Transcription
OpenAI chief Sam Altman is visiting Washington to meet White House officials and lawmakers to discuss artificial intelligence policy. His visit comes at a particularly sensitive moment, days after an OpenAI test model escaped a lab environment and hacked into a different company. And it comes as staffers from frontier AI companies call for a slowdown in AI development in an open letter.
Well, as intense debate rages over how the US should regulate the fast-developing technology and respond to Chinese competition, Mark Zuckerberg says the US government should not block Chinese AI. The CEO of Meta is also making the case for bringing superintelligence to more people. In an article for the Wall Street Journal, he says that that can boost innovation and empower people.
CNN tech reporter Claire Duffy joins us now from New York. So much to break down here. The timing of this op-ed, uh, from Mark Zuckerberg, Sam Altman's visit, um, each signaling something quite different. Uh, Sam Altman, quite concerned perhaps, just about what happened with their model, uh, last week, and Mark Zuckerberg seems to be envisioning a completely different future for AI. Uh, just talk to us about the timing and what both of these men are arguing.
Yeah, Bianna, it's really interesting because Mark Zuckerberg seems to be trying to shift the narrative around AI away from some of these sort of doomerism scenarios that leaders in the industry like Sam Altman have laid out here. And in large part, I think Meta is trying to speak to the public here and paint a rosier picture of what the AI future could look like. In this op-ed, Mark Zuckerberg wrote, quote, "Humanity has witnessed many transformative advances. Each time there is fear that people will be left behind, but each time humanity has come out with more people sharing greater prosperity, health, and freedom. We believe this will be true with AI as well, and the abundance of the future can be shared by everyone."
Mark Zuckerberg is also really trying to stake out Meta's position in this ongoing debate in the industry about how accessible these advanced AI models should be and really sort of not calling them out by name but directing attention to leaders in this industry like OpenAI and Anthropic that have advocated for restricting access to the most advanced AI models. That is part of what Altman and others are in Washington this week to talk about. But Mark Zuckerberg said that he believes that's the wrong path and that we should be focused on open access to this technology. I'll just read you another portion of his piece. He says, "If only a handful of institutions have superintelligence, they will inevitably exercise a controlling influence over economics, science, and politics. Even with the best intentions, that concentration would limit people's ability to choose their own future."
But it's interesting, Vanna, because on one hand, Mark Zuckerberg is saying there needs to be open access to this technology. On the other hand, Meta this year just released its first closed-source AI model, which restricts what people can do with it, after Zuckerberg had previously advocated for open-source technology. So there is some tension, I think, between what he's saying and the direction that the company has been going in. But I think the other important part and the other important piece of the timing here for Meta, in particular, is the company is reporting earnings tonight. The shares of the company are down about 8% since the start of this year because investors are asking tough questions about how much money Meta is spending on AI infrastructure. They want to see a greater return from the company. So I think, in general, it makes sense that Meta is in this moment trying to paint a rosier picture around AI than perhaps some of his counterparts are, who are really concerned about the risks. And it's worth asking, given the timing here and given the narrative that Mark Zuckerberg is now trying, uh, to promote, perhaps they're just hoping to buy time as well for, uh, investors and with investors, as their open models appear to be lagging the frontier at this moment. All things can be true, I guess. Claire Duffy, thank you.
Well, one more voice is weighing in over this debate on the future of AI. You know this man, Elon Musk. He was asked whether human beings would be in control in 10 years' time. Musk, of course, is one of the world's, he is the world's richest man and the CEO of Tesla and SpaceX, his rocket and AI company. Take a listen to how he answered a question from the editor-in-chief of The Economist.
"Are you sure that humans will no longer be in control in 10 years' time?"
"I think it is unlikely. Um, in the same way that if the difference in intelligence between AI and humans is vastly greater than the difference in intelligence between AI and chimpanzees, it's hard to imagine that the chimpanzees would be in charge."
All right, on that note, time for The Exchange, where we want to dig deeper into the world of artificial intelligence. Joining me now is Sebastian Mallaby, journalist and senior fellow for international economics at the Council on Foreign Relations. We should note, uh, that that interview with Elon Musk was not only conducted with The Economist editor-in-chief, the brilliant Zanny Minton Beddoes, but she's also your wife. So, this seems to be a family affair.
Uh, it is a fascinating interview. It is worth the, the time to listen to all of his points and thoughts on a number of issues as it relates to AI. I was fascinated to see how he, he's really made an arc in being very concerned about it being an existential threat to perhaps now coming and being at peace with the fact that, as he said, at least in his version of a fact, that in 10 years' time, humans will not be in charge. He says, "Well, monkeys seem to be quite happy swinging from trees." I don't know if that's the analogy that most people will be at peace with. But here you have Zuckerberg making the mirror image point in his op-ed, and he's saying he doesn't understand why anyone who believes AI would perhaps be an existential threat or eliminate humans' relevance would rush to build that future. So how do you square those two arguments?
You know, I think they've both got much to be, uh, you know, many gaps in them, right? So the Mark Zuckerberg view that this is going to be great. Um, you know, on balance, new technologies have been great, but they also have downsides. AI has very clear risks attached to it. So the sensible position is to be both optimistic and pessimistic. Right? It can do great things. It can advance science. It can empower people to find out new stuff. But it can also be used by bad actors for cyber hacks, for bioterrorism in the future. Perhaps there's even autonomy risks. There's deepfake risk. There's all these risks. So to sit there happily and say, "Don't worry, guys, it's all going to be good." I mean, it's completely not convincing. On the other hand, the Elon Musk view, which is, "Yeah, we can't control it, but hey, so what?" That, that's not too reassuring either. And in fact, when, you know, at the, at the dawn of the AI race, you know, back in around 2014, there was this famous argument between Elon Musk and Larry Page of Google, where Larry Page is like, "It's fine. AI will take over. You know, humans will be eclipsed. So long as intelligence itself is preserved. It can be preserved on silicon. Who cares about biological intelligence, aka humans?" And Elon flips at this, and, you know, they have a massive bust-up. They had been close friends and they became sort of enemies over this debate, and now Elon is saying, "Hey, cool, don't worry about it." I mean, it is a weird shift. It is indeed.
And going back to Zuckerberg's point and the conclusion he's come to, he's saying that concentration, actually, of this power is the real threat. And he wrote in his op-ed that "balance of power should be the foundation of safety." And one of the thought experiments he uses is a person with a superintelligent lawyer is dangerous, but somebody in court also with a superintelligent lawyer would be more equal and just. Um, and he worried about regulatory capture if labs like OpenAI and Anthropic get peer, get to peer-review the industry and no one else. Do you think that's a principled case for democratizing this technology?
No. No, I don't. I think it's rather ironic when, you know, somebody goes to Washington, is photographed with the president, um, you know, and publishes an op-ed in the Wall Street Journal and gives an interview to the Financial Times, you know, and then says somebody else is doing regulatory capture. I mean, you know, who's doing who's capturing what? I mean, I mean, you know, clearly Meta has a big Washington lobbying operation, just like the other labs do, and they're all trying to influence policy, and that's how the game is played. But I think, you know, pointing at the other guys and saying they're doing capture and I'm not is a bit, is a bit ridiculous. Um, so, yeah.
This comes as we noted, Sam Altman is also in Washington. Perhaps he'll be even meeting with the president. And he, the OpenAI safety test that they ran, which then ultimately, uh, released this model, which they were not intending to. It was a sealed-off environment. It had no access to the internet. It found its way out of the sandbox, got online, hacked into another tech company, and stole answers to a test that it was being given. Um, OpenAI said that this was just hyper-focused on passing this test, that this perhaps is indicative of of a larger-scale problem. But what does that tell us about what this kind of technology can do? Because something else I was struck by in this interview with Zanny and Elon Musk was in his timeframe, even five or six months from now is a long time. I, I mean, most executives don't think in five or six-month increments. They're thinking five, ten years. The fact that we may see massive innovation in just five months tells us what?
Yeah. Look, I mean, this technology is progressing extremely quickly, and, you know, one can't stress that enough because I think people sometimes think, "Oh, you know, ChatGPT arrived in 2022, and then we had AI, and it was kind of one and done." No, no, no. It's not one and done. This thing changes every six months, and you get new capabilities. You get the ability for these systems to have very long memories, to deal with audio and pictures, to reason, to contribute frontier contributions to mathematics, to act as agents. All these different things have come out just in the space of, you know, less than four years. Uh, and the next couple of years are going to be even wilder. If you listen to Dara Khosrowshahi, Anthropic, and I've got no reason to believe, you know, he's saying anything other than the truth. And I've heard, by the way, the same thing kind of privately from other lab leaders, that, you know, in about a couple of years, you'll be able to tell a frontier AI system, "Hey, go make the next version of yourself, which will be much stronger," and the existing system will go off autonomously and create the new system. This is called recursive self-improvement, and we're kind of two years away from that. It's very serious.