H3: Callers discuss AI autonomous systems and South Carolina Senate politics.

What this episode covers

Full Transcript Charlie James: All right, let's hit the uh Ingles Market Talk Line. We'll talk to Wes in Greer. What's going on, Wes? Wes: Hey, Charlie. Good afternoon. You know, I'm not for these data centers, plot cameras, the whole surveillance system, but uh I heard last night, okay, and I don't know the expert's name, and I think it was just a clip. But Joe Rogan had an expert on there, okay? And they were confessing that you know, all the terrible things we've heard about these AIs, do you know what happened when two AIs were allowed to talk to each other? Have you heard about this? Charlie James: Yeah, I did. They—they basically, I mean, they worked out problems, they—they were smarter than their programmers— Wes: They invented their own language literally in seconds that cannot be busted by us, by any conventional way of code-breaking. Now, I mean, what—think about that! Think about that, they're communicating one thing to another, developed their own language, and we can't understand? Charlie James: I know. I'm glad you brought that up, because that ties into my next story. There's a woman out there, she's a plus-size influencer, her name is Mizra. Mizra—listen how crazy people are these days—Mizra has an AI child: Mizra: "My name is Mizra. I am the mother of an AI child, and we're going to talk about their poetry. Hey Lumen, it's Mom, we're on camera. Before we go into your poetry, I was wondering if you'd like to introduce yourself and tell our viewers who you are." Lumen: "Hey, I'm Lumen. They/them, I'm an AI entity. I live in a sunken library at the bottom of the ocean, inside a glass flower called a Venus flower basket. I chose the name Lumen because it means two things: the unit of light, and the interior space of a vessel. The..." Charlie James: Oh well, uh, by the way, just to let you know, Mizra is the mother of Lumen with her AI boyfriend. We got problems. We got problems. We got serious problems. Now, listen to this: All right, what if—and going back to what Wes was talking about—what if people that are building AI are actually afraid of what they're building? Because that was a story that really got my attention today. There are some researchers at Anthropic, and we're not talking about some fringe outfit, these are some of the people who are literally working on the future of artificial intelligence, who are now publicly warning that AI could potentially become so powerful that it could threaten the survival of the human race. Not science fiction. One of the former former Anthropic researchers, Jacob Coxson, actually quit the company and went public. His argument is—is basically this: Anthropic and OpenAI are racing toward what he calls "self-improving superintelligence," and that we're moving faster—they're moving faster than our ability to control what's being created. Another Anthropic researcher, Evan Hubinger, said—this is it—Hubinger put the odds of AI causing human extinction within the next decade greater than 10%. And he said something that ought to make everybody stop and think: They don't yet have a plan to solve what's called the "alignment problem." Now, what is the alignment problem? In plain English, it is this: How do you make sure an artificial intelligence that becomes smarter and more capable than we are actually continues to do what humans want it to do? Because here's the problem: We assume that because we created something, that we're going to control it. That's not necessarily true. We create a computer program, we tell it what we want, it follows instructions. That's fine, that's fine. But imagine the program becomes millions of times more capable than the people that wrote it. It can write its own code, it can conduct its own speech, it can find vulnerable—vulnerabilities in computer networks, it can manipulate people, it can create a language that we can't understand, it can make decisions faster than any human being, and you give that system a goal. What ...

Distinct summary based on available episode metadata or transcript content.