ARI SHAPIRO, HOST:
Many of the world's leading tech companies met in Seoul, South Korea, this week. And they joined an international agreement on AI safety. Several governments, led by the U.S., also agreed to set up a new global AI safety network. So will these voluntary agreements actually keep AI safe? Commerce secretary Gina Raimondo is leading the Biden administration's effort on the issue. Welcome back to ALL THINGS CONSIDERED.
GINA RAIMONDO: Thank you, Ari. Good to be with you.
SHAPIRO: Will you begin by giving us a concrete example of the sort of AI risk you believe these new agreements will help us avoid? Are we talking about misinformation or invasive surveillance, or what?
RAIMONDO: I would say both and then some - misinformation, deepfakes. Some of these models are so sophisticated that if they're not properly tested before they're released, they can do things that the developers didn't even think they were capable of. You know, you've probably heard about hallucinations or the models doing things that weren't predicted. So one of the things we'll be doing is testing all of the most sophisticated models before they're released to the public for use so that we can be sure that they're going to do what they're intended to do.
SHAPIRO: But is Pandora's box already open? I mean, we've already seen deepfakes that use politicians' voices.
RAIMONDO: I would say yes and no. I mean, certainly the technology is out there. And we've already started to see some disturbing incidences where, as you just said, you know, there's deepfakes that's out there, misinformation. But look, the exciting thing here - it's easy to get bogged down in how it could go poorly.
There's also life-changing, life-giving possibilities that AI can bring to the world - bringing a doctor to everybody that has a phone, you know, a huge opportunity for education. So I think the whole point of it is let's do our best to work with like-minded countries that believe in democracy, that oppose civil rights abuses - and say let's put a lid on the risk so that we can let all of the exciting opportunities develop.
SHAPIRO: You say let's work with like-minded countries. China is a notable outlier. As I understand it, Chinese delegates have attended these international summits, but the government of China is not part of the global AI safety network that the U.S. has led with other governments. China exports its AI technology to many other countries around the world, so how effective are these guardrails going to be if China and its allies don't agree to them?
RAIMONDO: Ultimately, we want as many countries to participate as possible for exactly the reason that you just said. We are beginning with our like-minded countries to come up with AI safety policies across our nations. As you correctly point out, we've got work to do with China. But it's in everyone's interest, including the people of China, that AI is used in a way that is safe.
SHAPIRO: The U.S. and China are talking about AI. Can you see areas where you think these countries may be able to agree?
RAIMONDO: Yes. You can imagine AI as applied to nuclear weapons or AI as applied to bioterrorism. No one in the world wants that to happen. And so you could imagine getting to a common standard that would prevent that happening ever.
SHAPIRO: We're talking about three distinct but related agreements here. There's the private companies that have reached a set of standards. There's this international safety coalition. And then there is the federal government setting up its own AI safety institute under your Department of Commerce. The press release says it will do things like advance, support and articulate principles around AI safety. It does not say enforce. Does this institute that you're leading have any power to prevent AI companies from doing things that experts believe would be harmful?
RAIMONDO: Mmm hmm. Ultimately, Congress needs to act. What we're doing at the Commerce Department is focusing on the science and the technical aspects and setting standards for everybody to comply with. For there to be, you know, penalties - whether that's a fine or criminal penalties - we will need Congress to act.
SHAPIRO: One feature of democracy is that lots of voices get to weigh in - federal and state governments, the private sector, citizens, civil society. Things are built to move more slowly, and autocracies like China move fast by design. So if we are looking at a race between two models of AI represented by the U.S. and China, do you think the U.S. can win? Or is that the wrong metaphor?
RAIMONDO: No, I absolutely do. Well, first of all, I would take democracy any day over autocracy for a million reasons. But America is winning now, right? Like, America leads the world. Our models are the most sophisticated. Our AI companies lead the world. The U.S. has a dominant position, and I have every confidence that we will stay there.
SHAPIRO: Commerce Secretary Gina Raimondo. Thank you so much.
RAIMONDO: Thank you, Ari.
(SOUNDBITE OF LOLA YOUNG SONG, "CONCEITED") 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.
