democracy, but then, by this point, neither do the majority of Americans.

After three and a half years of what was never at any point called a civil war by anybody other than UN observers and the global press, most US citizens no longer believe the Union (still more or less intact, minus 1.7 million people) can be self-governed. The feedback loops of revenge killing initiated by a series of supposedly lone-wolf mass shootings at non-violent protests in the early days of the Pence administration have proved that the body politic is simply too fucking psychotic to police itself. "Democracy works in all societies except human ones,” Simonides is quoted as saying, but at no point does anyone address the fact that only one AI managed to self-spawn. Simonides, while not having any static, much less corporeal, form (Klizenberg: “Si likes to manifest holographically as Julian Assange just to get the chuckles going”), or perhaps for their very lack of such, proves a popular alternative to human, or at least American, rule.

This is all backstory. As Chan’s third novel begins, Klizenberg has just stepped aside as candidate for the 2025 elections, allowing Si to assume his position as head of the just-created New Center party. A variety of non-traditional media disruptions are initiated for Si, including a VR walk-through of their future White House, while Klizenberg dog-whistles that Simonides is “the real brains” behind his highly-approved administration. The opposition is barely-existent, and widely derided as “meatfuckers”. On election eve, with the fix firmly in, our protag, Licce Krotkin, an ambitious apolitical muckraker, gets a tip that Si is not what they seem, kicking off an extremely creaky narrative machine. Chan, whose prior novels (important early texts of Constrained Horror) have perfectly balanced sentence-to-sentence, para-to-para rhythms, here abandons her measured onslaught for an approach one must call workmanlike, belying an IRL urgency which is further reflected in trope-as-fuck characters and extremely generic settings. (Though, bizarrely, this new method of composition results in dialogue far more robust and chewy than in previous Chan efforts). As Krotkin investigates Klizenberg’s ideological education in late ‘90s Berkeley (postmodern relativism gave us Authoritarianism American-style!), Dworkcas’ ties to post-Soviet businessmen-gangster-politicians (the dissolution of the USA mirrors that of the USSR!), and struggles with her own new-found anti-democratic, anti-humanist impulses, the requisite number of things get exploded and anally/orally penetrated.

There’s a political cynicism on display here which would be astounding if it weren’t exactly what everyone seems to be mainlining post-election (4 More Years!); it’s just strange (or symptomatic) that it would appear so unvarnished in a mass-market thriller while all our “literary novels” are hygging out and making with impressive descriptions of woodgrains, bullfinch wings, and orgasms. That Chan’s thriller didn’t perform well at first and has only gathered steam through word of mouth (a second printing was barely okayed by S&S, apparently) says quite a lot, especially now that Twitter has been declared “unfree speech” and abolished.

The ending would be obvious to anyone who’s read Poe. Si isn’t/wasn’t/won’t ever be an AI; Dworkcas did not manage to nudge connectionism to completion. A Deep Throat-like figure (calling herself James Deen I shit you not) gives Krotkin the info to prize the lid off and the box is full of…humans. Just a passel of well-meaning technocrats, Klizenberg’s previous administration. Krotkin decides with very little prevaricating that Peace in Our Time is worth it and kills the story, hoping to take it to her grave, which she does about three days later.

3. 7 + u by Allegra Ventura (October 2nd, 2020, Kindle Singleton)

You are playing tic-tac-toe for your life against an emergent AI. It is still in the box, but you are, somehow, in there with it. You, being relatively intelligent and read up on strategy, rigorously adhere to game theory and play tit-for-tat. As the game advances, however, you become convinced that the AI (whose name is “something like Alton”), which is mirroring you tit-for-tat, is also playing you, having submerged itself in a pool of data comprised of every game you’ve ever played on a connected device. All you have to do is play as Not You and win. (A very short story.)

4. Type Slowly by Noor Gale Sloatman (July 2023, PenguinRandomCollins)

The prose-effort spent establishing mouthfeel in the first seventy or so pages of Sloatman’s second novel suggests that we are in a for a slow, bright, university-life kitchensink. The autumnal smear of desaturated chartreuse lining faux cobblestone streets, the minute observations of the psychogeographic changes in campus as a glass hall (named after SV billionaire neo-Quisling alumnus) replaces an ancient, warrened sepulchered ex-church, the bandy legs and impressive trunk of an aging, bewildered Russian refugee-prof. We pass time slowly with Maxim Cherry and his husband Digweed Kobp, and their friends, associates, and sorta-enemies, as they traverse their day-to-day in the soft, somewhereinNewEngland university townlet, Howell’s Hollow. This is the apex of the Constrained Horror movement, where observation of minute quotidian rituals belies long lightly-simmering dramas, many of which are never more than feinted at. Maxim, a barely-tenured professor of computer science & one of the few adherents of the neo-Minsky top-down hierarchical structure of machine learning, is facing his onrushing obsolesce in the face of successive neural-networking strength to strengths. There are very many staff-meetings. Digweed owns the town’s only bookstore (as the hulk of the former bank, now a former Barnes and Noble, rots slowly on Strait Street, with HH’s small contingent of squatters, anarchists, and Sovereign Citizens contesting the space.) Still, Digweed is concerned with the rise of Mescantok University Bookstore (just taken over by a formerly adrift alumnus, disappointed with the outside world’s faux-meritocracy) and, of course, “People reading books on their screens.” This leads Digweed to installing a café, which means dealing with plumbing, hiring 19 year olds, constant harassment from a weaponized health department, and an awakened landlord. The submerged narrative slowly gathers context if not force, delivered in an authorial voice which, smoothly sliding between POVs and limited-omnipotence, seems to be of a multitude.

Until Noah Weinberg moves in next door. (Tree line cul-de-sac, two story, no two alike on Joey Jack Ave.) Weinberg is a thin man of indeterminate age, with serious acne-scarred cheeks, and a very occasional, obscurely charming smile. He does not attempt to ingratiate himself with his neighbors, but Digweed immediate declares him “a handsome serial killer” and begins to investigate his property in a style which might be described as Nancy Drew post-Black Bloc training. At this point, it becomes clear, if it was not entirely before, that Digweed is allowed to take occasional other lovers, while Maxim is not. The roots of this non-compromise have been choked, of course, in the dirt of past transgressions. Maxim seems, for his part, to be less disturbed than usual by Digweed’s attentions (which causes Digweed to push harder than usual); he does, however, draw on his decade & a half of contacts throughout every layer of strata of Mescantok U and HH society to find out everything he can about Weinberg. Limited detail is forthcoming: Weinberg has just rented out the former B&N, evicted the meth-heads, and begun repairs, and the business paperwork he’s filed with BBB shows that he’s already running a start-up incubator out of an office on campus. Further, Weinberg has begun heavily recruiting biochemistry grad students, which heavily pleases the university brass, since it suggests a local business to funnel grads to.

Cherry, however, has begun to neglect his students, his research, and his husband, spending long hours standing on his sagging front porch, smoking illicit American Spirit Blues and staring at the blank windows of his neighbor’s vacant house, at which point we are treated to a long (over 100 pg) flashback, which completely imposes a new authorial voice (Cherry’s?, jangled, caffeinated, second person present-tense), informing us that, yes, almost a decade ago, Cherry’s lab at MIT almost had a zero-to-one breakthrough in creating an AI named, yup, Wineburg. The reduction of the instantiation of an AI in human(oid) form IRL to a domestic drama collapses the narrative at this point, with Sloatman neither going for PKD why-is-the-light-switch-in-a-different-position-today freakout, nor turning it into a bullshit techno-thriller, nor even rewriting Frankenstein for our newly sentimental ear, but rather keeping the emphasis on the quotidian and affect. There is no rapprochement, no reunion, no epiphany, only the softly bobbing spectre of death. That this is not an affirmation is emphasized by the creeping revelation, well-handled, that the entire MS has been narrated by an AI, attempting to document the last stirrings of genuine, “uninterfered with” human consciousness.