“The difference in perspectives between superpowers is
shaping the race for A.I. dominance.
The United States and China are really the only two
countries that matter right now in shaping the A.I. future. As President Trump
and President Xi Jinping meet in Beijing, there’s a kind of Cold War
atmosphere, with people talking about an A.I. arms race. But who is winning?
Are we even in a race at all? Kyle Chan, a foreign policy fellow at the
Brookings Institution, says it’s hard to call it a race because the U.S. and
China have very different A.I. goals.
Below is an edited transcript of an episode of “Interesting
Times.” We recommend listening to it in its original form for the full effect.
You can do so using the player above or on the NYTimes app, Apple, Spotify,
Amazon Music, YouTube, iHeartRadio or wherever you get your podcasts.
Ross Douthat: Kyle Chan, welcome to “Interesting Times.”
Kyle Chan: Great to be here.
Douthat: So at the moment, there are really only two
countries that matter for the A.I. future: the United States and China. Their
leaders are meeting in Beijing this week, and the atmosphere is sort of similar
to a kind of Cold War atmosphere, where people think and argue and talk about
them being in a kind of arms race.
You are an expert on China and A.I., and we’re going to talk
about that race: who’s winning, what winning even means, whether it even makes
sense to talk about the U.S. and China in terms of a race.
But I want to start with a basic question. How is China’s
current approach to A.I. different from the American approach?
Chan: It’s
quite different, actually. In the U.S., there’s a particular focus on
A.G.I. — artificial general intelligence — and to create something
approaching an artificial superintelligence, some kind of almost machine god
that can do virtually everything that any human can do ——
Douthat: And
more!
Chan: And more. That’s right.
Douthat: You want to get more. That’s the “super” part.
Chan: [Chuckles] Absolutely. And you can see that the amount
of spending, the amount of investment, the amount of effort that the American
Big Tech companies and their quote-unquote “start-ups,” like OpenAI and
Anthropic — which are now close to $1 trillion each — are pouring into this is
an indication that they’re making a big bet that they can get there at some
point, maybe in the near future. That’s the race to A.G.I. in the U.S.
China is
running a different kind of race. I would argue they’re running multiple races.
On the one hand, they are trying to produce better and better A.I. models.
They do want to try to keep pace with their American competitors, but that’s
not all they’re focused on.
They’re also focused on efficiency,
making these models smaller, cheaper to run, easier to deploy. That’s one area.
Another area
they’re focused on is diffusion — trying to get A.I. into the hands of
as many users as possible — and part of that strategy involves open source.
This involves kind of giving away your models for free. And that allows other
people around the world, including in Silicon Valley, to download Chinese
models and to also customize them and tweak them based on their own data, and
to make them work in a way that’s more tailored to their own needs. That’s the
advantage of open source.
Another
major area that China’s focused on is applications. Specifically, robotics
is a huge area of focus, both for the government and for Chinese A.I.
companies.
But you don’t really hear so much about A.G.I. You might
hear some of the Chinese tech founders talk about this, and they sometimes
sound a little similar to their counterparts in the U.S. But overall, they’re
much more focused on the nuts-and-bolts uses and applications of A.I. in
people’s daily lives. That’s the key priority.
Douthat: So if I went to Shanghai or Beijing right now and
spent a couple weeks there interacting with physical reality and digital
reality, do you think I would notice a big A.I.-driven difference versus life
in the United States? Just describe the everyday experience of this strategy,
to the extent that it makes a difference in how people are living.
Chan: Yeah. So in the larger cities in China, you might see
autonomous delivery robots dealing with package deliveries, food deliveries. In
a restaurant, you might see a waiter robot bringing your food. This is not
super, super widespread yet, but it’s starting to come about. Hotels, rather
than having room service be delivered by a person pushing a cart coming up the
elevator, it might be a delivery robot. You have of course self-driving cars.
You might even have drone delivery for coffee or food.
But it would be a subtle but probably surprising difference
to what most Americans experience in terms of their interaction with A.I. in
the physical world.
Douthat: Let’s just pause for context, because you talked
about the government versus the Chinese A.I. companies. I think most viewers
and listeners are accustomed to the American situation, where you have a set of
big companies, they have been extremely, lightly regulated by Washington, D.C.,
and just in the last year, we’ve started to get into dynamics where the
Pentagon especially seems concerned about their national security implications.
There’s talk about regulation, screening of models and so on, but basically
it’s been a very traditionally American capitalist environment. Not a Manhattan
Project or anything like that.
To what extent is China similar or different just in the
relationship between the companies and what is obviously a much more powerful
and often repressive state?
Chan: In China, the state is in charge. Or specifically, I
should say the party state. The Chinese Communist Party and the various
government agencies that they oversee, they’re the ones who set the rules.
They’re the ones who ultimately are shaping the trajectory of China’s A.I.
industry.
They have quite strict regulations — for example, requiring
A.I. models to be registered in advance. They have certain content and
censorship rules that must be followed. They have a whole host of ways to
enforce their rules and have leverage over Chinese A.I. companies. And there
are echoes back to a previous era where Chinese regulators cracked down on
Chinese internet companies, for example.
That’s the overarching relationship, but that doesn’t mean
that the Chinese A.I. labs themselves are just in lock step following whatever
Beijing says. Ironically, China tried a more top-down model to technology in a
previous era, and that failed miserably. It did not produce the kind of
innovation and flexibility and agility in the marketplace that you would need
to have cutting-edge technology.
Douthat: What era are we talking about with the more
top-down approach?
Chan: That was, I would argue, going back to the Mao era.
This is the classic ——
Douthat: Pre-Deng, roughly pre-1980s?
Chan: Exactly, yeah. That almost Soviet
command-economy-style approach.
So what you have is sort of a hybrid model in China, if I
could characterize it in a single word — a broader direction and guidance and
certainly support from the central government in China as well as local
governments on the one hand, but then also trying to create space for
competition and innovation from the Chinese A.I. labs themselves, whether
you’re talking about China’s equivalent of the Big Tech companies, like Alibaba
or Tencent, the maker of WeChat, the popular super app, or you’re talking about
China’s own A.I. start-ups, like Z.ai or Moonshot AI, which have actually
become quite popular around the world.
Douthat: What are the Chinese equivalents of an Anthropic or
an OpenAI right now?
Chan: That’s a good question. So maybe DeepSeek would be the
closest. And then you have the smaller start-ups. And by smaller I mean like on
the order of $40-to-$50-billion market cap. And those are some of the more
successful ones.
But it’s hard to find that kind of middle ground. DeepSeek
now is preparing to take in outside investment. Remember, they were actually
not originally an A.I. company. They were part of a hedge fund, actually, that
was trying to use A.I. to develop more sophisticated financial models. So
they’re sort of a category unto themselves.
Douthat: And all of these companies, though, are operating
under some basic constraints that don’t apply to U.S. companies right now,
mostly around chips. Can you describe the landscape of constraint in China and
what it means?
Chan: Yeah. I had mentioned earlier that Chinese A.I.
companies are trying to run different races, and one of those was efficiency.
Part of that is in response to the constraints that they’re under, in
particular around compute and chips.
So remember, right now the U.S. has export controls on our
most advanced semiconductors, basically made by Nvidia, and we stopped those
from officially being sold in China. We allow the sale of watered-down
versions, but the idea is that we keep the best and the most advanced chips for
American A.I. companies in the United States and for allies and partners.
For China, that means that they don’t have access to the
most cutting-edge A.I. chips. They have some Chinese domestic alternatives —
and this is a big part of the story. One of the leading players in the space is
Huawei, the heavily sanctioned Chinese tech giant that rose first in the
telecom space, branched into smartphones and is now in pretty much every other
industry — electric vehicles, clean technology and certainly now A.I. and
chips. So China’s trying to build up their own capacity for developing A.I.
chips on their own, not just designing them, but actually producing them.
The problem is, they’re just not quite as good as the Nvidia
chips. And without that, it does put a lot of constraints on what they can do.
So they’re trying to squeeze more out of very limited compute resources.
Douthat: Why aren’t their chips as good? I know this is a
simple-minded question. Is it just that Nvidia is so awesome at engineering and
China’s engineers, even if they have a Nvidia chip, can’t quite get there
themselves? Talk to me like a non-chip specialist.
Chan: This is the $5 trillion question, which is currently,
I think, roughly the market cap of Nvidia today.
There are a couple different aspects to this. One is
actually the chip fabrication that is producing the chips. Remember, Nvidia
doesn’t make their own chips. TSMC in Taiwan, they’re the ones that make the
chips.
Douthat: Conveniently located not that far from China.
Chan: [Chuckles] That’s right. To the consternation of
probably a lot of folks in Washington and maybe other folks dependent on those
supply chains.
But TSMC has been pushing the boundaries for increasingly
advanced semiconductors in a whole range of areas, and that includes A.I. And
Nvidia, by partnering with TSMC, can combine some of the best design work out
there with some of the best production capabilities.
For example, ASML, a Dutch company that maybe some people
have heard of, it’s actually one of the biggest tech companies in Europe now.
They make these extremely precise, extremely expensive lithography machines for
printing chips, basically. And they’re the only ones in the world that can make
this kind of machine.
They sell those to TSMC. TSMC can use that cutting-edge
technology, combined with their own cutting-edge manufacturing processes, and
work with Nvidia to produce these incredible state-of-the-art chips that keep
getting better and better.
Douthat: So essentially, when we talk about the U.S. not
allowing Nvidia to sell to China, we’re effectively talking about the U.S.
cutting China out of a larger supply chain that runs through Taiwan, through
the Netherlands, all around the world?
Chan: Absolutely.
Douthat: OK.
That’s interesting and very helpful. What does China have going for it then, in
terms of A.I. build-out, that the U.S. doesn’t have?
Chan: Energy
is absolutely huge in China. If you’re thinking about the broader A.I. stack —
that is, not just the chips or the models themselves, but deeper down on the
layer — energy is perhaps the most important and least talked about.
For the
U.S., this is a major bottleneck. It’s very hard now for data centers to build
out the power capacity to power all those chips that they’re putting together.
In China,
interestingly, they’ve been building out energy at a very rapid pace — clean
energy like solar, wind, batteries. They’re trying to leverage that ongoing
energy build-out to feed into their compute build-out, which then feeds into
their A.I. development.
So you see
really interesting sort of strategies that the Chinese are taking. For example,
they have this effort to try to build data centers out in the western
provinces, away from the high-population urban areas in China.
At first, that might not make any sense. Don’t you want to
have your data centers close to where people are actually using them? Don’t you
want to have that low latency, high response time?
What China’s trying to do is they’re trying to leverage a
lot of their renewable energy resources out in those further-off regions.
They’re also trying to just do sort of good old-fashioned
geographical redistribution, always concerned about having these poorer
provinces remain poor while the high-tech Shenzhens and Shanghais speed on
ahead.
So this is another area where they’re trying to leverage
some of their strengths to feed into areas where they’re maybe weaker.
Douthat: China is, to simplify, imagining a future where
they’re only a little bit behind the U.S. and — actually, say what that means.
People talk about the best Chinese models being three months behind the U.S. or
six months behind the U.S. How far behind are they, and what does that mean in
practice?
Chan: Overall, I think the consensus is that Chinese models
are somewhere between three, six, to nine months behind, depending on the time
of year, and which was the latest model that just came out.
What that means is that when you look at specific
benchmarks, specific evaluations for trying to understand how well these
perform on, say, math or coding tasks or even new agentic tasks, the Chinese
models that are released today are starting to get close to the American models
that were released a couple of months back. That’s what that lead time means.
The thing is, it’s not just about having the absolute most
cutting-edge model. You can have very, very strong models that can do a lot,
that can do a lot of useful agentic tasks, like maybe create a whole PowerPoint
presentation for you and do all the research and analysis that goes into that,
or answer your emails.
There’s this strategy, I think, right now in China where
they’re hoping that it’s not just all about having the very best models. It’s
about trying to figure out where to make this work, and also to build the
broader ecosystem for deploying these models, to integrate them into more and
more services, like food delivery or ride-hailing or, again, much more
practical real-world applications.
Douthat: In the U.S., obviously there’s just a lot of
anxiety around A.I., to a greater degree than any big technological change in
my lifetime, certainly. There’s apocalyptic fears, there’s economic fears about
job displacement, there’s social and cultural fears, there’s people who just
don’t want data centers built in their backyard. So there’s a whole range of
different moods.
If you were going to try and distill the mood in China, the
public mood around A.I., how would you describe it, and how is it different
from the U.S.?
Chan: I think the biggest anxiety right now in China is
around falling behind on technology. I think in the U.S., there’s a lot of
worries about job displacement, of A.I. being a net negative force in society.
In China, there are some of those concerns — and I can come back to that — but
I think right now, the fear among individuals and companies and workers is that
they’re not keeping pace with A.I., that they’re not using it enough, and
they’re not savvy enough with this new technology, so they won’t be competitive
enough in the labor marketplace.
It’s interesting: This anxiety at the individual level kind
of mirrors China’s anxiety at the national level. When ChatGPT first came out —
and in fact, you can even go back to when AlphaGo first defeated the human
world champion in Go — there was a lot of anxiety in China among China’s A.I.
industry and among policymakers in Beijing. They were worried that China was
also falling behind, that they were not making the most of this new
transformative technology.
So it’s interesting to see this kind of mirroring where it’s
not about how do I keep out this technology from my life? It’s about how do I
bring it in even more and integrate it and give myself that edge in a very,
very crowded marketplace?
Douthat: I see that attitude in the U.S., but it is a very
Silicon Valley and tech-adjacent attitude. It’s spreading, but you see it in a
pretty confined zone of the American economy. But are you saying that in China,
it is just much more widespread? That you don’t have to be working for DeepSeek
or working for Alibaba or something to have this “Am I falling behind? I must
add A.I. protocols” mind-set?
Chan: That’s right. It’s interesting that A.I. is hitting at
a time when China was already experiencing a whole bunch of anxieties around
labor markets, especially for young college graduates.
For example, the unemployment rate for young people in China
is basically double what it is in the United States. It’s somewhere close to 17
percent, which is extremely high. The number of new college graduates hitting
the job market this year alone is 12 million-plus in China.
These are all people competing for many of the same jobs.
They don’t want to work in the factories. They don’t want to have those
blue-collar jobs or delivery jobs. They want, in their minds, the “good jobs.”
And they’re worried that if they don’t keep up with A.I., they might not be
able to get those.
So it’s a longer-standing concern about this
hypercompetitive environment in China that has been there as long as I’ve been
going to China, but A.I. really sort of amplifies and accelerates those
anxieties.
Douthat: And part of the debate in the U.S. has also been
about the welfare state, and you have tech leaders talking about how the
welfare state has to adapt if there is A.I.-driven unemployment. You have Elon
Musk promising, not universal basic income, but “universal high income” — I
just like saying that.
China does not have a safety net to any degree like the
United States, or like Western Europe. Is there a welfare state debate in
China, a U.B.I. debate — anything like that?
Chan: Increasingly so. The great irony here is I was
speaking about the Mao era earlier. That is the era of the “iron rice bowl,”
the idea that you were a worker at a state firm, at a state organization, and
you basically had your job for life. And this idea of job security is no longer
there in China, unless you’re working for, again, a state-owned enterprise or
within the government.
So that concern is coming back. And there’s actually more
discussion now, including among policy folks in Beijing, about the potential
issues related to A.I. job displacement and what China should do about it from
a welfare and policy standpoint.
Douthat: Are there actual policy ideas in the wind? Is
U.B.I. under communist conditions?
Chan: [Laughs] It’s still early stages.
Douthat: “From each according to his ability, to each
according to his need” makes a comeback.
Chan: That’s right. [Laughs]
Douthat: To get rich is glorious, but also …
Chan: [Chuckles] But also … they are the Chinese Communist
Party after all.
I think it’s still early days for that discussion. And
there’s still a pivot that’s happening from the sort of all-in “hit the gas
pedal on A.I.” progress, including from the policymakers, where they were
emphasizing all the new jobs that would be created by A.I. — “don’t worry about
those other jobs that might be affected, that’s part of the industrial
revolution that’s happening now, Industrial Revolution 4 or 5.0” — but now that
conversation’s starting to shift.
Douthat: And what about the central government’s concern
about social effects of A.I.? One notable thing in China — you mentioned
earlier the crackdown on internet companies — there was and has been a deep
anxiety about the internet’s effect on social life. There have been attempts to
crack down on video-gaming among young men. All of the things that American
commentators worry about at a speculative level have actually sometimes been
actual policies in China.
And this is connected to the reality that China has a bigger
problem than the U.S. with falling birthrates, falling marriage rates. Are
China’s leaders looking at A.I. through that lens and worrying about the A.I.
girlfriend, A.I. boyfriend future?
Chan: Definitely. They are very worried about that. And in
fact, they are already rolling out policies and regulations around A.I.
boyfriends and A.I. girlfriends.
It’s so funny. They have a very negative view of wasting
time, basically, of what the folks in Beijing see as nonproductive activity.
And in that earlier era of a tech crackdown, they saw video games as not really
part of the Chinese vision for a high-growth, technologically powered future,
when everyone’s at home playing video games.
They also cracked down on the education market. So there was
a lot of private tutoring — ed-tech start-ups were sprouting up — and they also
saw that as wasteful because it was sort of a race to the bottom in terms of
preparing for exams and feeding into that kind of cutthroat academic
environment.
Right now, I think we’re seeing something similar happen
again, with worries that A.I. companions could end up being a big time sink for
Chinese youth when they should be engineering the future and building out the
start-ups and the future Chinese versions of SpaceX, for example.
Douthat: But is there also a sense that this is the solution
if China never fixes its birthrate, that robots are just the way that aging
low-birthrate societies compete? Is that also part of the theory or the
mind-set?
Chan: Definitely. That’s a big part of the story.
So China has a shrinking work force — I think their labor
force size actually peaked over a decade ago — and they’re heavily dependent on
manufacturing. They don’t want to let that go. They see that as the engine for
the whole economy. So how do you reconcile those two factors — when people
don’t want those factory jobs anymore, and young people want different jobs —
and there’s just not enough people to fill the factories?
One solution is robots. One solution is to increasingly
automate factory production, to put robots of many different kinds, whether
they’re your classic six-axis industrial robot arm that can lift up a car in
one go ——
Douthat: The classic six-armer!
Chan: Or now this big push with humanoid robots is seen as
being yet another potential solution, if not a perfect solution, to this
ongoing labor issue.
So China wants to continue to become more and more
competitive, to move up the value chain and to make better and more high-value
stuff, but they don’t have the work force so A.I. and robotics is seen as the
way to fill that in.
Douthat: It’s interesting, you mentioned robot waiters. One
thing that has been sort of encouraging, I think, to people worried about job
displacement in the U.S. is the extent to which robotics in restaurants, fast
food places, supermarkets and so on has not, so far, radically displaced human
workers. In fact, places like McDonald’s and Starbucks that have tried to
really move to kind of automatic ordering and so on have often found themselves
maintaining human staff beyond what they expected — or even expanding human
staff.
In a context, though, where the Chinese birthrate is maybe
two-thirds the U.S. birthrate at this point, depending on which stats you look
at, you’re just in a different landscape. Maybe you’re worrying less about
whether the robot waiter displaces workers and more about whether you have a
waiter at all, and so the robot waiter is welcome and necessary?
That seems like it could be a big point of divergence,
ultimately, between how the U.S. and China relate to robots.
Chan: Yeah, definitely. You’re going to have to err on one
side or the other. You’re going to have to err on the side of going too slow,
and then you may not have the ability to do all these things because there’s
not enough workers there. Or you might err on the side of going too fast, and I
feel like that’s more of the concern in the U.S.
Douthat: Let’s go back to the A.G.I. superintelligence
question. How do you think China’s leaders actually think about the American
fixation, or the tech world’s — Sam Altman, Dario Amodei — fixation on A.G.I.?
Two options — you can tell me if there’s a third. One option
is that the Chinese basically think that our tech companies are high on their
own supply. That there is never going to be some insane return to
superintelligence, and it’s always going to be fine to be three to six months
behind, but then you catch up.
Another option would be that China is actually worried about
superintelligence and is basically trying to figure out what their contingency
plans are if the Americans seem to be pulling much further ahead.
Does either of those describe China’s mind-set, to the
extent that you can read the tea leaves in Beijing?
Chan: Yeah. One interesting corollary question is: Is China
trying to do an A.G.I. Manhattan Project somewhere, buried underground in a
bunker, with data centers that can’t be seen by satellites and are powered by
——
Douthat: Yes.
[Chan laughs.]
Douthat: Yes. Are they?
Chan: And my inclination is no.
Douthat: Do you think they could do something like that
without the U.S. being aware of it?
Chan: I don’t think that they would be able to do that
without the U.S. being aware. I think that it would require such a scale of
production, of amassing resources and construction, that we would detect
something, and we would start to wonder what is going on.
I mean, we’re already watching everything about the nuclear
weapons build-out, for example, in China. I would be very doubtful that we
would miss something of that scale, because you really would need a massive
scale, in terms of compute and energy, to power something that would be like a
Manhattan Project for A.G.I.
Douthat: So they’re not secretly trying to win the race.
Whatever they’re doing, they are sort of accepting this position of being in
our draft on the racetrack, or whatever metaphor you want — for now. But is
that just making a virtue of necessity, or do they think that we’re deluding
ourselves in our race to superintelligence?
Chan: [Chuckles] I think they just see the technology quite
differently, and they just don’t have that kind of transcendent view of
technology.
I think that you can see this in other approaches that
they’ve taken to the internet, or to the I.T. revolution, which they were
obsessed with as well. They were really focused on just trying to integrate the
internet and I.T. infrastructure into basic services — education, health care,
government services — and I think they see something similar with A.I. now.
One kind of thought experiment I often think about is: What
would be the signs that they were trying to do a secret A.G.I. program? And one
of the signs I think would be about those Nvidia chips that I mentioned
earlier.
Right now, Trump has relaxed some of the export controls and
allowed H200 Nvidia chips to be sold to China. Those are better than what China
had gotten before, but not the very best. And China has basically said,
“Thanks, but no thanks.”
The A.I. companies in China, to be sure, really, really want
those chips. But here’s the divergence: Beijing doesn’t necessarily want to be
dependent on the U.S. They want to bolster their own semiconductor program.
So if they were really sprinting today for A.G.I., I think
they would’ve gobbled up those chips as quickly as possible, not knowing when
that window might close. That is one indicator that they are kind of seeing
this as a medium- to long-term bet.
Douthat: So there might be people at DeepSeek who believe in
the superintelligence future more strongly than people in Beijing?
Chan: Yes.
Douthat: The closer you are to the machine god, the more its
voice whispers in your ear.
Chan: [Laughs] That’s right. Yeah, I don’t think that
Beijing is A.G.I.-pilled.
Douthat: What about espionage? Which obviously played a big
role in the early Cold War arms race, with nuclear secrets. Is there an
equivalent spy-based solution for China if the U.S. seems to be pulling too far
ahead?
Chan: So there is something called distillation. That’s
where you take a weaker model and you actually train it on the outputs of a
stronger model. Distillation is a common practice for A.I. developers when it’s
done with full knowledge and full disclosure and total authorization.
What seems to be happening now is that some of the Chinese
A.I. labs seem to be distilling on American A.I. models without authorization,
and they’re using a number of different proxy accounts so they can get around
efforts to block these campaigns.
Douthat: But that doesn’t require stealing secrets from
Anthropic. It just requires using the Anthropic model in a way that you’re not
supposed to be able to use it.
Chan: That’s right. It’s sort of its own category. It’s not
quite outright I.P. theft — it’s not like taking the source code from Anthropic
or OpenAI. It harks back a little bit to an era where Microsoft was always
trying to cut down on black market copies of Windows and Microsoft Office.
Douthat: Does it work in the sense that you can have a
Chinese Claude distilled that works as well as Claude?
Chan: It can help somewhat, but you need to have that
foundation to start with. I think that this is probably one area where it’ll
still be hard to get concrete data on exactly what the net effect is.
I would say, if you or I were building a model from scratch,
we would not be able to use distillation as a way to catch up to the frontier.
But if you were one of the better Chinese A.I. labs, you might be able to use
some of this to improve your model, especially on areas where you’re weaker. On
coding, for example, you might be able to use Anthropic’s Claude models to
support your long-term coding capabilities.
So there is that aspect to this whole A.I. race.
Douthat: In a world where there is some kind of take off,
one of the theories that animate the American A.I. companies is the idea that
at a certain level, the A.I.s start training the new A.I.s, and you get this
acceleration where suddenly, being three or six months behind, it becomes
impossible to catch up. Again, this would be the theory.
Suppose that takeoff starts to happen. Does China just
invade Taiwan?
[Chan laughs.]
Douthat: Well, seriously. It’s just a fascinating
circumstance in which you have a kind of arms race. Maybe China doesn’t think
of it as an arms race, but it is sitting next door to a central hub in the
supply chain that makes the arms race possible. Is that the natural Chinese
move in the event that they seem to be falling incredibly behind?
Chan: Ironically, if that were really starting to happen,
taking over TSMC would be a move too late. The chips are already made and
installed and are already running and training the models and feeding into this
feedback loop in the United States. At that point, all bets are off, and you’re
kind of out of options for what to do.
The big question here is how fast that can happen and
whether this could happen without being detected. There’s always speculation
about whether there is a version of the latest A.I. models that hasn’t been
shared or even disclosed to the public in, say, the U.S. or maybe even in
China, where they have gotten the inkling of this recursive feedback loop that
will lead to this superintelligence explosion. So that question is sort of hard
to know, and then, how quickly can you actually get there?
Douthat: I want you to be prescriptive for a moment, because
we’re having a summit. We’ve been talking about what China is doing, how China
is thinking, and so on. What does all of this mean for the United States in
terms of our policies? Does it mean that we should treat China as a
fundamentally more benign actor than our current policy treats them? Or is it
an indicator that, in fact, our policy is working by shaping a Chinese
perspective that is not as engaged in the race as it could be?
Chan: I think at this point, what we should do is take a
step back from this all-out-race framework, because right now, that race
mentality is driving a kind of recklessness, I would argue, from the American
side.
To bring up the threat of Chinese A.G.I. — we should think
about that, but I don’t think that that’s what they’re so focused on. But if
we’re only focused on that, that means we need to get rid of the guardrails. We
need to not bind ourselves. We need to not have any kind of regulation or
restrictions. We need to have as many data centers as possible everywhere.
Right now, that approach is starting to run into some
problems in the United States. And whether you’re talking about the backlash to
data centers, or you’re talking about some of these models now getting so
capable that they might not be at whatever A.G.I. level, but they are at the
level of potentially causing greater damage, either in terms of cyberattack
capabilities or maybe even in terms of augmenting what a relatively
unsophisticated group could do with bioweapons.
There are all these sorts of questions that the A.I.
community has been talking about for a long time. But certainly, for the Trump
administration, if you recall JD Vance’s speech last year, where he said
basically we should not have hand-wringing over A.I. safety slow down the
progress of American A.I. development. In other words, in this trade-off — and
he viewed it as a trade-off — we should err on the side of going faster rather
than putting on a seatbelt.
Now we’re reaching that point where we need to think about
still making progress as fast as possible, competing with China, making sure we
do have the best A.I. models — we can keep that. But does it have to come at
the expense of wearing a seatbelt or having some basic safeguards?
Douthat: Would you also suggest that the U.S. should adopt a
more Chinese vision of the goal of diffusion and building the best possible
A.I.-enabled technology right now?
A different
way to frame this is that the U.S. and China are in a race, but China thinks
it’s running a race to build the self-driving cars and the robots that every
single country in the world will use. And the U.S. will be stuck sitting here
with its pretend machine god while China sells to India, Africa and Latin
America successfully.
Do you think the U.S., in being less breakneck, should also
be pivoting to a strategy of, essentially, integration and sales?
Chan: Yes. I
think we need to focus a lot more on deployment. One of those areas is actually
open source, which, because of the commercial incentives, is not a high
priority for the top American A.I. labs. They’re focused on selling access to
their models through subscriptions, through A.P.I.s.
The thing
is, that open source approach has been really, really powerful for these
Chinese A.I. models to gain adoption — not just in China, but around the world.
So it feels like right now, the U.S. is ceding a really important channel of
competition.
When it’s so expensive, it can be the most powerful A.I.
model, but you don’t want to pay for it. That can put limits on your growth.
Douthat: Do you think you get that shift organically if
there is a slightly stronger regulatory hand? Again, the U.S. has “industrial
policy” — I put it in quotation marks — but we don’t have the kind of steering
of economic strategy that China has.
So it’s not like you can say, “Oh, the United States should
be more focused on deployment,” and there’s a button to push in Washington,
D.C., that makes that happen. But do you think it would happen naturally if it
was a little bit harder and a little bit more challenging to maximize compute
and capacity for existing A.I. companies?
Chan: I think there’s a way to tweak the incentives in a way
that is not like the Chinese approach, that is not about a top-down steering of
the whole industry, but is more about trying to create maybe some of that
commercial or even research space for, say, open source models.
You can think about a number of different markets where this
is happening, where there’s a focus on the high end of the market, on consumers
or businesses that are willing to pay a lot, but there’s less focus on mass
adoption and that broader marketplace.
And we’re
seeing some of this. I should be clear that Nvidia is trying to release open
source models. They have a commercial incentive because the more A.I. gets
adopted, the more their chips are needed. So there’s that closed loop there.
And Google DeepMind has some relatively good open source models.
But the
commercial incentives as they stand are not quite there.
Douthat: Do you think we should sell more chips to China as
a sort of token of a different model?
Chan: It’s a very difficult topic because anyone who tells
you yes or no on chips to China is really flattening the whole story.
On the one hand, you do have real near-term effects on
China’s ability to produce the most cutting-edge A.I. models. So by limiting
chips, that does slow down China’s A.I. development in the near term.
And that can
be useful, for example, for giving our companies that edge in cyberattack
capabilities.
With Mythos coming out, even a few months of being able to
test on our own systems first is very useful, versus a Chinese model having
this capability and they’re testing on our systems. So that’s important.
At the same time, there’s the other side of this whole
equation, which is accelerating China’s
own chip development. That’s an area that they’ve been really focused on, and
they’ve been focused on because of our export controls. So it cuts both ways.
In the near
term, it will slow down their A.I. development. In the longer term, it could
speed up at least their ability to have a more resilient, self-reliant
semiconductor supply chain that is not as affected by U.S. actions.
Somewhere in there is a sweet spot, and it’s really about
where you draw the line rather than just saying more chips or less chips.
Douthat: And also, how short timelines are overall.
Chan: Absolutely.
Douthat: And I’m just going to make the hawk’s case against
your case and see how you respond.
The hawk says: Look, we’ve been at this for an incredibly
short amount of time. Since ChatGPT appeared during the pandemic, there’s been
tremendous acceleration. The people who have predicted acceleration keep being
vindicated.
And yes, if you’re talking about a 20- to 25-year time
horizon for the point at which you sort of hit maximum superintelligence
capacity, then yeah, you have a lot of room to figure out the optimal
regulatory balance and all of these things.
But if you’re talking about two to four to six years, then
maintaining a three- to six-month lead over your leading rival — who, by the
way, is an authoritarian government — seems like it may be really, really,
really important. And the slowdown that you’re advocating is one that could
give up that advantage.
So how would you respond to that kind of argument, which
seems to be the mind-set that certainly not just people at the Pentagon but a
lot of people in Silicon Valley have?
Chan: So that timeline comes up again and again in so many
different debates within the U.S. as it relates to the U.S.-China A.I.
competition. And fundamentally, it’s impossible to say how that timeline will
play out.
So, for example ——
Douthat: I’ve discovered that in interviewing people.
Chan: [Laughs] Yeah, people will squirm on the timeline
question.
Douthat: Yes, it is impossible to say.
Chan: I mean, then it really boils down to what your views
are about this A.G.I. timeline and how likely this is to happen.
Another factor that I will throw in there, as a thought
experiment, is: Imagine that China did have access to the most cutting-edge
American A.I. chips. Would they be more A.G.I.-pilled? Would Beijing be more
A.G.I.-pilled? Forget about DeepSeek or the actual tech founders themselves.
Even on that, I’m not so sure that they would be so
A.G.I.-pilled. My guess would be that they would certainly try to deploy better
models, but they would basically run their current playbook, just amped up a
whole bunch.
Douthat: But even their current playbook includes
cyberwarfare. You just mentioned the fact that an advantage of just three
months in the deployment of a cyberwarfare-capable model like Mythos makes a
big difference. So it’s not as though the current Chinese playbook is sort of
innocent of conflict with the U.S.
Chan: That’s right. That’s why I see it as different sets of
risks. One is this A.G.I. risk that you’re talking about. That, I would argue,
has been sort of overblown.
What I don’t think has been overblown — and in fact, maybe
even underestimated up until recently — is the cyberrisk and the biosecurity
risk. It’s kind of crazy to say this, but those are more medium risks relative
to the A.I. catastrophic total takeover by superintelligence.
Those more intermediate risks I do worry about, and I do
worry about U.S. competition vis-à-vis China. That would be, in my mind, a
reason for maintaining the export controls that we currently have, and not
fiddling with them or agreeing to these side deals with Xi Jinping, for
example. So that’s why I try to find that balance.
But in terms of the A.G.I. question, that’s where I’m just
less convinced that we’re really all in this sprint — that China’s really all
in this sprint — for A.G.I.
Douthat: But even on the medium risks — which I agree seem
to be the most plausible risks — you are then making a calculation where you’re
saying: What am I most afraid of? Am I most afraid of China with the capacity
to do unprecedented cyberwarfare against the U.S., or a rogue A.I. or
disastrous A.I. model that crashes the entire U.S. power grid for some
inscrutable A.I.-related reason?
It’s that balance that you’re worrying about.
Chan: Yeah, exactly. And it comes to this question, too,
about how the U.S. should engage with China about A.I. If we are focused just
on China’s cyberattack capabilities relative to our own, then you might say:
Don’t bother engaging. We’re both in this arms race, essentially, on
cybercapabilities.
But if you’re thinking about the rogue agent or, say, a
nonstate actor using either a set of American models, a set of Chinese models,
or maybe they do arbitrage — I mean, this is sort of like 4-D chess, where they
are deliberately playing this geopolitical competition against each other and
trying to distribute an attack across all these different models in order to
disguise their origins.
Those are areas where I do think that, one, it would be
useful to talk to the Chinese side about these, and two, where I think it would
be in the U.S. national interest. It wouldn’t just be about binding ourselves
and slowing ourselves down relative to China. It would be about this extra
third factor that we want to take seriously.
Douthat: And this is a good place to end, because a lot of
people in Silicon Valley will say: Oh yeah, in theory, we could engage with
China and negotiate a sort of mutual A.I. slowdown. But in practice, either
it’s not clear that China wants to do that, wants that kind of negotiation, or
it’s just unimaginably complex to verify some sort of A.I. control agreement in
the way that we did with nuclear missiles during the Cold War.
Do you think a Cold War-style ongoing A.I. control
negotiation with China is possible?
Chan: I think we should not have high expectations. I
certainly don’t.
I think that we should start by talking. We should start by
sharing our approach to A.I. safety and A.I. risk mitigation. We should try to
convince the Chinese to take this more seriously — and they are starting to
take this more seriously.
We should also have a discussion about open source models,
actually. As those get better, on the one hand, we want those to diffuse more,
but on the other hand, they could also pose a risk if they get into the wrong
hands.
So we can talk about all those areas, but I would be very
hesitant, certainly at this stage, to even think about binding constraints,
verification agreements, a kind of arms control treaty for A.I. between the
U.S. and China.
At this stage, it’s way too early. Let’s just start talking.
Douthat: If it’s too early for that, is it just because of
the sheer difficulty of imagining such a thing? Or is it a dynamic where
precisely because Beijing’s attitude is that we’re not in some Cold War-style
race, they’re actually less interested than they otherwise would be in that
kind of negotiation?
Chan: I think overall it really boils down to one thing,
which is an extremely low degree of trust between the U.S. and China, and an
unwillingness for either side to subject ourselves to invasive verification,
monitoring and surveillance by the other party.
And yeah, there could be interesting technical solutions
that would make that more feasible, but it boils down to this geopolitical
reality where we don’t trust them, and they don’t trust us.
So we might be able to make progress on areas that affect
both of us, but when it comes to letting, say, Chinese regulators come into the
U.S., or letting American regulators go inspect data centers in China, I think
that is pretty, pretty far out there at this stage.
Douthat: Do you think that that only changes on the far side
of some disaster, conflict, some sort of event? Because one theory that I don’t
just toy with, but I guess I hold, is that a lot of the negotiations around
nuclear weapons were only possible because they’d been used, and people were
aware of how destructive they are. Is there a world where the only way that the
U.S. and China come to terms is a world where something tragic has to happen
first?
Chan: Yeah, that’s a scenario I think about too. And I think
about what would be the level of incident, and what could the response be. You
can think about a most extreme case where you have some major cyberattack
incident, or even a bioweapons incident related to A.I., where there are real
lives at stake, for example. That could cause both countries to just
unilaterally put a pause on all of their A.I. development because they realize
that this is such a big issue with such huge risks. That is possible.
So I do wonder and I do worry that we might be waiting for
that incident to happen before we take action in advance, before we even start
to talk to each other about how to take action.
Douthat: All right. On that somewhat dark note, Kyle Chan,
thank you for joining me.
Chan: Thank you.” [1]
1. Why China Isn’t Worried A.I. Will Replace Its Workers:
interesting times. Douthat, Ross; Sophia Alvarez Boyd. New York Times (Online)
New York Times Company. May 14, 2026.
