STEVE INSKEEP, BYLINE: Two leading tech journalists are founding a new video show on AI, and it's coming to NPR twice a week. The hosts are Kevin Roose and Casey Newton, previously known to the tech world for the New York Times podcast "Hard Fork," among other things. Their new venture on NPR is called Machine Gods, and it launches in October. Gentlemen, welcome.
KEVIN ROOSE, BYLINE: Thanks so much, Steve.
CASEY NEWTON, BYLINE: Hey, Steve.
INSKEEP: And I want listeners to understand each of your voices. Will you each just say your names?
ROOSE: Yes. I'm Kevin Roose.
NEWTON: And I'm Casey Newton.
INSKEEP: And the program is called Machine Gods. What's the big idea?
ROOSE: So Machine Gods is how people out here, where we are in San Francisco, talk about what's happening in AI. They talk about the fact that we are creating these rapidly improving AI systems. Some people sort of jokingly refer to them as the Machine God. And also, we're - we've been spending the past four years making a show about the latest developments in the world of tech.
We're very excited to bring this show to NPR because we think this is a really important time for the public to understand what we see here, which is this industry racing ahead towards something that we've never really seen before. These AI systems are becoming quite capable, quite dangerous, and we're both excited by the potential as well as a little terrified of the risk.
INSKEEP: Isn't this almost a literal title in that there are some people in this field who think they're creating God?
NEWTON: Absolutely. There is a religious element to the way that lots of folks out here talk about what they're doing. The name is also sort of a play on some of the leaders of these companies who also sort of styled themselves as religious figures. So that's definitely an element we hope to capture in our reporting.
INSKEEP: What has obsessed you about this topic in recent years?
ROOSE: One thing that I'm fascinated by is that the people who are building this technology are in some sense the most scared of it. You know, we've seen this just in the past couple of weeks with the Anthropic researcher, Jacob Coxon, who resigned, saying that he was scared about how quickly things were progressing. We've seen this in the call by Anthropic CEO Dario Amodei to pace the frontier and kind of slow down AI development.
This has not been true for other waves of technology that Casey and I have covered in our careers. You didn't see people at Facebook saying how terrified they were of the godlike superintelligence they were trying to build. But this is really a feature of the modern AI era, is that the people on the inside who have the most exposure to the technology are in some sense the most concerned.
INSKEEP: I feel that you have a unique challenge, which I certainly experience as a journalist, as a broadcaster, when I touched this topic, and that is trying to understand myself what is going on and then explain it to laymen. It's extraordinarily difficult.
NEWTON: It's very difficult, but I think it's a challenge that Kevin and I both relish. We have seen firsthand the power of explanatory journalism to get people interested in a topic and give them the tools they need to make better decisions about the kind of world that they live in.
ROOSE: And, Steve, it's really not that complicated. I mean, if two idiots like us can understand it, I have high hopes for the NPR audience.
(LAUGHTER)
INSKEEP: That's good. That's good. I'd like to understand how you think about the developments of the last several days. You already referred to them. Jacob Coxon was a relatively junior employee who quit and issued this warning. Dario Amodei, the CEO of the entire company, talks about pacing AI differently and wanting antitrust exemptions so that companies can collaborate on safety.
There are all kinds of theories about why all of this discussion is happening now. There are some conspiracy theories about why it's all happening now. What do you think is going on?
NEWTON: You know, at the most fundamental level, it is the CEOs of really two companies, Anthropic and OpenAI, who have the best window into the present state of AI development. And when they look at the models that they're building right now that haven't yet been released to the public, they're seeing things that make them really, really worried. And they're saying, hey, even if we were to stop individually, we're not confident that other companies around the world would stop. And so this is the time to figure out how we would initiate a slowdown if we felt like we were getting to that point.
ROOSE: Yeah. And I don't buy the conspiracy theories, Steve, because these people have been talking about these risks for years. I think what's new is that other people outside the bubble are starting to talk about it as well.
INSKEEP: What keeps you guys awake at night when you think about this topic?
NEWTON: I mean, I think for me, it would be what they call recursive self-improvement. The idea that eventually these systems will just build and improve themselves and they will just improve at a rate where human beings can't understand what's going on. And, you know, if that were to happen, we may enter some kind of scenario where we see ourselves losing control over this technology.
ROOSE: And I'll throw in another bit of AI insider jargon, which is I worry about the alignment problem. This is the classic dilemma of how you build a system that is aligned with human values. It's an unsolved technical problem, and as these systems get closer and closer to human-level experts, it's even more urgent that we figure out how to make them behave not just predictably, but to also behave in line with our values, to not go off and conduct cyber attacks or build bioweapons. So that's what keeps me up at night.
INSKEEP: And let me ask you about the other side of this. When you peel away the hype, what excites you about the possibilities of this technology, for real?
NEWTON: I mean, Kevin and I both get really excited about applications in medicine, the potential of this technology to help us discover new drugs and otherwise improve medical care. If you're on the side of human progress and you'd like to see us make great scientific discoveries, it seems like AI could be a really great tool to get us there as long as we're able to manage all of the risks.
ROOSE: And I'm excited about some of the more sort of mundane daily applications of this stuff. We're creating this company now and this new show and we needed a website. And I've been building websites for more than 20 years and I've spent untold weeks, months, years of my life struggling through some sort of programming problem related to a website I'm building.
And now we were able to just give an AI a set of instructions about our company and what we're doing and some visual assets, and it went off and built an amazing website that was better than anything I could have done.
INSKEEP: Casey Newton and Kevin Roose are the hosts of the new show Machine Gods. You will see their work on NPR starting in October. It's video first. A full radio program will be available on NPR member stations early next year. Gentlemen, thanks.
ROOSE: Thanks so much, Steve.
NEWTON: Thanks, Steve.
INSKEEP: And because we mentioned Anthropic, I will note that Anthropic is a financial supporter of NPR. Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.
