October 31, 2024 • 36 mins
This week, Professor Werbach is joined by USC Law School professor Angela Zhang, an expert on China's approach to the technology sector. China is both one of the world's largest markets and home to some of the world's leading tech firms, as well as an active ecosystem of AI developers. Yet its relationship to the United States has become increasingly tense. Many in the West see a battle between the US and China to dominate AI, with significant geopolitical implications. In the episodoe, Zhang discusses China’s rapidly evolving tech and AI landscape, and the impact of government policies on its development. She dives into what the Chinese government does and doesn’t do in terms of AI regulation, and compares Chinese practices to those in the West. Kevin and Angela consider the implications of US export controls on AI-related technologies, along with the potential for cooperation between the US and China in AI governance. Finally, they look toward the future of Chinese AI including its progress and potential challenges.
Angela Huyue Zhang is a Professor of Law at the Gould School of Law of the University of Southern California. She is the author of Chinese Antitrust Exceptionalism: How the Rise of China Challenges Global Regulation which was named one of the Best Political Economy Books of the Year by ProMarket in 2021. Her second book, High Wire: How China Regulates Big Tech and Governs Its Economy, released in March 2024, has been covered in The New York Times, Bloomberg, Wire China, MIT Tech Review and many other international news outlets.
High Wire: How China Regulates Big Tech and Governs Its Economy
The Promise and Perils of China's Regulation of Artificial Intelligence
Want to learn more? Engage live with Professor Werbach and other Wharton faculty experts in Wharton's new Strategies for Accountable AI online executive education program. It's perfect for managers, entrepreneurs, and advisors looking to harness AI’s power while addressing its risks.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Angela, welcome to the Road to Accountable AI.
Hi, hello, thanks for having me.
Before we get to a conversation about AI in China, can you talk a little bit about thehistory of the tech sector there and maybe give people some framework for comparison to
the US or other countries they may be familiar with?
(00:22):
Well, China is really the only country other than the United States that have been able tofoster big tech companies so far.
in early 2021, like Alibaba, Tencent, I mean, these companies were on the verge ofreaching the one trillion dollar market cap.
(00:42):
And so soon would join this trillion dollar market cap club, although subsequently theyface a big crackdown.
They suffer severe market cap loss.
But seriously, we are talking about the world's largest big tech companies.
And partly because in the consumer tech sector, Chinese government imposed this, erectedthis great firewall that have made it very difficult for US social media businesses to
(01:15):
operate in China.
So in a way that that create entry barrier, very high entry barrier.
for US companies.
So far, only US firm that have been able to gain a little bit foothold is Microsoft.
But still, Bing has only 1 % of the market share in China's search business.
(01:36):
Many Americans that I talk to are under the impression that there's no difference betweenthe Chinese government or the Chinese Communist Party and private companies like Alibaba
and ByteDance and so forth.
Is that accurate?
Well, I think that's definitely an oversimplification of the situation in China.
Look, mean, Elon Musk also operates a hugely successful business in China.
(02:00):
And China is the reason why he became the richest man in the world, because the Chinesegovernment really helped Tesla to create a very, to have access to a very convenient
supply chain and also to
open up a huge production assembly line in Shanghai.
(02:24):
So the government was in many ways extending helping hands to its own tech firms and evento Elon Musk's company as well.
So in that way, that is something that you don't see in other jurisdictions.
And it's also true that companies operating in China, they exhibit
(02:47):
a much higher level of cooperation with the government.
You don't see them fighting with the government in courts, in anti -trust or big datacases.
Even Elon Musk himself, which sent poor emoji to the US government, will be really, reallyfriendly to the Chinese government.
(03:13):
He came and visited the Chinese government a few months ago.
And that's why the Tesla car now can drive Tesla car in some of the government compounds,which were restricted before.
mean, so I think it's definitely different type of relationship.
(03:34):
But I wouldn't say these companies are completely watchdog of the Chinese government.
You mentioned earlier the crackdown on companies like Alibaba and so forth.
So how did the relationship with the Chinese state and the tech sector change?
For a really long time, they were in the honeymoon period.
(03:56):
I would say from the early 2010s until late 2020, the Chinese government has played a veryimportant role propping up Chinese tech sector.
And in this period, I would see regulation more as a helping hand rather than a grabbinghand.
(04:16):
However, since late 2020, we experienced a massive crackdown of
China's own indigenous tech sector and big companies like Alibaba and Tansan, everybodyelse were affected.
And these companies really suffer a great deal, right?
(04:36):
mean, so since then, the government has really tightened its scrutiny over these farms.
However, that campaign only lasted for 18 months.
So the government relaxed.
its regulatory control in early 2022.
So now we are back to a lag stage.
(04:57):
So there's a lot of volatility in Chinese regulation and there's something that is veryuncommonly seen in other jurisdictions.
And that's something that I elaborated in detail in my book, High Wire.
was the rationale for that crackdown?
Was it similar to the antitrust motivation that we've seen in Europe, in the US, vis -a-vis the big tech platforms?
(05:24):
Yes and no.
OK, yes, because these kind of actions are actually way overdue.
And the Chinese tech sector is actually much more concentrated than the US and theEuropean tech sector.
China is home to really two duopolies, Alibaba and Tencent, which hold a massive hundredsof portfolio companies each in their own ecosystem.
(05:53):
created this erected this wall gardens that prevent companies, rival companies to accessthe ecosystem.
And so create all sorts of competition issues.
And however, because for a really long time, the government was trying to nurture thisindustry, regulators took a very lack stand.
(06:17):
So that kind of bureaucratic inertia or
agency inaction also partly further exacerbate this problem.
Now, in the late 2020, what really changed the game, however, is the government woke upand realized, look, I these companies, the Alibaba, Tencent, they're just not, they're not
(06:39):
just selling clothes or video games.
They are now, you know, controlling our own financial system because
At the end of 2020, Alibaba was, not Alibaba, Ant Group, which is a sister company ofAlibaba, and it's founded by Jack Ma, was on the verge of getting listed in Shanghai and
(07:05):
Hong Kong.
And this was supposed to be the biggest IPO of the year.
And one of the big parts of Ant's business is the...
the credit tag businesses, which Ant as an intermediary connecting ordinary users and thesmall and medium sized state -owned banks.
(07:32):
But Ant had very little skin in the game, so the Chinese government was quite concernedabout the moral hazard problem.
But what worries them more is that Ant got an extremely high valuation of 320 billion USdollars.
In other words, that valuation at the time put in at a higher, now put in as a bigger bankthan JP Morgan at the time.
(07:58):
I mean, so this was a 60 year old company that's been from Alibaba and it's grow to besuch a huge scale and size and with a big hidden problem of moral hazard.
And that really worries the Chinese government.
I mean, that explains why
(08:18):
They suspended Ant's IPO at 11 hour.
So a lot of people were looking at that whole saga as like, Jack Ma said something stupidat the financial summit.
And so Xi Jinping got very upset.
And then they initiated this huge crackdown.
(08:39):
But the story goes deeper than that.
mean, actually the tension went way back.
I mean, it's fundamentally.
It was this huge concentration of these big tech companies and their pervasive influencethat really alarmed the Chinese government.
And they realized, you know, these have threatened potentially financial stability,economic stability, and ultimately their political control.
(09:08):
Okay, let's pivot to AIs.
That's what I really want to focus on with you for most of our time.
You talked about the high barriers to entry for the US tech companies in China.
And similarly, the big Chinese companies with the possible exception of TikTok are notprominent in the US.
Is the same dynamic playing out in the AI sector?
(09:28):
Yes, in a way, yes it is.
Although AI, we are talking about many different services.
There are hardware services, software services.
Give an example of facial recognition technology.
This is the most widely applied technology in China.
And China really has become the biggest exporter of facial recognition technology.
(09:50):
But no Western countries want to purchase the Chinese tech product.
right?
Even though they are the most accurate, they are really efficient.
If you come to China, you will experience that.
It's everywhere.
The camera is everywhere.
And you basically go through all the security check seamlessly and highly efficient.
(10:11):
So I was quite frustrated when I live in US or in Europe.
You still need a lot of human control of the system.
But in China, you really experience extreme level of efficiency with this technology.
And similarly, in China, you can't access a track GPT, although it's not the government'schoice to block it.
(10:36):
It's more of a track GPT doesn't open AI doesn't want to operate in China, even block itsAPI in China.
But Microsoft operate in China and Microsoft continue to offer similar services throughits Azure to Chinese companies.
a little bit integration, but to the extent this dynamic, this kind of splinter netdynamics continues in the age of AI.
(11:06):
And what has been the Chinese state's approach to regulating AI?
Yeah, I mean, the Chinese government has a very ambivalent approach in regulating AI.
Well, I mean, it depends on which technology we're talking about.
mean, go back to facial recognition technology.
This technology has been so widely used in China in public and commercial settings.
(11:30):
And this has generated a lot of controversy.
And in fact, in some cases, even some lawsuits from some activist lawyers.
And I think there are growing discontent with some of the Chinese citizens against thiskind of public surveillance.
However, to this day, China has not adopted a comprehensive legislation to regulate thistechnology.
(12:00):
For sure, if China wants to have one, it can have it tomorrow.
And it has been drafting it for a long time.
So the reason for a delay is not that China does not have the capacity to write this lawor to adopt it.
It's more of other institutional and bureaucratic interests holding it back because theydon't want to stifle the commercial growth of this technology.
(12:22):
I mean, if an AI technology does not affect the fundamentals, the bottom lines of theChinese Communist Party, I don't see China is in any way in a rush.
to adopt a piece of legislation that might potentially hold back innovation that mightdeter investment and entrepreneurship.
(12:46):
Now, on the other hand, when CheckGPT emerged, China was the first one to rush to thescene.
And pulling back actually suspended a lot of the CheckGPT -like services in Apple's AppStore.
just three months after its launch.
(13:07):
And China was also the first one to promulgate quite comprehensive rules called interimmeasures to regulate genitive AI in just six months after CHKGPT's release.
And then a year ago, China already adopted this law way ahead of the EU's AI Act.
Now, what explained that proactivity there?
(13:30):
This is also quite intuitive because
This again goes back to the bottom line of the Chinese Communist Party because they areconcerned that AI, genitive output, particularly from this genitive AI technology may not
be aligned with the political value of the Chinese Communist Party.
So they call it the socialist core values, right?
(13:52):
mean, which is a very broad concept.
And...
And this is something that the government was most concerned about.
So you can imagine, right?
mean, so this type of regulation will be really targeted at content moderation.
But funny enough, you see the exact contrast in the US.
This is something that the US really lacks because of Section 230 and a lot ofconstitutional rights protection in the US, right?
(14:20):
But in China, content moderation is the single area.
which the government has imposed the most control.
So that's interesting.
what you're saying is some things like facial recognition and emotion detection and soforth are the things that are most heavily regulated in the US and Europe.
And some things like content moderation, which are the least regulated here or the mostregulated in China and place versa.
(14:47):
Yeah, I mean, I'm sure that autonomous driving and facial recognition, they will be stillsubject to some sort of sector regulation.
I once talked to one of the GC at general council at a leading Chinese automobile company.
Is that my day job, I spent half of my day job just meeting different government officialsand different government regulators and from various sectors.
(15:13):
I mean, they're not just dealing with one agency.
They're dealing with the whole bureaucracy here.
However, they're taking a consensual approach.
It's not like the government make a big announcement like how they deal with Alibaba.
I mean, suddenly, an announcement out of the blue shocked the whole huge investor, thehuge equity market.
(15:34):
It's more a consensual approach.
here are the things that you need to do and you need to tweak.
And here's what you can do to do compliance.
So it's more friendly approach, right?
mean, and so a more softer way of regulating these businesses.
(15:56):
Given the nature of generative AI though, where it's generative, it's you can't beabsolutely certain what it's going to create, is this extreme strictness about content and
presumably a great deal of fear among the companies if they get it wrong in terms of whatthe consequences might be, you much more so than if open AI has some bad output.
(16:18):
Is that going to be, you think, a fundamental limit on the success and the quality ofChinese large language models?
It definitely will hinder the development of the B2C products because in China, if youwant to introduce public saving, generative AI product and services, you need to obtain a
(16:44):
license from the cyberspace administration of China.
They don't officially call it license.
They said that you need to conduct a security assessment.
Basically, you need to go to CAC and then file registration and then
get some sort of green light before you can introduce your services to the public.
So in this way, company already voluntarily complied to conduct a self -censorship andimpose a lot of the filtering at the last stage of the development of the AI model to make
(17:15):
sure the content generated will not generate problematic, harmful, or inappropriatecontent.
And that will definitely affect the user experience because very often time you put it ina prom and then it tells you that they don't have the answer, this is not available.
(17:38):
So definitely affect user experience.
But let's also bear in mind that the government does not regulate enterprise use AIapplication.
In fact, interim measures of regulating generative AI only apply to public phasingservices.
So if you are selling to a company like Huawei, ZTE, or TCL, or whatever, you're notsubject to this kind of censorship control.
(18:12):
So to that extent, I don't think it necessarily hold back the development of thetechnology itself.
However, it definitely reduce user experience in a way that also make this product a bitless
interest attractive.
But to be honest with you, all these farms in China are struggling to capital to profitfrom this new genitive AI tools, right?
(18:38):
Company like Baidu, like Huashou, they, everybody is offering this kind of tools, butthey're burning every money every day.
To this day, none of them have come up with a sustainable business strategy planned.
as to how I can make money on this thing.
I mean, yes, OpenAI, can charge money fee.
(19:00):
But in China, people get used to free stuff.
And so there's a big problem.
Although this generated a huge hype, but it's not really bringing in a lot of profits yetfor this company.
It's still at the investment stage.
And it's very expensive because before, for this consumer tech business,
(19:24):
It's kind of like a fixed cost.
You develop this and you can use it.
But here, every time somebody put in an inquiry, you have to deploy your computingresource, which is hugely expensive.
And Chinese companies are really short of CPU at the moment to train the models, not tosay for deployment.
So that, again, that limit on China's GPU resources, again, also
(19:50):
kind of undermine user experience, right?
mean, so because companies don't want you to access the most powerful and most advancedmodel.
And so, yeah.
So you're talking about the US chips acts and other limits on Chinese access to the mostpowerful GPUs.
How much of an impact is that really having on the Chinese AIS?
(20:13):
It definitely has a very big impact.
mean, because this is really the biggest bottleneck on the development of Chinese AIfarms, especially the big tech companies.
You probably have read in recent weeks some of the reports in Financial Times and alsoJournal saying that, mean, this kind of expert control doesn't seem, the US expert control
(20:36):
program doesn't seem to be very effective.
Now it's even cheaper to buy CPU in China than in the United States because
There are these kind of export restriction laws are full of loopholes.
And by the way, it's not that difficult to smuggle Nivea's H100.
It's not that big.
There's a story that a lot of the Chinese students, everyone bring back H100.
(21:08):
But this is just one of the ways.
mean, I'm sure there are 100 million other ways that Chinese companies can get around it.
But it will be hard for the Chinese big tech giants like Alibaba and Tencent because thesecompanies, these are big companies, right?
mean, so they will put compliance.
They will highly value compliance.
They have business operation everywhere.
(21:30):
So this company will definitely...
suffer a great deal.
And in fact, Alibaba, which operates China's leading cloud business now, want to cut backsome of the cloud access from small and medium sized companies because even they are
running out of this capacity.
And that definitely has an impact.
(21:52):
Now, on the other hand, so in the short term, it seems the US program has worked.
But in the medium and long run, it's also creating a big problem for the US.
first because this kind of restriction, it's also very costly for the US in a way it'scounterproductive in a way because you are restricting your own companies to sell to a
(22:18):
very lucrative Chinese market while your allies may not be restricting it.
so in fact, a lot of the US toolmakers suffer a lot of, suffer tremendous loss because ofthat.
because companies in Japan and Korea, or Netherlands, they were not restricted in sellingto Chinese companies, while US companies are.
(22:44):
And by the way, I also need to mention that I think the US got really lucky with the AIboom.
If the US did not experience this huge AI boom, this kind of extra restriction wouldreally hurt.
the business operation of companies like Nvidia.
If you may recall a few years ago, Nvidia even felt bankruptcy because of all theserestrictions.
(23:10):
that's why, know, so a lot of the adverse impact of the US export restriction were notreally felt, to be honest, thanks to the AI boom.
So if we consider that counterfactual.
So first is hurting US business and in a way indirectly will harm US innovation becauseyou need
the Chinese market, you need that profit in order to supply and fry.
(23:34):
Just look at Intel, which recently cut, you know, suffered a huge loss in the equitymarket.
Because China accounted 35 % of its revenue and the US government want Intel to succeed inUS to build more factories in the US and develop more in the US, in a way cutting back the
(23:56):
business in China.
And the company really suffered.
Right.
mean, and I'm not saying that that was the single China was the single most importantcontributing factor, but it definitely had played a role in hurting Intel's business.
Right.
But more importantly, above all, all these export restrictions is incentivizing China tobecome more self -sufficient.
(24:18):
And you are seeing Huawei, for instance, you know, at the beginning did suffer a loss, butin the past couple of years, it's doing phenomenally well.
And it's it's
It's iPhone businesses back, right?
I mean, and people were teasing while Huawei was releasing its new tripod, try screens, aphone, vis -a -vis the Apple iPhone, right?
(24:43):
mean, and it also closely aligned the public and the private incentives.
because now companies like Alibaba and Tencent, these are the big tech companies, theywere not able to access Nivea chips.
They have to buy it from Huawei, right?
I mean, so in a way they are contributing to the profits of Huawei and that becomes like avirtual cycle that continue to nurture China's development.
(25:09):
Now, nobody knows at what point China will become really become self -sufficient andbecause the go -go post
keep moving because the US is also moving ahead very rapidly.
This is something that we don't know.
But from an academic standpoint, we are evaluating the trade -off between the short -termbenefits to the US and the long -term loss.
(25:31):
So there's a lot of variables.
But in a way, what we see the US is doing right now, we see very often a lot of theseexport control and restrictions.
They don't seem to be very well thought out and there's not a very clear feedbackmechanism to evaluate success.
And I think this is something that we should query.
(25:54):
Has the US government done enough to evaluate what it is doing and whether it's effectiveor not?
The other thing that is happening in the background is on the Chinese side, the shift inthe Chinese government's approach more generally to the economy since the pandemic.
And there was a fairly shocking statistic in the Financial Times recently that there wassomething like in 2018, there were 50 ,000 venture -backed startups created in China and
(26:22):
2023 it was 1 ,200 and still going down.
I want to ask you, first of all, is that accurate in terms of what you see in the marketthere?
And if so, what does that mean for the Chinese AI sector?
Right.
mean, that is definitely compatible with my research findings in my book.
actually have one chapter specifically devoted to assessing the impact of China's techcrackdown.
(26:44):
Is that because of this kind of really heavy handed intervention during that 18 month lawenforcement campaign, it had really undermined investor confidence and stifled
entrepreneurial spirit.
And as a result, both the domestic and international businesses
investors have retreated from China's consumer tech businesses.
(27:07):
And that explains this dramatic drop of funding into China's tech sector.
But on the other hand, a lot of funding were channeled into the high tech sector, the so-called hardcore tech businesses like semiconductors, like EV, high manufacturing, et
(27:31):
cetera.
Right?
I mean, but still overall, the whole macroeconomic environment in China in the past coupleof years was not very good.
Right?
I mean, the Chinese economy was doing phenomenally well in early 2021.
If we recall, China was the first one to recover from COVID and the first one to resume aproduction line where everybody else was suffering.
(27:57):
But since then, everything has gone downhill.
I mean, a series of policy missteps like the COVID pandemic control and this massive techcrackdown, as well as the property crackdowns, right?
I mean, has directly contributed to China's economic slowdown.
(28:20):
And that definitely reduced the appetize of the foreign investors in Chinese tech sector.
So let me ask you just one or two more questions and shift back to AI and regulation.
You talked about that China could, if it wanted to adopt a comprehensive AI law, buthasn't, but has some of these more narrow laws.
(28:46):
How do you see the development of either regulatory coordination or at least regulatorydiscussions across the world around AI, given what's happening in Europe and the US and
other places?
There's a lot of cooperation going on, but so far we see that the dialogue has remained atquite a high level on agreeing on very basic principles like accountability, transparency,
(29:19):
this kind of thing.
We haven't yet seen very meaningful and significant and concrete proposals as to how toensure AI safety.
And part of the reason is because, I think most important reason is because the twobiggest AI superpower, China and the US, are engaged in locking this intense tech rivalry.
(29:46):
And how could you persuade China to really cooperate when at the same time the US isdoubling down on export restrictions of AI chips to China, right?
mean, so.
Even though you have seen the two sides at the highest level actually have engaged in acouple of dialogues in San Francisco and then in Geneva, right?
(30:12):
But they seems to be talking without really doing anything.
So I would say lower your expectation of anything that will really generate from this AIinternational cooperation.
But I do think that it's extremely important.
Most importantly, I think the two sides should exert more efforts in understanding therisk of AI.
(30:37):
Because so far, there is a huge controversy and everybody think that this is quitespeculative.
So if the Chinese government come to realize that there are serious threat, because ifsomething is a serious threat to...
(30:58):
its citizens and the society and the government.
I'm pretty sure the Chinese government will deal with it very decisively, just like how ithas been able to manage COVID -19.
Xi Jinping was even touting its zero COVID strategy.
The big problem is nobody knows where these things will become an existential threat andhow likely it is.
(31:21):
So think at least on that point, I think the US and China can at least cooperate.
and engage more.
And finally, what do you expect for the future development of AI in China in the next fewyears?
I know it's a broad question, as you've accurately pointed out, there's different thingsthat we talk about when we talk about AI, but what should people look to in the next few
(31:44):
years in terms of the Chinese AI sector?
China is actually really traveling very closely behind the US.
Mark Zuckerberg was saying that Chinese companies are just four or five months behind us.
And others say, look, I with the large language models, China is maybe a few years behind.
(32:05):
But if we are talking about text to image, text to video, China is really on par.
You saw recently, Huashou released this text to video app.
It's really good.
they recently, but they have to kind of, they can't really introduce it to the marketbecause it's too expensive to deploy this thing, right?
(32:25):
Because they're running out of CPU, right?
I mean, so let's watch this space.
This is not like semiconductor where China needs decades of experience and support inorder to develop that sector.
China already is hugely successful, particularly in the consumer.
(32:46):
application side of AI.
And that's why, you know, TikTok can take over the US market so quickly, right?
I mean, this is a Chinese technology.
But with regard to the hardware application, like robotics, for instance, China is alsovery good, right?
Facial recognition, voice recognition, but everything else China is still lagging behind.
(33:10):
And query how much the semi -concept AI chips might be also holding
back China's development, although at the same time, people are saying that these daysthere's new technology that emerge.
You don't really need that much chips computing power in order to train very powerfullarge language models.
So I would definitely see Chinese as a very serious rival to the United States.
(33:33):
But at the same time, I would also highlight that there is also a lot of swing states thatare really catching up.
mean, even UAE is producing really
really good in cutting edge large language model.
And again, this is a foreign -taran country that are able to mobilize tremendous resourcesto develop this technology.
And they have money, they have energy, right?
(33:55):
mean, and then they don't suffer from export restrictions like China, right?
And in fact, US companies want to cooperate with G42 from the UAE, unlike in China, right?
I so there's other big players that should be under a close radar.
and closely watch as well.
(34:16):
Great, Angela, thank you so much for your time.
My pleasure.
Thanks for having me.
Angela, welcome to the Road to Accountable AI.
Hi, hello, thanks for having me.
Before we get to a conversation about AI in China, can you talk a little bit about thehistory of the tech sector there and maybe give people some framework for comparison to
the US or other countries they may be familiar with?
(00:22):
Well, China is really the only country other than the United States that have been able tofoster big tech companies so far.
in early 2021, like Alibaba, Tencent, I mean, these companies were on the verge ofreaching the one trillion dollar market cap.
(00:42):
And so soon would join this trillion dollar market cap club, although subsequently theyface a big crackdown.
They suffer severe market cap loss.
But seriously, we are talking about the world's largest big tech companies.
And partly because in the consumer tech sector, Chinese government imposed this, erectedthis great firewall that have made it very difficult for US social media businesses to
(01:15):
operate in China.
So in a way that that create entry barrier, very high entry barrier.
for US companies.
So far, only US firm that have been able to gain a little bit foothold is Microsoft.
But still, Bing has only 1 % of the market share in China's search business.
(01:36):
Many Americans that I talk to are under the impression that there's no difference betweenthe Chinese government or the Chinese Communist Party and private companies like Alibaba
and ByteDance and so forth.
Is that accurate?
Well, I think that's definitely an oversimplification of the situation in China.
Look, mean, Elon Musk also operates a hugely successful business in China.
(02:00):
And China is the reason why he became the richest man in the world, because the Chinesegovernment really helped Tesla to create a very, to have access to a very convenient
supply chain and also to
open up a huge production assembly line in Shanghai.
(02:24):
So the government was in many ways extending helping hands to its own tech firms and evento Elon Musk's company as well.
So in that way, that is something that you don't see in other jurisdictions.
And it's also true that companies operating in China, they exhibit
(02:47):
a much higher level of cooperation with the government.
You don't see them fighting with the government in courts, in anti -trust or big datacases.
Even Elon Musk himself, which sent poor emoji to the US government, will be really, reallyfriendly to the Chinese government.
(03:13):
He came and visited the Chinese government a few months ago.
And that's why the Tesla car now can drive Tesla car in some of the government compounds,which were restricted before.
mean, so I think it's definitely different type of relationship.
(03:34):
But I wouldn't say these companies are completely watchdog of the Chinese government.
You mentioned earlier the crackdown on companies like Alibaba and so forth.
So how did the relationship with the Chinese state and the tech sector change?
For a really long time, they were in the honeymoon period.
(03:56):
I would say from the early 2010s until late 2020, the Chinese government has played a veryimportant role propping up Chinese tech sector.
And in this period, I would see regulation more as a helping hand rather than a grabbinghand.
(04:16):
However, since late 2020, we experienced a massive crackdown of
China's own indigenous tech sector and big companies like Alibaba and Tansan, everybodyelse were affected.
And these companies really suffer a great deal, right?
(04:36):
mean, so since then, the government has really tightened its scrutiny over these farms.
However, that campaign only lasted for 18 months.
So the government relaxed.
its regulatory control in early 2022.
So now we are back to a lag stage.
(04:57):
So there's a lot of volatility in Chinese regulation and there's something that is veryuncommonly seen in other jurisdictions.
And that's something that I elaborated in detail in my book, High Wire.
was the rationale for that crackdown?
Was it similar to the antitrust motivation that we've seen in Europe, in the US, vis -a-vis the big tech platforms?
(05:24):
Yes and no.
OK, yes, because these kind of actions are actually way overdue.
And the Chinese tech sector is actually much more concentrated than the US and theEuropean tech sector.
China is home to really two duopolies, Alibaba and Tencent, which hold a massive hundredsof portfolio companies each in their own ecosystem.
(05:53):
created this erected this wall gardens that prevent companies, rival companies to accessthe ecosystem.
And so create all sorts of competition issues.
And however, because for a really long time, the government was trying to nurture thisindustry, regulators took a very lack stand.
(06:17):
So that kind of bureaucratic inertia or
agency inaction also partly further exacerbate this problem.
Now, in the late 2020, what really changed the game, however, is the government woke upand realized, look, I these companies, the Alibaba, Tencent, they're just not, they're not
(06:39):
just selling clothes or video games.
They are now, you know, controlling our own financial system because
At the end of 2020, Alibaba was, not Alibaba, Ant Group, which is a sister company ofAlibaba, and it's founded by Jack Ma, was on the verge of getting listed in Shanghai and
(07:05):
Hong Kong.
And this was supposed to be the biggest IPO of the year.
And one of the big parts of Ant's business is the...
the credit tag businesses, which Ant as an intermediary connecting ordinary users and thesmall and medium sized state -owned banks.
(07:32):
But Ant had very little skin in the game, so the Chinese government was quite concernedabout the moral hazard problem.
But what worries them more is that Ant got an extremely high valuation of 320 billion USdollars.
In other words, that valuation at the time put in at a higher, now put in as a bigger bankthan JP Morgan at the time.
(07:58):
I mean, so this was a 60 year old company that's been from Alibaba and it's grow to besuch a huge scale and size and with a big hidden problem of moral hazard.
And that really worries the Chinese government.
I mean, that explains why
(08:18):
They suspended Ant's IPO at 11 hour.
So a lot of people were looking at that whole saga as like, Jack Ma said something stupidat the financial summit.
And so Xi Jinping got very upset.
And then they initiated this huge crackdown.
(08:39):
But the story goes deeper than that.
mean, actually the tension went way back.
I mean, it's fundamentally.
It was this huge concentration of these big tech companies and their pervasive influencethat really alarmed the Chinese government.
And they realized, you know, these have threatened potentially financial stability,economic stability, and ultimately their political control.
(09:08):
Okay, let's pivot to AIs.
That's what I really want to focus on with you for most of our time.
You talked about the high barriers to entry for the US tech companies in China.
And similarly, the big Chinese companies with the possible exception of TikTok are notprominent in the US.
Is the same dynamic playing out in the AI sector?
(09:28):
Yes, in a way, yes it is.
Although AI, we are talking about many different services.
There are hardware services, software services.
Give an example of facial recognition technology.
This is the most widely applied technology in China.
And China really has become the biggest exporter of facial recognition technology.
(09:50):
But no Western countries want to purchase the Chinese tech product.
right?
Even though they are the most accurate, they are really efficient.
If you come to China, you will experience that.
It's everywhere.
The camera is everywhere.
And you basically go through all the security check seamlessly and highly efficient.
(10:11):
So I was quite frustrated when I live in US or in Europe.
You still need a lot of human control of the system.
But in China, you really experience extreme level of efficiency with this technology.
And similarly, in China, you can't access a track GPT, although it's not the government'schoice to block it.
(10:36):
It's more of a track GPT doesn't open AI doesn't want to operate in China, even block itsAPI in China.
But Microsoft operate in China and Microsoft continue to offer similar services throughits Azure to Chinese companies.
a little bit integration, but to the extent this dynamic, this kind of splinter netdynamics continues in the age of AI.
(11:06):
And what has been the Chinese state's approach to regulating AI?
Yeah, I mean, the Chinese government has a very ambivalent approach in regulating AI.
Well, I mean, it depends on which technology we're talking about.
mean, go back to facial recognition technology.
This technology has been so widely used in China in public and commercial settings.
(11:30):
And this has generated a lot of controversy.
And in fact, in some cases, even some lawsuits from some activist lawyers.
And I think there are growing discontent with some of the Chinese citizens against thiskind of public surveillance.
However, to this day, China has not adopted a comprehensive legislation to regulate thistechnology.
(12:00):
For sure, if China wants to have one, it can have it tomorrow.
And it has been drafting it for a long time.
So the reason for a delay is not that China does not have the capacity to write this lawor to adopt it.
It's more of other institutional and bureaucratic interests holding it back because theydon't want to stifle the commercial growth of this technology.
(12:22):
I mean, if an AI technology does not affect the fundamentals, the bottom lines of theChinese Communist Party, I don't see China is in any way in a rush.
to adopt a piece of legislation that might potentially hold back innovation that mightdeter investment and entrepreneurship.
(12:46):
Now, on the other hand, when CheckGPT emerged, China was the first one to rush to thescene.
And pulling back actually suspended a lot of the CheckGPT -like services in Apple's AppStore.
just three months after its launch.
(13:07):
And China was also the first one to promulgate quite comprehensive rules called interimmeasures to regulate genitive AI in just six months after CHKGPT's release.
And then a year ago, China already adopted this law way ahead of the EU's AI Act.
Now, what explained that proactivity there?
(13:30):
This is also quite intuitive because
This again goes back to the bottom line of the Chinese Communist Party because they areconcerned that AI, genitive output, particularly from this genitive AI technology may not
be aligned with the political value of the Chinese Communist Party.
So they call it the socialist core values, right?
(13:52):
mean, which is a very broad concept.
And...
And this is something that the government was most concerned about.
So you can imagine, right?
mean, so this type of regulation will be really targeted at content moderation.
But funny enough, you see the exact contrast in the US.
This is something that the US really lacks because of Section 230 and a lot ofconstitutional rights protection in the US, right?
(14:20):
But in China, content moderation is the single area.
which the government has imposed the most control.
So that's interesting.
what you're saying is some things like facial recognition and emotion detection and soforth are the things that are most heavily regulated in the US and Europe.
And some things like content moderation, which are the least regulated here or the mostregulated in China and place versa.
(14:47):
Yeah, I mean, I'm sure that autonomous driving and facial recognition, they will be stillsubject to some sort of sector regulation.
I once talked to one of the GC at general council at a leading Chinese automobile company.
Is that my day job, I spent half of my day job just meeting different government officialsand different government regulators and from various sectors.
(15:13):
I mean, they're not just dealing with one agency.
They're dealing with the whole bureaucracy here.
However, they're taking a consensual approach.
It's not like the government make a big announcement like how they deal with Alibaba.
I mean, suddenly, an announcement out of the blue shocked the whole huge investor, thehuge equity market.
(15:34):
It's more a consensual approach.
here are the things that you need to do and you need to tweak.
And here's what you can do to do compliance.
So it's more friendly approach, right?
mean, and so a more softer way of regulating these businesses.
(15:56):
Given the nature of generative AI though, where it's generative, it's you can't beabsolutely certain what it's going to create, is this extreme strictness about content and
presumably a great deal of fear among the companies if they get it wrong in terms of whatthe consequences might be, you much more so than if open AI has some bad output.
(16:18):
Is that going to be, you think, a fundamental limit on the success and the quality ofChinese large language models?
It definitely will hinder the development of the B2C products because in China, if youwant to introduce public saving, generative AI product and services, you need to obtain a
(16:44):
license from the cyberspace administration of China.
They don't officially call it license.
They said that you need to conduct a security assessment.
Basically, you need to go to CAC and then file registration and then
get some sort of green light before you can introduce your services to the public.
So in this way, company already voluntarily complied to conduct a self -censorship andimpose a lot of the filtering at the last stage of the development of the AI model to make
(17:15):
sure the content generated will not generate problematic, harmful, or inappropriatecontent.
And that will definitely affect the user experience because very often time you put it ina prom and then it tells you that they don't have the answer, this is not available.
(17:38):
So definitely affect user experience.
But let's also bear in mind that the government does not regulate enterprise use AIapplication.
In fact, interim measures of regulating generative AI only apply to public phasingservices.
So if you are selling to a company like Huawei, ZTE, or TCL, or whatever, you're notsubject to this kind of censorship control.
(18:12):
So to that extent, I don't think it necessarily hold back the development of thetechnology itself.
However, it definitely reduce user experience in a way that also make this product a bitless
interest attractive.
But to be honest with you, all these farms in China are struggling to capital to profitfrom this new genitive AI tools, right?
(18:38):
Company like Baidu, like Huashou, they, everybody is offering this kind of tools, butthey're burning every money every day.
To this day, none of them have come up with a sustainable business strategy planned.
as to how I can make money on this thing.
I mean, yes, OpenAI, can charge money fee.
(19:00):
But in China, people get used to free stuff.
And so there's a big problem.
Although this generated a huge hype, but it's not really bringing in a lot of profits yetfor this company.
It's still at the investment stage.
And it's very expensive because before, for this consumer tech business,
(19:24):
It's kind of like a fixed cost.
You develop this and you can use it.
But here, every time somebody put in an inquiry, you have to deploy your computingresource, which is hugely expensive.
And Chinese companies are really short of CPU at the moment to train the models, not tosay for deployment.
So that, again, that limit on China's GPU resources, again, also
(19:50):
kind of undermine user experience, right?
mean, so because companies don't want you to access the most powerful and most advancedmodel.
And so, yeah.
So you're talking about the US chips acts and other limits on Chinese access to the mostpowerful GPUs.
How much of an impact is that really having on the Chinese AIS?
(20:13):
It definitely has a very big impact.
mean, because this is really the biggest bottleneck on the development of Chinese AIfarms, especially the big tech companies.
You probably have read in recent weeks some of the reports in Financial Times and alsoJournal saying that, mean, this kind of expert control doesn't seem, the US expert control
(20:36):
program doesn't seem to be very effective.
Now it's even cheaper to buy CPU in China than in the United States because
There are these kind of export restriction laws are full of loopholes.
And by the way, it's not that difficult to smuggle Nivea's H100.
It's not that big.
There's a story that a lot of the Chinese students, everyone bring back H100.
(21:08):
But this is just one of the ways.
mean, I'm sure there are 100 million other ways that Chinese companies can get around it.
But it will be hard for the Chinese big tech giants like Alibaba and Tencent because thesecompanies, these are big companies, right?
mean, so they will put compliance.
They will highly value compliance.
They have business operation everywhere.
(21:30):
So this company will definitely...
suffer a great deal.
And in fact, Alibaba, which operates China's leading cloud business now, want to cut backsome of the cloud access from small and medium sized companies because even they are
running out of this capacity.
And that definitely has an impact.
(21:52):
Now, on the other hand, so in the short term, it seems the US program has worked.
But in the medium and long run, it's also creating a big problem for the US.
first because this kind of restriction, it's also very costly for the US in a way it'scounterproductive in a way because you are restricting your own companies to sell to a
(22:18):
very lucrative Chinese market while your allies may not be restricting it.
so in fact, a lot of the US toolmakers suffer a lot of, suffer tremendous loss because ofthat.
because companies in Japan and Korea, or Netherlands, they were not restricted in sellingto Chinese companies, while US companies are.
(22:44):
And by the way, I also need to mention that I think the US got really lucky with the AIboom.
If the US did not experience this huge AI boom, this kind of extra restriction wouldreally hurt.
the business operation of companies like Nvidia.
If you may recall a few years ago, Nvidia even felt bankruptcy because of all theserestrictions.
(23:10):
that's why, know, so a lot of the adverse impact of the US export restriction were notreally felt, to be honest, thanks to the AI boom.
So if we consider that counterfactual.
So first is hurting US business and in a way indirectly will harm US innovation becauseyou need
the Chinese market, you need that profit in order to supply and fry.
(23:34):
Just look at Intel, which recently cut, you know, suffered a huge loss in the equitymarket.
Because China accounted 35 % of its revenue and the US government want Intel to succeed inUS to build more factories in the US and develop more in the US, in a way cutting back the
(23:56):
business in China.
And the company really suffered.
Right.
mean, and I'm not saying that that was the single China was the single most importantcontributing factor, but it definitely had played a role in hurting Intel's business.
Right.
But more importantly, above all, all these export restrictions is incentivizing China tobecome more self -sufficient.
(24:18):
And you are seeing Huawei, for instance, you know, at the beginning did suffer a loss, butin the past couple of years, it's doing phenomenally well.
And it's it's
It's iPhone businesses back, right?
I mean, and people were teasing while Huawei was releasing its new tripod, try screens, aphone, vis -a -vis the Apple iPhone, right?
(24:43):
mean, and it also closely aligned the public and the private incentives.
because now companies like Alibaba and Tencent, these are the big tech companies, theywere not able to access Nivea chips.
They have to buy it from Huawei, right?
I mean, so in a way they are contributing to the profits of Huawei and that becomes like avirtual cycle that continue to nurture China's development.
(25:09):
Now, nobody knows at what point China will become really become self -sufficient andbecause the go -go post
keep moving because the US is also moving ahead very rapidly.
This is something that we don't know.
But from an academic standpoint, we are evaluating the trade -off between the short -termbenefits to the US and the long -term loss.
(25:31):
So there's a lot of variables.
But in a way, what we see the US is doing right now, we see very often a lot of theseexport control and restrictions.
They don't seem to be very well thought out and there's not a very clear feedbackmechanism to evaluate success.
And I think this is something that we should query.
(25:54):
Has the US government done enough to evaluate what it is doing and whether it's effectiveor not?
The other thing that is happening in the background is on the Chinese side, the shift inthe Chinese government's approach more generally to the economy since the pandemic.
And there was a fairly shocking statistic in the Financial Times recently that there wassomething like in 2018, there were 50 ,000 venture -backed startups created in China and
(26:22):
2023 it was 1 ,200 and still going down.
I want to ask you, first of all, is that accurate in terms of what you see in the marketthere?
And if so, what does that mean for the Chinese AI sector?
Right.
mean, that is definitely compatible with my research findings in my book.
actually have one chapter specifically devoted to assessing the impact of China's techcrackdown.
(26:44):
Is that because of this kind of really heavy handed intervention during that 18 month lawenforcement campaign, it had really undermined investor confidence and stifled
entrepreneurial spirit.
And as a result, both the domestic and international businesses
investors have retreated from China's consumer tech businesses.
(27:07):
And that explains this dramatic drop of funding into China's tech sector.
But on the other hand, a lot of funding were channeled into the high tech sector, the so-called hardcore tech businesses like semiconductors, like EV, high manufacturing, et
(27:31):
cetera.
Right?
I mean, but still overall, the whole macroeconomic environment in China in the past coupleof years was not very good.
Right?
I mean, the Chinese economy was doing phenomenally well in early 2021.
If we recall, China was the first one to recover from COVID and the first one to resume aproduction line where everybody else was suffering.
(27:57):
But since then, everything has gone downhill.
I mean, a series of policy missteps like the COVID pandemic control and this massive techcrackdown, as well as the property crackdowns, right?
I mean, has directly contributed to China's economic slowdown.
(28:20):
And that definitely reduced the appetize of the foreign investors in Chinese tech sector.
So let me ask you just one or two more questions and shift back to AI and regulation.
You talked about that China could, if it wanted to adopt a comprehensive AI law, buthasn't, but has some of these more narrow laws.
(28:46):
How do you see the development of either regulatory coordination or at least regulatorydiscussions across the world around AI, given what's happening in Europe and the US and
other places?
There's a lot of cooperation going on, but so far we see that the dialogue has remained atquite a high level on agreeing on very basic principles like accountability, transparency,
(29:19):
this kind of thing.
We haven't yet seen very meaningful and significant and concrete proposals as to how toensure AI safety.
And part of the reason is because, I think most important reason is because the twobiggest AI superpower, China and the US, are engaged in locking this intense tech rivalry.
(29:46):
And how could you persuade China to really cooperate when at the same time the US isdoubling down on export restrictions of AI chips to China, right?
mean, so.
Even though you have seen the two sides at the highest level actually have engaged in acouple of dialogues in San Francisco and then in Geneva, right?
(30:12):
But they seems to be talking without really doing anything.
So I would say lower your expectation of anything that will really generate from this AIinternational cooperation.
But I do think that it's extremely important.
Most importantly, I think the two sides should exert more efforts in understanding therisk of AI.
(30:37):
Because so far, there is a huge controversy and everybody think that this is quitespeculative.
So if the Chinese government come to realize that there are serious threat, because ifsomething is a serious threat to...
(30:58):
its citizens and the society and the government.
I'm pretty sure the Chinese government will deal with it very decisively, just like how ithas been able to manage COVID -19.
Xi Jinping was even touting its zero COVID strategy.
The big problem is nobody knows where these things will become an existential threat andhow likely it is.
(31:21):
So think at least on that point, I think the US and China can at least cooperate.
and engage more.
And finally, what do you expect for the future development of AI in China in the next fewyears?
I know it's a broad question, as you've accurately pointed out, there's different thingsthat we talk about when we talk about AI, but what should people look to in the next few
(31:44):
years in terms of the Chinese AI sector?
China is actually really traveling very closely behind the US.
Mark Zuckerberg was saying that Chinese companies are just four or five months behind us.
And others say, look, I with the large language models, China is maybe a few years behind.
(32:05):
But if we are talking about text to image, text to video, China is really on par.
You saw recently, Huashou released this text to video app.
It's really good.
they recently, but they have to kind of, they can't really introduce it to the marketbecause it's too expensive to deploy this thing, right?
(32:25):
Because they're running out of CPU, right?
I mean, so let's watch this space.
This is not like semiconductor where China needs decades of experience and support inorder to develop that sector.
China already is hugely successful, particularly in the consumer.
(32:46):
application side of AI.
And that's why, you know, TikTok can take over the US market so quickly, right?
I mean, this is a Chinese technology.
But with regard to the hardware application, like robotics, for instance, China is alsovery good, right?
Facial recognition, voice recognition, but everything else China is still lagging behind.
(33:10):
And query how much the semi -concept AI chips might be also holding
back China's development, although at the same time, people are saying that these daysthere's new technology that emerge.
You don't really need that much chips computing power in order to train very powerfullarge language models.
So I would definitely see Chinese as a very serious rival to the United States.
(33:33):
But at the same time, I would also highlight that there is also a lot of swing states thatare really catching up.
mean, even UAE is producing really
really good in cutting edge large language model.
And again, this is a foreign -taran country that are able to mobilize tremendous resourcesto develop this technology.
And they have money, they have energy, right?
(33:55):
mean, and then they don't suffer from export restrictions like China, right?
And in fact, US companies want to cooperate with G42 from the UAE, unlike in China, right?
I so there's other big players that should be under a close radar.
and closely watch as well.
(34:16):
Great, Angela, thank you so much for your time.
My pleasure.
Thanks for having me.
Popular Podcasts
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.