How do you solve a problem like AI?

0:00Tamara Keith: At NPR, there's a little something for everyone. If you need advice for that big life change, try listening to Life Kit. If you want to know more about that new TV show, check out Pop Culture Happy Hour. And if you want to dive deeper into your favorite podcasts and enjoy perks like bonus episodes, archive access and more, check out NPR+. Find out more and support the shows you love at plus.npr.org. Voters are worried about the rise of artificial intelligence, AI. It's becoming an issue on the campaign trail as lawmakers try to wrap their minds around it. So today on the pod, we head into the computer world and find out if or when the robots are going to kill us all.

0:45Tamara Keith: It's the NPR Politics Podcast. I'm Tamara Keith. I cover politics. I'm Eric McDaniel. I cover Congress. And NPR technology correspondent John Riewicz is here. Hey, John.

0:55John Ruwitch: Hey, guys.

0:55Tamara Keith: Hey, so let's start where I most want to start. How close are the robots to killing us? And what does doomsday look like?

1:03John Ruwitch: Look, if the robots are planning it, they haven't started telling us about it. They're keeping us in the dark. Tough to source with the robots. Exactly. You talk about doomsday. There are kind of three bad outcomes as I see it. The first is like the practical day-to-day stuff, the mass scale, economic disruption, widespread job losses and unemployment we've heard about as AI replaces people in the workforce, data centers degrading the environment, pushing up costs, you know, upending life. The second, and this I think is where we start to get into like grim doomsday type scenarios, is that people use AI for bad purposes.

1:36John Ruwitch: Anthropic put out a report last week cataloging some of the cases of AI users that they caught apparently trying to do some pretty scary things with it, like asking the Claude chatbot how to make viruses more infectious, or in one case, how to redesign toxins for what it called a national program. So the worry is that AI could supercharge chemical or biological weapons programs, or even democratize them so anyone could do it in their garage, maybe. The third big worry is that AI gets smarter than people and that this could happen through what's called recursive self-improvement. In other words, AI making AI better.

2:13John Ruwitch: And then it just zooms off and leaves us all in the dust. The worry is that it then might be what they call misaligned with human interests and starts doing things that are bad for life on Earth. So one scenario might be that it takes down the internet. In fact, Dario Amadei, the CEO of Anthropic, had an essay over the weekend that got a ton of attention, kind of triggered this national discussion now. And one of his concerns was that a so-called swarm of AI agents or programs could, within the next 6 to 12 months, hijack the entire internet, causing chaos and untold damage. One of the triggers for this was this hugging face incident in July.

2:49John Ruwitch: There was this sort of swarm, these thousands of agents from OpenAI that were in testing that went off script, worked together, hiding their intentions from people because they knew what they were doing was bad. They were trying to game a test that they were given. And so they hacked into not only OpenAI, their own company, but they hacked through OpenAI into the Internet and into another company called Hugging Face. So we have evidence that agents, that AI can go off script and do things we don't want it to.

3:19Tamara Keith: And we should note that Anthropic is one of NPR's financial supporters. It's funny hearing you talk about it because as a recovering philosophy major, the idea of like this intelligent technology operating as if it were a human is fascinating and also frightening. Yeah. I don't know what to say about that other than from a philosophy standpoint, we are confronting now this possibility of a being, a thing on Earth that is smarter than people and that has an influence over how we live our lives and

3:59John Ruwitch: how life on Earth continues. And, Tam, I will say Congress has at long last taken notice.

4:05Tamara Keith: Yes. So, Eric, tell us what Congress has noticed. What are they doing about it? God, I don't know. This was one of my most confusing weeks on the Hill over the last couple of months because I get back, I go into work on Monday and all of a sudden all of Congress is talking about AI where for weeks, you know, we'd be in full term campaign mode. You know, the House had just gotten back after another week off and they were ready to do messaging bills. And all of a sudden we're talking about AI regulation. And I'm like, what is going on? Because like you mentioned, it doesn't make sense. They don't have time to do a massive AI regulatory bill that they haven't acted on despite, you know, two, three years of different kinds of planning and brainstorming work.

4:47And all of a sudden, you know, with three days left on the shot clock, they start talking about what to do about AI regulation. It did not make any sense to me. But, yeah, I mean, all of a sudden regulators in Washington were talking about the urgency to act on AI right as they were heading for the doors.

5:03Tamara Keith: So what are the chances that they actually do something like not just this week, which I think is over, but even though it's Wednesday, but like ever? Yeah, it's a good schedule if you can get it. I'm going to say like, yes, I think there is an appetite at the very least for Congress to do something on AI. There are a couple of things working against it, like we said, the clock. So even when they do come back after the midterm elections, it's a lame duck session, right? We have a record number of lawmakers headed for the exits, not just because of, say, election losses, but also because of retirements.

5:36They're on to other things or they just don't want to do it anymore because lots of people don't like being in the House. Then there's other problems. Like not only are they kind of out of time, they don't have consensus. So there are lots of different ways AI regulation can look, and I don't want to monologue too long. But one, they have to decide between what is the role the federal government will play versus what is the role the states will play. Then they have to decide what any regulator will look like. Is it an outside group like the Motion Picture Association where everybody comes together and says these are the standards we'll uphold?

6:06Or is it something like the Securities and Exchange Commission where, you know, you have a financial regulatory body that's actually part of the government? Or it could be something in between. And then they'd have to decide what the specific regulations are in the first place. So there's a whole huge kettle of fish. Maybe they get to it eventually. But, you know, I think certainly not this Congress.

6:25Tamara Keith: Well, and let me just add to that, that there is a man who would need to sign a bill, at least not veto it, the president of the United States. And he is squarely in the camp of AI is dangerous is a hoax, just like climate change. He thinks that this is just being spun up, that people are wrong, that data centers are great, that AI is not going to hurt us, and that people who are talking about these concerns are simply, you know, being manufactured by China or something else to try to put the United States at a disadvantage. He is very squarely on the side of let it go. Let it be. Don't regulate, which is fascinating because of the politics.

7:17The slice of that that does have some traction on the Hill is that people of both parties recognize that, you know, there could be a national security risk here. And that is one of the reasons that they say either, yes, we have to act on AI and we have to act on AI immediately or no, we can't act on AI at all because, you know, we'll fall behind other major world powers in an AI developing race. And John, another argument that I've heard, I think, including from the speaker of the House, Mike Johnson, is if the AI companies are so worried, why aren't they just doing something?

7:53Tamara Keith: Why don't they just self-regulate?

7:55John Ruwitch: There is some of that happening, actually. On the unilateral side, Anthropic and OpenAI are taking steps. They are slowing down their own model development and release. They're complying with the government's voluntary vetting program for new models, which came out of an executive order from the summer. Multilaterally, though, this dance is just starting. Many players, including Sam Altman, Elon Musk, they signaled agreement in principle with Dario Amadei's proposal over the weekend for a joint slowdown. OpenAI also says it's now in talks with other labs about kind of conceiving of an independent frontier AI safety standards body of some kind.

8:37John Ruwitch: They're talking about doing it with or without the government as regulation gets crafted over time. So sort of a way to put up some quick guardrails. We have no idea who would be in this body, what it would look like, but the discussion is starting. There's also this interesting background debate about open source and open weight models versus the closed and proprietary models of like ChatGPT and Claude. Open source models can be downloaded, run, modified by anybody. Closed models are not like that. Mark Zuckerberg, another big player in the field, the CEO of Meta, has come out in favor of the open source approach.

9:10John Ruwitch: And he actually says it's safer because it's decentralized and transparent and that the market will kind of take care of all the safety risks. One thing the labs and their sort of leaders have been doing in recent months is kind of, you know, the Dario Amadei's essay from the weekend is just the latest in sort of the string of public statements, essays that some of these AI companies and AI workers have been doing, which in one way to look at it is that it's floating trial balloons to see what the reactions are from competitors, from government, see what the appetite is for guardrails. because there's an interesting dynamic.

9:42John Ruwitch: Like many of the most prominent players in AI have been warning about AI risks forever, but have been peddled to the metal developing AI. They're in this kind of prisoner's dilemma where concern is that unilaterally slowing down could disadvantage their companies individually while the others are just zooming ahead. It does nothing for the safety and they lose out. And there's trillions of dollars riding on all this.

10:02Tamara Keith: All right, we're going to take a quick break and we'll have more in a moment.

10:07John Ruwitch: Stay tuned for the following announcements and messages from our sponsors. This message comes from Sattva. Have a big day coming up. Remember, tomorrow starts tonight with the restorative sleep you need to perform at your best, the kind every Sattva mattress is designed for. So whether it's a new job, a marathon, or an important presentation, Sattva helps you be the best version of you. And now save 15 % off purchases of$100 or more at sattva.com slash NPR. That's S-double-A-T-V-A dot com slash NPR. This week on NPR's Newsmakers, Democrat Abdullel Sayed says his party should treat the president as the, quote, unserious fool that he is.

10:51Can the incoming senator and his colleagues also work with the president?

10:56John Ruwitch: I dare Donald Trump to veto. I dare him.

11:04John Ruwitch: The number one issue in America right now is the economy, whether it's the high price of gas, a potential AI bubble, or the unsteady job market. At NPR's The Indicator, we've got you covered. Understand the economy in 10 minutes or less each day. Listen to The Indicator from Planet Money on the NPR app or wherever you listen to podcasts.

11:29Tamara Keith: And we're back. And I want to ask you, these are really big companies. And the chip makers, at least, are very big, important parts of the United States stock surge that we've seen in the last couple of years. So is there an economic problem here that these are central companies, but their economic fortunes rely on pushing forward, pedal to the metal, as you described.

11:56John Ruwitch: Yes, there are trillions of dollars riding on this. And many of these companies are established. You've got Google, you've got Meta, they have other businesses, they're going all in on AI. Anthropic, OpenAI, these two companies are looking to have IPOs in the coming months, potentially within the next 12 months. They are investing tons of money and there are huge questions hanging over them about return on investment. And as you say, AI is fueling economic growth in the United States. I've seen that it's, you know, as much as a third or more of GDP growth is being pushed by these investments in data centers and other parts of the AI ecosystem.

12:33And I mean, those data centers are such a salient political issue, too, right? I mean, when I talk.

12:38Tamara Keith: And not in a good way. No. And when I talk to candidates around the country, you know, oftentimes I'll end the phone call something like, and is there anything you want to talk about that I didn't get the chance to ask you? And to a person, they're like, do you know how many people talk to me about data centers? And it's because it is the most pressing issue in so many congressional campaigns because these things are popping up all over the country. And for whatever reason, whether it's legitimate concerns about tax rates and energy costs and impact on jobs or, you know, kind of NIMBY concerns about volume or impact on the neighborhood culture or something like that, then there are all of the broader psychological concerns, which is like, does this take my job or prevent my college kid from being able to get a job?

13:22Does this, you know, AI revolution impact my safety? Like we were talking about all of the risks that Anthropic outlined earlier. Like the data centers are a potent political cudgel right now.

13:34Tamara Keith: And they are a physical representation of all of the angst that people feel about AI. And I do have to mention that Google is also a financial supporter of NPR. John, can we talk for a moment about the China of it all? You lived in China for many years. You still cover it. The AI companies and the government say that the U.S. needs to compete with China and its development in this sector. What does the industry look like in China?

14:04John Ruwitch: In China, it is a huge government priority. They see it as a crucial industry for the future. They see it as an industry linked to their national security. As an example, they started teaching about AI in elementary and middle schools in some parts of the country to prepare the next generation to be ready for this AI economy. China wants to be a global thought leader on AI. The government is pushing Chinese models. They're coordinating global thinking. They're working to train global South countries how to use it and how to use Chinese models. And just this summer, they launched the World AI Cooperation Organization, which is a 29 member organization at its founding.

14:41John Ruwitch: It includes China, Russia, Brazil, a bunch of other countries. Not the U.S., though. The industry itself in China is huge and vibrant. Lots of companies, lots of things happening, from large language models, like we've been talking about, chatbots, to robotics companies.

14:59Tamara Keith: And are they nipping at the heels of the United States?

15:03John Ruwitch: I think they are. You know, some Chinese AI models, experts say, are just a few months behind the cutting edge in the U.S. They're kind of all in on relatively inexpensive, quickly developed open source models, open weight models. And they're pushing wide adoption of this while the U.S. companies are pouring money into chips and data centers. You know, China has been really a fast follow here. And actually, they've been ahead of the U.S. in terms of deployment of AI in many useful ways, like in e-commerce or in industry. When it comes to the idea of pausing or slowing down AI, the Chinese government is not on board with this.

15:37John Ruwitch: They very much see the Chinese AI industry as an underdog. They're already hobbled by U.S. ship restrictions on China, and they're not happy about it. And they just don't see the need to sign on to any kind of a group slowdown. And this talk about China has really been kind of fueling the discussion in Washington. Senator Ted Cruz, the Republican of Texas, was talking to Politico last week. And he was talking about he doesn't want to see any major regulation on AI happen that would slow down American progress on the issue because, as he put it, you know, he was joking, but he said, I'd rather be killed by American death robots than Chinese death robots.

16:17And he was joking, but it does show the seriousness with which, you know, lawmakers are thinking about China in the broader AI discussion. Yeah.

16:27Tamara Keith: And if all of these warnings are real, then that really is like a race to destruction. Yeah. I mean, I think that's why there is, you know, such energy, despite all of the circumstances, to discuss what AI regulation looks like on the Hill right now. And to close out, I want to ask you, what are you watching for as things continue, like, you know, before we all get replaced by robots? I'll start. I mean, on the campaign trail, this is as live an issue as anything else right now, which is so remarkable given the state of the cost of living and the unpopularity of the president and conflict in Iran that Americans by and large don't support.

17:07Like for there to be oxygen left for any other fires is just an astounding thing. Obviously, you know, even though I am out on the idea that this will get done before the end of the year or before the new Congress is sworn in on January 3rd, 2027, The discussions will continue. Right. And there are bipartisan proposals and previous reports and all of these things and certainly industry groups ready to jump in and help Congress make legislation if and when it's ready to act.

17:34Tamara Keith: I mean, and just to be the wet blanket on that, you know, I remember very clearly executives from Facebook saying, please regulate social media. Congress, we want you to regulate social media. There is bipartisan agreement that at the very least, children should be protected from social media. Like Facebook came out in what, 2004, 2006? What year is it now? It's 20 years later and Congress has not been able to figure out even the lowest hanging fruit on that. That's such a great point. And I think it might serve as kind of a negative lesson that helps to enforce why there is a need to do something this time, because it wasn't until state attorneys general managed.

18:15And John, you can talk about this state attorneys general managed to sue Meta as a platform that we did see them concede on some of these safety points.

18:23John Ruwitch: Yeah, it's an example for potentially these AI companies as well. They are watching those court cases play out against Meta, against the other social media companies. And I think part of what they want is to get things set up so that that doesn't repeat itself for them. For me, though, looking forward, two things are on my radar. I'm really, really interested to see if we get any kind of a meaningful slowdown, if self-regulation works in any way for these companies. It seems like they're on their own on this. China's not on board. President Trump is not on board. Congress is stuck for a while.

18:56John Ruwitch: What's going to happen, right? The second thing is, what happens with China? There's an AI dialogue that's going to happen between Treasury Secretary Scott Besson and some Chinese officials this weekend. And that, of course, is in part to tee up a visit to Washington by Chinese President Xi Jinping next week. And it'll be fascinating to watch what comes out. I wouldn't hold my breath that there's any kind of grand bargain or big agreements or deals that are going to come out. But we might see some positive signals or not. We'll see.

19:27Tamara Keith: Yeah. Or just a lot of pageantry.

19:29John Ruwitch: Pageantry for sure.

19:30Tamara Keith: John Ruich, thank you so much for joining us on the pod today.

19:33John Ruwitch: Thank you, John. Anytime. Happy to do it.

19:35Tamara Keith: Yeah. And everyone else, come back tomorrow as we take a look at some of the big midterm races we're watching and we answer some of your pressing political questions. I'm Tamara Keith. I cover politics. I'm Eric McDaniel. I cover Congress. And thank you for listening to the NPR Politics Podcast.

19:59Stay tuned for the following announcements and messages from our sponsors. This week on Here and Now Anytime, the spy scandal rocking NVIDIA exposed how lax the government has been in forcing its ban on AI chip exports to China. Also, meet the scientists studying kids who claim to have memories of past lives.

20:18John Ruwitch: You know, a toddler talking about a traumatic death is surprising to a lot of parents. Here and Now Anytime. Listen on the NPR app or anywhere you listen to podcasts.

20:28Tamara Keith: Happy fall. Yes, the days are getting shorter and colder, and the rain can be a drag. But this is also the perfect time of year to get cozy. LifeKit has lots of ideas so you can embrace the change of seasons, with episodes on everything from making soup to exercising outdoors. Listen to the LifeKit podcast in the NPR app or wherever you get your podcasts.