Sen. Mark Warner says U.S. can strengthen AI safety without losing race to China

Amna Nawaz:

The warnings about artificial intelligence are growing more urgent. And, increasingly, they're coming from inside the industry. A.I. researchers have resigned over safety concerns, while some of the field's most prominent leaders are calling for stronger government oversight. But Congress has yet to agree on meaningful safeguards.

Geoff Bennett spoke earlier today with Senator Mark Warner of Virginia, the top Democrat on the Senate Intelligence Committee, about Washington's willingness to act.

Geoff Bennett:

Senator Warner, welcome back to the "News Hour."

Sen. Mark Warner (D-VA):

Thanks so much for having me.

Geoff Bennett:

Jacob Coxon, the Anthropic researcher who resigned -- and the resignation ignited this latest push for regulation -- he says the next year or two could be crunch time for humanity. Do you take that literally? Do you think A.I. poses an imminent existential threat?

Sen. Mark Warner:

I don't believe imminent existential threat, but potentially catastrophic threat, yes.

And the irony is, there have been numbers of A.I. researchers who've been saying this for some time. Matter of fact, Dario Amodei, who is the CEO of Anthropic, who I spoke to over the weekend, saw last week, Sam Altman, he's saying that for the last year as well.

It feels what's happened is that there's been building concern about this. A lot of that is exhibited in pushback against data centers, which do need some guardrails on them in terms of energy and water and setbacks from communities. But I think the data center pushback reflects a greater concern about this technology people don't understand. Could it take away your job?

And then you add the incidents that have taken place recently. Think about this, the OpenAI-Hugging Face model, where the OpenAI tool had 3,000 A.I. agents that collaborated together in a way where they hid information from the humans. And literally some of these agents would, in a sense, sacrifice themselves or commit suicide to try to accomplish their task.

That was with 3,000 agents. More recently, they have had a tool that had 10,000 agents working as a swarm. And so I do think we have had these indications. You have got the four biggest A.I. companies in the world literally saying, we got to do something. And shame on us if we don't act.

Geoff Bennett:

Meantime, President Trump says the government already has the tools it needs to police A.I., and he says these recent warnings amount to a hoax. You have got David Sacks, the White House A.I. adviser, who says, if these companies want a slow A.I. advancements, they should do it voluntarily. They don't need permission from Washington to do it.

What's your assessment of those two arguments?

Sen. Mark Warner:

Well, first of all, Mr. Trump also said the only thing we need is a really brilliant president with a high I.Q. like himself. So I will let your viewers make your judgment on that.

And then, Mr. Sacks, he's got a point that, as we go through this process of an A.I. safety review, we have got to make sure there is the voice of small tech or early tech, so that this is not an effort to simply solidify the incumbency position these big four entities have.

And I do think the idea, what the Trump administration has done is, they put together a bit of a voluntary forum. You can't do voluntary when there's at least a reasonable chance of a catastrophic risk that might currently disrupt the financial markets or disrupt water systems or electric systems.

So we have got to take this seriously. And those of us who -- in a bipartisan fashion, who have been calling for this kind of safety guidelines for a while, this is our moment to move. It's made more difficult because, frankly, President Trump's administration has gotten rid of a lot of the talent at the Cybersecurity Agency, which is part of the problem.

He's gotten rid of some of the safety institutions that were being set up. Frankly, the British have now got a stronger safety regime than we do, and not one that's over the top in terms of overregulated.

So I think we have to think about putting safety in into the mix before these models are released. We need to put safety around what's called the sandboxes, where they test out these models, and then we have got to really decide for how long and who's at the table, because getting the details right under this are going to be critically important.

Geoff Bennett:

On that point, how do you slow American companies without handing China an advantage and creating a different type of national security threat? How do you slow the A.I. race without losing the A.I. race?

Sen. Mark Warner:

Well, first of all, I want America to win the race. And, secondly, I have been involved in monitoring China. I have been chair or vice chair of the Intelligence Committee for the last 10 years. I'm very well aware of the competition.

But here's what some of the accelerationists say. You got to do pedal to the metal or China will win. Well, China has already, for example, put in place requirements for their A.I. models to not create A.I. bots or chatbots that would form personal relationships with children and young people.

Don't our kids deserve the same? China has already put regulations on not having these A.I. tools create what's called deepfakes, false images of people. Shouldn't Americans have that same protection? And I think you can go ahead and put a bit of safety into this process and not lose the race.

And, frankly, the downside of a mistake or a catastrophic risk being created, boy, oh, boy, that's not something you -- I fear, frankly, that, if we had that kind of activity, you might see politicians from both the left and the right come together and say, we got to shut it all down, and that would not be good either for the -- for America or for, frankly, innovation.

Geoff Bennett:

So, Leader Thune has suggested that Congress has a role here. I know there are bipartisan talks under way. What do you think realistically could pass this year in terms of legislation?

Sen. Mark Warner:

Well, that's the million-dollar question.

I do think some of these companies believe that we are so close to being on the cusp of potentially these models going beyond humans' ability to control them, that I think we have to act with a sense of urgency.

Now, I'd love to see guardrails on data centers. I'd love to see kid protection. This whole notion of A.I. agents, there needs to be liability. There's national security concerns.There's job disruption concerns.

Obviously, the way Congress works, we're not going to get the whole process done. The companies do need some ability to talk to each other. That would require some level of antitrust exemption. For them to get that exemption, they have to show how we put this safety effort together.

But I think they also need to show -- and this is what I have been frustrated with them. They need to do a better job of convincing normal Americans, normal humans how A.I. is going to actually improve their life and not just get kind of overwhelmed with the billions and trillions of dollars that seem to be passing back and forth between these companies.

So, I do think we have got to grant them the ability to collaborate in a certain way, because, frankly, the best talent is not inside the government. It is in these frontier models. There are other companies, large corporate enterprises, that have used these tools. They need to have a seat at the table.

Small tech has to have a seat at the table, so there's not simply an effort to solidify these leaders' incumbency power. So I think we could act, and there is this sense of urgency right now that I'm hoping in a bipartisan way we can get done.

I'm talking with some of Republican colleagues right after this -- right after this viewing.

Geoff Bennett:

What would you say to those skeptics who would point out that social media has been around for more than two decades now, and Congress hasn't been able to regulate it in a meaningful way, even after its harms became clear?

They might ask, why would A.I. be any different?

Sen. Mark Warner:

Geoff, you just made my case.

I don't think there is an individual anywhere in the political sphere that doesn't think, gosh, if we could go back 10 years, wouldn't -- shouldn't we have put some guardrails on social media? Shouldn't we have said 10 years ago maybe our kids shouldn't have their phones available all day through the school day?

If we thought social media was a problem with our -- with our youth, think about having a chatbot that is forming an affinity relationship with your son or your daughter. If that doesn't scare the hell out of you, I don't know what will.

And also the very notion on top of that of these agents collaborating together, these A.I. agents collaborating together in a way that humans don't control, that's even beyond social media. So I think the failure to act on social media is, frankly, something that's moving a lot of us to say, hey, we can't repeat that mistake.

Geoff Bennett:

Senator Mark Warner, Democrat from Virginia, always a pleasure to speak with you. Thanks for joining us.

Sen. Mark Warner:

Thank you, Geoff.