Flattened by Genius
Thoughts on why the best novel I’ve read recently ended in disappointment.
AlphaGo, the Go program du jour in 2016, defeated Lee Sedol three games in a row. Even though the first player to win three would be declared the winner, Sedol was contractually obligated to play all five. He stole the fourth. Lee Sedol is considered one of the best humans to ever play Go, if not the best, and the move he made in game four, solidifying his, by then, unexpected win, has been described as divine and touched by God and a move so unfathomable not even the supercomputer anticipated it.
Sedol’s placement of his 78th white stone—a wedge—caused a delusional fit in AlphaGo. In the end, four to one, AlphaGo.
The stakes couldn’t have been higher. Man versus machine (back when it didn’t happen daily). One South Korean man from humble, human beginnings turned world champion, pitted against DeepMind, the technology company backed by millions of dollars (purchased by Google for half a billion dollars) whose purpose was to create a computer program incapable of losing at the 2500 year old Chinese board game.
After the match, Sedol said he learned a lot from AlphaGo. Go moves he would have previously considered to be creative or masterful now felt conventional, perhaps unnecessary, in the grand scheme. For Go players everywhere, the program had unlocked a new way to play the game, the boundaries had been pushed.
AlphaGo’s style of play was unconventional because it was fine with appearing to lose early on in a game, by seemingly giving up territory or sacrificing stones, so long as its probability of winning increased with every move.
“If AlphaGo must choose between a scenario where it will win by 20 points with 80 percent probability and another where it will win by 1 and a half points with 99 percent probability, it will choose the latter, even if it must give up points to achieve it.” - DeepMind research scientist Thore Graepel
In a way, AlphaGo can see into the future, thus eliminating the drama inherent to competition.
At age 36, Lee Sedol retired from professional Go, he said, “Even if I become the number one, there is an entity that cannot be defeated."
DeepMind’s goal beyond AlphaGo is to create a program that can learn and, therefore, think.
Benjamín Labatut’s The MANIAC is a loose triptych that closes with the AlphaGo saga. The book mixes fiction and non-fiction, never indicating explicitly its status as a novel or book of essays. The opening of The MANIAC is about a man named Paul Ehrenfest. The book’s first line:
“On the morning of the twenty-fifth of September 1933, the Austrian physicist Paul Ehrenfest walked into a Professor Jan Waterink’s Pedagogical Institute for Afflicted Children in Amsterdam, shot his fifteen-year-old-son, Vassily, in the head, then turned the gun on himself.”
The Ehrenfest intro only lasts 25 pages while the AlphaGo essay takes up the last 80 of the 350-page book. Despite the precision of Labatut’s prose and the way his sentences stretch and contract like a puppet master’s strings, connecting the dots of history from Albert Einstein to Robert Oppenheimer and everyone in between with an exacting, yet propulsive pace, there is a looseness to the book. The events and characters are real, but the scenes and dialogue are made up–probably. A statement from the author claims: This book is a work of fiction based on fact.”
The MANIAC’s strength is in its ability to keep the stakes of the story upfront. Murder-suicide, World War 2, Mutually Assured Destruction, the fate of the human race… well-trodden territory, but Labatut is able to colour the space between the monumental events with the utmost drama.
Paul Ehrenfest lost faith in physics as he watched it stray from its classical origins toward quantum mechanics, and he felt scientists were losing their sense of reason in pursuit of what was rational. Labatut is at his best when the historical figures in his works are at their most hysterical:
“Paul detested the likes of John von Neumann, that Hungarian wunderkind, with his "terrifying mathematical guns and unreadably complicated formula apparatus," as much as he despised the indigestion that the "infinite Heisenberg-Born-Dirac-Schrödinger sausage-machine factory" caused him. He lamented the attitude of his younger students, who "no longer noticed that their heads had been turned into relays in a telephone network for communicating and distributing sensational physics messages" without realizing that, like almost all modern developments, mathematics was hostile to life: “It is inhuman, like every truly diabolic machine, and it kills everyone whose spinal marrow isn't conditioned to fit the movement of its wheels."
“Sausage-machine factory” and “it kills everyone whose spinal marrow isn’t conditioned to fit the movement of its wheels” stand out. The book is filled with these scathing observations.
But, Ehrenfest was depressed, and before committing the above act, says to his mistress Nelly:
“...in utter helplessness I crave your proximity, and if that craving gives me neither warmth nor strength, then I am overcome by desolation. Love is such a mightily divisive element. All the suffering it brings! Surely it's one's duty to put an end to one's life as soon as possible, before causing hideous destruction."
His wife divorces him when he is unable to split with his new lover. And perhaps having waited too long, Ehrenfest ends two lives by way of the hideous destruction he so fears.
After this harrowing introduction, the actual story begins by introducing John von Neumann (AKA Janos, Jansci, Johnny, etc.,) the Hungarian mathematician, physicist, computer scientist and engineer who died in 1957 but not before theorizing, quite accurately, about the future of advanced computing and artificial intelligence. This middle section, which feels most like a novel, is an oral history of John von Neumann, as told by the people in his life, including his two wives, daughter, a childhood friend and classmate, teachers, fellow scientists and mathematicians, colleagues from the Institute for Advanced Study, colleagues from the Manhattan Project, an American military officer, etc.
“There are two kinds of people in this world: Jancsi von Neumann and the rest of us.” - Eugene Wigner
“Whatever judgment awaits me, I cannot deny that I saw it before anyone else. What he could do. It was so rare and beautiful that to watch him was to weep. Yes, I saw that, but I also saw something else. A sinister, machinelike intelligence that lacked the restraints that bind the rest of us. Why did I remain silent, then? Because he was so superior. To me, to all of us. I felt slightly ashamed in his presence. Belittled and debased. Nothing but a foolish old man with foolish ideas.” - Gábor Szegö
“I was instantly enamored, and stayed at his side enjoying the lunatic pleasure of seeing all those people destroying themselves, while he scribbled numbers and solved equations on a large piece of paper spread out in front of him, as though he was a schoolboy in detention. When he was down to his last chip, I picked it up from the green felt and walked over to the bar, where we proceeded to get delightfully drunk, although I had to pay for all his drinks, of course, since he was absolutely broke. It was over there and then for me: we were hooked, and after that, things went so fast that to me it seems as if that roulette wheel had not yet finished spinning when we were already married and miserable, halfway around the world.” - Klára Dán
Von Neumann’s legend grows over the course of the book, and so too does the reader’s understanding of his intelligence. His many irons stoke the fire of a rapidly accelerating 20th century. He contributes to the development of the bomb, game theory and economics, the development of the Mathematical Analyzer Numerical Integrator and Automatic Computer Model (MANIAC for short), and more. He comes across as a heartless rationalist but is redeemed for his incredible efforts to solve the unsolvable, not really motivated by riches or fame—though he welcomes all that—but more so by a need to know everything.
Gábor Szegö goes on:
“And now, older still, I recognize that, in spite of his callousness, he was trying to understand this world at its deepest level.”
There form chosen here is apt. We experience von Neumann entirely through the eyes of others. He isn’t the narrator, we aren’t inside his head, so this authorial choice keeps him distant, like an alien.
The 250-page middle section of The MANIAC is also where the book’s form is at its loosest.
With the dozen characters charged with telling us about von Neumann, the prose, and therefore the voices, can’t help turning into something of a chorus. With the exception of the computer engineer Julian Bigelow and his sections, which stand out from other chapters because of their sentagraph style, and a few lines from the Marina von Neumann and Richard Feynman, most of the voices sound the same, a sameness similar to the voice found in the sections about Ehrenfest and AlphaGo.
Julian Bigelow:
“This was 1951.
So we had to use war surplus parts and vacuum tubes that would fail
without warning.
In summer the room got so hot, tar would splatter on the machine. Months of work ruined in an instant.
And the memory was incredibly fragile. Someone wearing a woolen sweater could wipe it clean.
Passing cars and planes did that too.
And this mouse crawled inside it once.
Chewed on some wires and was burned to a crisp.
We rescued the machine but never managed to get rid of the stench.
It always smelled of charred meat, singed hair,
and burned whiskers.”
Feynman:
“We fought over the chessboard in Los Alamos, you know? Then someone brought in a Go board and we started playing on that too. Vicious, endless games with no time limit against some of the smartest guys I ever met. It was nerve-wracking. It ate away at me. 'Cause I'm competitive, see, I like to play and I like to win, so I couldn't help myself. We had nothing better to do!”
A few instances of you know and see and exclamation points do just enough to give Feynman the New York accent he may or may not have had.
Marina von Neumann on her step-mother Klára Dán (Neumann’s second wife).
“Klari, Klari, Klari. Klari the figure skater, the witch, the shrewd, the nag. Klari the MANIAC’s wet nurse. She programmed it, Father’s machine at the institute.”
But even these three, out of the dozen characters, have a tendency to fall into the meticulously deployed verbs and run-on sentences Labatut employs so well.
Here are a few more excellently written passages that may or may not be from different characters:
"The mathematical universe is built much like the pyramids of the ancient pharaohs. Each theorem rests on a deeper and more elementary substrate. But what supports the bottom of the pyramid? Is there anything solid to be found there, or does it all float on the void, like an abandoned spiderweb blowing in the morning wind, already unraveling at the edges, held together merely by frail and thinning strands of thought, custom, and belief?”
and
“Even the simplest programs can lead to fabulous complexity. And the opposite is also true: you can erect a sprawling, many-leveled tower of ciphers that produce nothing but sterility, a barren unchanging landscape where no water will ever fall. [...] While I know [von Neumann] to be a scoundrel, I cannot deny that his mind was unique, and that he had a special, almost filial relationship with his computer, his MANIAC.”
and
“The hydrodynamics were simply too complex for an individual to handle. It was godlike mathematics, equations so intricate that not even Fermi or von Neumann could see them through to the end, at least not unaided. But I could tell that it was exactly that the hopeless impossibility, the irreducible complexity of the calculus that seduced von Neumann. Here was an unsolvable problem, and he was slobbering! He couldn't help himself. It was obscene. […] we couldn't get von Neumann to take his hands off them. It seemed that he had forgotten all about the war, all about the bomb. Knowing what he did later, it makes sense, but back then I was confused.”
and
“The horror of experiencing the deterioration of his mental powers was too much for him to bear. He was only fifty-three years old when he was diagnosed, still in the prime of his life, and he retained his reason and extraordinary faculties almost to the end. But he simply could not accept what was happening to him. Terror of his own mortality crowded out all other thoughts. Try as he might, he could not visualize a world that did not include himself thinking within it, so he had none of the grace that some people exhibit when they finally accept their fate.”
These quotes come from Theodore von Kármán, Nils Aall Baricelli, Feynman, and Marina von Neumann, respectively.
Is this a problem? The book doesn’t claim, in any way, to have distinct voices. If these oral histories are the product of interviews (within the apparatus of this novel) it would stand to reason that the book would be edited by a single editor. Furthermore, most of the characters would, technically, have been articulate writers part and parcel with their work as successful academics (theorizing, writing papers, lecturing, etc.). But it doesn’t matter much. Labatut’s writing sings with grandeur, eloquence, rhythm, and ease1.
The plot remains gripping as it builds a case toward our contemporary, maybe contemptuous, relationship with technology, ultimately, arguing in favour of the truth—whatever that may be. This is the hot spot. The book’s main source of tension lives in the gray zone where we try to identify the differences in perception of what stands to reason as true and rational. Are you Ehrenfest or von Neumann, Lee Sedol or DeepMind? The voices, homogenized into a sort of Labatutian chorus, can only sit in awe of the destructive beauty of superintelligence. Or is he just a maniac?
Oskar Morgenstern, von Neumann’s co-author of “Theory of Games and Economic Behaviour” questions, in retrospect, von Neumann’s almost puritanical rationalism, making the case instead for human irrationality.
It’s an interesting moment in the book. Morgenstern claims that while, in theory, treating conflict like a game makes sense, as the mathematics of it can be proven, but life isn’t a game. Von Neumann, arguably the smartest person to have ever lived, develops theories that essentially force human behaviour into non-human frameworks, treating behaviour as a set of inputs and outputs, in essence, to be more like him, as he tries to force all of humanity into a von Neumann-shaped box.
And then, today, in the case of DeepMind, we have a technology company spending billions of dollars to force what we know about humans, our ability to learn and think and create, onto non-human machines in hope of some similar outcome—artificial intelligence forced and formed within a box in the shape of human consciousness.
Von Neumann’s daughter finds an unpublished memoir written by her stepmother, Klára Dán, in which she describes life with von Neumann:
“…she confessed that she felt "like a tiny little speck, an insignificant insect who had just been chirruping around to see where the most fun could be had, when I was swept up by the hurricane force of international events and global minds." No matter what she achieved, she could not measure up to her husband's accomplishments, or become the sole object of his interest, as she so keenly desired.”
Such is life with a computer. The genius flattens all those around him. But maybe we can thank God he was human, at least John von Neumann had the capacity to desire.
I watched the AlphaGo documentary and was disappointed to find out that Labatut pretty much retold the story verbatim—from the beginning, with the origin of DeepMind’s founder, to the detailed nuances of face off between Lee Sedol and AlphaGo—including the smoke breaks and the branding of move 78 as divine.
Discovering this regurgitation let some of the air out of the inflated sense of accomplishment I’d associated with the book up until that point. But I also realized, this feeling of disappointment is how I feel about content produced by generative AI.
And, yet, if artificial intelligence and large language models are now tasked with setting the example for the potential of human language and thus storytelling, it’s apt that Labatut’s bookend on AlphaGo and Lee Sedol is basically a regurgitation of the documentary2 on AlphaGo and Lee Sedol. Your everyday LLM, as far as a consumer is aware, doesn’t seem to understand language so much as it is able to mirror it, producing text derivative of the text it was trained on3.
In the book’s final section, AlphaGo continues to defeat the world’s best human players, but its progress slows down. DeepMind eventually removes all human experience data from future AlphaGo iterations—it “learned” to play against humans—and within days (hours, even) the new program known as AlphaZero has played hundreds, if not thousands, of matches and defeated all other existing computer programs at our most complex games; unbeatable at chess, shogi, and Go4.
Can LLMs be programmed to never lose at writing? I suppose current models barely, if ever, lose, and outside of the odd hallucination that goes viral, base level chat bots are able to string together text in increasingly impressive ways. However, writing isn’t about being undefeated.
Labatut may have had an easier time creating 12 distinct voices had he fed a few interviews into an LLM and asked for as much. But why would a writer want to remove themselves from the writing? Writing is not a game with winners and losers, though it is a competition with oneself—not the reader, not the culture. A work of art does not defeat the one who experiences it; the novel is collaborative, more of an extension of the person who wrote it. It is the original artificial intelligence.
LLMs might eventually come to “understand” language in a way that moves far beyond mimicry, becoming genuinely able to teach us new ways of understanding literature, as was the case in Go—but what makes writing and storytelling profound is its mystery. We can theorize as to why a certain book affects us the way it does but with so many independent variables (reader, writer, subject, aesthetic, etc.,) it can be hard to figure out exactly why (this is most obvious when re-reading a book later on in life). I think, deep down, every writer knows the answer to why is that there is no answer, not one worth knowing, anyway, it’s the pursuit of trying to know that is worthwhile. By my estimate, this is what we lose when writing with an LLM (at least so far). Writing is not a game and it has no 9-dan masters, the making of art is the writing of a new program every time. (And, for writers—just like Lee Sedol—they don’t need half of a billion dollars to do it.)
More compute is the future. Where AlphaGo aimed to defeat the human, an LLM has so far been designed to pander. But how will we know what it looks like if and when we come across literature whose boundary has been pushed into new territory by AI? Go has victory. Literature does not. Even our best writers “lose” most of the time in their attempts to render the unrenderable.
I don’t want to have a closed mind to what I can’t even possibly imagine, but when art truly excites you, you experience a feeling of awe, an automatic acknowledgment of its creator. “How’d they do that?” or “I can’t believe someone wrote that.” Part of you wants to know how a person—a person just like you—could come up with something so moving. And, so far, it’s a feeling I don’t experience when I interact with AI-generated “art”; I have no skin in the game and the voice in my head says something more like “of course it can do that, of course AI could write that”—it’s got more compute than us, it’s programmed to do anything (and is it any surprise that the most valuable companies in the world, after years of collecting personal data with or without the public’s consent, are able to deliver a product that truly speaks to us?). I’m more impressed with the engineers or the person pulling the strings, prompting the AI to do its bidding—but even then, I’m no longer interested in the so-called art.
I asked ChatGPT how LLMs might help us push the boundaries of literature. One of its suggestions was to write “The Polyphonic Biography” wherein “A “biography” of a fictional figure told through 100+ different narrative voices—each generated and refined using LLMs” could be shaped as “literary fiction with a postmodern twist, appealing to fans of David Mitchell, Jenny Erpenbeck, or W.G. Sebald.”
I asked if it could pitch me a polyphonic biography about John von Neumann:
Title: The Mind at the Edge of Time
Pitch: A dazzling, kaleidoscopic portrait of John von Neumann, the twentieth century’s most brilliant and enigmatic mind, told through the conflicting voices of those who knew him—or thought they did. From fellow Manhattan Project scientists to exiled Hungarians, Cold War spymasters to chess hustlers, each narrator offers a different truth about the man who helped invent the computer, game theory, and the nuclear age. As their accounts diverge and intertwine, the reader must decide: was von Neumann a visionary prophet, a dangerous architect of the future, or something stranger still?
I asked it to write the first page:
The Mind at the Edge of Time by [Your Name]
Voice 1: A Colleague (Los Alamos, 1944)
They said Johnny could do a week’s worth of calculations in the time it took you to find a pencil. He’d pace, muttering, filling the air with numbers as if they were incense, and then—snap—some new theorem, some shortcut through the impossible. The rest of us were proud, jealous, terrified. I swear he could feel the future under his fingertips, like Braille.Voice 2: His Sister (Budapest, 1912)
My brother was not brilliant. He was monstrous. At six years old, he asked me if I remembered the exact shape of my dreams. When I said no, he looked at me with disappointment so profound I nearly wept. Our parents praised his cleverness, but I knew it was something else—something cold.Voice 3: An Enemy (Princeton, 1955)
He wanted machines to think like us. No, worse: he wanted us to think like machines. Don’t let them fool you. Von Neumann wasn’t saving the world with his computers—he was mechanizing it, grinding down the soft flesh of humanity until only logic remained. A dangerous man, dressed in affability.
I asked if its pitch resembled any novels that already existed and it told me about The MANIAC (and provided ideas for how to write “my” version so that it wasn’t so similar)5.
I’m willing to bet, if you made it this far, you zoned out while trying to read the ChatGPT excerpts. I know I did. I hesitated to even include them. We crave humanity.
I watched an interview with Labatut and in it he claims The MANIAC is 95% true, 5% fiction. This could mean, in a sense, he is somewhat of an LLM himself.
But I would guess that he’s showing some humility here, knowing the story isn’t exactly his (whose story ever is their own?) and minimizing the work it takes to string together a narrative. But it’s that crucial 5% that separates him from a program.
The problem will be when we can no longer tell what’s human-made and what’s not.
The silver-lining here, and I know it’s a reach, is that the smartest person in history, John von Neumann, spent his life trying to understand what previously couldn’t be understood for, what he thought was, the betterment of mankind. Realistically, the importance of his accomplishments were magnitudes greater than anything tangibly offered by literature. So, I can only hope that if, or when, the machines are finally able to think, they will spend their infinite time on better things and leave the literature to the lesser minds.
The MANIAC is the best book I’ve read this year and, despite its greatness, it caused me a feeling of disappointment when considering art in the wake of artificial intelligence. It wasn’t the novel itself, but what its content means for the future. I have questions and no answers.
In another interview, Labatut references that old quote about belief [being] the death of thought. I will continue to think about AI in spite of the big wigs who want from me nothing more than to cede my thinking altogether.
If you are an artist in the year 2025, it’s time to go maniac mode, before it’s too late.
This is Labatut’s English language debut.
Labatut cites the documentary in the back matter of the book.
Mostly stolen text.
10 million hours is the new 10 thousand hours.
I didn’t write any of this, by the way, simply asked a few questions. The LLM was quick to suggest I put my name on its content.
If I was gonna read Labatut, which book of his do you recommend? I enjoyed the article, especially the last part with your thoughts on LLMs, but the book itself sounds like it might not be for me.