AI superpowers: China and the U.S.

Below is a complete transcript of the China Corner Office Podcast with George Shen.

Chris: Hi everyone. Thanks so much for joining us today on China Corner Office, a podcast powered by The China Project, the New York-based news and information platform that helps the West reach China between the lines. I’m Chris Marquis, a professor at the Cambridge Judge Business School. And today, we are joined by George Shen from IBM. George is a tech exec and thought leader who has a longstanding focus on [artificial intelligence] AI and cloud technologies spanning both the U.S. and China. While George has been in the U.S. for decades, he grew up in China and attended Tongji University in Shanghai and was an AI researcher, even as far back as the mid-1990s. Before we dive into the discussion, George wanted me to point out that the views he expresses here are his alone and don’t represent IBM’s point of view.

At the outset of the show, George provides some helpful context into the general development of artificial intelligence technologies. We then discuss details and the similarities and differences between the AI strategy of the U.S. and China. One key difference that George highlights is China’s whole nation approach that is exemplified by the 2017 China National AI Strategy, which we go through in detail. The U.S., conversely, does not have any sort of national overarching strategy, and the strategies are fragmented across the private and public sectors. We also, however, discuss the recent report from former Google CEO, Eric Schmidt, that does give suggestions and recommendations for national advancement of AI.

One aspect of that report that has gotten some media attention is Schmidt’s prediction that China will overtake America and AI by 2025. We discuss the likelihood of that prediction, and George provides some helpful insights on the strengths and weaknesses of China’s path toward being a global leader in AI. In particular, we focused on the use of AI in different industries in China, from facial recognition to public surveillance, as well as leading players in the field from large well-known tech firms like Baidu and Alibaba, to more specialized up and comers like Megvii and Hikvision, who also have been in the news recently.

We also discussed differences in public reaction to mass adoption of AI in China, and also in the U.S. This actually led us into a discussion of ethics of AI and how cultural and social factors across the two countries result in different attitudes. For instance, in China, people being much more tolerant towards AI technologies like facial recognition than in the U.S.

Thanks so much for listening and enjoy the show.

Chris: George, welcome to The China Corner Office.

George: I’m glad to be here, Chris.

Chris: First, it’d be great to just dive into a little bit of background. I mean, I think that’s always a nice way to start off. Our topic is AI and talking about the development of AI in both China and the U.S., differences, similarities. How did we get here in some ways? Because if you think about it, just probably five, 10 years ago, AI was mostly talked about in movies, but now it’s really part of the daily discussion, political discussion, and obviously a lot of economic consequences too.

George: Maybe I can start with some of the background on AI, how we came to be.

Chris: Sure, that’s great.

George: And as you said, there’s a lot of popular literature and the popular culture has AI. In human imagination, I think that some sort of intelligent being can perform some sort of activity or exhibit some kind of intelligence is always in human imagination. If we think about the English word, “automaton,” the etymology is Greek. So, it means some sort of self-operating device can perform some task. And since ancient Greek, we can also find some equivalent automaton concept in ancient Egypt as well. Fast forwarding to Middle Ages, as well as Renaissance period, you have seen the concept really evolve from sort of like a dumb device, to a more intelligent kind of being. I’m sure you notice that in the da Vinci sketch, he had actually a great sketch, a very detailed design about a kind of a knight. And a lot of people believe he actually made a model in humanoid form, which still has a replica still available. So, the idea is always there from folklore, from fairy tales, from mythology. People always imagine a kind of supernatural or some kind of device that can perform intelligence tasks.

Now, the word “artificial intelligence” was actually coined by the great computer scientist and also engineer, John McCarthy. He taught in Dartmouth College, and also moved to Stanford after he founded a MIT lab, AI lab with another great mind in AI, whose name is Marvin Minsky. So, both of them actually led a workshop in the summer of 1956. Lot of people, very smart people, attended this workshop. In this conference, basically, McCarthy coined the term artificial intelligence. Basically, he defined the field as artificial intelligence, which could comprised of a lot of different things.

So, the world of artificial intelligence is really big. It actually has a lot of different fields, sub branches, if you will. I can point out a couple of key branches within the term artificial intelligence. First, computer vision. If you think about how computers can see things like human do, this is widely used in autonomous driving. So, computer vision is number one. Number two is language processing. How computers can understand what you say, what I say, then respond. This is called natural language processing. Another field is something called knowledge representation. I think about how humans have a lot of common sense. Where does common sense come from? And how common sense is represented in our mind is also a big research field. This is called a knowledge representation. Another field is called automated reasoning, how humans can use logic. Since ancient Greek time, humans are able to use logic to derive, to infer, to deduce. So, this is a capability that we very much want to compute to learn as well. The last thing I want to say is more popular today is part of machine learning.

Chris: Machine learning’s also in the news quite a bit.

George: Right. Exactly. We will dive into, what sort of application machine learning can make? And this is a very interesting field right now. How machines can learn either by something called supervise the learning, or unsupervised the learning, or reinforce the learning. There’s different methods to teach machines how to learn. Machine learning is a really big field. But in essence, this is really about pattern recognition. How do you recognize the pattern in a problem so that you can solve a new problem based on what you learn?

Chris: One thing implicit, a lot of the writing about China AI is that pattern recognition in machine learning, and the idea being okay, if you have these huge data sets from whomever, Alipay or whomever, this actually is a big resource for machine learning. Can you say a little bit more about the specifics of AI today and in the recent past in China, in the U.S., and how maybe there’s different strategies that the different countries or companies in the countries are adopting?

George: I will say this, I actually get this question quite a lot. Especially when it comes to where is China vis a vis the U.S. when it comes to AI research and development? I actually like to divide our topic a little bit into two subtopics, if you will. One is about research and development. Another is about application and adoption. To me, these are pretty different kinds of paradigms, if you will. In terms of R&D, as we just discussed, U.S. pretty much pioneered the idea of artificial intelligence in modern days, right?

By the way, I want to mention that, before artificial intelligence, people always think that computers predated artificial intelligence. That actually is not quite right. As we said, the idea of artificial intelligence, maybe it’s not called artificial intelligence, but the idea has always been there since ancient times. Now, once computers were invented, actually exactly 200 years ago, by a professor of Cambridge, in fact, whose name is Charles Babbage. He is basically credited with the invention of a mechanical computer in 1822, which led to more advanced electronic computers first for analog than nowadays to digital.

The idea of artificial intelligence really is the history of computer science because all the computer scientists want to use computers to mimic or imitate or simulate how humans do things naturally. Since the ‘50s, we talk about ‘60s, ‘70s, ‘80s, ‘90s, even to a large extent, the U.S. led the R&D pretty much single-handedly. Once we get into this century, we see a lot of other countries starting to catch up, right? Like the U.K., Canada.

Chris: Can I ask you a question? When you say the U.S. led, I mean, was it companies like IBM or was it actually researchers at MIT, Caltech? I mean, was this industry led, or university led, or defense department led? Who was doing this work?

George: Good question. Actually, I would say both. In the ‘60s, ‘70s, by large, it’s more about government sponsored programs, like the DoD (Department of Defense) program, especially under DARPA (Defense Advanced Research Projects Agency). They sponsored a lot of small people like John McCarthy and Marvin Minsky. In fact, if you just look at it, who attended the Dartmouth Conference? The two giants I just mentioned, both of them actually worked for universities. Marvin Minsky worked for MIT for a very long time, and was a professor at MIT. John McCarthy is doing a lot of research work at Stanford. But there’s one guy whose name is Nathaniel Rochester who actually worked for IBM, was also in the conference. There’s another big heavyweight whose name is Claude Shannon, who is basically known as the father of information theory. Very, very famous. He worked at Bell Labs. So, I would say, if you look at all these people’s backgrounds, I would say the combination of mostly research institutions, universities, some industry labs like Bell Labs, as well as corporations like Toshiba, IBM, and a lot of these big names.

Chris: Can you say then… So, okay. We’ve come up U.S. as this early pioneer decades of fundamental research work. And so, now China has really risen and is being seen as sort of an AI superpower. Can you describe the China trajectory then and how that may differ as far as who’s involved and what sort of subfields they’re focused on?

George: Let me finish the story on the U.S. side and quickly dive into the China side. After the 60, 70, 80, most of these university research, R&D led effort will not see a lot of real-world application during this period of time. Maybe one exception is the expert system, which is widely popular to use in DEC like digital equipment corporations. Basically, it’s a kind of a heuristic system, right? Use a bunch of rules to define what you can do, what you cannot. It’s a kind of decision tree, where you can derive your outcome. But that system is outdated nowadays.

But more specifically, if we look at the 90s and the 2000s, the corporations gradually caught up. Corporation like IBM, which actually invest a lot of money in artificial intelligence. One of the achievements is in 1997, IBM Deep Blue beat Kasparov, the best human champion player in chess. And then in 2011, again, IBM had a system called Watson, played the jeopardy game and beat two human champions, Ken Jennings, and Rutter. The two human champ was beaten by a computer. This actually is a field called natural language processing. Then, fast forward to 2016, DeepMind is a U.K. company which was acquired by Google, played one of the best human players in Go, and also had a large board, much bigger computation than chess. In 2016, the DeepMind algorithm called AlphaGo was able to beat one of the best human players in the world whose name is Lee Sedol.

In fact, 2016 is also a very important year for China. If pivot to China, I would start with 2016. 2016 is really the pivotal moment. We can even call it the historic moment for China in AI. Why? Because after the five-game theories, which computer one four, Lee Sedol, could manage to win one game. At that time, this sent basically a shock wave to the research and scientists, and not only in the area of AI, but as well as the people who caught the news and were shocked by this news, and how AI was progressing so fast, to be able to surpass the human best player in Go. At that moment, the Chinese government immediately stepped in to make a lot of directives, dictate the strategy, using the whole nation approach. We call it jǔguó tǐzhì 举国体制. Basically, concentrating all the resources, using the government, managing all the resources, dedicated to one task, which is to advance AI in the field of research and development, as well as adoption than compete with U.S. This is a critical moment for China, but if we look at how China, in terms of technology, advanced, China actually had a very interesting history in this whole nation approach. A lot of people in the U.S. have the idea of philosophy, I would say. That the whole nation would probably not work, we were diving into, what kind of problem maybe can work for this kind of approach? Maybe other problems do not work so well.

Just in terms of the technology, China has some success, although you can argue it’s limited success. In the ‘60s and the ‘70s, China had a success in technology such as, we call it liǎng dàn yì xīng 两弹一星, which is two bombs and a wind satellite, which is the China’s own version of the whole nation approach to create nuclear bombs, the atomic first, then the hydrogen bomb. Then the West satellite, which means the satellite was sent to outer space successfully. Or by the whole nation approach. Then you can argue that, “Hey, China had some success with this kind of approach.”

Fast forwarded to the ‘90s. Once the internet came to China and China was able to catch up a lot of the progress, you see a lot of startups in the dotcom era in China. You also have a lot of AI companies that started in the early 2000s catching up with the idea of computer vision, facial recognition, as well as different applications in video and audio in natural language processing, in image recognition, video perception, this sort of application. Interestingly, if we talk about AI, we cannot avoid talking about the chips. A lot of advanced chips used AI. If you look at the chip industry in China, very quickly, in 2000, China had its first important chip manufacturing company called SMIC, Semiconductor Manufacturing International Corporation. Although it is registered in Cayman Island, it’s a Chinese company, no doubt. It has manufacturing plants in Shanghai, Tianjin, Wuhan. And I think they’re also starting new plants in Shenzhen as well. So, China is trying to catch up with the whole nation effort in hardware, in chips, in software, in algorithms, in research and development. That’s kind of the whole picture if we go and see. So, China, after the 2016 historic moment for China, adopted, very early on, a nationwide AI strategy in 2017.

Which is way earlier than the U.S. The U.S. had an AI plan in 2020. Then it commissioned a study led by Eric Schmidt, led a very well-documented research study, which is called the National Security Commission for Artificial Intelligence. The report itself is like almost a hundred pages, called out a lot of interesting strategies for the U.S. government to adopt.

Chris: George, can we actually go back a little bit to this China? You said you had this national strategy. What areas specifically, and obviously facial recognition is a famous area, machine learning. These are things I just read about in the media. Facial recognition is frightening to think about from a Western perspective. Maybe is it overhyped or is that actually an area that China has really been focused on?

George: Let me answer your first question first, which is about the national strategy. China, as I said, was one of the first countries to release a nationwide strategy. In fact, Canada did it even before China. What’s interesting is that the U.S. did not have a national strategy as of now. You can call out the Eric Schmidt report, which tried to help, and asked the U.S. government to adopt the strategy he laid out. But whether or not the U.S. government will do that, that’s a different question. But coming back to the Chinese national strategy, I want to call out a couple of things. First, the strategy identified as the national champions. Called lǐngxiù 领袖.

Chris: Okay. And who are the lǐngxiù 领袖?

George: So, they are more than a dozen companies. First you have Baidu. Baidu is going to be responsible for autonomous driving. And even search engines use AI as well to come up with different algorithms. The second one is Alibaba. Alibaba is responsible as a national champion for smart city. Smart city is actually a term used a lot in the West as well. In China, it actually has a different connotation, which I will dive into a little bit later. The third one is Tencent. Tencent is actually, will be responsible for healthcare, medical imaging, which also uses artificial intelligence to detect, diagnose certain cancer or some other diseases in medical imaging technology.

There are also a lot of other companies, such as, a company called Kuangshi, Megvii, it’s really used for facial recognition. It used to be called Face++, its first product. I think they renamed the company to Kuangshi, but the company is focused on facial recognition technology, cooperating with the Chinese government to adopt this technology widely used in surveillance and public safety, public security arena. Another company is called Shangtang. They’re responsible for something called smart vision. Very similar to Megvii. These two companies had a lot of similarities. They overlap a lot. Has image recognition, smart vision. Huawei obviously, right? Huawei actually has its own semiconductor chip design subsidiary called HiSilicon. HiSilicon designed a lot of the water glass. I would even say some of the chip design capability, very close to matching the U.S. capability. Very close to match. But this is before the sanction of Huawei. After the sanction of Huawei, the EDA sanction, which basically cut off the software used for designing the Silica chips. It will be very interesting to see how Huawei will survive, how HiSilicon will survive as well. The last company I want to point out maybe is Hikvision. Hǎi kāng wēi shì 海康威视. The company specializes in video surveillance technology. It’s a national champion for video perception.

So, those are the key companies. Of course, other companies as well, but these are the key companies. I also want to call out a three program, which is not widely reported in the U.S. First program is called a safe city. Safe city is, in Chinese, called píng’ān chéngshì 平安城市. Safe cities have a lot of similarities to smart cities. Basically, it is very much focused on surveillance, public safety, public security, traffic management, city emergency, reaction, response, disaster alert. So, this kind of small city capability is very much embedded in this program. The second program is called a Skynet, basically called Tiān wǎng 天网. It is a program basically focused on surveillance networks, the hardware, the cameras, the network supporting the surveillance, as well as the software used for control. This program is applied at the provincial and the city level. The last one is called bright project. But in English, I guess the best translation is called sharp eye. Xuěliàng gōngchéng 雪亮工程. There’s a slang phrase in Chinese called qúnzhòng de yǎnjīng shì xuěliàng de 群众的眼睛是雪亮的, the people’s bright eyes can catch the criminal or the spies. And it is very much focused on the surveillance camera security monitoring software. And it is supposed to be widely applied and adopted in smaller cities, rural townships, counties, and villages.

Chris: One of the questions I had is really, so understanding how the priorities differ and the extent to which those programs really are so focused on public safety surveillance, really pops out at me. In the U.S., I can’t imagine that the programs are so focused on such public security. Can you say a little bit about the U.S. plan or the Eric Schmidt plan?

George: Let me touch upon the U.S. side in terms of public reaction to technology like facial recognition. You would see a lot of pushback from the general public in the U.S. In the very privacy conscious mind of the public, such as the U.S., it is very concerning. It will cause a lot of backlashes. Even just the topic of it, let alone implementation. The facial recognition is more nuanced, I think, in terms of the discussion. You see a lot of the debate in the U.S. First, you see public resentment towards the police use of facial recognition to catch criminals. And they are research scientists in MIT discovered that the kind of facial recognition is very immature nowadays. And if you use that in a real-world scenario, you very much will misidentify people, particularly people with dark skin tone. It was a concern that potentially you could criminalize the innocent.

There are cities, San Francisco, Cambridge, that are banning it, pushing the regulation, and banning the usage of facial recognition in police departments. Some of the articles you can see nowadays, you also see pushback on that notion as well. Police are saying that, “Oh, look, we are shorthanded. We don’t have enough resources to monitor all the areas which traditionally have very high rate of criminal activities.” So, what do you do about that? There’s still lot of debate. But in terms of national strategy and national regulation, there’s none in the U.S.

But also, I want to point out that if we talk about at the state level, there are regulations. There is a website called the National Conference of State Legislature. It attracts all legislation related to artificial intelligence. It has like hundreds, if not thousands of legislation proposed at the state level. People are talking about how we can, using legislation, to force the use of AI, which is more moral, ethically acceptable.

Chris: We talked about Eric Schmidt and his Foreign Affairs article, and I know that he has warned China will overtake America in AI by 2025. What do you think of that? Does that statement ring true to you? Are there qualifications for it? I’d love to hear your opinion on that.

George: I have a great respect for Eric Schmidt. Eric is a great mind. He was the former CEO of Google, and his leadership made a lot of progress, building a very capable company, as well as a lot of AI ability. A lot of his point of view on target. However, when it comes to China’s AI ability, I think there is a lot of hype in his assessment. I don’t want to come across as somebody who underestimates China’s capability, but I think in order to sell yourself, overstating the problem is a way you can get government funding.

But I will point out that one point he may have a valid point on, which is the Chinese capability of adopting AI to quickly scale at a speed that the U.S. wouldn’t be able to match. So, if we talk about facial recognition, surveillance technology using AI, population control, that area, I think, China is in uncharted territory, in my view. In that respect, Eric is right. China is adopting very aggressively all these technologies we just mentioned. But in terms of AI research and development, R&D, I think he’s incorrect. I want to call out that U.S. is very much still the leading nation. China, I wouldn’t say even close. However, by some measures, China may be leading. He could point out to, let’s say China is leading in AI paper publication, right? In a number of AI papers, if you look at it, in 2017 or ’18, by some measures, AI is ahead of the U.S. in fact. But I don’t think that is a credible measurement to look at holistically in terms of the power and the capability of a nation in AI. I will say AI, China is behind the U.S., behind the UK even, behind Canada even. U.S., U.K., Canada, Japan would be the leaders. China is not in the first tier. I wouldn’t even say that.

Chris: This idea of sort of using number of papers, number of patents as indices of innovation and power can be misleading. Actually having governmental programs that have the funding KPI, so to speak, as patents and papers. So, that just results in actually high numbers, not necessarily real innovation.

George: Yeah. Indeed. Let me conclude with one point. Eric maybe is right about China’s ambition could rival the U.S., maybe even surpass the U.S. That’s why the U.S. still does not have a national strategy in AI. Let me quickly call out three things in terms of AI strategy, which is outlined in the National Strategy paper we just mentioned. First phase, by 2020, which we already passed now, China should be in line with its competitors on AI. By 2025, China should reach the world leading in some of the fields of AI, whether it’s facial recognition, you name it. And the same report also calls out, by 2030, China should be the primary center for AI innovation in the world. So, that is very ambitious. If you look at that, Eric is certainly right.

Chris: We started talking about how in the U.S., there’s a lot of social resistance, a lot of legal resistance to a variety of AI. How about in China? Are you seeing any pushback? One of my few experiences in China with AI directly or sort of experiencing AI, was when I took some students to a demonstration site of an IoT company that had installed a bunch of fire sensors. And along with those were cameras. In a village not that far from Beijing. And we went there and went to their control room, and it was, really was surprising to Westerners. A student asked, they said, “Well, do people mind at all the fact that they are being monitored all the time?” And the person that was giving us a tour responded, “Well, no, actually the opposite, they really appreciated because one woman, for instance, she lost her four-year-old son, and she was able to come to this control center and then they could actually find him where the son had sort of wandered off to.” How are some of the Chinese population or media thinking about AI?

George: Let me touch upon the cultural aspect of your question. Culturally, I think you are right. Majority of the general public will not push back very hard on this. The reason being that culturally, historically speaking as well, traditionally, Chinese people don’t have the same sense of privacy and security as the typical American would have. China certainly has more tolerance, I would say, right? And some would even say, “Hey, as long as I’m a good citizen, I have not done anything wrong, why I’m afraid of being watched over.” But I think in general, you are correct to point out that the society tolerance is very different when it comes to the privacy issue, I think.

So, we are talking about legislation. One question I also get a lot is, what China is doing because of regulations? The first law, as you may know, in 2017, China published the first cyber security law which is really relevant to what we are talking about. Then in 2021, it enhanced, on top of that, cybersecurity law to new laws. Data security law, which essentially defines who can access this data, defines the data sovereignty concept, who can control it, who can access it, who can manage it, who can possess. Stipulating that data generated in China cannot leave China. The third one is called Personal Information Protection Law, PIPL. What corporations can access or can gather, or can collect, and how you can manage your access to the data. But lastly, just this year, which is the regulation on algorithm recommendations, how you can implement your algorithm in your programs to make recommendations. Government wants to make sure that the companies adhere to social norms, morality, and code of conduct. When I look at this law, it is very broadly brushed and it’s very difficult to implement, in my view. Because how do we define moral? Subject to a lot of debate or definition or criteria. So, the law itself did not define clearly, but in any case, the law is already pushed out.

Chris: Well, unfortunately we’re out of time, George. This has been super interesting, enlightening. Thank you so much, George, for joining us on China Corner Office.

George: Thank you, Chris.