Anthropic CEO sounds alarm on AI risks: What to know

No warnings within the artificial intelligence industry have revived a long-running debate over whether advanced AI could escape human control and ultimately threaten humanity's survival. It's all prompting a conversation about doing more and whether governments should get involved and regulate AI. Man oh man, it's a conversation that has just taken over every dinner table. Everybody's reading about this right now. It's probably the biggest story in the world. So we wanted to bring in a local voice on this. Justin Gramans is the managing partner at Recursive Awesome, which is an artificial intelligence consulting and training firm based here in St. Paul. Justin, thanks for coming on the show. Yes, thanks. Thanks so much for having me on. OK, so over the weekend, the CEO of Anthropic said he thought the industry needed to reduce the speed of its work, warning that AI might be able to take over the Internet in six months to a year unless safeguards are put in place. OK, obviously a hotly contested issue. You know, do you put in safeguards? Do you not? I just I want you to respond to this, these warnings and then this proposal that maybe we should rein it in. Yeah, no, it's it's a very, very touchy debate here. I mean, just think back to when the Internet came out and there was a lot of thinking with regards to how can we put regulations and other frameworks in place so we make sure that the Internet doesn't actually do harm. And so we're sort of at these crossroads again, I think, right now with artificial intelligence. The game is definitely higher stakes than it was back then, arguably, because there are a lot of things that it's artificial intelligence can do. And I think, you know, what you saw this past weekend with regards to the post by the CEO of Anthropic was was really trying to set an alignment with regards to not only Anthropic, but also OpenAI, also also Google, Microsoft. You're seeing kind of trying to build a coalition, I guess, of these companies so we can all sort of address this problem together. Listening to one side versus the other side, you're going to see all sorts of things. You know, everyone's sort of motivated by different factors with regards to how they can kind of capitalize, I guess, on what's going on in artificial intelligence. And so, you know, what what's happening right now is I think the industry is trying to come together and then obviously bring in enough government regulation that it doesn't slow us down. Well, and that, of course, is the concern. You'll hear it again and again. But if we slow down, China will not. Yes. And that's a completely fair assumption. And one of the things that that was has been brought forth is is let's make this be an international, essentially, pact. Right. So we're looking to not only the United States, but also China and other leaders of these other Chinese AI companies to sort of get together and sit around the table and figure out this is the best for humanity. Right. And that's really what what Dario is saying in his in his framework that he starts to outline here is, is that we'll do it if other people do it. And that kind of remains to be seen. But it is sort of viewed as something that it's a global framework that we're going to have to work on together. Well, it has generally been believed that China would never get on board with something like that. But The New York Times is now reporting that China may view AI as a major threat to the ruling Communist Party. I don't know if you've read that article, but I'm interested to get your take on that. Well, yeah. I mean, again, like I say, regardless of if you're a communist or capitalist or whatever society you live in, we're talking about, you know, ultimately human lives. Right. And so I think any any government wants to make sure that its people are protected from harm. Right. And so I think cooler heads will prevail in this particular case, if we look at it through that lens, that that that this is something that it's not only, you know, you think about nuclear weapons. Right. I mean, as I was thinking about this, they can be used for all sorts of bad things. But yet, you know, we decided to kind of put together, put our differences aside with the Soviet Union and figure it out. OK, this is the right way for us to come together as a world. And of course, there are, you know, there are theories out there, too, that maybe some of these large A.I. companies have concocted all this so that these regulations are put in place and then it slows down growth and they don't have more competition out there. Do you have any thoughts on that? Yes. I mean, there's always conspiracy theories out there. You're always going to get people that are going to kind of hear what they want to hear in the news and ignore what they don't want to hear in the news, as you maybe did see. And open A.I. was very public when they came out. They basically had models that were running internally within their labs that got out. Right. And they were very, very open with regards to the hugging face hack that happened. And so that that is true. I mean, I don't I believe I don't believe that an organization will come out and fabricate something like that. So the question is, is can we go from something like that that happened to a larger leap, which is kind of what happened over this past week is, oh, my gosh, what if. And that's what I feel like is happening right now. It's we had a couple of cases, you know, obviously, where models kind of went off the rails and went off and did some pretty malicious things, right. Some things that we didn't really expect them to do and kind of got outside of their bounds of what we thought that they would do or what their capabilities were. And now we're saying, honestly, we're saying, what if what if this got taken to the extreme? And I think that's really what's been blowing up is is really, you know, potentially kind of trying to forecast the future. And no one knows the future. Well, it feels like Hollywood in real life. It feels like the Matrix, the movie is just playing out before our eyes right now. I know that that's been a parallel that many people have made in the last couple of days. But it's certainly certainly an interesting discussion. How do you feel about safeguards? I mean, do you think that they would work? I mean, when we're talking about A.I., I have seen, like you mentioned, some models suggesting that they could take over in six months, that it could get out of control and out of our hands in a matter of months. Do you think at this point, as far along as we are, could safeguards even stop it? Yeah, I know. So a couple of things. Number one is the population is seeing a certain level of models. Right. And so we're looking to these labs to kind of give us a preview into what could happen in the future. So I take sort of half of, you know, half of the truth with regards to they obviously know things that we don't know. Right. They've been using models internally and they've been able to train these and see things that we possibly haven't seen. Now, when these models end up getting out into the public and regulations happen and actually what Dario is suggesting is actually independent audits happen. Right. Within all of these large language model providers. Right. These frontier labs. I personally think that this is a problem that we can solve. Right. And I think there is a lens that people are looking at this. With regards to, oh, my gosh, look at what happened. What if we don't do anything? And they're taking it to the extreme on the far side with regards to this could potentially, again, wipe out human civilization. I believe that we will be able to put enough safeguards in place. I believe that we will have enough authority and enough of a, I guess, a window of time here that we can actually make sure that we do this the right and proper way. Now, that's not to say that we can't just sit back and we should just sit back and let things go. And that's kind of what these people are starting to say is we need to have the conversation now. But I don't believe it's as quickly as people are saying. And they're probably saying that because they want action and they want things to move quickly. But I also think there's a little bit of, I would say, fear mongering in some ways. Yeah, I mean, the president of the United States responded today saying it's a hoax. You know, the attack on artificial intelligence, on data centers. It's all a hoax, implying that maybe this was that this is all being said and done to slow down our progress. It sounds like you don't think that that's the case. No, I don't. I don't think that's the case. And again, you've got just widely varying opinions on this. And and I think, you know, I don't think there's any reason as to why we would want to slow down the progress as an economy and as the as the United States, per se, wants to lead. I think we all have that goal of the United States leading in this new technology because it's it's not going away. Right. And so, you know, I think more and more transparency from these labs, which I will say being an artificial intelligence, you know, thought leader in the space have been been working in this for more than a decade. I think we're not so good at explaining to the general public sort of what can happen, how the systems work and some of the guardrails that we are looking at and have been looking at to put into place. So I think as an industry, AI professionals like myself need to get better at educating people on kind of like really what's possible and and what's not possible. Yeah, it really makes me think about what the Anthropic CEO said, which is he thinks that the industry really lied to consumers about the dangers. Yeah, I don't I'm not sure if it was lied, per se, because frankly, his his words, not mine. Yeah, no, no, no. But I guess what I would say is just to maybe support what anyone is saying at that level from any one of these, you know, large frontier labs is I mean, just rewind the clock back. We couldn't have predicted where we're at, you know, even 12 months ago. I think even these researchers that are working on this, they are literally learning new things every day. So I think to sort of point out a particular company and say, you messed up, you didn't understand how the future was going to be. I think I give them a little bit of a pass on this because while they are leading and creating, I think we are in like the early, early phases of this. And it's very, very difficult for any company who's even working in this space to be able to look out 12 months and say, you know, this is exactly what's going to happen. Oh, it's so interesting. And there's so many different sides to this debate. And like you said, the everyday consumer just really doesn't understand it. Yeah, yeah. And there are some great things and that we're using with A.I. today and everyone is using it as a component of what they're doing. And I think, you know, the message to those folks is, is, you know, keep keep using it how you're using it. Right. And and it will be more and more embedded in the work that you do. And there are some I think the positives drastically outweigh the negatives with any new technology, you know, whether it was nuclear weapons, whether it was the Internet. There's always going to be questions with regards to, you know, good actors and bad actors using it. So it's a phase that we're going to go through. And again, I continually kind of remain optimistic that these are things that we will collectively come together and solve. Well, it's all anybody's talking about today besides the Vikings win. So we appreciate you coming on the show. Justin Graham is the managing partner at Recursive Awesome. Thanks for joining us. I hope you'll come back. You certainly know a lot more about this than I do. So appreciate you. Yes. Thank you so much. Appreciate the opportunity. OK.