The U.S. and China are pouring hundreds of billions of dollars into artificial intelligence as political and business leaders frame the technology’s development as a race between world powers.
But Jason Furman, Aetna Professor of the Practice of Economic Policy jointly at Harvard Kennedy School and in the Department of Economics, doesn’t think the race has a finish line.
In this edited conversation with the Gazette, Furman explains how competition among AI companies differs from earlier battles among internet and social networking companies and how much having the best AI model matters.
Gazette: What is the current global AI economy like right now? Are there two major players or is it more complex than that?
Jason Furman: There are two major players in the AI economy: the United States and China. When it comes to frontier models, Europe has an OK one, but it’s not in the running. No one else does. But the open question will be how much bigger the AI economy is than just the frontier models themselves.
What does it mean to win the AI race? Are there certain metrics these frontier models and countries are aiming for in medicine and science?
First of all, I don’t think there is a finish line, so there’s no such thing as winning the race. It is going to be sort of constantly racing.
One big open question is: Are the frontier labs just building something like a car that can drive faster and faster, and so they’re increasing the speed from 300 miles an hour to 400 to 500 miles per hour?
In that case, the fastest speed will matter to a few people, like race car drivers, or in the case of AI models, will matter for cybersecurity and certain cutting-edge research, but for everyone else, the difference between a 100-mile-an-hour car and a 500-mile-an-hour car is irrelevant.
And so, I think one interesting thing in this race is it may be a race for a smaller and smaller segment that needs the very best, while everyone else can increasingly get by with something that falls way short of the very best.
“When we think about U.S.-China, we have to accept that they’re going to learn about the things that we figure out and be able to copy them. And by the way, they’ll figure out some things we don’t figure out, and we’re going to be able to learn and copy from that.”
Is there an example of a small sector that you see these companies aiming for?
Their valuations and their capital expenditures are what you would expect from something trying to do a “winner-take-most” that dominates everything, and that is the previous model with the internet. You know, Google with search, Meta with social networking.
The network externalities in those previous technologies were incredibly powerful. If you were on Facebook, I needed to be on Facebook.
Today, if you’re on ChatGPT, I can still be on Claude, and that works just fine. Moreover, it’s so far relatively easy to switch from Claude to ChatGPT. So, some of the economic “winner-take-most” premise that we’re so used to in the technology space — I think it is an open question and far from certain that it will apply to these companies.
What advantage does, say, the U.S. or China get for making the pre-eminent model in a certain area? Isn’t a lot of the technology easy to switch between?
There’s sort of two things that matter for where the frontier is located.
First, for cybersecurity and military espionage — national security-related issues — having the very best is really important, and the second best just may not work.
Second it matters for market capitalization, which determines how many people are getting millions of dollars or billions of dollars because they’re part owners of these companies.
But then when it comes to the economy as a whole, what’s going to matter much more is how the technologies are used in jobs, how they’re adapted. And so some of the broader-based effects probably depend much less on which country got to the frontier first.
How legitimate are claims by U.S. politicians and leaders who say we can’t slow down AI development because of China?
I mean, implicitly, that is accepting something that I think is at least partly correct, which is that on cybersecurity and military capabilities, what matters is not how capable the system is, but how its capabilities compare to the capabilities of your rival. That’s part of what the argument is.
But I think the argument may be part excuse as well for companies that don’t really want to slow down or policymakers that have a deregulatory mindset or think this technology is good for other reasons. So I think there’s one part sincerity in that argument and one part convenient rationalization.
Given the recent high-profile incidents of AI going rogue involving OpenAI, Anthropic, and Google, wouldn’t China be on the same page in terms of slowing it down?
I do think there is enormous common interest between the United States and China on AI. There’s also enormous divergent interests, so we have both. But it’s a mistake to think that it’s all divergent interests.
I was in China in May talking to a lot of people, and there’s a lot of sincere concern about AI safety. There’s a lot of worry about job displacement. If you talk to professors at Chinese universities, they’re worried about how it impacts teaching on their campuses.
So, a lot of the same issues that we’re struggling with and worrying about and talking about in the United States are exactly the same issues on AI that they’re worrying about and struggling with in China. And so the more dialogue we can have with them, the more we can work on this with them, the better.
What would happen to our economy if AI isn’t as transformative as expected?
So far, we’re seeing very big effects on demand from the AI build-out. We’re building lots of stuff.
We’re not seeing very large effects on supply, which is to say, productivity and the like. If we don’t see the productivity, then we’re going to see a lot of financial problems because much of the financing of the system was built on the premise that number one, we’re going to get the productivity, and number two, that companies will be able to profit from the productivity.
If either one of those is wrong, we don’t get the productivity, in which case, you can’t profit. Or even if you do get the productivity, if you can’t figure out how to monetize it and keep other people from competing the prices down, then I think we’d have a bear market, and we’d have some real problems in our credit system.
I don’t expect that it would be anything like the global financial crisis because this is in different pockets of the financial system that are a little bit better able to absorb risk than what we had there.
But it wouldn’t be pretty because, yes, right now we’re betting trillions of dollars on the premise that this is all going to work out.
Anything else you want to add?
I think the last thing I’d say is it’s hard in general, and it’s almost impossible, to prevent knowledge in this area from spilling over. And so, when we think about U.S.-China, we have to accept that they’re going to learn about the things that we figure out and be able to copy them.
And by the way, they’ll figure out some things we don’t figure out, and we’re going to be able to learn and copy from that. And that’s, again, why I think it is foolish to think you can completely contain this. Maybe you can protect a small edge.
But ultimately, I think trying to work together while not being naive about what that work consists of is the best path forward.