MICHEL MARTIN, HOST:
President Trump will visit China this week, where he will meet with Chinese leader Xi Jinping. One of the big things they're expected to talk about is artificial intelligence. Both the U.S. and China have amped up development and adoption of AI in order to get ahead in the global race over the technology. NPR White House correspondent Deepa Shivaram reports on who is winning and for how long.
DEEPA SHIVARAM, BYLINE: Trump says the U.S. is leading on AI, and by some measurements, he's not wrong. Here's Kyle Chan, a fellow at Brookings, a centrist think tank in Washington.
KYLE CHAN: If you're looking just at sheer performance of the top models, then yes. Models from OpenAI, Anthropic, Google DeepMind are widely regarded as the very best.
SHIVARAM: And having the best models is critical, Chan says, for things like complex coding tasks and for national security and preventing cyberattacks. But having the best AI product isn't the only measurement of success when it comes to the global race over this technology.
CHAN: While you can have the very best models, if you don't know how to integrate them into your economy, into your different sectors and services, then it matters less how good they are.
SHIVARAM: The data isn't always clear. But this is an area where China may be outpacing the U.S., says Chan, by further integrating AI into their society and everyday life - like healthcare, education and government services - even if the technology isn't as good as what the U.S. is making. The other factor with Chinese AI models is that they're open-source. That makes them free or very cheap, and it means that developers can adapt and fine-tune and customize the models in a way that fits their needs.
U.S. models, on the other hand, are essentially locked and are more expensive but better quality. And that's working for now, but it might not in the long term, says Michael Horowitz. He's a professor of political science at the University of Pennsylvania and a former Pentagon official. He compares the decision between Chinese and U.S. AI models to buying a TV.
MICHAEL HOROWITZ: The top-of-the-market television is, like, certainly better than a mid-market television. But a mid-market television that you can get for under $1,000 now is, like, still pretty darn good.
SHIVARAM: So in other words, U.S. AI models are the top of the market. But...
HOROWITZ: The risk is that the impact of that lead is overcome by actors around the world choosing to work with Chinese AI models because they're cheaper and easier to access.
SHIVARAM: Experts, including Horowitz, estimate that the lead the U.S. has on its AI models is about six to 12 months. But the recent controversy over Anthropic's Mos has some in Washington worried about whether that buffer is enough. Anthropic deemed their AI model Mythos too dangerous to release because of cybersecurity concerns. AI is developing so rapidly that it brings up questions about the security of the U.S.' AI lead, like this one from Chan.
CHAN: Does having a lead in AI frontier capabilities, even if it's just a few months, give the U.S. a chance to deal with some of these cyber vulnerabilities before China has a chance to try them out on the U.S?
SHIVARAM: What measures the U.S. can take to widen that lead will likely be what Trump and his administration try to advocate for in his meeting with Xi later this week.
Deepa Shivaram, NPR News, Washington. 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.
