How Should We Approach A.I. in 2026?

0:05Welcome to The Political Scene from The New Yorker. I'm Tyler Foggett. 2025 was the year that artificial intelligence pushed its way into our politics and into all of our lives, whether we liked it or not. Today, we're bringing you a conversation that I had with the writers Charles Duhigg, Cal Newport, and Anna Wiener at the New Yorker Festival in the fall. We looked at the current state of AI, the kind of world being created by it, and how all of us might adapt or recalibrate to the technology. I hope you enjoy the conversation.

0:40So I guess I thought we'd start by talking about the two competing narratives about technology and AI that we often hear about. So you got the first narrative, which is that, and this is the narrative that's being sold to us by the tech giants, and that's that these tools will make us better. They will make our lives easier, more efficient. AI will eliminate our busy work, and it'll lead to incredible discoveries in science and medicine. And then there's the other narrative, which is that these tools are ruining us. AI is going to replace us and our jobs entirely. It might even kill all of us if it decides to be evil.

1:16You know, social media, meanwhile, and smartphone dependency are destroying our attention spans and our ability to think for ourselves. We were told that these tools would connect us and instead they're leading to a fractured, helpless, sad society. So my question for you guys is, which of these narratives feels more correct right now? Or do you think it's like a mix of the two? Do you want to start, Charles? Sure. So I think the thing that's really interesting is we've fallen into this pattern where we talk about this as a binary outcome, right? There's this instinct to see this as good or bad.

1:52And if you think about most of the technology that's influenced our lives, this is pretty typical historically, that when a technology comes into being, we tend to look at it in a binary way, but then the application of it is incredibly non-binary, right? It's much more in between. It's much more of a normal distribution. Telephones is one of my favorite examples. The number of articles that appeared when telephones became popular, saying that it would destroy society, that people would never have a real conversation on a phone, that it would pollute children's minds so much that they became mute and could not communicate face-to-face, all of which ends up being a little bit true, right?

2:33Like, we don't feel great about our kids on our phones today. But I think that for me, the thing that I keep thinking about is there's this instinct to see this as good or bad. And it's much more natural for us to use this thing in a way that it's neither good or bad. It's both. Well, so there's another axis, I think, about this, it could also be useful. So if you're watching the rhetoric coming out of the AI companies and then the rhetoric of people who are talking about the AI companies, I like to think that we have just passed in the last two months from Act I of the Gen AI revolution to what I now think of as Act II.

3:11So in Act I, there's a lot of focus on predictions. Hey, this is an amazing technology. Let's react to what it could do. So there's a lot of discussion about, so what will happen if and when it automates half of our jobs? Like what will happen when this industry, this industry, the industry disappears? So it was a lot of looking at the predictions about what this could do. There was a pretty astonishing shift, I think, in the discourse right around the August release of GPT-5, where we moved past saying thought experiments, what this could do and reacting to that, to, well, what does it do right now?

3:46And I think this is where we are now in the last couple of months, has been starting to grapple with, well, what specifically do these tools do? Where are they good and where are they bad? Which I think is closer to the way you're talking about it, Charles. I think this has been or is going to be a very healthy transition because up until now, it was very difficult to really have a pragmatic conversation about AI because we had super intelligence and job automation. We had Sam Altman talking about how the release of GPT-5 was going to be the biggest cultural event since the Trinity test of the first atomic bomb.

4:17We had Mark Zuckerberg saying superintelligence was now within sight. He was just making that up. We had Dario Amadei saying probably half of new white-collar jobs will be gone within three to five years. We were entertaining thought experiments. And I think right now the discourse in tech journalism is now saying, oh, forget that. Like, what exactly are we using these tools? Oh, that's good. That's bad. Oh, here we might want to control that a little bit. So I think we're now in this more pragmatic grappling phase of the new technology curve. I think it's a better phase that we're in. Anyways, that helps me think about what's been shifting recently.

4:51I really love the way that Sam Altman, Mark Zuckerberg, the way that they talk about this technology, just the pros style is really unique to me. I think that where I fall on all of this actually has less to do with the tech itself and more to do with how the sort of social structures around it, both that will kind of facilitate its normalization or its use. and also how well-equipped are we to deal with potential repercussions. So I'm a little more interested in that than in, I mean, I don't, just to be candid, I don't actively seek out generative AI tools personally. They're interesting to me on a sort of cultural level.

5:28But yeah, like how well-equipped are we to absorb the effects of this technology? What needs to change? Who will it benefit? Like who will suffer because of this? and how do we mitigate that suffering or save it off at the past? So those are some of the questions that kind of animate my feeling about it between the marketing and the Doomer perspectives. Cal, Charles, how often are you guys using generative AI tools? And I'm wondering if you can speak about your personal experience using these things or spurning them. I use it probably every other day and I just learned how to vibe code. I went to this thing and they taught me how to vibe code, which is basically like just like having a conversation It's not that different.

6:08But I'd say at least two or three or four times a week and sometimes five or six times a day. I mean, I'm mainly using it as an enhanced Google occasionally. I feel like this is what gets me trying more tools. So right now I'm in the middle of something where I'm using an AI tool to help clear my inbox. They're very good at filtering messages. So figuring out this is important or this is not important. But if not for that, it probably wouldn't have a big footprint yet in my life. Can I ask a question to Charles? Yes. What did you vibe code? What do you not want to vibe code? You can vibe code anything.

6:45Like, did you make something? Yeah, yeah, yeah. So I made something that told me if there was a, I was using RepLeaf if anyone uses it. And I said, build me an app using this data set that's available online to tell me if there's going to be a thunderstorm in my area. Because I was just trying to figure out what to use it for. And it is cool. But this actually gets to a point that I think you raised about the absorption, which is if you think about the last 25 years, last 30 years, the big things that changed our world were not necessarily just advances in technology, right? It was Jeff Bezos realizing that now that the internet exists, you can use it to sell everything.

7:28And that was true for the 15 years before Bezos started doing that. or it was the guys at Google realizing the page rank, that what's more important is figuring out if other people link to a page to start the search engine. And so I actually think that there's a cognitive barrier that no matter how advanced the technology gets, it'll be used by a very small number of people and they might be terrorists creating Ebola viruses, so that's bad. But for most of us, the technological breakthrough is not necessarily gonna be the technology. The technological breakthrough is gonna be a new concept of how we use that technology and the possibilities of the world.

8:03But I'm wondering if you guys disagree. I mean, I think that's right. I think if you look, let's use the search engine example, right? If you look pre-Google, there was something like two to three dozen search engines that were out there. So they weren't innovating the concept of this technology exists, which can crawl the web and return search results. And really the primary breakthroughs of Google were infrastructure breakthroughs, figuring out how we can use low-cost computing components in a very distributed way that can be very fault tolerant. and allow us to actually deal with like way more data that there's a computer science story there.

8:34But that infrastructure changed the experience of this technology that already existed in a major way. I think social media was similar. We have social media, web-based social media for a while. Then Instagram comes along and begins to figure out some of the technology needed to make this native on a phone because it was a phone-based app and it had to be native on the phone. And this was like infrastructure breakthroughs that allowed attention harvesting and attention engineering to occur on an exponentially bigger scale. We started using our phones way more. So I think that's interesting. We're looking now at like chatbots that show that generative AI exists and it has a lot of power in processing text and producing text and other types of media.

9:11But yeah, maybe we haven't. We don't know yet what the sort of infrastructure breakthroughs, app architecture breakthroughs are that makes that a killer app in a way. You're right. I think we might not be able to predict that yet. We often get that wrong. I mean, it does seem like there's a sense of inevitability that's already pervading this conversation. And we all sort of seem to tacitly agree that this is a technology that is here to stay and will be revolutionary, even if we haven't yet reached that point. And I guess I sort of want to challenge that assumption that seems to be underlying the conversation so far.

9:43Let's start there. I guess you guys see this as an inevitable, massive breakthrough that will transform the next hundred years of, you know. Can I ask you a follow-up question on your question? Yeah, yeah, yeah. So I guess we should – what technology are we talking about, I guess? Right. Well, you were talking about generative AI and specifically text. Yeah. And I was thinking, you know, we haven't really even talked about any – the other forms of media generation yet. And, you know, we might see that maybe like image generation or video generation accelerates faster than certain kinds of data processing.

10:22And so, yeah, I mean, you probably have more insight into the actual. I mean, I can get my one minute computer science view of like what I think is going to happen with the technology, because I think what's happening now is there was this belief and we did a piece about this in August. There was this belief that you could take something like a text-based LLM, like one of the GPT family of models, and that if you trained it on enough information and you trained it long enough on enough information, that just this text producer would end up internalizing enough logic and understanding of the world that it could be generally intelligent.

10:57A language model itself, if you trained it big enough, would have something like artificial general intelligence. That's a storyline that within the industry came to an end last year, and the public realized this sort of last summer, that actually that's not going to work. So text, my theory is there's a lot of work to be done to how do we actually take this idea of we have a tool that can understand text really well and integrate this into all sorts of products and that it's going to be really useful. But my vision of artificial general intelligence is that I think it's going to be distributed in the sense that we are going to have many, many different types of systems that use many different combinations of different types of models, some of them using breakthroughs that happen with language models, but some using other types of models, more neurosymbolic AI as well.

11:41And it's not going to be one oracle that does everything for us. It's going to be, here's a system that's really good at doing exactly this task. It can do it a little better than people. Here's one that can do this task. It's its own startup. It can do that a little better than people. And you multiply that by a thousand different tasks. And then you look around like, oh, I guess we kind of got to artificial general intelligence, not with like a HAL 2000 that does everything, but just we have figured out task by task how to sort of reconfigure and pull together these ideas to get systems to handle it.

12:09The industry does not agree with me on this, by the way, but that's sort of my vision. I don't know if it's going to happen. That's where I am now. So I want to answer your question. And I think this ties into what you just said. So if you think about, again, going back to telephones, when telephones were invented, what's interesting about the phone is that the scale of invention there, the scale of transformation was relatively low. It replaced the telegraph. And with the telegraph, we could basically do the same thing that's useful in our lives with the telephone. But what it did is it eventually democratized it.

12:40Then in comparison, there's the Internet. And I would argue you could actually say the internet is more powerful than the invention of the telephone because the internet has slowly but eventually transformed everything. It's transformed how we date, how we buy things, how we decide on what jobs or colleges to attend. I think that AI is closer to the internet than it is to the telephone. I'll make that argument. I'll invite you and everyone else to disagree with me. You mean language models in particular or AI more broadly? I think AI more broadly. AI more broadly, but neural nets, let's say. If this technology proves to be useful and eventually we have hardware robots that can use this technology to move through the world and do things for us, I think it's more akin to the internet.

13:24And I will say, much like the internet, if in 1995 you tried to forecast how that technology would be transformative, the best thinker on earth would get probably 20 % of it right. It's moved so much more slowly and so much more differently through our lives than we imagined initially. And so I think that's why it's going to be profound. Do you agree or disagree? Yeah, where are you? Seems reasonable. Would you consider it a platform? No, I think of it more as like a tool. But maybe we're using words differently. Like, what do you mean? I don't know. I'm just curious from your description. I mean, I think, Cal, I sort of land where you are in terms of it being, I think your approach is really pragmatic.

14:08It's also obviously informed by your background. You know, I live in San Francisco and the narratives, it's like a very storytelling oriented industry. I think there's a lot of enthusiasm for a larger than life omniscient entity, whereas I see this as more of a sort of an additive technology that, like you're saying, will improve certain things probably from the bottom up quite slowly. There's actually an East Coast, West Coast feud going on among technologists. So like on the West Coast, you have technologists based mainly in industry who are very utopian about this. And on the East Coast, you have a lot of sort of dour East Coast computer science professors who are looking at the technology and saying like, I don't understand how this is going to do this or that and are being curmudgeonly about it.

14:57And we're not getting along right now. Sort of interesting. When I moved to Northern California from Brooklyn in 2020, and I used to say, and I think this is still a little bit true, that in California, all anyone talks about is AI and psychedelics. And in New York, all anyone talks about is AI. That's pretty much the difference. And that might explain that divide between the two coasts. Okay. I've been in New York for 10 days, and I've only had one conversation about AI except for this one, and that was my fault. So I feel like it depends on the company you keep. That's true. It can be found.

15:31That is absolutely true. Yeah. Well, can I throw out a provocation about what you were saying, Charles? So I'm very interested when we have these new technologies that become potentially revolutionary and they completely transform things, what the story is when they arrive. Like what is obvious and what's not obvious, right? So I feel like there might be daylight here between the current crop of AI technologies and the Internet. because when the internet was coming along, there was a very clear story to be told there, which is anyone can publish information, anyone can read it, right? So this was a hugely democratizing, but also like very easy to understand, like this is very different.

16:09Like never before in history could it be because we have shared protocols that everyone uses, anyone can put up information, anyone else can read it. And so we were wrong in our predictions of exactly what that was gonna lead to, But that felt like this is like printing press in the sense of, oh, now we can produce books 10x fast or 100x fast or whatever. It feels a little murkier still right now with the current AI tools. It feels a little bit murkier about is it the it can generate text, it can understand text, it can generate video, but it doesn't yet have that self-evident feel yet. I talk about email with something similar.

16:47Like when email came along, it was like self-evidently useful to everyone. I was like, this is great. This is like a much more efficient way to like send information. And I feel like AI hasn't, it's still new enough and we don't have the killer apps yet. And so there's something different. There's like a bit of a different valence that like, it feels like artificial intelligence as a concept should be revolutionary, but it's a little bit harder to put our finger on unless we go into just pure thought experiments. Like, well, it can automate whatever, you know, anyone is doing, but it feels like there's this is what's happening now as we're trying to figure out, well, what exactly is going to be the key drivers of the near future revolution of this?

17:21So I completely agree. And I think we are in this very plastic moment, much more so than at the birth of the internet, with one exception, which is the investment that's happening right now in AI, right? That there's this absolute certainty about how much should be invested. So just to put this in scale, because particularly in New York, I think you see less of it. The Apollo program, putting people into space, took 10 years. And in today's dollars, it costs about$400 billion. The AI industry is spending $500 billion a year building out infrastructure. So we're doing the equivalent of 10 years of Apollo spending basically every 10 months.

18:00And there's a real legitimate question. Is that a bubble? Is this going to be all for naught? Is it wasted money? Is it real money? But the truth is that we are building out an infrastructure that will be, I think, transformative regardless of what the transformation looks like or what it's used for, just because we're building so much capacity. It's like fiber in the late 90s. Exactly. Which a lot of those companies went out of business, but then we use all the fiber today.

18:29The Political Scene, live at the New Yorker Festival, will be back in just a moment.

19:04Thank you.

19:20do you think there's ever a point at which these companies you know who are sinking tons and tons of money into this project that they would ever admit that we're just never going to have a super intelligence or is that kind of impossible given how much money is going into this well i think what they would quibble with is what's the definition of super intelligence so But like you guys have seen Meta say like they're building super intelligence, which is like even everyone in the valley is like, that is such BS. Like it's basically a marketing message that you want to attract people to work on cool problems.

19:50But so I think that there's an interesting question. The problem is we use these phrases like super intelligence or AGI, artificial general intelligence, without actually defining what they are. And the best way to do it, I think, is to root it in analogies of things that we are familiar with. Like does a dog have general intelligence? I would argue yes, but other people could argue no. And if so, the things that make that dog intelligent, can a machine replicate that in a way that doesn't particularly concern us? So I don't think that companies are actually having that debate as much as the debate that they are having is, for the last 15 years, everything in tech has been winner take all.

20:32and if we don't invest 500 billion dollars we won't be the winner and someone else will and we'll go out of business yeah i mean i guess it's it's very hard to define what a super intelligence even is but i kind of feel like i know what it's not and that's what we're getting right now which is like i try to use you know chat gpt for a very specific kind of google search and sometimes it can be very good at that and sometimes it gives me just like it totally just hallucinates and And even when I'm asking it, something very simple. And so I'm like, am I doing something wrong? Or is this just like a fundamental problem?

21:07And like, why is it so hard? Or if you use Alexa, it's super unintelligent, right? Like, it's like Alexa is like, I'm not worried about the robots. Yeah. I mean, the entire, the story of super intelligence, I think is nonsense. The entire technical story for super intelligence is this idea of takeoff. So it's this idea of recursive self-improvement, that once we make a machine smart enough to improve itself, then that next machine will be smarter and it will improve itself to be even smarter. And then that machine will improve itself to be even smarter. So this rapid takeoff scenario, they call it, that's at the root of the super intelligence thought experiment, which is we have an AI so smart that we don't understand it and then it can conquer us.

21:44This is the new book. If we build this, we'll all die. That's the concept. From a computer science standpoint, that recursive self-improvement, language models can't do that. It's just nonsense, right? I mean, they cannot solve the problem of producing a program that has like a brand new theory of intelligence and how to build this into a computer program. I think that's nonsense. But you will see that the rhetoric has changed. If you can put a timeline on the CEO's rhetoric. 2023, early chat GPT, it was a lot of super intelligence rhetoric. It's like, we got to be worried. The whole world's going to change.

22:17Sam Altman saying we're going to have to have an equity tax because there'll just be two companies left and everyone else will have to have a way to eat. And they really believe that. Then they all shifted at some point. They're like, oh, this is scaring investors. So on mass, at some point in 2024, everyone started talking about job automation. Now, this is what the power is going to be. It's not super intelligence. It's going to automate jobs. And we have to grapple with this in the economy. And that was a better story for investors. Because if your company is automating jobs, your company is going to be worth a lot of money.

22:45I want to invest in you. And now they've shifted again because the year of the agents never happened. We have the new MIT study that found 95 % of the deployments they studied of AI and corporate environments had zero return. And so now they've shifted the rhetoric again. So I feel like there's, I mean, you know these leaders, you probably have a better sense of them than I do, but there's like a sort of coordinated shifting that keeps happening. And they want there to be something to emphasize that makes it seem like this is an inevitable, powerful enough technology you should invest. But they keep sort of shifting what that target is in a sort of like semi-coordinate.

23:18Now they're talking about bubbles and about, so it's interesting, like they keep shifting the way they talk about. So they haven't talked about superintelligence as much as they used to. I was just going to ask you a question, actually, because you write very beautifully about the importance of human connection and how people connect to each other, I felt like, in Uncanny Valley. And you've been talking to people, I think, who are using AI for connection. We've been talking about very practical applications, but do you think that there might be a transformation in how people feel about themselves and the world because of a relationship they develop with AI?