AI Book Club recording of The MANIAC by Benjamin Labatut

by Tom Johnson on Aug 16, 2026 comments
categories: ai • ai-book-club • podcasts

Note: These shownotes are AI-generated.

• Audio-only version
• Links mentioned
• Topics covered in this podcast
• Narrative essay version of the conversation
• Transcript

Audio-only version

Links mentioned

• The MANIAC (Benjamin Labatut — the book discussed in this session; also on Audible)
• Notes and discussion questions for The MANIAC (the discussion doc used to prep this session)
• Luis Gurovich’s essay and notes on The MANIAC (a breakdown of the book’s core themes, contributed ahead of the discussion)
• AI Book Club: A Human in the Loop (the book club home page, with the full reading roadmap and past recordings)
• AI Book Club discussion of The Infinity Machine by Sebastian Mallaby (the previous session, also covering Demis Hassabis and DeepMind)
• I Am Not a Robot: My Year Using AI to Do (Almost) Everything (Joanna Stern — the next book, for September 20, 2026)
• Docs as Tests: A Strategy for Resilient Technical Documentation (Manny Silva — considered as an alternate pick; see also my earlier podcast with Manny)
• A Beautiful Mind (Sylvia Nasar — the biography of John Nash recommended during the discussion)
• Copenhagen (Michael Frayn — the play about Niels Bohr and Werner Heisenberg mentioned as another example of art interpreting science)
• AlphaGo versus Lee Sedol (background on the 2016 match and Move 78, which the last third of the book covers)
• NotebookLM (the research tool used to generate audio overviews of books)

Topics covered in this podcast

Here’s a list of topics we talked about.

• Fiction, imagined biography, and the work of interpretation — The club’s first work of fiction turned out to be genre-busting rather than cleanly fictional, and that ambiguity changed how it had to be read. Non-fiction can be judged on whether the argument holds; here the reader has to assemble the argument from anecdotes that are never explicitly connected.
• The opening chapter as the book’s thesis in miniature — The book begins with Paul Ehrenfest shooting his son Vasily and then himself, not in a rage but as a calm deduction about reducing a future burden on his family. Starting there frames everything that follows: the horror is not that reason fails, but that it works.
• Logic carried to its irrational ends — The same pattern recurs across the book. Game theory applied to nuclear strategy yields mutually assured destruction, a world balanced on a hair trigger, arrived at by impeccable steps. Devotion to reason keeps producing conclusions that any ordinary person would recognize as madness.
• Why fiction gets closer than a biography could — A Wikipedia entry on von Neumann can list the inventions but can’t put you inside the mind of someone that strange. Fiction lets a reader hold two anecdotes close enough together to spark a connection, which is why a dry essay on the same material wouldn’t land the same way.
• The death of play — The final third profiles the AlphaGo match against Lee Sedol, a man who spent his life on a game refined over three thousand years and saw it as art made jointly by two people. The devastating detail isn’t the loss; it’s that afterward he stopped wanting to play at all.
• Move 78 and the case for unpredictability — In the fourth game, Lee Sedol played a move so improbable that the machine’s evaluation collapsed, and the room cheered even though the tournament was already gone. The moment reads as proof that human creativity can still surprise a system that has otherwise surpassed us.
• Why we still want the human on the other end — Part of the pleasure of reading is reaching for the author: what they think, what they want, how they grew up. None of that applies to a machine-written book. Playing tug of war against a brick wall isn’t a contest, and it isn’t art either.
• What happens to a game after machines win it — Competitive Scrabble offers a preview. Tournament funding thinned once software could brute-force optimal plays, yet a devoted core still follows the game, and a player like Nigel Richards is still fascinating precisely because he’s at the ceiling of what a human can do.
• The garage door, and the alienation of the geniuses — Klara von Neumann wanted help with a broken garage door while her husband worked on what he considered the most important problem in the world; the confrontation ended in a breakdown and a miscarriage. Figuratively, we’re all standing at that door while the labs pursue abstractions nobody asked for.
• “It would have to play, like a child” — Asked near the end of his life what it would take for a machine to think like a human, von Neumann answered that it would have to grow, understand language, and play like a child. The irony is sharp: we’re teaching machines to play, at a million iterations to our dozen, while playing less ourselves.

Narrative essay version of the conversation

If the podcast were an article, this is what it would read like.

The Madness at the End of Reason

A novel about the men who built the computer and the bomb argues that logic, followed far enough, arrives somewhere monstrous. The uncomfortable part is how ordinary each step looks along the way.

In 1933, a physicist named Paul Ehrenfest shot his young son, who had Down syndrome, before turning the gun on himself. What makes the scene unbearable is not violence but arithmetic. Ehrenfest could not reconcile himself to the irrationality of the new quantum physics, he was certain his own death was coming, and he calculated that leaving a disabled child behind would grind his family into the ground. Killing the boy was, on those premises, the only practical possibility remaining. A book that opens this way is not warming you up. It is stating its thesis in the first paragraph and daring you to follow the reasoning to see where else it goes.

It goes, as it turns out, everywhere. Game theory, developed by minds of unprecedented power, was applied to nuclear strategy and produced mutually assured destruction: a permanent world-ending posture that every rational actor is obliged to maintain and none can safely dismantle. The pattern running underneath the book is not that brilliant people go insane. It is that a particular kind of reasoning, pursued with enough discipline and enough insulation from anyone who might object, reliably generates conclusions an ordinary person would recognize immediately as lunacy. And the people producing them are never the ones who notice.

The last third moves to the 2016 Go match between AlphaGo and Lee Sedol, which is where the argument stops being historical. Go had been refined over roughly three thousand years, and its masters understood it as something two people make together, a piece of art produced by an exchange of ideas. What the book captures is not the defeat but what came after it. Lee Sedol lost interest. The desire to sit down and play was simply gone, extinguished by an opponent that could not be beaten and could not be joined. Call it the death of play. It lands hardest on anyone who has noticed their own writing slow to a crawl, sentence by tedious sentence, in the presence of a machine that could produce a competent version in seconds.

Then comes the detail that closes the circuit back to the opening scene. To make the system stronger, the team deleted the human games it had learned from, cutting its only direct tie to the tradition it had inherited. Stripped of that anchor it did not weaken; it became untouchable. Removing the human element was not an unfortunate side effect of the engineering. It was the engineering, and it worked, which is exactly what Ehrenfest concluded in his own domestic calculation.

A smaller scene may be the book’s sharpest. Klara von Neumann wanted her husband to come out of his study and help open a broken garage door so she could drive somewhere. He would not, because he was close to solving what he considered the most important problem in the world. She eventually tricked him out, locked herself in, threatened to burn his papers, and collapsed into a breakdown that ended in a miscarriage. Most of us are standing at that garage door. Surveys keep finding that ordinary people distrust this technology, worry about the water and the electricity and their jobs, and mostly want the tedious parts of their lives handled, while the labs are elsewhere, working on the most important problem in the world.

The case against despair is real, though, and it has history behind it. Photography was supposed to end painting, and instead painting abandoned realism and became more interesting, because what people wanted from a canvas turned out to be evidence of a human hand rather than accuracy. Machine-generated audio overviews were supposed to end podcasts, and podcasts are thriving, because listeners want the banter and the mannerisms and the imperfections. Nobody at this table wanted to read a novel written by a model, and the reason is not nostalgia: reading is partly an attempt to reach the person on the other side and work out what they think and why, and there is nobody there to reach.

Von Neumann, dying at Walter Reed with radiation eating away the most famous brain of his century, was asked what it would take for a machine to think and behave like a human being. He took a long time, and then whispered that it would have to grow rather than be built, that it would have to understand language, and that it would have to play, like a child. We are now building exactly that: systems whose entire method is play, iterating a million times where a person manages a dozen, unconstrained by the pattern of what humans have already tried. Whether that produces something wonderful or something we cannot follow is genuinely open. The harder question is what kind of people we will be by the time we find out, and whether we will have kept any of the playing for ourselves.

Transcript

Tom: Hi, my name is Tom Johnson. This is a recording from the AI Book Club, A Human in the Loop, recorded August 16, 2026. About a half a dozen of us are discussing Benjamin Labatut’s The MANIAC. This discussion lasts about an hour, we cover a lot of topics from the book, and it’s a great lively discussion. If you would like to get involved in future book clubs, go to my site idratherbewriting.com/ai-book-club or just click AI Book Club in the top navigation. The next book we’re reading is I Am Not a Robot: My Year Using AI to Do (Almost) Everything by Joanna Stern, scheduled for September 20th. So you can see a list of all the upcoming books and all the previous sessions as well. There’s over a year’s worth of recordings on all kinds of fabulous books with notes and other kinds of things. In addition to the recordings, some have blog posts and so on that review the book. Jump in, participate in our Slack group, add your suggestions. We’d love to have you. All right, here’s the recording.

Tom: Welcome to this book club. We’re discussing The MANIAC by Benjamin Labatut. Man, I was gonna look up how to pronounce his last name before this book club and I didn’t, but I think it is a hard t. Has anybody got the pronunciation correct on his last name? Okay, I initially thought it was gonna be like Home Depot where you don’t say the t, but then I think I read that it is a hard t. But anyway, let’s — did you like this book? This is a work of fiction, or imagined biography. It’s kind of a genre-busting sort of book. It doesn’t really fit neatly into fiction or non-fiction. It’s got elements that are heavily biographical, or not biographical but that relate to the actual lives of the people. It’s not just like pure fiction, but of course it is imagined. But did you like having a work of fiction, or do you like more traditional non-fiction? Just curious. Lois.

Lois: I actually find it difficult, and this is a problem for me with movies as well. Like when I know the real life events that occurred, reading it in a fictionalized way is often very jarring for me and I get upset that it’s not matching the reality. So to be honest, I did not finish this book. But I still thought it would be an interesting discussion, that’s why I’m here.

Tom: For me, it made me realize that fiction requires more from me to interpret it. With non-fiction you can say, hey, the author didn’t make clear what their argument was, or they didn’t support it well. But here you have to figure out what is the author trying to say. Because he’s compiling anecdotes from so many different scientists, and portrayals of their lives from different angles, that it makes it unclear what the larger goal is sometimes. So it really forces you to interpret, which is not the mode that I’m usually in when I’m reading non-fiction books.

Do you like that mode? More of a puzzle to figure out, like what’s this author trying to say, why is he putting this story with the next story, and why is he having this polyglot perspective on this character, and why do we suddenly move into AlphaGo? It’s never spelled out in a straightforward hey, this is what I’m doing and this is why.

Nathan: Yeah, I found it really engrossing but very confusing also at times, to understand the reasoning behind each story and what was described in each chapter. I was especially confused at first about the first chapter where Ehrenfest killed his son because he couldn’t comprehend the new science that was coming out. And I went into Claude and asked it if it could give me an analysis of why it was there. And it was basically describing the difference between somebody that could accept the changes in our future with the way science was changing, and somebody that couldn’t and actually killed his son because of it. So that choice of having that story in the first chapter, I think, really presented how you could view the rest of the book.

Tom: For sure. The book begins with such an explosive first story that it has to be some kind of larger intentional message, a microcosm, kind of like hey, this is where we’re going. So you interpreted that as a scientist who couldn’t reconcile his worldview with this irrational world of quantum mechanics, or whatever it was that was driving him mad. Any other interpretations on that? That opening anecdote where Ehrenfest kills his son Vasily, who’s crippled and has Down syndrome, and the scientist calculates in his mind that once I kill myself, I’m going to leave this crippled child behind, it’s going to be this incredible burden on the family, it’s going to drive them into the ground, and the only logical action I need to take is to also kill him. Any other interpretations on that?

Sharon: I agree, I think it probably has many dimensions. But there’s another sort of interpretation, in that maybe the author is saying this is what happens when you take logic to its ultimate ends: it ends in this irrational action. And that is a theme that echoes throughout the book, where these scientists who are so devoted to reason and logic and understanding, they’re ending with these crazy ideas.

For example, game theory has it that when you apply it to the atomic bomb and nuclear strategy, the idea is that everybody has to have enough nuclear arms to blow the other person to smithereens, and you have to have all these bombs readily aimed at your opponent. And once you have that, so that the world is basically on a hair-trigger about possibly ending, then everybody takes the course of not acting, because reducing mutually assured destruction is worse if they do act. It’s kind of like, gosh, is that really — that seems kind of a madness, a mad conclusion.

In the same way that this scientist thinks, hey, I’ve got this son with Down syndrome, obviously I should just murder him instead of making him a burden to somebody else.

Tom: Yeah, I feel like the author is saying that the pursuit of all this logic, the development of AI, answering these big questions, is pushing us off a cliff. And really, ultimately we are figuratively that Vasily character who will be removed, because as AI becomes more and more powerful, you just need to remove more of the human elements.

The last section with AlphaGo, the training of AlphaGo to beat Lee Sedol, the champion — there was a section where Demis Hassabis, in one of the training rounds for AlphaGo, removes all human elements and input in order for AlphaGo to learn entirely on its own and kind of surpass any influence of weaker human thought. It kind of echoes this idea of killing Vasily, where like, hey, you’ve got to remove the human element from the equation.

Or not, I don’t know. (laughs) Like I said, because it’s fiction, you have to interpret and try to guess what exactly is the purpose of his anecdote. Lois gave a trigger warning about that initial opening, where you have a scientist just blowing out the brains of his son, as being really potentially off-putting. So it can be interpreted in many different ways with many different reactions. Definitely an interesting way to start a book. Molly?

Molly: Tom, I basically agree with you, and I think as fiction it really helps us connect these dots ourselves. And maybe the answers aren’t perfectly clear and outlined in a way that makes the author’s argument interpretable to us, but to me that’s the beauty of poetry and of fiction and of art, where as humans in this equation, we’re not just downloading facts about John von Neumann. We’re holding these points close enough together to make an electrical connection on our own.

And so to start the book with that story about Paul Ehrenfest — and I mean really the first paragraph is him killing his son and himself, which is just an amazing way to start a story — that gets us involved. Instead of like, you know, in 1928 this person was born or whatever. And I think as a storytelling device it’s a really powerful way to start things.

And just making these connections between the three pieces which, Luis, that essay of yours mentioned that this is a triptych. I hadn’t thought of that when I was reading the book. I really thought of the book as in two parts at first, but then when I went back I was like, oh, it is in three parts.

But thinking of the three parts of the book together and really trying to put on my human interpretation hat, it’s really saying something really scary and profound, I think, to your points, Tom, about this pursuit of rational ends, right, with the atomic bomb and then with AI.

I don’t think non-fiction could really make this argument or inspire these kinds of interpretations and thoughts, at least for me. A dry essay could help me start thinking about these things, but to really see into the madness of these characters, I don’t think we could get that from a non-fiction perspective. And as far as von Neumann goes, he’s such a complex, weird, genius person that I don’t think a Wikipedia page could really help me understand that in the way that this book did. So fiction wins, I think, for this story.

Tom: That seems like a very good analysis and an argument about fiction as the right mode. Because you’re right, I’m trying to imagine this book as non-fiction and what would that read like. I mean, it would be reading a biography of John von Neumann. I didn’t even know about this person before this book. I can’t believe I’d never heard of him. He’s apparently the founder of so many different important things. But yeah, does fiction allow us to get into the psychology of madness in a way that non-fiction wouldn’t? I mean, probably.

Lois: I would recommend reading A Beautiful Mind, about John Nash. I mean, it sort of goes along with this, and there could very well be some — I definitely see the argument for fiction, but that’s an interesting memoir that is non-fiction, as much as such things can be.

Tom: I have seen the movie, and yeah, that’s kind of a trip into the mind of somebody who’s got all kinds of things going on there. And it’s linked to von Neumann as well. Anybody else, do you think that fiction is the way that this author makes his argument much stronger about the irrationality of the current direction of these obsessed scientists?

Sharon: Well, for example, a few years ago I saw the play Copenhagen, which was about a meeting between Niels Bohr and Heisenberg. It was speculating as to what might have happened during this meeting that they had during World War II, and did this stop the German atomic bomb, for example. And yes, I definitely found that very helpful, to bring the arts in as a way to interpret science and scientific events.

Tom: Yeah. Thanks for the recommendations, you’ve mentioned a couple of things now. I was also thinking of the movie, and probably the book, Oppenheimer that came out, where it really got into a lot of these same themes, like why are people building hydrogen bombs? It’s insane.

But also think about this last third of the book that profiles the match between Lee Sedol and AlphaGo. If you’ve been following other books, this has appeared. But we’ve never had such a forceful, in-depth profile of potentially what’s going on in Lee Sedol’s mind and Demis Hassabis’s drive to win. I mean, that’s pretty intense. There’s no way you’re pulling that off in standard non-fiction modes. And that to me was actually the most powerful part of the book.

I know that a lot of critics didn’t like that, they felt that the last third of the book took it in a different direction or didn’t cohere. It definitely felt like a third story, the triptych.

But for me that last third was all about the death of play. You have this person who’s a Go master who has learned to love Go. He sees it as making art between two people when they’re exchanging ideas and strategies and building something mutual, and he’s like, this is a game that’s been honed over 3,000 years. This is a game with deep history, and you’ve got a person who has spent his entire life learning it, loving it, and now he’s coming to grips with the fact that he can’t beat this machine.

This machine is indestructible, it’s just ruthlessly beating him. He wins one game, but due to a fluke in the AI, it goes haywire for some reason, almost like a glitch.

But it’s after this game that Lee Sedol just loses interest in Go. He just doesn’t want to play, he doesn’t have the drive, he’s lost that sort of love and delight in the game.

And for me this jumped out because I wonder about the same thing. Am I losing this sense of delight in play in writing, for example? I’ve noticed I’ve written less on my blog in the past couple of years since AI has come out. Every time I sit down to write, it feels so painfully slow and tedious, having to type out sentence after sentence and then to revise it. It takes forever, and I know it’s going to be way worse than a machine that can quickly interpret massive amounts of text and understand the right analysis in a deeper way. What do you think, do you feel like this sense of play is at stake, and that this is what this last part of the book is about, the death of play?

Nathan: Thinking about that story with Lee, I think the one saving grace was that fourth game that he played, where he put down a play that was so obscure that it basically destroyed DeepMind’s ability to understand how to play it, knocked it off its pedestal. And he was cheered so much at the end of that, even though he had already lost the tournament. People were looking for that sense of individuality that humans can possess, and the genius behind creativity. So I think that’s what we’re always trying to reach for, finding that spark that we feel is inside all of us. It’s hard to find our place in AI now, but I think that’s what we’re trying to reach for.

Tom: Thanks for bringing that up. That is a great point, and definitely the author makes a big deal about move 78, is what I think it’s called, this God-like move, divine move, something. Where yeah, it’s like a 1 in 10,000 chance that a human would actually make that move, and somehow that makes AlphaGo go haywire and it starts flipping out. So you’re saying that that is a redeeming view about humanity’s fighting chance, or not even fighting chance. Is that our move, like figuratively, should we all be making move 78? I don’t even know if I understand what that would mean. Like for me to do something so unpredictable and out of the box that it’s just something computers couldn’t predict, I don’t know.

Nathan: I’m kind of just thinking about what we’re saying, and in the AlphaGo situation, it’s this kind of adversarial relationship that a human could have with an AI, right? And I mean, off the top of my head I can’t think of anything we have for that outside of a kind of fictional relationship, outside of games.

Like is there a situation in the rest of the world where we have this adversarial relationship with AI? I mean, maybe in a marketplace or something like that. But I guess as far as writing goes, and I say this as a writer, and I love writing and books and that kind of stuff, there is use beyond even the communicative act, that we can write just for pleasure, you know?

And I think we wouldn’t play tug of war with a brick wall, and we wouldn’t necessarily love to see a movie created from the mind of AI, right? There’s something very important about a human being involved in that, in tug of war and in art and communication and those kinds of things.

And I think for me, and maybe for others on the call, when we’re in work mode, AI makes perfect sense there. We’re kind of digging a hole for someone else, figuratively, we’re trying to put words out there on behalf of these organizations for whom we work.

But in maybe a more private or individual context, we can use ourselves to dig that hole, or play tug of war, that sort of thing. So I’m just wondering if the AlphaGo situation, or chess or whatever, is very impressive, but they’re just games that are useful for demonstrating the prowess of AI instead of having any real social utility, I would say.

Tom: Yeah. Agree, agree.

Sharon: I think that’s an excellent point. Just sitting here thinking about, do I want to read a book written by AI? I think part of the pleasure in reading a book is trying to connect with the author. What are they thinking, what are their motivations, what are they trying to tell me, how did they grow up, all sorts of things. You’re having a relationship with the author, and there’s nothing — if I’m reading a book by an AI I’m not thinking any of those things. Yeah, I think it’s well said, very very good point.

Tom: I know that when I’m writing a blog post outside of work, just doing something fun. For example, before this I was trying to write my review of this book, something I’m trying to get myself to do, to review a book in order to wrestle with the themes in some meaningful way.

And I’m thinking, well, what do I bring that’s unique to this? What is it that an AI-written thing couldn’t do? And it is as you say, Sharon, and Nathan, you bring your humanity to it. How does the book impact you, how does it affect your experiences, your viewpoints, emotionally how do you connect with it. That’s something that an AI is not going to be able to do.

And you’re absolutely right, I’ve been overplaying the prowess of AI in things, but I’m mostly thinking of computer coding and documentation, not necessarily more creative endeavors. I wouldn’t want to read an AI-written book either. There are many, many, many of these now, and for some reason AI hasn’t quite cracked the code in figuring out how to win at that game.

So yeah, AI is probably a lot more dominant in things like computer games where you have a win, or coding where you can compile something, or maybe even math perhaps, where you’ve got an equation to solve. But for sure, writing and the arts are outside this realm of a very predictable sense of winning.

Let’s see, anybody else have any other thoughts on that matchup between Lee Sedol and AlphaGo? God, I keep hearing about this in every book we read, I feel like it’s becoming a bit of a tiresome thing to be like, oh, this again. But Lois?

Lois: So I can give a little bit of perspective from a different game. I’m sort of a defunct competitive Scrabble player, and I follow developments, and what has happened is that our tournament funding has really dried up to a large extent, and I wonder if that’s due to AI. I mean, even before LLMs, just because of the ability of computer programs to brute force this type of thing. We already had humans — like the top tournament players basically have the dictionaries memorized, they don’t even necessarily know English, a lot of them don’t really, and they’re also able to do basically combinatoric mathematics in their mind to figure out the best plays. But of course a computer is going to do better than that. But still, there still is a core group of people interested in these tournaments and they still go on.

Molly: Lois, your example makes me wonder, when AI can reliably beat a human in a certain game, at a certain point won’t people just stop caring about it and get back interested in humans being able to be really good at the game? I’m just wondering if with games in general, it doesn’t seem to be the case, or at least there’s a circle of devotees who will continue to analyze every endgame and look at every possible play that could have been made at that point and why it was or wasn’t the right point. So it may definitely reduce the number of people who are interested.

Tom: Yeah.

Molly: Yeah, just like with Go. I think we know that a computer can beat a human in Go, so can we just go back to enjoying the game?

Lois: I mean, there’s this one person named Nigel Richards from New Zealand, who’s like the best human in the world at Scrabble, and he plays in multiple languages that he doesn’t know. He’s basically at the very very top of what a human can do, and so his plays are still of interest to people, despite the fact that he could probably be beaten by a top AI.

Tom: It’s interesting to think about games and where people get the motivation. I mean, I play basketball, I watch football, basketball, whatever sport you like, and it’s not a scene where there’s an AI that’s beating everybody. But even so, I love to play basketball but I know that the equivalent of an AI would be like NBA players who nobody will beat. And yet I will still go out and play basketball knowing that I could never beat even a college player. So, why? I mean, there’s just pure enjoyment in some things, it’s just fun to do. And that’s why I do it, even if I’m just playing in a small pond, as one of the authors said.

Hey, there’s another anecdote in the book that jumped out at me the most. This is in the von Neumann section, I think either the first or second, with all the different perspectives. And by the way, I listened to the Audible version and the voices were amazing. The narrator — I mean, I don’t really know if certain voices are accurate, but it sure sounded good.

And so the story that’s jumping out at me the most is Klara, von Neumann’s wife. She wants him to open the broken garage door so that she can leave, drive somewhere, I don’t know why she’s trying to get out and go somewhere. And von Neumann is sequestered away in his study, working on what he says might be the most important problem in the world, that he’s close to solving or something.

And she’s just like, man, you can’t take five minutes to open the garage door for me. She finally somehow tricks him to come out and then she locks herself in the study, she threatens to burn some papers, I can’t remember if she actually does. And von Neumann is banging on the door, and Klara’s inside. She transitions into this hysteria where she’s really in a breakdown, and also ends up having a miscarriage out of this, I think.

I really like this. I think personally it connected a little bit with me. Sometimes I’m sitting at my computer and my wife wants my help with something and I’m just like, you know, later. And I’m not working on the most important problems in the world, for sure.

But I do think the larger message is that these scientists as a whole, and these technologists, feel as if they’re working on the most important problems in the world, and figuratively, we’re kind of that Klara who’s like, hey, I just want my garage door fixed, man, I just want to get out and go drive around. I don’t need you solving whatever you’re trying to solve.

There’s definitely a disconnect between AI research and science and what people actually want. You see it in the polls where so many people have a negative view about AI. It’s not matching the promises, it’s not bringing abundance, it’s threatening not only jobs and livelihood but the whole economy is balanced on a few stocks.

So yeah, do you think that there’s more to this anecdote, about it being a larger metaphor for this disconnect between, hey, you guys in your research labs, you’re not actually solving problems that people want?

Nathan: I totally agree with you. Continuing on with that story, closer to the end of his life von Neumann was wanting to understand how he was going to be contributing to the rest of the world, and what he was leaving to the world. And he was concerned because he didn’t have any children to continue his biological essence. And that story of his wife having a miscarriage because he was locked in his room is very telling, because he could have had a child to pass on his genes if he wasn’t focusing on that one problem, and if he wasn’t being so selfish. He could have been able to do what he wanted at the end of his life, which was passing on something to the world.

Tom: Yeah, that’s a great point there. He did have a child, right? Marina? She ended up — is that right?

Nathan: Oh, that’s true, he did have a daughter. Oh.

Lois: But with Klara — it’s confusing, I had to make a little family tree kind of thing. He had a daughter, I think they were somewhat estranged, she ended up working for General Motors and was, among other things, someone who had a professional career. But this miscarriage was with his second wife.

But I think your point still stands, in some way, because I think with all of these people, von Neumann and even Kasparov, there’s this kind of alienation in all of their work. Like they’re so invested in the work but yet still disconnected from something.

And that’s like, Tom, you bringing in the AI research labs, I think we see a lot of this. There’s kind of this hubris, right, where they believe they’re changing the world and revolutionizing the world, but they’re still so alienated from it in so many ways, where it’s like, well wait a minute, what about the things people actually need? People don’t want to play tug of war with a brick wall, you know what I mean? We need ways to be more human, basically.

Tom: Go ahead, Molly.

Molly: I was just gonna add to that, because of the idea of them being kind of alienated from the rest of us. The chapter that was Oskar Morgenstern, who I think was the person who worked with von Neumann on game theory, the chapter ends with that character saying the only person he’d ever met who was exactly like that, thinking in the way of game theory, was von Neumann, and normal people aren’t like that at all. And it really supports that, that it’s designed by people who are disconnected from normal people. Like, life is so much more than a game. I’m just reading the part here: it’s full of wealth and complexity, it cannot be captured by equations, no matter how beautifully or perfectly balanced. Yeah, I thought that part was really powerful.

Tom: I think there’s another part that I’m trying to remember, near the end. Since we’re talking about von Neumann and his legacy, there’s this passage at the end, I’ve got it pulled up. “Before he became unresponsive and refused to speak to his family and friends, he was asked what it would take for a computer or some other entity to begin to think and behave like a human being. It took a long time before answering, and then he said in a whisper, it would have to grow, not be built, it would have to understand language, to read, to write, to speak,” and he said that “it would have to play like a child.”

This part of the book is almost the saddest part, in my opinion, because it’s essentially saying that in order for these machines to fully achieve their AGI-like state, they have to learn to play and be like a child, which is everything that the machine is taking away from humans. This delight in play and learning and being like a child, experimenting.

And yeah, I think that would be the ultimate tragedy, and this connects to the alienation aspect. Why do people dislike AI? It’s taking away all these things that we love, this delight in play and learning and experimentation and going about things like a child. Did that passage jump out at anybody? Was that one that you thought was particularly interesting in any way?

Nathan: I thought it was very — yeah, go ahead, please. Really quick, I thought that that section probably was what Hassabis was really excited about, because he was trying to understand how a thinking computer could be created, and he was really influenced by von Neumann’s work. So I think that’s a really telling section.

Tom: Yeah, thank you for bringing up Hassabis. You’re right, he totally focuses his strategy on having AI learn by itself, learn on its own from experimenting and so on rather than programming. In fact, I can’t remember which book we read, but at some point there was a big shift in AI strategies. Instead of trying to program the AI with all the knowledge it needs to operate, instead people said, well, we’re gonna let it learn on its own from a bunch of data and kind of trial and error, experimentation, whatever goals you give it. And it turns out that that self-learning was way more powerful than trying to give it all the knowledge it needs. So yeah, that totally connects with Hassabis.

And yet, Google’s frontier models still struggle. (laughs) Sorry. Yeah, it’s like everybody’s like, where is Gemini 3.5 Pro? Where’s 4.0 Pro? We’ll see, I don’t know.

So I probably shouldn’t bring that in, but yeah, it’s very difficult to figure out how do you get to that next level on the frontier AI jump and up-leveling leaps. What is it? Previously Demis had great insight about, well, we’re gonna model human psychology — or actually I don’t know, this is Demis, this is Geoffrey Hinton, we’re gonna model AI after the way the human brain works.

And then Hassabis gets a PhD in neuroscience because he feels like, hey, if I understand neuroscience, I’m gonna be able to understand how AI can learn, right? And then he has this multimodal learning and other kinds of strategies. But yeah, people are trying to figure out how do you get AI to be smarter, and now we come back to von Neumann: you let it play. Really powerful.

At the same time humans are playing less, perhaps. Or more, I don’t know. I’m sort of mixed about the whole play thing, because on the one hand I’ve done a ton of stuff recently just on my own personal site that I never had time to do, and which I wouldn’t consider play, but a bunch of drudgery. I upgraded to the latest version of things, I updated analytics, I reiterated the numbering schemes on past podcasts. None of this is really play, and I was happy for AI to do it all.

But at the same time, there are other aspects where now that I can just direct AI to do stuff, I do probably play less in other ways.

We lost Luis before we could get his insight. I think he’s a professor, I’m not sure, I need to look back. Maybe we’ll get him back. What else in this book do you want to talk about?

Sharon: This is sort of peripheral, but getting back to this issue of play and playing like a child. Currently I don’t have very many, or any, small children in my life, but from what I’m hearing it’s just absolutely ridiculous with kids and screens, giving a one-year-old screens. Does anyone have any knowledge of what’s going on with this and how people are counteracting it?

Lois: I know my brother has a — she’s almost two, and screens are banned, they’re gonna ban screens for as long as they possibly can. So I’m not sure how long that’s gonna last. I always give them my two cents, you know, flip phone until she’s 16 and things like that. I think it really depends on the family. Like I see some families where it’s really prevalent and then others have no real rules about it. I think it really depends, from my experience.

Tom: I’ve got four daughters between the ages of 15 and 25, and yeah, we’ve tried to ban screens in the past, it never wins out in the long run. But it’s been amazing to see my youngest daughter and other kids grow and develop in different ways.

She has this best friend that she became really close with this past year, and this other friend never learned to ride a bike, never learned to swim, sort of a deprived child. And so my daughter was teaching her how to ride a bike, and it was so amazing to just watch them. They went to a track and she was riding around for like the first time and it was kind of amazing.

So this sense of play in these kids is fun to watch. They still watch a lot of TV and they’re on screens and they’re doing stuff, but they’re also playing a ton. They’re playing in the kitchen, cooking, making cookies, they’re literally playing video games together on Roblox, and then they’re riding their bikes and they’re paddleboarding. I don’t know, I’m jealous in many ways of this childlike play, and I look for my own hobbies and so on. But yeah, I don’t know. Kids, if you can ban screens and ban sugar, more power to you.

Nathan: Something about the von Neumann quote about the machine learning to play, I thought it was really scary, and it is about all these things that you’re talking about. There was a leaked phone call within the last couple months at Meta where the executives are saying, oh well, we put keylogging software on all of your computers, of course, because you’re the smartest engineers in the software industry and we want the smartest people to train the AI, right? So as the Meta engineers are on their computers, the AI can see what they’re doing and then use that for its own training.

And I think it could happen where we see some AI company putting sensors on babies to understand how babies are playing and interacting with the world. That just freaked you out.

But yeah, I think that is within the logic and within the rationality of training all of this, right? Because from the perspective of any AI company, it’s like, yes, we want AI to understand what it’s like to be a human, and really what better way to do that than using a baby for training, you know what I mean?

And how does a baby — so maybe the AI will start putting things in its mouth, figuratively, and learning the world that way. I don’t know, but there’s all of this, just like with the atomic bomb, that is sort of within this kind of warped, quote unquote, rationality that we have in our culture. And I wouldn’t really put it past any company to start using babies for training their models.

Sharon: Is that a real thing, like the baby sensors? I’ve never — you just made that up? You just made it up. But we might see it, you know? It makes perfect sense. Who’s gonna volunteer their babies for that? I want to know. I mean, they’re probably going to pay them handsomely.

Lois: It’s like the Skinner box, if you’ve read about that experiment, the Skinner box. So just expand beyond that.

Tom: What is that? I haven’t heard of that. Is that the behavioral experiment?

Lois: Yes. Yeah. That happened in like the 50s, right? Something like that.

Molly: Yeah.

Sharon: Yeah. Anyway, I don’t know. That’s dark.

Tom: I do think, yeah, Nathan, you’re really onto something about how if we start teaching machines how to play, that might be game over for humans, because they could play at such a faster iteration. They could play something a million times, and we play something like a dozen times. So they’re going to learn everything and accelerate so much faster. When they can break out of any kind of predictable algorithm and just experiment, it’s going to be wild. I mean, first it’ll be wild because they may go in so many different directions that you’re like, what are they doing, or this is going haywire, and maybe they’re going haywire for a while. It’s weird, like sometimes I’ll be using an AI model and it does go haywire underneath, it’s fixating on some word or something.

But that kind of glitch might be the same glitch that evolutionarily is what helps us evolve. It’s like you have a thousand iterations and that one iteration is that thing that gives you that advantage that makes you biologically advantageous and so on. You accelerate that at a much faster loop, and gosh, it’s kind of scary to think about that. All right, any other ideas on this book? This has been a good discussion. I wasn’t really sure how this was gonna go, because fiction — Nathan, you mentioned at the beginning the whole interpretation thing, and it was bringing me back to my English major undergrad days 30 years ago. But this has been a good discussion. Geoffrey?

Geoffrey: Oh, so I have an art background, I went to school for design, and one of the stories that connects with this understanding of how we are going to evolve with AI: back when photography was first invented and people were able to see actual pictures of whatever they take pictures of, maybe most of the time it was still lifes because they needed to stay still for a long time. That’s when art paintings changed from being realism to being more abstract and conceptual paintings. When people realized, hey, this picture is so perfect, we don’t need people to make still lifes anymore, they didn’t have the interest in still life paintings. So people started doing abstract paintings to say, oh, this is true, a person created this. We’re looking for the imperfections that only humans can do now, because photography is showing exactly what’s in front of us.

Tom: That’s a really good point. Yeah, I think didn’t a lot of people think that photography was gonna kill art at some point? It’s like, you can take a picture, why are people going to paint anymore? Yeah, it’s really good to consider the ways that we change direction based on other influences there.

Geoffrey: Yeah, like I know that Google has a tool, NotebookLM, that can create a perfect podcast, where if you just tell it what you want to learn about, it’ll just recite it back as if it’s a podcast to you. And at first people were concerned that podcasts would disappear, but no, they’re thriving right now because people like the uniqueness and imperfections that each host has, and their own mannerisms and stuff like that. So it wasn’t as impactful as people thought it was going to be.

Tom: Yeah. I actually listen to those podcasts all the time, by the way. I probably make half a dozen of those NotebookLM ones. It’s now, I think, just like Google LM or Gemini, what is it? They changed the name and I should know it. Gemini Notebooks. But yeah, I have hundreds of these. I usually create a podcast when I want to learn something specific, and it works quite well for that. But yeah, I want to hear actual humans and their banter and their points of view and their unique experiences, for sure. I listen to mostly human podcasts but also the AI stuff. They’re actually really good, those AI ones. I don’t know, Gemini Notebook is amazing.

If you ever wanna quickly get up to speed on a book, search for the PDF of the book — Ocean of PDF has, I don’t know, probably all the pirated books in the world — find the PDF, upload it to Gemini Notebook, and suddenly you have a perfect resource to just ask the questions. It can consume the text seemingly in an instant, it can create podcasts and so on. It’s a great learning tool, definitely a unique tool. I don’t think any other AI company really has something similar. Hey, we’ve got five minutes left and I wanted to take a poll on something. So I’ve been split about the next book. Hold on, I’m trying to get this window up. There we go. Okay.

So as far as the next book, we have a couple of options, right? Hopefully you can see my screen. Yeah, there we go. So I had I Am Not a Robot up here, which is a book by Joanna Stern. I’ve heard her on podcasts a lot, she seems like a really good author. I think she writes for — I can’t remember who she writes for, Wall Street Journal maybe, or New York Times, one of these, I think it’s Wall Street Journal. And it looks kind of like a fun read, to be honest. It looks more lighthearted, or a light read. I’ve also started putting Audible links here just because I know a lot of people like Audible, including myself.

There’s also another book by Manny Silva called Docs as Tests, which I actually reviewed an earlier copy of, at least part of it, and it’s pretty on point in terms of relevance to AI. It’s quite impressive how current it is. I thought if you write a book about AI it’s going to be dated next month, but no. It’s good. The problem is, this book is very slanted to tech docs, costs 40 bucks, there’s no Audible version, and it’s also got a lot of stuff about doc testing. Which book would you rather read? I Am Not a Robot, Docs as Tests, or something else entirely? I’ve got some others listed here, but any thoughts?

Nathan: Could you explain a little bit more what Docs as Tests covers?

Tom: The main idea is that you should be testing your docs, because now if the wrong information gets into AI, it’s amplified so much worse, a thousand times, because now people are incorporating incorrect information directly from their AI interfaces. So it’s kind of a better argument that we should be testing all these docs. He says docs are really just lists of assertions. If you look at a page of docs, it’s asserting things, and you gotta figure out which assertions you can test, you develop these tests for them, and then you run that. And by having more tested and accurate docs, your accuracy goes up in the AI tools. That’s my read in a nutshell.

But you know, I’ve also found that the most enjoyable books are books with big ideas, and ones that have a more mainstream audience. They kind of are better for discussion and just more interesting. I like books that scratch an intellectual itch, you know, give you something deep to think about, like this MANIAC book we read.

And I’m not sure that more industry trade-specific books fulfill that. That might be more of a read-this-at-work kind of book. So, anyway.

Nathan: I prefer, just for my own selfish interest, I prefer Joanna Stern. I really like her videos. She has a really good sense of humor about technology, and she brings technology into the main zeitgeist, so that it’s easier to understand it through her eyes.

Tom: Cool, yeah, it’s good to hear other feedback about Joanna Stern. I know she’s done a lot of research on this. Yeah, well, unless anybody else has any strong opinions, I’ll just leave it as is and leave I Am Not a Robot. Okay. There was a study group in Write the Docs Slack that was digging into Manny Silva’s book, and it’s definitely on a lot of podcasts. Maybe I’ll feature something on my blog about him. But he is a good thinker, he is a good guy. I think we interviewed him — yeah, we interviewed him for a different podcast.

If you have any other recommendations, these are all tentative. Lois, I really appreciate your book recommendations. Actually this Reboot one is from one of the links you posted. So if you have any ideas, please let me know, in the roadmap. And next month we’ve got I Am Not a Robot. Listen to it on Audible, read it, however you want. Thank you so much for this discussion, you are sharp thinkers and I really appreciate your insights. I’ll send out the recording. There’s a lot of people who listen to this even afterwards. Each video gets around 70 views and each MP3 gets several hundred downloads. So there’s more than just five people here, there’s a lot more. Thank you so much, have a great weekend.

Group: Thanks everybody, take care. Bye. Thank you. Bye bye.

Comment on LinkedIn