MICHEL MARTIN, HOST:
For more on President Trump's AI summit, we've called Aalok Mehta. He directs AI projects and research at the Center for Strategic and International Studies. He previously served as the responsible AI policy lead at Google. Good morning. Thanks for joining us.
AALOK MEHTA: Thanks for having me.
MARTIN: As we just heard from Emily, President Trump said that AI leaders agreed that there should be, quote, "tremendous self-regulation" at AI companies. And he said that there's already regulation by the FBI and DOJ. Can you tell us what that might look like?
MEHTA: Yeah. So in this new accord that the AI companies signed, they agreed to four levels of oversight. So essentially, they'll implement robust internal controls. And then they will have progressively layered sets of oversight, including new measures like having an independent evaluator inside the companies and having more oversight directly from company boards. And what this is doing is building on top of a number of sort of measures that companies have already taken voluntarily to address risk from their models.
MARTIN: But - OK. But what's the enforcement mechanism, other than President Trump saying it would be, quote, "morally binding" for companies to monitor their AI models?
MEHTA: There are laws on the books. So the primary law that would apply here is that the FTC has authority to regulate unfair and deceptive trade practices. And so what happens is when a company publicly commits to doing something, they have then sort of signaled to the world that that is their intent. And the FTC can then say that if the company doesn't follow through on its promises, it is acting in a deceptive way. And so this is the primary enforcement mechanism that I'm aware of that can be used when a company says it does something and then doesn't follow through on those commitments.
MARTIN: I understand. But what is the company's incentive to do that, other than - oops?
MEHTA: They have lots of incentives. I think the biggest incentive is that they want people to use their technology. They want people in the U.S. to use their technology, and they want people around the world to use their technology. And people will feel more comfortable doing that if they trust that the product they're using is safe and that they trust the company that is issuing that product. And if you look at polling around the U.S. and in other places in the world, like Europe, right now people are very skeptical about AI. They're skeptical about - that it will make their lives better. And so part of this is driven by the fact that they want to build trust in the underlying technology.
I think part of it is also because a lot of people at these companies genuinely worry about these things. In many cases, they have access to models that are more advanced than the public. And so they're seeing things that we haven't seen yet. And so that is also part of the - part of what is driving these decisions and part of what is underlying their thinking.
MARTIN: So - well, a couple questions about that, though. More broadly, is the government equipped to independently evaluate AI model safety and effectiveness?
MEHTA: In this case, they are largely relying on the companies to sort of implement these measures themselves as a form of self-regulation. And as we heard from the Anthropic CEO, there are lots of details to be figured out. So part of the work that will happen over the next months and years is for companies to figure out how to implement this in a way that is robust and useful and actually addresses safety issues. I think the thought is maybe in the future, this could become enshrined in some sort of regulation or law, but that is not contemplated right now in the near future.
There are various parts of the government that can help provide technical insights. So there's, for example, an institute within the Department of Commerce that has technical talent and does do testing on AI models. And it's unclear how that testing will factor into what the companies are going to be implementing.
MARTIN: Well, before we let you go, I think you note - and polling has indicated - that there's a lack of public trust in this technology and in these companies at this moment. But there were no AI safety groups or independent regulators present at the White House meeting. You know, there's a saying in Washington - personnel is policy. But what does that tell us?
MEHTA: It tells us - it's really consistent with the way that the Trump administration has approached the AI industry basically since the administration started, which is that its primary concern is competition with China. And it really wants to remove obstacles to the development of AI because it - the administration is concerned that any regulations will slow down the development of U.S. technology and put the U.S. at a disadvantage.
MARTIN: All right. That's Aalok Mehta, director of the Wadhwani AI Center at the Center for Strategic and International Studies. Thank you so much.
MEHTA: Thank you. 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.
