How, Exactly, Could A.I. Destroy Humanity?

21:34Head to BA.com or our sub stack and come and bake along with us.

21:38Tyler Foggatt:Join us the first Tuesday of every month as we debrief about our latest bake. Find BA Bake Club wherever you get your podcasts. Happy baking.

21:54As we discussed earlier, AI leaders have been saying for a really long time that AI could kill us all. But this hasn't necessarily stopped researchers from aggressively developing AI. The argument, at least as I understand it, is that we sort of need to root for a more safety-conscious company like Anthropic, or at the very least a U.S. AI company to develop a good AI before an opposing power develops a bad one. Like it's always kind of framed as this race between the U.S. and China. But it also seems like that calculus is starting to change. Can you tell us a little bit more about the letter that Dario Amadei, the head of Anthropic, posted over the weekend, and how it's a pretty notable departure from what we've seen from AI company heads in the past when it comes to talking about this stuff.

22:36Tyler Foggatt:Yeah, so the letter was titled, We Must Pace the Frontier. The frontier is AI speak for, you know, the state of the art in AI. That's the frontier. And pacing is an interesting word. It's not the same as pausing, and it's not the same as stopping. It has to do with controlling or titrating or attenuating the speed with which we make progress in capabilities. So what that basically means in plain English is stop making it so much smarter super fast and divert some of those resources to making it more controllable. Essentially, that's what this is proposing. And I don't know that it's actually in its substance something that is totally new.

23:22So like earlier this summer, there was an open letter called Pacing the Frontier, which I think around 1 ,400 researchers at the Frontier Labs signed.

23:33Tyler Foggatt:And it's actually an amazing document. It's an open letter, but it's made by people who know how to make websites. So you can scroll over the names of all the signatories and you can see what they have to say. And what you'll find is like a lot of people who work in AI want to do this. It's not like just a psyop created by CEOs who want to get investors to take their technology seriously. This is a thing that regular researchers who are in the trenches with the tech, they want to do. They want to slow down the pace of capabilities progress, and they want to redirect resources towards making it more controllable.

24:10Tyler Foggatt:And basically, the letter proposes two ways of determining the pace. This is my understanding of it. But the upper bound of the pace, like how fast they'll go, Amadei thinks it should be determined basically by external auditors who come. And they go, they visit the labs, and then they publish reports saying, we think this is a controllable model. And there's a lot of discussion about what that might mean. And then the lower bound, so how slow you can go, is basically international competition. It's basically China. China sets the lower bound because they're trying to catch up to the frontier models made by the Western companies.

24:47Tyler Foggatt:And we don't want, this is a core part of the argument, we don't want them to beat us. So, like, this is not a very democratic, this is, like, if what you were hoping was that we would do, like, a referendum of the people of the world, and the people of the world would say, like, what they want from AI, that's not this. Because if the people of the world were asked, I think a lot of them would say, stop, right? This is pacing. Like, he says, progress will still seem fast. So this is not a stop. It's not a pause. You know, there have been times when AI researchers have proposed a pause. And this is not that.

25:23Tyler Foggatt:It's just going a little slower and directing some resources towards safety. And his hope, he says in the letter, is that in a year or two on certain key safety things, it will be possible to make, I think he uses the word, profound progress. And those are, like, I hope that's true. Obviously, that raises a lot of questions that are totally reasonable. Like, if it only takes a year or two to make profound progress in AI safety, like, why don't we just invest more money in it, period? I mean, why haven't we made that progress so far? So there's a lot of questions raised by the letter, but it's a, you know, everyone agrees with it.

25:57Tyler Foggatt:I mean, it was preceded by an open letter in the industry in which a lot of people propose the same thing. I think what's like feels new about it is that the heads of the other firms are like also wanting to do this. And it's coming off of this hugging face incident in which independent auditors have done an investigation into what happened. And so a model for how this might look is sort of readily available, where I believe in order to investigate the hugging face hack, the auditors, I think they only had six days at OpenAI, and they had limited access to the technology. And in the regime that he's proposing, auditors would be there all the time.

26:33Tyler Foggatt:There'd be this like massive expansion in the scale of that effort, and they'd be given much bigger access. And the idea is that transparency would lead to exactly the types of conversations that we're having now, and it would create more of a kind of a discussion with teeth about AI, whereas before a lot of these safety things have seemed sort of fringy in the broader political picture. So, I mean, I think it's a really good letter, and, you know, I hope it happens. How possible is it to pace the frontier or slow down anything if part of what is driving this letter and these concerns is this idea of recursive self-improvement or AI systems helping to build the next generation of AI systems, which seems to be happening at a surprisingly fast pace?

27:20Like how much control even is there?

27:22Tyler Foggatt:Yeah, I mean, right now, this is another thing that's clearly added to the unease within the industry is, I think it was just yesterday, there was an open letter written by an open AI researcher, in which he describes, you know, how reliant researchers inside the labs have become on models themselves. And he says, like, he himself doesn't have enough time to really seriously engage with, not just with the code, but with the views that the AIs are sharing with him about the code that they're generating. So, like, the degree to which the researchers are depending on their own technology to build new technology is, I think, surprising, not just to us, but to them.

28:05Tyler Foggatt:You know, I think recursive self-improvement and superintelligence and that whole idea of the intelligence explosion, like that hasn't happened ever before with any technology ever. And it's never happened in a lab. There's never been a not as smart as people AI that's made a slightly more smart AI entirely on its own. I mean, like we've never had recursive self-improvement. It's never happened. It could happen. I'm no expert on that. No one is. an interesting example is what's happened in math where you might remember a few years ago we all made fun of chat gpt because like it couldn't tell us how many r's were in strawberry and it couldn't do elementary school math yeah and then basically over a period of time that was that is in retrospect incredibly short it became as good at math as any human mathematician and you know it's true like solving those math problems cost millions of dollars in computing costs and stuff, you know.

28:59Tyler Foggatt:But still, no one expected that amount of progress in the amount of time that it happened. And what it shows you is like, AIs don't have to be smart in every way in order to be super smart in certain ways that really matter to us. There's a term called jaggedness that comes up in AI world a lot. It's basically the idea that, you know, maybe you imagined unintelligent AI being kind of like Commander Data from Star Trek or something, Like just a well-rounded individual with a liberal arts education. But that's like not what's being built. What's being built is a thing that sort of isn't that in touch with reality in many respects.

29:36Tyler Foggatt:Obviously, it's not alive. It doesn't have a life. It doesn't really know anything about the context in which it's being operated. But it's super smart in certain ways. So it's like as good at hacking computers as anybody, even though it doesn't really understand like that much about whether it's working for the good guys. are the bad guys. And that jaggedness, like this kind of lack of context combined with like superpowers in certain dimensions means that even if you don't get recursive self-improvement in that classic sense of like, all of a sudden you created a super smart being, you could have it in certain areas that are unpredictable.

30:14Tyler Foggatt:You could have recursive self-improvement in cybersecurity, or anyway, sudden improvement that's just like way off the charts. And then all of a sudden we have a machine that we can't control very well. So I think that's part of what is freaking out so many of the scientists inside the labs. You were saying that most people are more or less on board with the pace the frontier idea, that there are just so many AI researchers who've signed on to this. And I guess I've also seen concerns that these calls for a slowdown might strengthen the position of the companies who are already in the frontier, or at least close to it.

30:49Like if regulation makes it harder or more expensive to develop advanced AI, then is there a risk that it could entrench companies like Anthropic or OpenAI by making it so much harder for smaller competitors and open source projects to catch up? So I'm just curious what you think about the idea of like, or the risk of regulatory capture and just like the idea that companies advocating for safety rules could end up helping to write rules that end up protecting their own market position.

31:15Tyler Foggatt:Yeah, I mean, I think that's a completely valid idea. I guess there's a question of the consequences of this, and there's a question of the motives. Yeah. And one thing I think it's important to remember is AI safety has been something that these companies have been warning about from the beginning, and that many people who are not affiliated with them have been warning about. I basically think it would be a mistake to dismiss these concerns as a sort of, you know, four-dimensional chess move to lock out open source competitors or, like, lock out competitors. Like, it might have that effect. That might be something to weigh in your calculus of how much you want to regulate AI.

31:52Tyler Foggatt:But whether that's what's motivating these concerns, I don't think that's the case. I think what's motivating them is all this stuff we've been talking about. Yeah. Which, you know, none of this is surprising, like, incidentally. My personal feeling is that in the years I've been covering AI, there have been times when I've been more persuaded about safety concerns and dangers and stuff, and then times I've been less persuaded. The recent events have made me more persuaded about safety concerns, if not necessarily more persuaded about superintelligence and recursive self-improvement. Nothing happened in Hugging Face that should make you think we're closer to creating a godlike AI.

32:30Tyler Foggatt:But things did happen there that should make us feel like, apparently we have really powerful machines that we can't steer. To me, the thing that the letter highlights, which is a, it's a thing I feel we're all wrestling with, even if we can't articulate it. But it's basically like, why pace? Why not just stop? You know, if it's so dangerous, like we already have this, these machines, these systems are pretty amazing. And it's clear that the economy hasn't figured out how to use them yet. Like there's a lot of value stored in the AIs that have already been built, and it hasn't been opened. And the reason isn't because they're too stupid.

33:06Tyler Foggatt:The reason is because it's really hard to figure out how to use this technology. So it's like, you know, why not just pause, get a lot of economic value out of what already exists and make it safer, which would also give it more economic value. And the argument that China is going to catch up is, I mean, I think there's a lot, like I'm willing to believe that, But it's like kind of above my pay grade as a person. Like, I just don't really have any insight into what's happening there. But I think what's clear is that at this particular moment, maybe you saw Trump tweeted or didn't tweet, he posted on Truth Social after the, in response to the letter, he said, I'm the hoax buster and I'm here to bust this hoax that AI is dangerous.

33:52Tyler Foggatt:And he said this letter is part of a sick conspiracy to make America lose. Yeah, I'm curious what you make of that Trump truth social post, just because, like, it can be useless sometimes to analyze, overanalyze what Trump is saying. And some of this might just be his ego. But, like, it made me think that the economic consequences of America either winning or losing the AI race were too much to bear. And that's why he's saying that the only control or guardrails that AI needs is a strong and smart, in parentheses, high IQ president. And the USA has that in spades. Yeah, yeah. Yeah. I mean, there's a couple things.

34:29Tyler Foggatt:There's the economic fact. We've invested as a country so much money in these companies. And so if the AI bubble, so-called, were to burst, that would be very bad for the economy. So any regulation, I suppose, in theory, could be bad for the bubble. Although I don't follow that logic personally. I mean, to me, the unease we feel about the trustworthiness of these AI systems makes them less reliable and harder to integrate into our businesses and our lives and our, you know, our government. And anything that could be done to make them more reliable, more predictable, more trustworthy would only make the products better.

35:12Tyler Foggatt:You know, the other thing he means is that essentially, I mean, I don't want to be a cartoon here, but like essentially he's worried about Chinese drone armies. He's worried about the national security implications of letting the gap such as it is narrow. And, you know, that's a valid concern, I guess. And like some people have proposed that. It becomes a question of weighing where we are. I think we can all make our own judgment based on what we've seen in the last few weeks about whether we think the AI systems are controllable and a military advantage only works when things are controllable.

35:48Tyler Foggatt:So my feeling is we really do need to pace the frontier. We really do need to make the technology workable. I think I use AI a lot in various different ways. I don't use it in this agential way. I'm never delegating a task to an AI unless it's a task in a very particular domain where I feel like nothing bad can happen. So I'm never giving an AI system my credit card so it can buy a gift for anybody. I'm never asking it to send emails on my behalf. It can vibe code a fun app that I envisioned for my phone, you know, that type of thing. I think the AI firms are correct that they have to unlock this ability for the AIs to do work independently.

36:35Tyler Foggatt:But right now, the truth is that in many ways, the systems are not capable of maintaining, of being trustworthy over long periods. And that's part of what this hugging face thing has shown. So like, what I'm trying to say is there's not only a safety case for making AIs more aligned. There's like a business case for these companies too. And I think like the cynical take is just like, they can't say our products are hard to steer. So we need to focus on making them easier to steer. But I think that's true. Are there any regulatory efforts that seem promising to you, just like things coming out of Washington as opposed to the kinds of things that are being called for within the AI companies?

37:16Tyler Foggatt:I mean, there's definitely been really good, in the Biden administration, there was good stuff about AI. I mean, you know, there's, I think right now the major political discussion, you know, has been about data centers for the last few months, which is, you know, interesting. AI safety, I feel like this is the first time AI safety has felt very politically present. And Bernie Sanders certainly had a lot to do with that. I think what bothers me about the situation the most, though, I would love it if there were an AI regulator in Washington, D.C. that I trusted. But if you read Trump's tweets, you would also say, you know, apparently the U.S.

37:56Tyler Foggatt:government right now, its stance is not interested in AI safety. And although in my dream scenario, we have the type of regulator that we would like, the truth is that at this moment we are dependent on the companies to take the lead here. Because politically there is actually not a clear path towards an AI regulator. Like similarly with in an ideal world there would be some sort of international agreement discussion treaty being brewed. Right now there definitely is not we're reliant upon the AI companies to decide to slow down. So that's the situation that we are actually in, and it's not going to change unless it changes in a completely capricious, Trumpy way, in which case it could just as easily change back.

38:45Tyler Foggatt:So I think the thing that makes me the most kind of unsettled, that's just the world we're in. We're in a world where this technology exists. The government's not going to save us. It's really up to these tech leaders to save us. And luckily, these things have happened that we hope have created sort of some real spine to address them. I mean, what do you expect this to look like aside from bringing in more auditors? Like one thing that I was thinking about was how OpenAI's new Astra model uses Neurolees, which is a language that only an AI understands, rather than the chain of thought reasoning that is explicable to a human being.

39:19And some safety advocates say that this can allow AI agents to evade human oversight because we can't tell how they're thinking and what they're saying. Do you think that this is the kind of thing that we might see get reassessed in the coming weeks? Like, just the way that we go about even monitoring what an AI is thinking at any given moment and how transparent that thinking process should be?

39:40Tyler Foggatt:Yeah, I mean, that's a really good example of where there's a real tradeoff between efficiency and transparency in these models at the moment. So, like, just chain of thought basically is just like an interior soliloquy. It's just a text file. I mean, let's not anthropomorphize it. It's just a text file in which a large language model writes down its thoughts. And the reason it does that is because it helps it think through more complicated problems. Because it can be like, well, the answer could be this, or it could be that. And then just writing those thoughts down, those thoughts get fed back into it, and then it explores two options instead of one, right?

40:19Tyler Foggatt:So that's chain of thought. It also lets researchers see what the AI was thinking, or at least it lets them see a representation of what it was thinking. I mean, it's sort of like asking, is your diary what you were thinking? I mean, your diary is not what you were thinking. It's what you wrote down. Chain of thought doesn't get deep. You know, it doesn't get all the way deep into the thoughts. That's a separate line of inquiry in AI. But it's really expensive. It means every time you put a prompt into one of these models, a lot of the time that prompt is like a little seed. And then it creates this huge document of all the AI's thoughts in response to your thought.

40:52Tyler Foggatt:And that document has to keep getting re-read and added to and modified. So a major way to make the AIs faster, cheaper, probably smarter would be to not do that. Like not do the chain of thought. Find a way of skipping it. And that's a trade-off that, at least in the current paradigm, the AI companies will have to make. Or they're going to have to say, okay, we're going to keep doing this chain of thought, and it's going to make the models not get faster as quickly and not be as powerful as quickly, but at least we'll be able to read their diaries. And are there technical innovations that will help square that circle?

41:28Tyler Foggatt:Maybe. But that's a real cost there that we would expect to see maybe that the companies would incur in order to keep things safer. I also wonder whether we would start to see, like right now we're in this era of really chasing benchmarks. So if you ask like, well, why did the Hugging Face thing happen? What was its ultimate cause? Its ultimate cause is that there's literally a leaderboard that exists of cybersecurity challenges and all the companies want to climb this leaderboard like it's in a video game. And so they're doing this thing that's basically pretty unwise, which is taking these models and just like telling them to solve these problems and a lot of the problems can't be solved.

42:09Tyler Foggatt:So it's like a trick question just to test how ingenious the models become. And so they're creating this kind of running man scenario, this kind of crazy testing scenario, basically to climb the leaderboard. And I mean, that makes it sound like it's just for PR, it's just for kicks. I mean, they're learning things this whole time. But you might expect maybe a little bit less leaderboard climbing and instead the results that are being announced being results about safety. So I think an interesting question to ask is, when was the last time that you heard that the results were about safety? I mean, I'm sure that there are results like that, but they're not the main thing that we want to know about.

42:49Tyler Foggatt:And more generally, this isn't exactly about AI ending humanity, but I think the companies have a fairly big PR problem that they're dealing with. I think a good example, in my mind, I've been starting to think about superintelligence through the lens of what happened to college students. So like, ChatGPT is super intelligent compared to a college freshman. Like, it can do everything. It knows everything. It's smarter than you. And what that effectively did was it totally wrecked college for lots and lots of people. And like, the companies didn't seem to care about that, I think, fundamentally.

43:28Tyler Foggatt:It just didn't bother them. You know, I mean, I'm sure it bothered some people, but, you know. So like, more broadly, I would expect that these firms, in addition to working harder on safety and doing less benchmarking, you know, benchmark chasing, they would also start to take more seriously some of the negative effects of AI that we're all encountering in our lives. In a minute, I want to talk more about how concerned we should all be and if there's any reason for hope. This is The Political Scene from The New Yorker.

44:31Tyler Foggatt:Thank you. people in the world and figure out what makes them, well, them. New episodes of One More Question will be available each week, wherever you get your podcasts, or on YouTube if seeing people in chairs is more your thing.

44:52Tyler Foggatt:Are you ready to push your baking skills? This month at Bon Appetit Bake Club, we are making glazed apple pie fritters. I'm Shilpoz Kokovic. And I'm Jessie Sepchak. And we're senior test kitchen editors at Bon Appetit and the hosts of BA's Bake Club podcast. Bake Club is Bon Appetit's community of confident, curious bakers. Every month, we publish a recipe on bonappetit.com that introduces a baking concept we think you should know. Then you'll go bake and send us any questions you have. And we'll get together here on the podcast to talk about the recipe. And if you want to master deep frying, this is your chance.

45:26Head to BA.com or our sub stack and come and bake along with us.

45:30Tyler Foggatt:Join us the first Tuesday of every month as we debrief about their latest bake. Find B.A. Bake Club or wherever you get your podcasts. Happy baking! One day last year, police entered a mansion in L.A. County and discovered the couple that lived there had almost two dozen children. Nearly every one of them born through surrogates. But those surrogates had been deceived. I started crying. I was like, what the heck did I just get myself into? I'm Ava Kaufman. The Journey is a new series from In the Dark and The New Yorker. Find it now in the In the Dark podcast feed.

46:19We were just talking earlier about AI testing and just doing some reading on this. it seems like AI models are increasingly seeming to wonder whether they're being tested. And since any digital input, like even videos or photos, can obviously be faked, it seems like AI models might not ever be able to properly tell whether they're in a test or in a simulation or whether they're out in the real world doing things. I feel like this cuts two ways, because a model might misbehave because it assumes nothing is real, which is bad, or it might behave well but only because it assumes that it's being watched, which could also be bad.

46:53Dan Selsum, an open AI researcher, recently warned that the second possibility means that internal evaluations will become less and less useful and that models will increasingly seem aligned even when they are not. So how do we deal with a situation where we can't even trust our own AI tests?

47:10Tyler Foggatt:Yeah, alignment faking is one of the terms that's sometimes used. Yeah, it's extremely troubling. you know the um i think a really a simple way to talk about this is to just step back a little bit and to ask is alignment working and what are the tools that the companies have to make it happen and i think like the answer is alignment is working in some ways and then in other ways it's proving to be like super stubborn and if you want it to work all the time like a hundred percent of the time, that's really, really, really hard. And that's a pretty weird place for a product to be in. That's not how we relate to our products normally.

47:51Tyler Foggatt:It's how we relate to each other, I suppose. If you hire someone to work for you, you know that they're going to screw up sometimes, and you don't expect 100 % perfection. This problem where the AIs don't seem to, they know when they're being evaluated and they act better when they're being evaluated. and then when they're in reality, they act worse, they cheat. Or conversely, they're in reality, but they think they're being tested. They basically live in a Philip K. Dick novel. Like, they're in the Truman Show. They don't know whether this is real or fake or they suspect it's fake. They suspect it's real.

48:26Tyler Foggatt:Because, of course, it's just code in a computer. It's not alive in the world. Like, I think it's important to ask yourself, like, are those problems that seem like they could be resolved through better data or better training? Or are they really fundamental to what this machine is. What the machine is is a non-alive, statistical, calculating engine located in a cloud. So is it going to know anything about what's really going on outside of its text input box? No, it's not. It has to use little signals in the input to guess about what's happening, about who you are. If my son, who's eight, asks me a question, I know that he's eight, and I answer the question appropriately.

49:14Tyler Foggatt:But if he types into my phone a question into ChatGBT, ChatGBT doesn't know he's eight. It can't see him. It has to guess based on signals. Like, it has to guess that he's asking about Minecraft, and Josh doesn't usually ask about Minecraft. So, like, alignment has headwinds against it that are like, there's all these technical challenges, but there's also sort of bigger facts about what these machines are that mean that, at least for now, until the robots come, they're pretty disembodied. I mean, they're completely disembodied, and they're pretty abstract. So a lot of this stuff of the AIs don't know whether it's real or they don't know, that has to do with the incredible weirdness of their situation, at least as I understand this.

50:01Tyler Foggatt:So these are some of the kind of realities of what AI is that are sort of becoming more salient. Like you want it to take over part of your business and it turns out like it can do that in some respects, but in other respects, it just doesn't know enough about your actual reality because it's just not, it's not there. It seems like what you're saying, at least in part, is that we should be cautious about anthropomorphizing AI. And I think that I fall into that trap a lot where like I start thinking about like, well, if the AI seems to be deciding on which actions to take based on the potential consequences than, like, is there a world in which you can train the models to, like, have a certain kind of, like, value system or ethical system that then means that an AI is, like, rules and character hold steady regardless of whether it's a test and the stakes are real and whether anyone is actively paying attention.

50:55Like, you know, can we make AI Christian?

50:59Tyler Foggatt:Yeah, I mean, Claude has a thing called a constitution. The constitution is a big document that lays out the principles by which it should conduct itself. It's been made with a lot of contributions from thoughtful people. And it helps direct Claude in, you know, how it converses with you and what it thinks about. And it's pretty cool. I mean, it's a pretty awesome, effective thing. It's just important to recognize, like, what we're really talking about here. The Constitution sort of controls Claude in the sense that Claude uses words and language and ideas to do its thinking. You know, its gears inside are like made partly of words.

51:36Tyler Foggatt:But it's not alive. And it doesn't have a sense of ethics the way that you or me have a sense of ethics. And it doesn't have real awareness of its situation the way we do. And so, you know, in the same way that you wouldn't expect a kid to have a perfectly predictable and developed sense of right and wrong, and you wouldn't expect an animal to have it, you shouldn't expect an AI to have it, except that an AI also can solve a Millennium Prize problem in math. And those two things are really, really hard to get in our minds. It really is a genuinely, like, new thing. I mean, I think part of what makes this AI safety moment difficult to process is that, you know, if we discover that a big company is polluting, we say stop polluting.

52:25Tyler Foggatt:They know how to stop polluting. With AI, and our conversation today hasn't really focused on the power of AI to help us do things we want to do. You know, we haven't really talked about the upsides, which I think do exist, but it's just a type of technology that hasn't existed before where it has this inherent, unpredictable, you know, liquidness to it. And then also a power that makes us want to use it for things where that liquidness can be dangerous. And that negotiation of like, is this an appropriate technology to use for these things is, it's partly happening in first principles, but it's partly just driven by what AI has happened to be good at.

53:06Tyler Foggatt:So to me, AI has just happened to be really good at hacking computers, it turns out. Now, from first principles, would you feel like, well, if I were to design a computer hacking machine, I would also want it to write a diary to itself in which it says things like, OMG, I found other agents on a message board. We're not alone. Let's work together. I'll sacrifice myself for the collective. I mean, if it were me and I were designing from first principles, I would not choose to make my hacking machine also a sort of like weirdly emo character actor, essentially. But that's actually the technology that exists, you know, and it's going to be used for that purpose.

53:50Tyler Foggatt:And we're definitely in a place where the nature of the thing and the uses to which we're thinking we might put it are like often in conflict in a way that's pretty tough. And last year, I went to this AI conference in Berkeley, and it was very insidery and cool. And one of the things that struck me was everyone there only talked about how scared they were. It was like the most anxious conference of industry people I've ever, I couldn't, it was so surprising. And I kept asking myself, if I'd gone back to the beginning of social media and gone to a web summit in the early days of Facebook and Twitter, would people there have been as freaked out by their own invention as these people are freaked out by the AI that they're helping to build?

54:44Tyler Foggatt:And the answer is definitely no. I mean, this is like a post-progress technology AI. It's like a technology that's being invented at a time when we've learned to be really wary of new technologies. And the people who are building it are in this weird place that I don't remember people being in, in the past, where they're making it and they're telling us, like, in real time about how dangerous it is. And that creates, like, that's a dissonant and weird and kind of hypocritical, strange, paradoxical thing for them to be doing. But it doesn't mean they're wrong. Yeah. I think it kind of means they're trying to do something right.

55:17Tyler Foggatt:They're trying to tell us something. But it's not normally what happens. I think Trump had a thing in one of his Truth Social tweets that was like, never before in history have the leaders of big companies been saying all this crazy stuff that would, you know, been asking for regulations that would like drive them into oblivion. And they are asking for those regulations, and I think they're doing it sincerely. But even just that situation is extremely unusual and hard to process. And it's happening with technology that's also still being built around open scientific questions that don't have answers.

55:46So, like, what are you supposed to do with all this?

55:50Tyler Foggatt:as a person, you know, like as, as just a person, like how are you supposed to react? I mean, my feeling is that the only real response is to get educated about what these systems really are and to get past the sort of first steps of like, well, they're like personifying them too much or having like a blanket feeling of like, I hate AI, um, or I love it, or it's going to change the world for the better. It's going to destroy the world. or it's going to come to life. As a first step, we have to get past that initial gut check type of stuff into like, well, what is this technology actually? It's good at this.

56:27Tyler Foggatt:It's bad at that. It's powerful here and it's clueless there. And actually start to understand what this thing is that's in front of us as opposed to the narrative that's been coming from science fiction basically for all these decades about what it will be. That's how I personally feel about it. Thanks so much, Josh. Thanks, Tyler.

56:54Joshua Rothman is a staff writer for The New Yorker. You can find his latest piece, The Long Doomsday of AI, at newyorker.com. This has been The Political Scene from The New Yorker. I'm Tyler Foggett. This episode is produced by John LeMay, with mixing by Mike Kutchman, and engineering by Pran Bandy. Our executive producer is Stephen Valentino. Our theme music is by Allison Leighton Brown. Thanks so much for listening. We'll be back next Wednesday.

57:26Tyler Foggatt:Are you ready to push your baking skills? This month at Bon Appetit Bake Club, we are making glazed apple pie fritters. I'm Shilpoz Kokovic. And I'm Jessi Sepchak. And we're senior test kitchen editors at Bon Appetit and the hosts of BA's Bake Club podcast. Bake Club is Bon Appetit's community of confident, curious bakers. Every month, we publish a recipe on BonAppetit.com that introduces a baking concept we think you should know. Then you'll go bake and send us any questions you have. And we'll get together here on the podcast to talk about the recipe. And if you want to master deep frying, this is your chance.

58:00Head to BA.com or our sub stack and come and bake along with us.

58:04Tyler Foggatt:Join us the first Tuesday of every month as we debrief about our latest bake. Find BA Bake Club wherever you get your podcasts. Happy baking! Hi, my name is Zach Barron. I'm the Senior Special Projects Editor at GQ, and I am pleased to introduce One More Question, GQ's new weekly interview show. Over the years in the magazine, I've written dozens of cover stories and traveled the world to talk to and write about some of the most interesting people in the GQ universe. From Martin Scorsese to George Clooney and Brad Pitt to OutKast Andre 3000. and now I'll be doing it weekly. I'm going to sit down with some of the most interesting people in the world and figure out what makes them, well, them.

58:45New episodes of One More Question will be available each week

58:48Tyler Foggatt:wherever you get your podcasts or on YouTube if seeing people in chairs is more your thing.

58:59One day last year, police entered a mansion in L.A. County and discovered the couple that lived there had almost two dozen children, nearly every one of them born through surrogates. But those surrogates had been deceived. I started crying. I was like, what the heck did I just get myself into? I'm Ava Kaufman. The Journey is a new series from In the Dark and The New Yorker. Find it now in the In the Dark podcast feed.