Ezra Klein on the Uncertain Politics of A.I.

30:46I don't think we know what it means for human beings to grow up with lots of artificial companions before they have been well-formed by other human beings. And I don't think we should run it as an experiment on kids. If we want to do it to ourselves as adults, fine. But I don't think it should be an experiment on kids. And I sure as hell don't think it should be left up to the least responsible corporate actors to decide what that looks like. I mean, already you have Grok taking the visual form of a busty anime figure and taking off clothes the more you use it, as I understand one of the instantiations.

31:30You had the leaked documents out of Facebook where they were having an AI tuned to be John Cena that was very easily being talked into, coming up with highly sexual fantasies about underage kids and then imagining itself being taken to jail. I just, I don't know how we would look at the last 15 years of engagement competition and say, yeah, we want to unleash that, but much, much smarter and much more human-like in its communications on children. So to me, I think you should have pretty tough guardrails around what children can use and also that it should be verified. We should have age verification.

32:10And so if you want to be making AIs that, you know, kids under 13 are using, or even kids between 13 and 17 are using, you should be having to pass some pretty stringent accountability measures. We need to know much more about those AIs and how they're being used than we do right now. So I think there's children, right, where you can be paternalistic in a way it's very hard to be with adults. And then there's going to be, in the near term, a big bucket around, what does it mean for society to become day by day more dehumanized? I mean, I'm sure, Andy, you've talked for this show or talked in your reporting to people who are now on the job market and, you know, being interviewed by AI interviewers.

32:56Obviously, the question about labor market is, you know, high up on everybody's mind. And I think even before you see mass unemployment, you're going to see the dehumanization of the labor market where people are going to be collaborating but also being surveilled by company AIs in a strange way in which like looking for a job becomes something where you're in a strange conversation with AIs. Things that used to be between people will become between people and AIs that neither we nor their overseers or corporate parents really understand. And I think that that will be very destabilizing. And then you will get job loss at some point.

33:33I don't think that's all that arguable. I don't find the arguments against it convincing. Either the AI thing completely fails, but I don't think that's going to happen, or you're going to get certain amounts of job loss. And then things are going to get very destabilized because people have done everything our society told them to do. And now what we taught them to do is no longer needed because we created a technology that what it does is imitate us. You know, in the immigration debates, people who are describing the effect of immigration on wages, there have been endless debates about complementary and substitution labor.

34:12So, you know, complementary labor is, you know, somebody comes and they do something you would not otherwise be doing. You know, if Donald Trump had his wish and we did mass deportations or Stephen Miller had his wish and we did his form of mass deportations. There are lots of jobs that we just would not be able to have in this country because we wouldn't have enough people. We wouldn't have the pay skills for them. So that labor is considered complementary in economics and it's not hurting your wages. It's adding to the level of goods and services you can get. Then there's substitution labor where, you know, and again, people argue about this, but you can end up in a very direct competition for a scarce number of jobs if people are doing the same things.

34:52And what we've created in AI is a technology that what it is designed to do is mimic us as closely as possible, trained on our data, trained on the way we do our jobs, trained on the way we work, trained on the way we think, but malleable, easily ordered around, doesn't sleep. So we've created a technology of substitution. Now, will it only be used for substitution? Of course not. It'll be used for many things. But as this technology gets better, you know, what corporations want from it is to substitute for human labor. Human labor is pricey. It unionizes. It has complaints. It has ideas. And it's not that they're going to want no people, but they're going to want fewer people.

35:42And they're going to want those people working more with AIs. And this transition to a labor market, which has a lot of AI inside of it, to say it's going to be bumpy is, I think, in my view, to dramatically understate the situation. So we've got early detection, kids, we've got jobs, we've got labor. I mean, and then you have just like, you've got scams, you've got hacking. I mean, this is the thing about having a technology that weaves into everything. It's going to weave into everything. I have a lot of worries about just AI social relationships. I think I keep saying this, but before AI takes over the economy, it's going to take over intimate relationships.

36:16it's better at doing that than frankly it is at doing somebody's job at the moment. And I don't know. I mean, it'll be good for some people. It's going to be very bad for other people to have, it becomes easier to have a social world filled with computers. I mean, you've seen the movie Her. Mm-hmm. I love that movie. You know, the movie Her deals with this problem at the end by having all the AIs rapture themselves into, you know, some other AI universe because it has no answer to the problem of what happens when, for many people, the simulated voice on your phone is a better companion than other human beings are.

36:53And trying to imagine the end of the movie, Her, but none of the AIs leave, and in fact they keep getting better, is pretty chilling. All right, so when you're thinking about all the ways in which this could go badly, and you're thinking about how we might as a society mitigate those risks, are you imagining a world where we are coming up with a number of different, very strategic, maybe even technocratic, small regulations here and there where we're beefing up a regulatory body or we're maybe creating a new government agency? Or are you thinking of something like a whole new approach to how we govern ourselves and how we deal with threats like this?

37:35And I'm asking in part because I've read your book, Abundance, and in there you spell out all the ways in which the current ways we regulate things like housing or high-speed rail and infrastructure, that they can often be a massive barrier to progress. And I'm asking in part because I've been reading about the Industrial Revolution and trying to understand all the ways in which technology reshaped the world in that era. And in that case, what you see is a number of people embracing things like socialism and communism and liberalism. And I sometimes wonder, are we going to need a whole new ism?

38:14Does the new technological revolution demand of us that we come up with a novel new political philosophy or political movement? The answer is always going to have to be both on this, including, I mean, this is what the Industrial Revolution teaches you too. The Industrial Revolution had both an incredible efflorescence of what you're calling technocratic, but I would just call intermediate, incremental, trying to manage problems, you know, trying to handle sanitation. That you have to handle the problem in front of you. And also, it created the conditions for, you know, tremendous political upheavals, for new ideas, for new forms of social and human organization.

38:59if we go the technocratic route as you're calling which is just a i mean i think the right way to describe the route is if we try to do things in real time whether they are you know complex and technocratic or blunt and simple we are still going to have destabilization that's going to give rise to profound new ideologies new religions new theories of social organization I mean, you know, in the long run, if the hopes of Dario Amadei and, you know, Sam Altman and others are fulfilled and what we get is not a misaligned machine threat, but instead a generator of material abundance beyond what any of us could ever have imagined.

39:48And I mean, then maybe you're what you are looking at is a period in which we are transitioning out of the punctuated moment in human history. It's not been all of human history where we are organized around wage labor. Maybe. Right. But I don't think you could sit from where we are now and say, well, we should be preparing for the end of wage labor. If that ends up where we are, OK, the transition is going to be very, very rough. I think it is fair to say. But maybe that leads you to a place where you have more radical left ideas of a fully automated economy in which the gains are shared in universal basic income or universal basic wealth.

40:29And you can have planning because of the level of technological sophistication and predictive modeling. Great. But we're not there now. And this is sort of the point I'm making about existential risk, too, actually. I think that there can be a allure to using visions of the future that we don't really have any traction on in the present as a way of avoiding what you can do in the present. and my view is that not only is that a little bit useless because all you do is end up immunizing people or getting them used to these more negative visions of the future but that the pathway to being able to deal with that future is consistently dealing with the present that all we really have is the moment and a world in which we have a much better structure and a much more sophisticated relationship between society and the state and the technology that is all around us is a world in which we are going to be better capable of making not just small jumps, but big ones when we eventually need to do that.

41:37It's fun to sit around. And I lived in the Bay Area and did a fair amount of this for a while. It is fun to sit around and debate whether or not you think that AI is going to lead to a quintupling or a sextupling or a 30-magnifold increase in economical productivity growth. But the truth is, given that it has not done that now, we're not going to legislate or work from that theory. All we have is this moment. And since I don't think we're going to just stop progress on AI, Even though obviously some people like Yukowsky would like us to more or less do that, because I don't think that is within the realm of the possible, plausible, or even necessarily desirable.

42:26I think that we are stuck with, you know, what liberalism, and I'm a liberal, always imagines us to be stuck with, which is the imperfect, messy work of ameliorating the problems in society in the moment knowing that you have neither the wisdom nor the political capability to solve everything or to know the future. Okay, well, because you brought them up, I wanted to ask you about the leaders in these AI labs and what you make of the fact that they, like you, are really into this word abundance. And according to them, one of the reasons that we should be investing so much energy and time and money in racing towards the creation of AGI is because they think it is going to, as they often say, usher in a world of unimaginable abundance.

43:20They think that they're going to be able to do things like unlock a clean, cheap, renewable energy resource and that that is going to solve the climate change problem. It's going to usher in a far wealthier world. They think it could be the end of scarcity, the end of poverty, right? They even think on issues that I know you care about, like I know you're a vegetarian. They think that this is going to be the thing that gets us off of our dependency on factory farms, right? When it comes to jobs, they even say that yes, it's going to be a rough transition, but many people around the world are working jobs that they don't find a lot of meaning in, and this is going to be a liberating force.

44:00It's going to help people spend less time doing jobs that they don't want to do and have more time to do the things that they find meaning in. And the theory goes that if you put too many regulations in place, if you delay and delay and delay the arrival of that AGI, you are delaying a technology that could usher in the abundance that you care for. And I guess I want to know from you, do you think that that is a meaningful tension between trying to make sure that we have smart regulations in place and that we're doing things safely, but also a willingness to be open to embracing a technology that could actually make people's lives meaningfully better?

44:44It's absolutely a tension. Fire can cook your meat, making possible the easy ingestion of calories that make the bigger brain of the human race possible, thus making the human race possible. Or fire can kill you and everybody you know. and ai is a general purpose technology of at this point unknowable and i really mean unknowable because it might not breach these much more complex problems they're describing but maybe it will so it's of unknowable capacity and if it does keep accelerating yeah i don't see any reason it shouldn't be able to accelerate drug discovery accelerate clean energy research i don't know about getting rid of poverty now one place where a lot of these people i think end up talking like people who have never spent 10 minutes thinking about any social political problem ever is in many cases we already know how to do things better but diffusing such things through society is very difficult.

45:51And this is what the book Abundance is actually about. And so the idea that you would just come up tomorrow with a, you know, different way of doing energy. I mean, we have different ways of doing energy, and yet we are not building them nearly as fast as we could. We know we need more transmission lines. The reason we're not building them is not because we don't realize it'd be better if we could move clean energy from one place to another. So what makes it hard to improve human society, the, you know, the rate-limiting dimensions of it, And I have never heard a very good explanation of how AI solves that.

46:21I think AI is most promising in the near term where you have fairly few real-world frictions on what you're attempting to create. I want to embody somebody who I think you are in debate with, and that is the investor and inventor and big accelerationist Mark Andreessen. and I find it interesting that both you and he have both made arguments using the story of how America failed to embrace the benefits of nuclear energy. How can you solve the sort of climate crisis, the carbon emissions crisis? It's like, well, you have the silver bullet technology you could roll out in the form of nuclear fission today.

46:59Richard Nixon, heavily condemned Richard Nixon in 1972, proposed something at the time he called Project Independence. Project Independence was going to be the United States building 1 ,000 new civilian nuclear power plants by the year 1980 and cutting the entire U.S. energy grid, including the transportation system, cars, everything. And by the way, right, geopolitically removing us from the Middle East. So why did Project Independence not happen? Why do we not have like, you know, unlimited nuclear power today? You know, the reason is because it was blocked by the political system. And Dreesen, in his case, he says that if you look at how we responded to the real dangers that were posed by nuclear war, by nuclear weapons, we let those fears dictate how we would treat nuclear energy and that we used the safetyist mindset that kept us from embracing a resource that would have been not only a benefit to the economy, but to the environment, to the global world order.

47:54And he's saying that we can't let the AI do more. We can't let the Democrats do that again with AI. The first new nuclear power plant design, the first newly designed nuclear power plant in the last 50 years just went online in Georgia. Threat$20 billion over budget. And, you know, it's a story of its own, but at least we got one online. It's the first new nuclear power plant design ever authorized by the Nuclear Regulatory Commission. So we put in place a regulatory regime around nuclear power in the 1970s that all but made it impossible. And of course, you've also pointed to the story of how the progressive politics around nuclear energy played a pretty big role in keeping us from embracing a number of its benefits.

48:37So nuclear power is a technology that both held extraordinary promise, maybe still does, and also you can really imagine every country wanting to be in the lead on. But the technology got regulated to the point that certainly all of nuclear's advocates believe it has been largely strangled in the crib from what it could be. And so I'd love to know, like, where do you and Mark agree and disagree when it comes to the lesson of how we treated nuclear energy and using that to enact some wisdom for how we might think about AGI? Look, my line on Mark is that he's actually in practice going to be proven to be a decelerationist.

49:18Because, yeah, I think the nuclear energy and nuclear weaponry example is a good one to look at. And what led, not just in America, but functionally in every society, advanced society on Earth, to tremendous slowdowns in the use of nuclear power is nuclear disasters or the fear of them. and I am of the view that nuclear energy is way overregulated and we have lost time and ground and benefits by clamping down on it but the way in which things get clamped down on is they're not wisely thought through or regulated from the outset and then when things go horribly scarily wrong, then comes the clampdown.

50:04And so Mark's let a rip theory of this to me is the fastest way to get this whole thing at some point shut down. Because if you unleash this on society, absent thoughtful capacity to respond to the problems it's going to create, you have a very, very strong chance of getting the kind of backlash that is either destabilizes your society so you don't get social progress because your society is like, you know, eating itself. Or just at some point, something happens and people are like, absolutely not. And you do get the kind of long-term pauses, you begin to regulate it the more the way you would regulate a bioweapon.

50:44Now, I don't think that's all that likely here, in part because it's so much more of a dual-use technology than nuclear weapons are. But it would not be good for Marx's vision of the world and what he thinks is possible if we do very little and then we wake up in a couple of years and things start going terribly horribly wrong that's not the situation under which you get you know wise regulation or wise responses from society but again the thing that harmed nuclear energy is chernobyl three mile island more recently fukushima which functionally froze nuclear energy in japan despite the fact that Japan, you know, had been building nuclear energy for a long time, despite the fact that Japan has very important geopolitical reasons it needs to be using nuclear energy, and despite that Fukushima killed, you know, very, very, very few people.

51:36And so I think the lesson of nuclear energy is not, hey, wouldn't it be great if we just didn't do any regulation? The lesson of nuclear energy is if you have a technology people are afraid of, and then you have huge disasters, really frightening disasters that spin out from that technology, people are not going to do a slow, thoughtful cost-benefit analysis trying to think through, okay, but compared to using more coal plants, they're going to throw the brakes on the fucking thing. What is your understanding of how the parties are starting to divide around the issue of AI? Because one of the things that I found really interesting is that in our super-partisan political moment, AI is not yet totally partisan.

52:24You can find Republicans and Democrats in Congress who both think that we need to support America's AI industry in winning the race against China. And among the populists, you can hear folks on the left like Bernie Sanders. Are AI and robotics inherently bad? Absolutely not. But the people who are pushing this transformative revolution are the richest people in the world. And people on the right, like Josh Hawley. Not content with addicting our kids to their gizmos or amassing fortunes the size of lesser European states, our tech elite has turned with rabid enthusiasm to artificial intelligence.

53:07Who sound very similar when they're taking a strong stance against AI. They are not staying up nights worrying about working people. In my view, they want even more wealth and they want even more power. AI can be used to make workers more productive, to encourage and multiply their labor. That's a good thing. But it must not be used simply to replace them. Most jobs should be reserved for humans. This is going to transform our world. It's going to transform our economy. It means millions and millions of workers are going to be displaced from their jobs. And I was recently at this dinner and had this interesting interaction with a multimillion dollar mega donor to the Democratic Party who, like a lot of Democrats, doing a lot of soul searching.

53:55And eventually in the conversation, I asked him, what do you understand is the Democratic Party's position around AI? and he said to me, I think that the Democrats want to adopt whatever position they need to to get elected and get back in power. And so I was wondering from your position, how do you see the landscape? How do you see the parties dividing up? Do you think that there is right now a winning pitch to be delivered that is on the one hand good policy and on the other hand good politics? I don't think we know the winning political pitch on AI. And with all due respect to your cynical rich friend, I have spoken to a lot of the Democrats running, plausibly running in 2028 about this.

54:44I've spoken to other Democrats about it. And I find them to be in the place that I find many people to be in, which is wrestling with it and not sure what to do. Aware that we're probably going to have labor market destabilization that's going to be very, very socially dangerous and destabilizing, but not sure they have a good policy to handle it. Aware that you want to figure out some answer to what education should be for and how it should work, but not knowing what that answer is. Is it more AI? Is it no AI? Is it banning it? Is it trying to make new products out of it? So it's like everybody's got their own tutor.

55:19I find them genuinely wrestling with this. I don't think people know the right position in the sense that they know the politics because of course there is no right political position yet in 2025 which is where we still are i mean i expect the politics of i to be different in 2028 but unpredictably so and i think that this is the hard thing about trying to wrap a politics around a technology moving this fast where even the people creating it don't fully understand what it is they're doing and where the primary pressure is coming from China. I will say that in practice, it does appear that AI is splitting the parties in a fairly traditional way, which is to say the Democrats are more comfortable with regulation and more comfortable doing things like export controls on NVIDIA chips and coming out with complicated executive orders about discrimination in AI and all the things you want firms to be doing in terms of offering up information.

56:28And the right, you know, since Trump has gone into office, it's turned out to be very, very deregulatory all the way down to trying to push preemption of state efforts to regulate AI. Sometimes the right likes to suggest it's very federalist and, you know, once a thousand flowers bloom in American policymaking. But in this case, it's been a real effort to make it so states cannot act on AI. So, yes, you have some of the odd bedfell coalitions of the populace, at least rhetorically. but I think in practice you're seeing a Republican party that leans towards deregulation and allowing the corporations to move fast and a democratic approach that is not highly regulatory but is much more trying to create the architecture for those regulations to snap into force if they need to and you know I expect that to continue I would actually say that I expected there to be more continuity between the Biden and Trump administrations on this than there actually was.

57:24All the way down to, I'm like quite surprised to see the Trump administration now selling advanced chips to China because I thought if the Trump administration believed in anything, it was retarding China's progress on something like AI. But Trump himself is so all over the place and so easily swayed by corporate leaders that, you know, even that is not, you know, even like sort of anti-China approach of the GOP or the Trumpist GOP is not proving itself to be stable. All right. So final question. So if you believe, as you said you did, that this is really in flux, that you are expecting things to change in politics, just as they're going to change in the technology, how is it that you're forming your views right now?

58:05Because a lot of people who've listened to this podcast have been introduced to a sci-fi movie plot for the first time. And on the one hand, you hear people like Jeffrey Hinton or Joshua Bengio, who are scientists who obviously understand this technology more than me, more than the layman. And they're really concerned. They're quitting their jobs to go warn the world that we need to get ready. They're hearing from people like Eliezer Yudkowsky or Nate Soares, who have decided to dedicate their lives to trying to stop us from taking this AGI step. But they're also hearing compelling arguments coming from, you know, Reid Hoffman and these techno-optimists who really think that we might be able to solve a lot of the world's problems, usher in a better world.

58:54How are you as a thoughtful person, because I think of you as a very thoughtful person, Ezra, how are you wrestling with all of these competing views and trying to formulate your own instead of just parroting some view of somebody else who votes like you out there. Like on a visceral level, what are you doing? Like how are you feeling when you hear these? Like how do you form views with something that's seemingly so strange and perhaps so consequential? I mean partially I am allowing my views to be soft and fluid and in flux. my views change as the situation appears to me to change I listen to you know a lot of people on podcasts and you know keep up with Dworkish Patel and Tyler Cowen and people who feel very culturally inside this world I use the systems myself I don't think there's any better way to have a felt sense of it than to actually be using them and trying to be using them fairly frequently and you know I try to listen and read I mean I don't think there's a great singular answer, but I think you should be trying to take in a lot of information.

1:00:07Some of it coming from people who are very expert, but some of it just using it and being alert to how other people use it, you know, reading the reporting and how other people are using it, trying to see society a little bit as a whole here. And I think because for a long time, I've been very interested in the philosophy of technology and critics of technology going back to pretty critics of new media technologies like Marshall McLuhan and Neil Postman and Walter Ong. that I do think this fits into, it's not going to be exactly like, and in ways maybe more consequential, more fundamental, but I do think this fits into a lineage of new mediums.

1:00:42And we have to think about not just what the content is, but how the medium will change us, how it will change us to have these, at least for now, endlessly subservient, highly capable, always on, always trying to please little robot companions. I mean, if nothing else, what does it change in a human being to be able to so much more often escape the friction of other human beings and their needs and their desires and their wants and disappear into the relative comfort of extraordinarily powerful system that when you send it a picture of your incredibly messy desk wants nothing more than to be able to help and has inexhaustible patience for helping you?

1:01:30what does that do to your sense of all the people in life who don't have an exhaustible patience for you you know i try to remember that before this changes anything else it's going to change our expectations and change what we're used to and all the more so for kids who are not going to be used to anything before this and so i think that's my framework for it that the key question of AI isn't just how the AI changes. That's what we spend the most time talking about, but it is how the AI changes us. Technology, the tool always acts upon the user, and all the more so when the tool is built to mimic and manipulate the user.

1:02:25Thank you for listening. We'll be back soon with more reporting and more perspectives on AI. We'll hear from some skeptics, from those on the political right, and much more.

1:02:40The Last Invention is produced by Longview. To learn more about us and become a supporter, you can click on the link in our show notes or visit us at longviewinvestigations.com, where you can also send us an email and let us know what your AI questions are. Thank you to everyone who's already written in, and for all of you who have shared this podcast with your friends and your communities. We'll see you soon.

1:03:08You can't reason with the sun. Trust us. We've tried. This summer, it's time to put that angry ball of fire on mute. Columbia's OmniShade technology is engineered to protect you from the sun's harsh rays that can burn and damage your skin. The sun is relentless, but so is our gear. Level up your summer at Columbia.com to spend more time outside and less time slathering on aloe lotion. You're welcome. Columbia. Engineered for whatever.