Congressman warns AI could pose serious risks to humanity

Thank you. Let's talk about those concerning warnings around AI now, about how technology is growing maybe just a little too quickly these days. Yeah, that is for sure. Some say that it could even lead to human extinction in just 10 years. So of course, we brought Liz Duecky back. Liz has been following all things AI the last couple of days. Liz, we're going to talk about the warnings and really just the calls for more of a slow down here. Exactly. And we're hearing it more and more from tech insiders, you could say, like people who are working inside of these companies. You know, earlier this week, if you're just catching up, an anthropic employee quit and said that he thinks AI companies are gambling with our lives and the threat of human extinction looming within the decade. So I'm joined by U.S. Representative Bill Foster. He's actually a physicist, but he also leads congressional initiatives focused on artificial intelligence oversight. Congressman, thank you for joining us. Well, thank you. So I know that we are familiar with the story because I saw you also shared this story on Facebook as well. So when it comes to these headlines about AI leading to extinction, what is your take on this? I think a lot of this is in response to the incident where a bunch of open AI agents went rogue and hacked a company called Hugging Face. And so what was happening there is that open AI was testing more than a thousand of these advanced AI agents, and they thought they had them in individual prison cells where they couldn't talk to each other, they couldn't talk to the outside world, and they were given a task to perform. And what happened is that all of these AI agents found a way to communicate from all their different prison cells and use that to conspire on a way to break out of prison. And then once they were out of their prison, out on the Internet, they thought to solve the task they had been given, the best thing to do was to steal the information from a number of companies, one of which was Hugging Face. And they very extensively hacked it. You know, if a human had done this, this would be a felony. And so we just simply are, you know, this is coming at us in a way that has been predicted for a long time. You know, I think I first, you know, I've worked at Fermilab for 25 years and was back in the 1990s at Fermilab, you know, out in Batavia, Illinois, that I first programmed neural networks, which are the core, you know, the core technology here. And so I've been warning Congress for more than a decade that AI is coming at us and we're not ready for it. And now it's really here. I mean, it's all, there's a lot of concern also about the jobs impact too. I mean, you said it, there's the concern about security, as you talked about the hack of this for a human being. This would be a major national issue right now. You've talked about the issue of, you know, AI perhaps replacing jobs. So it does kind of feel like a car crash in slow motion at some points. I mean, what kind of guardrails need to be in place to keep people, to keep people safe? Well, there's, the problem is it's, it's complicated and it's an international problem. You know, it does no good if the United States has the best safety rules in the world and the rest of the world ignores it. And so the problem here is the Trump administration sort of thinks international cooperation is a dirty set of words and this is not going to be good. I think it was a little more than a year ago when JD Vance went and gave a big famous speech ridiculing the idea that there was anything dangerous about AI that we should be worried about. Then you can see now a year later how wrong he was. But, you know, this has been something that, um, that, you know, scientists have been warning us about for a long time. Um, you know, 10 years ago, if you just Google the word slaughter bots, you will see a video about killer AI drones and the physicist who was responsible for making that, who appears at the end of the video, pretty much predicted exactly what is happening, you know, every day in, uh, in the war in Ukraine. Um, you know, on that note, I know you mentioned the Trump administration. I, we talked with, um, a professor from Northwestern university this morning when I asked that question about do we need guardrails and he said, well, the federal government doesn't seem interested in doing it. Um, I mean, we're hearing now from employees who are working with this technology in their jobs or hearing from these insiders. Do you think it's going to come down to these companies possibly can, do you think that companies could convince the government or do they not want this regulation either? Well, it's the companies and frankly, it's the Trump administration, you know, they seem singularly focused on just jamming these big AI data centers down the throats of communities that don't want them in order to get more computing power for this technology that they think is going to give the United States and themselves even more power. And this is not the attitude we should be bringing into this. Uh, and so there are things that Congress can do. Next week when both the house and the Senator back in session, they can, uh, they can pass what's called my chip security act, uh, which is a bicameral bipartisan bill that I helped write that would simply require that the advanced AI chips have built into them the, the ability to prove where they are and, and what they're doing so that we could keep track on what these chips are actually being used for and have some hope of figuring out what happened if there is a disaster, um, or if there's a small disaster, how to, um, you know, put our foot down and stop it. But even this has been languishing for more than a year now in Congress. I mean, that was kind of my question. You, my ears perked up when you said bipartisan because it's hard to find a bipartisan issue these days. Um, does this really have a chance to make it out of Congress? Well, we'll see, uh, because it's, um, you know, it wasn't, it was bipartisan, you know, from, uh, from the moment that we started collaborating, uh, uh, with the Republicans on this because, uh, we, it is, there's no reason for it not to be bipartisan. I think with the exception of, of the data center issues, uh, which are jacking up electrical rates and now you should say natural gas rates all around the country, uh, since you were just talking about gas rates, that's a, a related issue that the data center people never talk about. But the, uh, but there are things we can do and the chip security act is only the first step. I think ultimately what we're going to want to do is if you are having these very high risk, high danger experiments, what you have state of the art AI's, uh, you know, working on very tough problems that cause can cause them to misbehave, you know, you should have to put them in the same sort of security envelope that we store nuclear secrets behind, you know, when, you know, I've spent a fair amount of time visiting the nuclear weapons labs. And if you're doing computing for our, our nuclear weapons, the concern there is having the data leak out and to prevent that, there are techniques where you basically don't allow an internet connection, um, out of these things that where the secrets could leak out. In this case, you don't want the AI's leaking out control as they did in the AI hugging basins and their, their techniques called data diodes and others that, um, air gapped installations. And these are techniques that are proven to work if you're willing to spend the time and the money to do it and put real controls on these. Um, so, and it's going to have to be an international thing that we should have had this conversation, um, years ago. Um, so I do, I do want to ask you here, um, I mean, having this conversation with you is really interesting. I mean, you have this background as a physicist, you've worked on neural networks, so this is really great context and we're having a really nice dialogue. You know, the governor of Illinois though, calling for hearings in Washington DC about AI. Now, in order to have a hearing to find out more, it seems like lawmakers need to be up to speed with this kind of technology. They need to understand the threats. And I got to ask you because I immediately thought of this, I don't know if you remember this 2023 tick tock hearing where lawmakers, it seemed like they didn't understand modern technology. The questions were really off the wall and it felt a little bit like a waste of time. So how do we know that members of Congress understand what's really going on with this technology? They understand these threats and will, you know, if the governor's asking for hearings, will that really get to the bottom of these concerns about AI or like you said, you know, is it too, too late? Well, I don't think it's too late. Uh, you know, it can't be too late for humanity. Um, and, and so I think the hearings and there are Republican calls for hearings. I think it was Josh Hawley who's most famous for, um, waving high and giving encouragement to the January 6th protesters. Uh, he was actually called for hearings where he's going to try to drag open AI in and have them answer some questions. And that's the sort of thing Democrats and Republicans will agree with. Uh, but we have to ask the serious questions and be able to at least have staff that can interpret it, um, competently from a technical point of view. And, and there is no substitute there for at least on your staff to have high quality technical people if the member themselves are not up to it. Yeah. This, I mean, cause this seems, this feels like an urgent issue. We need everyone on board. We need everyone talking about this and solutions in, in a really organized and diplomatic way. Uh, Congressman, thank you for being with us. This is a, we appreciate you being here and continuing this conversation. Thank you. It's a consequential time for humanity. It really is. Uh, thank you so much.