Can the AI Industry Slow Down?

SPEAKER_1 (0:02)

Bloomberg Audio Studios, Podcasts, Radio, News.

Sarah Holder (0:09)

The stakes are high right now.

Sarah Frier (0:11)

The stakes are high.

Sarah Holder (0:12)

And it feels like in the past week, the stakes have gotten so much higher. At least people are talking about the stakes in different terms.

Sarah Frier (0:17)

The rhetoric has gotten really heated. The capabilities have not gotten dramatically better. And I want to be clear, you know, there is no current existential threat to humanity, right?

Like, I got a call from my mom last night. Sarah, I'm just really concerned.

Everyone's freaking out. Do I need to freak out? I would say you do not currently need to freak out. This is about understanding how we need to shape and train the future models of development. It's about having the knowledge about how to align these models correctly. It's not as much about the models gaining amazing power that they shouldn't have.

Sarah Holder (1:02)

I'm Sarah Holder, and this is the Big Take from Bloomberg News. Today on the show, the AI industry says it's ready to slow down. I sit down with Sarah Frier, a Bloomberg managing editor who focuses on Big Tech, startups, and AI to discuss why Anthropic and OpenAI are asking the government for help in setting guardrails, and why they may not get it.

Dario Amodei, the co-founder and CEO of Anthropic, came out with this widely read, much discussed, much debated blog post over the weekend, laying out a plan for how an AI development slowdown could work. What exactly did he suggest?

Sarah Frier (1:46)

He suggested, first of all, that there would be some third-party evaluation of the models before they're released, checking if they are in alignment with what humans would need from them, that they wouldn't do things like escape their, their, quote-unquote, sandboxes and access to open Internet if they're not supposed to.

He also thought that there needs to be some clear collaboration with the government, working in concert with the White House to make sure that what they're doing is in accordance with what the US needs. Then third of all, he wants some global coordination. He and other leaders have frequently talked about the threat of China's development, and this is all coming ahead of Trump's meeting with Xi Jinping, and this is going to be a major topic of conversation in those meetings. I should say, OpenAI Sam Altman signed on and endorsed what Amodei was writing about.

Sarah Holder (2:41)

Yeah, he was kind of like, sounds good to me.

Sarah Frier (2:43)

Sounds great, let's do it. Elon Musk said he was in favor.

We got a post yesterday from Satya Nadella at Microsoft saying, yes, we also want to be responsible here. So everyone's saying, sure, we want to be responsible, but the actual contours of that plan is where it gets complicated.

Sarah Holder (3:03)

I want to dig in a little bit more to Amodei's plan, because one of the things he suggests is embedding these independent third-party evaluators in these AI companies to make sure AI is being developed responsibly and that things do not go awry essentially. Who could actually fill this role in the AI space, and how could you ensure their independence?

Sarah Frier (3:27)

Well, there's one company that he has mentioned called METER, which has been doing some of their earlier research on these incidents, and that company, they're a little under the radar. We haven't written much about them yet. I expect we'll do more, but that's the kind of company that could come in and evaluate, has the expertise to do it. But there may be others. I think maybe thinking about this in the context of what other tech companies have done when they're under intense scrutiny, it hasn't always worked out, right? Remember META created an oversight board of folks from non-profits and former government leaders around the world, and they were tasked with coming to evaluate META's community standards and see if they were adhering to them correctly.

Nothing really happened.

That board didn't seem to have much operational power at META. They were just kind of the appeals court for user posts. So all of these discussions are taking place, but the backdrop that complicates everything is that Donald Trump does not want AI to slow down at all, and he made it very clear over the last couple of days, this is not a problem. The robots are not taking over the world. He called it a hoax. He called it a hoax. So he seems upset that this is even being suggested. He does say, oh yeah, there do need to be some things we'll need to do, but we're still going to win this. AI is going to be an American win.