There’s this assumption that because you have an AI plan, you’ve solved the problem. Whereas actually central forecasting, and planning about complicated, uncertain technological developments is oftentimes flawed and doesn’t solve anything. It may actually be counterproductive.
The State Council of China’s 2017 AI plan was the starting point of China’s AI planning; China’s approach to AI is defined by its top-down and monolithic nature; China is winning the AI arms race; and there is little to no discussion of issues of AI ethics and safety in China. How many of these ideas have you heard?
In his paper ‘Deciphering China’s AI Dream’ today’s guest, PhD student Jeff Ding, outlines why he believes none of these claims are true.
He first places China’s new AI strategy in the context of its past science and technology plans, as well as other countries’ AI plans. What is China actually doing in the space of AI development?
Jeff emphasises that China’s AI strategy did not appear out of nowhere with the 2017 state council AI development plan, which attracted a lot of overseas attention. Rather that was just another step forward in a long trajectory of increasing focus on science and technology. It’s connected with a plan to develop an ‘Internet of Things’, and linked to a history of strategic planning for technology in areas like aerospace and biotechnology.
And it was not just the central government that was moving in this space; companies were already pushing forward in AI development, and local level governments already had their own AI plans. You could argue that the central government was following their lead in AI more than the reverse.
What are the different levers that China is pulling to try to spur AI development?
Here, Jeff wanted to challenge the myth that China’s AI development plan is based on a monolithic central plan requiring people to develop AI. In fact, bureaucratic agencies, companies, academic labs, and local governments each set up their own strategies, which sometimes conflict with the central government.
Are China’s AI capabilities especially impressive? In the paper Jeff develops a new index to measure and compare the US and China’s progress in AI.
Jeff’s AI Potential Index — which incorporates trends and capabilities in data, hardware, research and talent, and the commercial AI ecosystem — indicates China’s AI capabilities are about half those of America. His measure, though imperfect, dispels the notion that China’s AI capabilities have surpassed the US or make it the world’s leading AI power.
Following that 2017 plan, a lot of Western observers thought that to have a good national AI strategy we’d need to figure out how to play catch-up with China. Yet Chinese strategic thinkers and writers at the time actually thought that they were behind — because the Obama administration had issued a series of three white papers in 2016.
Finally, Jeff turns to the potential consequences of China’s AI dream for issues of national security, economic development, AI safety and social governance.
He claims that, despite the widespread belief to the contrary, substantive discussions about AI safety and ethics are indeed emerging in China. For instance, a new book from Tencent’s Research Institute is proactive in calling for stronger awareness of AI safety issues.
Why Cold War analogies don’t work with AI
So I think there will be competitive dynamics as it relates to specific AI enabled military applications, [but] I don’t think it will be on the same scale as the arms race in the Cold War. I think in the Cold War there were a bunch of unique dynamics, but one aspect that was unique is there was something that was countable about ICBMs, nuclear weapons, that made it so we were always thinking that there were some missile gap between the US and the Soviet Union. For AI, it’s not necessarily the case that we’re just counting how much AI is in the military. So in that sense it’s more about a system-wide transformation about how you upgrade your military across a whole range of factors from information, logistics, communications. So I think that’s a poor analogy. And then the other difference is that it’s not just about the weapons applications that will be competitive, it’s about these system-wide applications: who will be able to adopt AI across different manufacturing systems to enable wider growth in the entire economy.
China’s social credit system
One thing that I’ve been reading about a lot lately is the social credit system. So this ties into a lot of the things that we’re talking about in which people who have researched what’s actually happening in the social credit system, for example, Jeremy Daum at Yale Law School, he says that when Western servers look at what’s happening in China and the social credit system, they’re looking through a glass darkly. That they’re projecting their own worst case fears about what the social credit system could become and then saying that’s what’s happening in China. Whereas actually other people who have been researching the social credit system say it’s a very low tech system of black lists, credit scoring that isn’t using machine learning algorithms for now, even though there are vague claims that they’re going to integrate AI and big data into these systems. So I think that is another misperception. The idea that the social credit system is a very high tech AI empowered system where maybe that is part of the vision or part of these vague proclamations, but it’s definitely not what’s happening on the ground right now.
Should you learn Mandarin?
My advice would be if you have some of the language already, or you have a strong passion to learn the language or you like learning languages, to invest in learning Mandarin as much as you can, especially if you want to learn about tech in China. Because for me, I read probably 60 to 40 of my stuff, in terms of learning about China’s AI development, is Mandarin to English. So I think some of the best stuff is still in English, but there’s just so much more stuff coming out every day in Mandarin about what is happening in China’s AI scene. It’s like if you’re a German, you’re not going to read Der Spiegel about what’s happening in terms of Silicon Valley. Even if Der Spiegel has a correspondent in Silicon Valley, sure, they may have like three people covering Silicon Valley. In the US, you’re going to read MIT tech review, you’re going to read Bloomberg news, you’re going to read The New York Times. So if you want to learn about what’s happening in China’s AI scene, sure those New York Times correspondents, Financial Times correspondents across China, they’re doing great work and they’re getting stories that you can’t get in some of China’s censored media. But for a massive scale effect, you should be reading Chinese language news or translations of Chinese language news.[However], honestly, the neural machine translation is getting really, really good and it can augment translation capabilities to the point where you can build pipelines, and I think the Center for Security and Emerging Technology, one of the organizations I work with, is trying to build a more automated industrial scale pipeline for translation of these materials. I’m sure these exist across other institutions and I think it can get to the point where you can be someone who knows none of the language, but you have read all the best English language work on this and then you can say, “I want a translation of this thing” and then use that. Feed it through the pipeline and then work from there. Yeah, so I’m a little bit agnostic between the choices, but just know that there is a trade off.
The Interview with Jeffrey Ding
Robert Wiblin: Today, I’m at EA Global London, speaking with Jeffrey Ding. Jeff is the lead China researcher at The Centre for the Governance of AI, which is part of the Future of Humanity Institute at Oxford University. He produces the ChinAI Newsletter and his work has been cited in the Washington Post, South China Morning Post, MIT Technology Review, Bloomberg News and Quartz among others. Jeff grew up in Iowa, studied economics, has worked at the US State Department as well as the Hong Kong Legislative Council and is currently a Rhodes Scholar studying International Relations at Oxford. Thanks for coming on the podcast, Jeff.
Jeffrey Ding: Thanks Rob for having me. It’s good to be here.
Robert Wiblin: Let’s talk about career specializing in China and what people get right and wrong about AI in China, but first, what are you doing and why do you think it’s really important work?
Jeffrey Ding: Yes, so I’m a DPhil researcher at The Centre for the Governance of AI where I look at China’s AI development. I think it’s important because for one, AI is a really central technology to economic development, military competition, and we should understand how China, as a rising power, is approaching this technology. I think it’s also important because it’s become a very hot topic and it’s also become an area where we project a lot of things about how we perceive China, how we perceive China’s model of technological development. So oftentimes areas where there’s a lot of misperceptions are areas where you can make a lot of impact with more grounded research.
Robert Wiblin: Yeah. Makes sense. I’m interested to find out how much of the views that people have about China and AI are just predictions of our own fantasies or fears. First, you grew up in the US basically entirely, right?
Jeffrey Ding: Yeah. I was born in Shanghai and then moved to Iowa City when I was three.
Robert Wiblin: Okay, so your parents spent much of their life in China?
Jeffrey Ding: Yeah, they came here for grad school.
Robert Wiblin: Okay.
Jeffrey Ding: They came here, as in the United States, for grad school.
Robert Wiblin: Yeah, we’re recording in London. Yeah. Have you spent much time in China? I guess I’ve got so many questions about China in general. I’m wondering like maybe I should talk to your parents rather than you.
Jeffrey Ding: You should definitely talk to my dad for computer science stuff because he’s a software engineer, but I have gone back to visit grandparents near Shanghai and a neighboring town every two or three years and then studied abroad for a semester at Peking University. I’ve done a couple internships as well and now go back for research every now and then.
Robert Wiblin: And you’re a fluent Mandarin speaker because you’re doing this translation newsletter: ChinAI?
Jeffrey Ding: Yeah, I don’t know what fluency really means nowadays, but fluent as in I can plug things into Google translate and then correct them afterwards.
Robert Wiblin: Yeah. When you go to China, do people kind of regard you as American or Chinese or some kind of confusing mixture?
Jeffrey Ding: It’s funny because my experience at Peking University was pretty tough because a lot of my friends who had more American looking faces, the bar for their language was much lower, whereas I was given no breaks. So it was a tough experience, but at the same time, I think it’s weird that they also view me as an insider of sorts. So they’re more trusting with information. Like you might understand. Those other people don’t understand what it’s like to be Chinese or to understand what China is doing.
Robert Wiblin: I have a friend who’s Chinese parents that grew up in the United States almost completely, and speaks Mandarin reasonably well, but not that well. But apparently whenever she makes mistakes in China, apparently the people give her a really hard time and it’s almost a betrayal somehow not to be able to speak Mandarin perfectly.
Jeffrey Ding: Right, you have ‘dishonored your mother nation’.
Deciphering China’s AI Dream
Robert Wiblin: Yeah. Alright. So last year you wrote this paper ‘Deciphering China’s AI Dream’, which got a whole lot of attention. Yeah. In brief, what were the key messages of that paper?
Jeffrey Ding: Yes. So I basically looked at four descriptive things about China’s AI development. The first is, what is China actually doing in the space? What is the context behind China’s AI development? The second is, what are the components, different levers that it’s pulling to try to spur AI development. Third is, what are its AI capabilities? And then fourth, what are some of the implications of that for national security, economic development, AI safety and social governance.
Robert Wiblin: Yeah, and so there were four myths you wanted to highlight and try to correct with that paper. What are those four?
Jeffrey Ding: Yeah, so along each of those four categories, I used one of the myths to ground some of the analysis. The first in the context is that China’s AI strategy did not appear out of nowhere with this AI development plan issued by the state council in July 2017. It’s part of a long trajectory of focus on science and technology development overall. It’s connected with other plans like the Internet Plus plan that’s trying to develop an Internet of Things. It’s linked to a history of strategic planning for technology in areas like aerospace and biotechnology. And it was not just the central government that was moving in this space. Companies were already pushing forward in AI development. Local level governments already had their own AI plans. So, in a sense the central government was following their lead in the AI space. In the second category of components, I wanted to deconstruct the myth that China’s AI development plan is based on actions of a monolithic actor that’s just issuing a top-down central command to develop AI.
Jeffrey Ding: Whereas you do have local governments pursuing their own interests. You have bureaucratic actors, sometimes conflicting, you have companies that are setting up their own strategies and those sometimes conflict with central government as well. In the third category of capabilities, I wanted to deconstruct this notion that China’s AI capabilities have surpassed the US or are the world’s leading AI power. So I analyze trends and, capabilities along the lines of data, hardware, research and talent, and then finally the commercial AI ecosystem to kind of rebut that myth. And then fourth, I wanted to rebut the idea that there were no discussions of AI safety or ethics in China. So those were the four main arguments in the report.
Robert Wiblin: Yeah. Is it the case that China has lots of different plans for development in lots of different technologies and so the fact that they have something about AI is not especially scary or surprising. It’s like, of course they do, they probably have something about all these different manufacturing areas too.
Jeffrey Ding: Yeah. There’s a really interesting misperception in this space where a lot of Western observers think that China was in the lead in issuing this national AI plan and that we now have to catch up to China to have a good national AI strategy. When you read what Chinese strategic thinkers and writers were talking about, they were thinking that they were behind because the US white house, Obama administration 2016, had issued a series of three white papers. They thought that Japan was already a robot superpower and had issued a series of plans to develop the robotics industry. So in that sense there is a lot of misperception in this space.
Robert Wiblin: Yeah, that’s interesting. There seems to be a general phenomenon here that when it comes to technology, everyone feels like they’re falling behind or not good enough. Do you know what’s going on there? I know that there’s this incentive for, I guess especially government organizations to say that they’re falling behind because that’s a way that they can appeal to get more funding for their project. Is there anything else going on here?
Jeffrey Ding: Yeah, so I think that is one component which is bureaucratic interests. Interest groups have an incentive to say “We are falling behind in this space, give us more funding to pursue this type of technological development”. I also think that there’s something deeper going on here. There’s an anxiety about technological development. It’s something that when we think about AI advances, it’s something that’s so abstract. It gets tangled up with really complicated things like disinformation campaigns, automation, what does labor mean? What does intelligence mean? And sometimes our responses, we just need a plan for that. And there’s this assumption that because you have a plan, because you have an AI plan, because you have a nanotech plan, you’ve solved the problem. Whereas actually most of the academic literature on this subject says that central forecasting, planning about complicated, uncertain technological developments, which AI definitely fits that bill is actually oftentimes flawed and doesn’t solve anything. It may actually be counterproductive.
Robert Wiblin: Yeah. So you must’ve written some of this paper two years ago, I guess it came out 18 months ago. Is there anything that you feel in retrospect, “Yeah, I nailed this ahead of time and I was ahead of the curve”. Is there anything where maybe you feel like you got it wrong with the benefit of hindsight?
Jeffrey Ding: Yeah, I don’t know how much I’ve nailed. I think the discourse has just matured as more people are asking questions like, “What is AI”? What are we talking about when we talk about China’s AI capabilities? So, one lesson I’ve learned is during that paper there was some debate at The Centre for the Governance of AI over whether we should release this AI potential index that I included in the paper to benchmark China’s AI capabilities with US capabilities. And it was a very rough first shot approximation where we came up with different indicators to proxy for a country’s data advantage. And the proxy for data in that edition two years ago was just a rough mobile user’s measure where now we have to get to the point where we disaggregate this notion of AI capabilities into different subdomains. So it’s not the case that the number of mobile users makes any meaningful impact on which country is going to have a better head start on autonomous vehicles, for instance.
Jeffrey Ding: So that is one of the big things that I would want to change. And I’ve tried to change in a written testimony in front of the US-China commission where to analyze national AI capabilities, I try to cut up this very fuzzy abstract notion into, “Where in the AI value chain is China ahead or the US is ahead or the UK is ahead”. Is it at the fundamental technology layer, like who is producing more and developing more of the AI open source software or who is producing more of the application and products? Who’s ahead on the fundamental NLP research, natural language processing research, who is building more smart speakers and capturing the market in smart speakers. And then also differentiating in terms of different subdomains of AI.
Robert Wiblin: Yeah. In my interview with Helen Toner about AI and partially about China, we were joking that China has so many mobile users. There’s so many people using Tencent, WeChat and things like that and so we should be very worried that the People’s Liberation Army would just have this fabulous ability to make excellent product recommendations to Americans if they tried to. Is there any sense in which it’s concerning that China has access to more data? I guess Helen thought that it was vastly overstated, this issue of China having a data advantage in ML training.
Jeffrey Ding: Yes. So I think the implications of the data advantage for capabilities comparing who is going to derive more value out of AI is oftentimes overstated. I think a lot of times what’s happening there is people are mixing up advantage for authoritarian governments to control social media and control the population versus advantage of authoritarian governments to increase their overall economic prowess. So I’m not sure that a country that is better at recognizing faces is going to generate that much of a meaningful economic impact. Whereas I think a lot of the main economic impacts, and this goes back to your interview with Helen, where she was talking about a good analogy for AI might be electricity, where it’s more about how AI can transform a range of productive industries, say smart manufacturing, transportation, logistics, some of the less sexy consumer facing applications of the technology.
Analogies for thinking about AI
Robert Wiblin: Yeah, so in that conversation we also said that people often draw analogies between AI and nuclear weapons. There’s going to be some nuclear arms race with this stuff. And Helen thought that this is actually a pretty terrible analogy and maybe a better one would be electricity, though there’s like some problems with that as well. Like for example with software it’s much easier to potentially like scale it up very quickly than it is to build lots of electrical infrastructure. When we were trying to like scale it that way for city, do you have any ideas for what, what are, what are even better analogies or how we ought to think about AI?
Jeffrey Ding: So I think electricity is useful in the sense that there are similarities in terms of electricity as a power source that will affect a bunch of different industries. So consumer facing ones like illumination through lamps, manufacturing facing ones like powering devices that factory workers used to power sewing machines. And AI in the same sense will be consumer facing in a broad range of different applications and will also be manufacturing facing in terms of improving the processes by which we make things. So I think it’s useful along that lens. I think it’s also useful because when we talk about electricity, we have a better sense of electricity as a technological system. Electricity in the 19th century, the electrical power system, and I’m drawing from the work of Hughes who wrote this book called ‘Networks of Power’ where he talks about the electric power system as not just the dynamo or the generator or the transformer that converts the current, but it’s about a system of power generation, distribution through transmission lines, electric utilities that are providing the central power and then being transferred to a factory production system where workers can now run machines at different levels of speed and they’re not connected to a central steam generator anymore.
Jeffrey Ding: So that is one conception of an electricity technological system in the sense of manufacturing. Electricity as a whole is a much broader concept that might also include communications like telegraph messages and how messages are being sent. But electricity offers us a way to talk about a specific electricity driven manufacturing system that evolved during the 19th century and I think we need to get to that level of depth and specificity for AI. What are the technological systems that AI will enable rather than just throwing around the term AI anytime something somewhat related to software comes up.
Robert Wiblin: Yeah. Do you think that there will be some sort of AI arms race between the US and China in coming decades or is that just a misunderstanding?
Jeffrey Ding: I think that it goes back to exactly what we’re just talking about, which is, what are we saying when we say AI? So I think there will be competitive dynamics as it relates to specific AI enabled military applications. I don’t think it will be on the same scale as the arms race in the Cold War. I think in the Cold War there were a bunch of unique dynamics, but one aspect that was unique is there was something that was countable about ICBMs, nuclear weapons, that made it so we were always thinking that there were some missile gap between the US and the Soviet Union. For AI, it’s not necessarily the case that we’re just counting how much AI is in the military. So in that sense it’s more about a system wide transformation about how you upgrade your military across a whole range of factors from information, logistics, communications. So I think that’s a poor analogy. And then the other difference is that it’s not just about the weapons applications that will be competitive, it’s about these system-wide applications: who will be able to adopt AI across different manufacturing systems to enable wider growth in the entire economy.
How do prominent Chinese figures think about AI?
Robert Wiblin: You mentioned that it’s kind of a myth that there’s not thinking going on inside China about AI safety and risks from artificial intelligence. Yeah. Tell us about that. What perspective do Chinese thinkers have on this?
Jeffrey Ding: Yeah, I think the first starting point is the myth is partly a product of the fact that we are not as familiar with China’s media ecosystem and not reading Chinese language materials. So I came to a lot of writings about Chinese views on strong AI and artificial general intelligence by reading a 500 page book put out by Tencent and the Chinese Academy of Information and Communications Technology, which is a government sponsored think tank. So, in that book they reference the AI Asilomar Principles and provide a breakdown of those principles. What they mean. I’ve also translated work by one of the leading contemporary Chinese philosophers, Zhao Tingyang, and he talks about the long term worries of superintelligence. Zhou Zhi-Hua is a professor at Nanjing University. He leads one of the strongest fundamental AI research labs in China at Nanjing University. And he talks about how we shouldn’t touch strong AI at any costs in the China Computer Federation’s weekly journal where top computer scientists discuss the issues of the day.
Jeffrey Ding: So I think there’s a lot of discussions about risks of artificial general intelligence. And that’s only one slice of what we mean by AI ethics and AI safety if we’re talking about discussions about robustness, technical standards to ensure that there’s no malfunctions and improve reliability of AI systems. There’s much broader discussions happening there and also about applications of AI and privacy and other ethical considerations. Now, there are areas that are off limits because of censorship and political sensitivity. So there’s not discussions about facial recognition and the ethical implications of disproportionately targeting ethnic minorities with facial recognition, for instance. So I think it’s careful not to take this too far, but I think it’s also important to recognize that Chinese people are people too and they can care about ethics and they care about what advances in AI will do for their livelihoods and their lives.
Robert Wiblin: Yeah, it sounds like the conversation might actually not be that different and I suppose maybe there’s quite a lot of information flowing over from the West because so many of the people involved can read English. So Nick Bostrom’s writes his book and maybe people can read that if they want to.
Jeffrey Ding: Yeah. I do think that there is much more information transfer from the top Chinese thinkers reading the top work in English in this space than the other way around.
Cultural cliches in the West and China
Robert Wiblin: Yeah. So, I guess in the West we’ve got kind of all of these cliches of artificial intelligence. I guess some of them come from movies, some of them probably come from like deeper older culture, like Terminator, Blade Runner, all of these preexisting concepts that people have. Are the ideas or the preexisting kind of cultural cliches similar in China or are they different in any important ways?
Jeffrey Ding: Yeah. One thing I’ve noticed is there’s a lot of discussions about AI through the context of space for some reason. So Brian Tse and I translated an article by Fu Ying who was China’s ambassador to the UK actually and now leads a group at Tsinghua University looking at the implications of AI for international security. So in her essay she talks about a community of shared future for mankind and the idea that AI could become the Mars invasion challenge that unites China, the US, Russia and the rest of the world. Recently there was a sci-fi movie called ‘The Wandering Earth’ based on a book by Liu Cixin, one of China’s leading science fiction writers. And in that movie, the astronauts are fighting against this AI named MOSS that controls the ship and has turned against humanity. So some of these discussions happen through the context of space.
Robert Wiblin: Yeah. Sounds like 2001: A Space Odyssey. So is there just a lot of interest in space in China in general? I suppose maybe it’s a bit like how the US was in the 60s when there was a greater sense of progress and China’s building stuff so quickly they can imagine going to space, whereas it seems like the US is sometimes flat out just building a block of apartments. So the idea of actually going to Mars is a little bit more of a stretch.
Jeffrey Ding: Yeah, I think that could be part of it. I think there is something valuable about studying science fiction just because it seems to be a good medium for us to play out these long term scenarios, futuristic scenarios. So I have translated interviews by Liu Cixin, where he also talks about AI development which seems to be something that he’s very interested in. So I don’t know if there’s a meaningful cultural difference there, but I do think science fiction is an interesting medium to see how ideas about this space are being played out.
Robert Wiblin: Yeah, I guess that seems kind of hopeful to me. I’d rather have people dreaming about how AI can help with space colonization than about like, I guess conflicts on Earth is maybe the worst way of thinking about it, whereas with space it’s kind of empty, it’s more hopeful. We can all go and colonize it as a species or it seems to give you like a broader, less adversarial perspective perhaps.
Jeffrey Ding: Yeah.
Robert Wiblin: So in the US and the UK there’s been particular people who have been raising fears about risks from AI, like Eliezer Yudkowsky, Stuart Russell, Nick Bostrom, writing various books about this. Are there any particular figures in China who are familiar with the technical side of AI who are worth knowing about and keeping your eye on?
Jeffrey Ding: I think Zhou Zhi-Hua at Nanjing University. He comes from a technical background, leads an AI lab, publishes in some of the top forums and he has written articles about how we shouldn’t touch what he calls ‘strong AI’. I don’t think there’s as mature of a discussion about AI safety as some of the names that you just mentioned and sort of the groups that have developed in the UK, US and across Europe. But I do think that there are some people in China thinking about this stuff.
Robert Wiblin: Yeah. Interesting. Whenever we talk about China on the show, I’m a little bit torn between wanting to be like just realistic and frank about the fact that the Chinese government does things that I disapprove of and that I imagine most listeners would disapprove of. And also thinking sometimes there’s a bit of a double standard that we have between things that China does and things that kind of we’re doing. We find it more critical because it’s another country and it seems so foreign. I was thinking, is there kind of a double standard do you think between us worrying so much about Chinese surveillance but then maybe giving a free pass to the NSA or Palantir or the sort of surveillance that people are just used to within the United States or the UK?
Jeffrey Ding: Yeah, I think that there has to be a distinction made between the structures and the vehicles surrounding surveillance. So in China, there’s not much of a community that can push back against some of these more invasive surveillance efforts. There are some think tanks like the Nandu Personal Information Protection Bureau that are pushing for more protection of personal information. But most discussions of privacy are about protections from companies losing financial data and then your bank records being leaked. Whereas in the US yes, there is something that Shoshana Zuboff at Harvard calls ‘Surveillance Capitalism’. And you mentioned stuff like the NSA and wiretapping, but there is a system in place where you do have, you do have opportunities for whistleblowers to come forward and civil society to push back against these efforts.
Robert Wiblin: Yeah. So there’s a sense in which maybe the technical capabilities are not so different, but then we might have more trust that that power is less likely to be abused in the United States.
Jeffrey Ding: Yeah. I would say that’s largely right.
Coordination with China on AI
Robert Wiblin: Do you think it would be possible for a future US administration to reach out to China and establish like really deep cooperation on AI safety issues or is that just not going to be practical?
Jeffrey Ding: I think it’s already happening depending on how broadly you define AI safety. So right now there are standards alliances among different firms that include Chinese banks who are trying to implement things in the area such as facial recognition for personal identity authentication. So you have US companies, Chinese companies, international companies working together to come up with what is the best standard to balance a bunch of different things in terms of allowing facial recognition to substitute for us typing in passwords and forgetting passwords and losing passwords. And in one sense that isn’t what we think of when we talk about AI safety and controlling superintelligent AI agents. But in another sense it serves as a kind of toy example of how we’re trying to think about what is the facial recognition system that’s going to be the most accurate. How’s it going to balance between a bunch of different things like costs, computing cost, efficiency, how many facial features do you need? And all of these are attributes of systems that companies and people will need to agree on. They may have implications for broader AI safety issues in the future. So I guess what we would call AI security issues… like to have a good facial recognition identity authentication system, you also need to have things that protect against cybersecurity risks, right? So those would all be things we’d be concerned about for larger, broader issues of AGI safety. And these discussions are already happening right now.
Speaker 2: So there was some recent talks on AI ethics and standards which I heard that they stalled because the US wasn’t keen to negotiate with China supposedly while they were using their technology for authoritarian purposes. Was that a good move or should the US be willing to negotiate on these issues regardless?
Jeffrey Ding: Yeah, I haven’t heard what discussions were stalled. I think that there’s a lot of stuff going on behind the scenes in technical standards bodies that don’t often get covered in the media. And there’s a lot of stuff happening in track II, track 1.5 dialogues that don’t get covered in the media because there’s no news angle hook to saying, “Oh we had a nice bureaucratic exchange meeting where we talked about things”, but there is a nice hook that says, “We’re in a cold war and all things are screwed”.
Robert Wiblin: Yeah, interesting. There’s an issue with what gets covered can create the reality. I do worry that because it’s a more exciting story and I suppose also just because it’s like so easy to tap into people’s xenophobia for other countries and China, in particular, is a rising power, that will start out as a myth and then it will become reality. Is that something I should worry about?
Jeffrey Ding: I think there is some sense where it becomes a self fulfilling prophecy so that’s why I’ve been very strong to come out against this new tech cold war meme. Been very strong to actually question this idea of decoupling as well. I don’t think anyone has come up with a good measure of decoupling but we just assume that decoupling is happening and the internet is splintering. So I think that oftentimes the difficulty here is it’s also an issue with understanding the technical details of AI. So when something becomes so abstract you can cherry pick anything and say, “Oh look at this one domain of AI, there’s some decoupling happening here. That means that the entire system has decoupled”.
Robert Wiblin: Yeah. Before we started recording, you were saying that one of the problems is that people talk about China and AI in China in such abstract terms I suppose because they lack any concrete details. What kind of effect does that have and maybe could we just actually gather more like empirical information and that would potentially massively improve the conversation?
Jeffrey Ding: Yeah, I think you just have to be more precise about your writing. I think there is a trade off here because on the one hand you want to be parsimonious and you can’t write a headline that says “X company in China has made a specific improvement in this algorithm.” You have to write a headline in some way. So I’m understanding of the need to be parsimonious. I think then the obligation is for people who do have more flexible word counts to be more precise with their language. So in my writing for the newsletter and for my reports, I try to limit the use of AI as much as possible and then also whenever I reference China, I try to reference either the Chinese Communist Party leadership or specific local governments or a specific Chinese company or a Chinese think tank rather than take one person’s words as a proxy for the entire country of a billion people.
Private companies vs. government research
Robert Wiblin: Yeah. How much is AI research in China being led by private companies versus the government and is the distinction between the two sharper or less clear than in the US?
Jeffrey Ding: On research, very much private companies. So the top labs are at Baidu, Tencent, Alibaba, ByteDance. There are some labs in the Chinese Academy of Sciences, much closer tied to the government, USTC, much closer tied to the Chinese government: University of Science and Technology of China. But I think most of the leading fundamental research is happening in the private labs.
Robert Wiblin: Yeah, interesting. I guess the same is true in the US. I suppose in both cases maybe it’s just private companies doing most of the work would you say?
Jeffrey Ding: Yeah, I think in terms of the fundamental basic research, for sure, in private companies and in universities.
Robert Wiblin: Yeah. And then the government I guess sometimes does applications, although I can’t think of that many cases of the US government using AI maybe outside of intelligence.
Jeffrey Ding: Yes. So there’s a cool dataset from R&D magazine where they come out with 100 of the top innovations each year. And I was looking through some of these data sets and actually there are some AI related advances and many of them do come from US labs. So the Lincoln Laboratory, Argonne, for example. And yeah, many of them are more about specific applications. I think I saw one about defense against cyber attacks, cyber defense systems empowered by AI algorithms. So I think the US national labs are also doing a lot of interesting work in this space.
Robert Wiblin: Yeah. You mentioned that during the nuclear arms race, one of the things that pushed it forward very aggressively was this thing that both sides thought that the other side was way ahead and so they were both like plowing lots of money into it, trying to like avoid falling behind. I guess one way we could tap down on that phenomenon in AI is to have a better measure of the capabilities of either side. So like, what is the least bad way of trying to measure like what actually are China’s capabilities in artificial intelligence? And I suppose vice versa as well. How can China like have confidence in what the US can and can’t do?
Jeffrey Ding: Yeah, I think in some sense it’s much easier to measure who is leading in integrating AI in the economic domain. So for that it’s relatively straightforward to come up with an approach to do that, and I think consulting companies like McKinsey, PwC, they can all do that, and it requires differentiating among both breadth and depth. So looking at AI across a bunch of different domains because you could have a world in which China leads in one sub domain, so Chinese natural language processing is different from English natural language processing and it makes sense that Chinese companies will be ahead in Chinese NLP and Chinese NLP applications. In the military realm, I think it’s much harder to measure because a lot of the applications will not be tested in real world battle scenarios. So you can do simulations of, say, drone swarms but you don’t have real world data for that. Whereas in the economic realm you do have real world data for what’s actually happening.
Robert Wiblin: Yeah. I guess one way that people have suggested of trying to quantify AI capabilities is looking at just the amount of compute power that people have. So how many processors do they have, which is potentially going to be quite a limiting factor on how much people can actually apply ML algorithms even once they have them. How do you think things are going to play out with compute? It seems like there’s a bit more sabre-rattling now about export restrictions on processors. Do you think that in the future it might be the case that there’s a lot less movement of compute hardware between the West and China or China and other countries?
Jeffrey Ding: Yeah. I do think that is a very salient lever of influence: compute capabilities. I think that that will depend on efforts China’s making to develop its indigenous semiconductor industry. I think there are also open questions as to whether if you just have enough money, can you eventually just buy as much compute as you want? At the same time, that makes it that you can still track how much compute people have. So I think it’s definitely an issue to keep an eye out for in the future.
Robert Wiblin: It seems to me like China fell way behind on semiconductor manufacturing despite the fact that they are at the cutting edge of manufacturing lots of different items. They’re behind the US and Taiwan in producing like… the kind of fabs that they have for doing that. How did that happen? It’s kind of a surprising turn of events to me.
Jeffrey Ding: Yeah, I think it’s a long history. They’ve been trying to catch up in making chips for I think around 40 years. Did a translation where we talked about the long history of these failed efforts and I think part of it is the cultural revolution took away a whole generation of leading academics and technicians and scientists in this field and had them go sweep toilets and clean up toilets and you lost a generation that was also going to train the next generation. So, there’s a lag there. It’s a much higher tech, higher barrier of entry field than other fields of manufacturing that you mentioned. So there’s a lot of learning by doing gains where the firm that is able to get an advantage in one generation is going to have the lead in making the next generation, and there’s constantly technical iteration in the field to get thinner and thinner wafers. So I think that sometimes locks in the winners and gives TSMC that incumbency advantage. So I think there’s a lot of factors there. Yeah. It also requires billion dollar capital investments and long term payoffs. So if you miss with a couple of plants, there’s more, I think, variants of misguided policy in this area as well.
Robert Wiblin: Yeah. Interesting. So it sounds like that might persist if it’s persisted for the last few decades or maybe China will keep kind of catching up to where things were five or 10 years ago and then by the time they get there now they’re behind what’s new.
Jeffrey Ding: Yeah, that has been the trend. I think that there’s a system wide variable here which is just the degree to which technical iterations in the field increase the efficiency and I think that is starting to diminish where, at some point, maybe seven nanometers is not that much better than 15 so that could provide an opportunity for Chinese companies to catch up to some extent.
Robert Wiblin: Because the progress is just going to slow down because we’re hitting kind of physical limits and so eventually everyone ends up at the same place.
Jeffrey Ding: That could be it and then it could just be about more about cutting costs or efficiency in some other way.
Robert Wiblin: Yeah. So you mentioned this expression, “A community of shared future for mankind”, which I think has been promoted by Hu Jintao and then Xi Jinping. I think it’s meant to be an underlying principle for China in terms of how it conducts its international relations. What does that mean and is it something that people really pay attention to in practice or is it a bit more of just a slogan?
Jeffrey Ding: I see it as more of a slogan. I do think that slogans and rhetoric to some extent do matter as we talked about before. But I see it more as a slogan and I don’t think it means that China’s not going to care about relative gains in a wide variety of domains. But I do think that it is a concept that we should be paying attention to and then also that there are domains where things that benefit China will also benefit the world. So we have this idea that everything here is a zero sum game, but there are things like improving the applications of AI for drug discovery and cancer research that’s going to benefit China, but if China does good in that area, it’s going to benefit the entire world as well.
Robert Wiblin: Yeah, so I guess the idea of it was that China was now going to think more about how the rest of the world was going. It’s now more integrated with the rest of world, so it has to take an interest in making sure that the world as a whole is functioning because we’re all in it together. I guess that’s the angle. I suppose there’s positive angle on that which is China will contribute to solving global problems. I guess there’s also the angle that now it has to take a greater interest in other countries for its own sake which could potentially have a negative side to it if you’re Vietnam or the Philippines or something.
Jeffrey Ding: Yeah. Yeah, the negative angle is an interesting point that I hadn’t considered. I think the positive angle you could see in climate change where it’s very clear that a lot of Chinese cities and areas, especially on the coast, are going to be devastated if climate change continues and you see subsequently more investment in global climate change initiatives.
Robert Wiblin: Yeah. Well there’s this other slogan that’s “One world under heaven” or “One kingdom under heaven”: Tianxia?
Jeffrey Ding: Tianxia.
Robert Wiblin: Yeah, what does that mean? I’ve read the Wikipedia article and I was still a little bit baffled having gone through it.
Jeffrey Ding: Yeah, so this also comes from Zhao Tingyang who I mentioned earlier is a Chinese contemporary philosopher. I think he’s at Tsinghua and he wrote the article about the short term concerns and long term worries about AI and he brings up that term again: Tianxia. I also don’t really understand it completely. I think it’s the idea of world government and applying and taking some idea that was important in Imperial China about how the emperor was the ruler of all under heaven. And it’s the idea that this world government will be able to unify the world and will be able to solve a bunch of public goods and collective action problems.
Robert Wiblin: Yes, I had the same reaction. Like is this good or is this terrifying? I suppose it’s not actually that different perhaps than how a lot of people think of the United States. The US takes a big interest in what happens everywhere because it regards itself as a global power and has an interest in things everywhere going well for its own economic sake. Suppose we shouldn’t single China out here by any means.
Jeffrey Ding: So to clarify, the world government would not be like a Chinese run global government under Zhao Tingyang’s conception. A cynic might say that because you’re using a concept rooted in China’s history that talks about the Chinese emperor there may be other–
Robert Wiblin: It’s giving that impression. Well I suppose that you have to use concepts from your own culture otherwise it doesn’t make sense. There’s this book, “AI Superpowers” by Kai-Fu Lee, which paints this picture of Chinese AI researchers and entrepreneurs as maybe working and studying even harder than their counterparts in the West. Do you think that’s accurate or is that maybe just a stereotype?
Jeffrey Ding: Yeah, my usual reaction to these things is that cultural differences are usually essentialist and structural factors are probably a better explainer for these things. It seems that the more people that I talked to in the venture capital space and the tech space, they consistently talk to me about how Chinese researchers and Chinese entrepreneurs just do work harder. I don’t know if that really makes that much of a difference. So I’ve also heard stories about people clocking in on this nine-nine-six schedule where they work from 9:00 AM to 9:00 PM for six days a week, but for two of those hours they’re watching TV and they just have to be there because they clock in.
Jeffrey Ding: There’s also been recent protest against the nine-nine-six culture on GitHub from Chinese workers. So I think there is some backlash to some of this.
Robert Wiblin: Yeah, it’s interesting. I guess when I was in China, I had this very strong sense that people were working harder, or at least the service culture was a lot more intense. Like people were just willing to do anything. So I think we’ve got this Airbnb place and we said, “Oh there’s this room that someone’s gonna be sleeping in. It’d be great if there was more sound insulation for that room” and like within a day they’d managed to install an extra wall.
Jeffrey Ding: Okay.
Robert Wiblin: I’m not sure whether that’s typical. That really made an impression on me just arriving in the country. Like you install a wall within a day. That’s insane.
Jeffrey Ding: I’ve talked to people who have similar stories.
Robert Wiblin: Yeah. But I suppose especially in knowledge work, it seems like working more hours does not always necessarily make you more productive.
Jeffrey Ding: Yeah. I also just don’t know if it’s sustainable. I mean, from a very, very rationalist perspective, maybe because you just have more of a labor supply, you can just work your population harder.
Robert Wiblin: Yeah. But I suppose as people get wealthier, maybe they want to take more leisure time and I guess there’s a bit of a gold rush perhaps on at the moment as well. There’s like lots of opportunities to make a lot of money right now that perhaps might settle down once the country reaches technological maturity to a greater degree.
Jeffrey Ding: Yeah. I do think some of this can be explained by a development gap. So there are not that many data labeling factories in the US, but there are rings of them across third tier, fourth tier cities, rural towns in China because people just are more willing and have to do that sort of hard labor where for 12 hours of the day they’re labeling ladders as ladders, as movable or not movable. So yeah, that is just a development difference.
China’s social credit system
Robert Wiblin: Yeah. So obviously we’ll stick up a link to this paper, “Deciphering China’s AI dream”. Are there any misunderstandings about China and AI that we haven’t covered that you think might be listeners heads that would be good to clarify?
Jeffrey Ding: Yeah. I think one thing that I’ve been reading about a lot lately is the social credit system. So this ties into a lot of the things that we’re talking about in which people who have researched what’s actually happening in the social credit system, for example, Jeremy Daum at Yale Law School, he says that when Western servers look at what’s happening in China and the social credit system, they’re looking through a glass darkly. That they’re projecting their own worst case fears about what the social credit system could become and then saying that’s what’s happening in China. Whereas actually other people who have been researching the social credit system say it’s a very low tech system of black lists, credit scoring that isn’t using machine learning algorithms for now, even though there are vague claims that they’re going to integrate AI and big data into these systems. So I think that is another misperception. The idea that the social credit system is a very high tech AI empowered system where maybe that is part of the vision or part of these vague proclamations, but it’s definitely not what’s happening on the ground right now.
Robert Wiblin: Yeah, that’s definitely my understanding from people who have looked into it more closely is that in many ways it’s not that different than what we already have in the West. We have a credit score for financial things and of course the government keeps track of people who commit crimes and don’t pay their fines and things like that. We’re not shocked by that. Do you think that people should be worried about what it could evolve into though? It does seem perhaps there is more of an appetite for turning it into something that’s more of a system for social control than people would tolerate in the United States.
Jeffrey Ding: Yeah, I definitely think that there should be concerns about what it could turn into. So even today, I think already an application… I think New York times reporting has revealed this, a New York Times report by Paul Mozur, which states that facial recognition systems are being used in neighborhoods not just in Xinjiang, but also outside of Xinjiang in different provinces where they’re counting up the number of faces identified as Uighurs or ethnic minorities.
Jeffrey Ding: And if it’s a certain number threshold that’s exceeded within a certain time frame, then an alarm goes off. So this is something that there would be strong pushback against in other countries. So I don’t want us to come away with this saying that there are no differences. I think we should just be more precise and specific about what the differences are and where the similarities are.
Relationship between China and other countries beyond AI
Robert Wiblin: Alright. I’d like to move on and talk a bit more about the relationship between China and other countries, looking beyond just AI. So it seems like over the last few years, I guess the US in particular has taken a more adversarial stance towards China. Do you think that is a good move, that there’s just been these frustrating issues that the US maybe hasn’t been addressing with China over the last decade or is it kind of ultimately futile because China is not going to be moved on questions of intellectual property or its relationship with its neighbors and basically we’re just going to damage the relationship without much gain?
Jeffrey Ding: Yeah, I think it’s a good move to be more adversarial in certain areas. I don’t think that the move to consider China as our existential enemy where we have to be against them in every issue is the right direction to go forward in. I think that there’s good evidence that being more aggressive and pushing for more in negotiations can lead to China agreeing to things. So when we have won WTO disputes against China, they have had to comply. And when China has won WTO disputes against us, we have had to comply. So I think that having dispute settlement mechanisms with teeth like the ones in the WTO would be helpful.
Robert Wiblin: So yeah, countries like the US face this trade off perhaps between fostering a cooperative international culture and relationship with China and also wanting to speak out about bad things that they see happening in China, like the treatment of Uighurs or things that are happening in Hong Kong. I suppose it’s actually quite a stark trade off potentially, or it’s not obvious what you ought to do, because speaking out about the Uighurs, it’s not clear that that’s actually going to have any positive effect and it might damage relations with China on other important issues. Is it worthwhile for the EU and US to speak out about human rights issues within China or is that just maybe not actually going to have any practical impact?
Jeffrey Ding: I don’t know what the practical impact will be, but I think it’s hard to say what the best strategy is going to be. I don’t think there’s a lot of research comparing, for example, speaking out versus trying to backchannel things, and I think when people try to say that speaking out is bad or that speaking out trades off with something, I think they’re misinterpreting the notion that if there’s a problem, a bunch of people can have different roles. So, I think that there is a big role to be played for speaking out. I think there’s a big role to be played for trying to backchannel things. Some people just don’t talk about certain issues so like some people are just working 12 hours a day and can’t just take up these concerns as their guiding force in life.
Jeffrey Ding: So for me, I’ve tried to at least do research on what I think my contribution can be, which is looking at what is happening with regards to facial recognition and applications of facial recognition in Xinjiang. I think it’s probably worth it for governments to speak out. I mean before governments spoke out or also activists and media reports and other people spoke out, I don’t think the Chinese government even acknowledged that these re-education camps existed and now they acknowledge them and they frame them as political re-education camps, rather than internment camps or concentration camps. But at the very least, they had to acknowledge that these things were happening. So there’s a world in which not speaking out, it would have just been hidden even more. So yeah, I think it’s probably, on balance, good for governments to call this out.
Robert Wiblin: Yeah. So there’s a fair bit of concern I guess in the US and in some parts of Europe about Huawei building 5G networks, and I saw this interesting tweet pointing out that if you can get the NBA and Blizzard to bow to Chinese government concerns about exactly what they’re saying, you know, what would happen if you needed a Chinese company to be delivering kind of all the replacement parts you needed basically for your entire phone network. It leaves you somewhat vulnerable to being told what you can and can’t say because otherwise perhaps your 5G network is going to stop functioning after a couple of months without the ability to repair it. Do you think that is a legitimate concern or is that just scaremongering?
Jeffrey Ding: Well, I think it’s a legitimate concern. I think there’s always dependency concerns when you rely on a sole source supplier for key inputs. I think that there’s also a concern for Huawei because if they would ever capitulate to Chinese government demands on this issue, then all of their services become suspect across the world, and they have interests across, I think, 180 countries. So I think it cuts both ways.
Robert Wiblin: So I guess it’s something that they maybe do once, but then the concern would be it’s very financially costly and then people just won’t ever trust them again.
Jeffrey Ding: Yeah, exactly.
Robert Wiblin: That’s interesting. So yeah, there’s this book I really like, “Destined for War”, I think by Graham Allison. He is an international relations specialist and he’s talking about relationships between the US and China in 21st century and concludes that there is no fundamental interest that the US has that’s in conflict with a core interest of China and vice versa. So even though there’s like lots of areas where there could be friction, fundamentally these two countries should be able to get along because the issues where they conflict are secondary concerns for both countries. Do you think that is right in the big picture and maybe should be talked about more if that’s the case?
Jeffrey Ding: I guess the key question would be what is his dependent variable? Is the idea just that these disputes will never escalate to a full on nuclear war?
Robert Wiblin: Yeah, I guess the argument is that, for example, Germany and Russia or Nazi Germany and the UK were much closer to begin with and there’s a lot more tension and there was lot more pressure I guess for war because the things that one side wanted were things that were very important to the other side. I guess for example, one of the tensest things between the US and China might be over Taiwan. The thing is there is that that’s something that’s very important to China and it’s kind of important to the US, but it’s not at the top of the list of the US’s priorities in the scheme of things. You wouldn’t think it’s something that they would be willing to risk a nuclear war over, kind of the independence of Taiwan. And I suppose it’s a slightly hopeful thing I think that he was trying to say that although there is some pressure that will create conflict between the two countries, it shouldn’t escalate because ultimately like one side should always be willing to capitulate on these issues because there’s no one thing where both of them care just overwhelmingly about it.
Jeffrey Ding: Yeah. I don’t know if I see that as that hopeful of a message, that the US would just be willing to capitulate if China tried to conquer Taiwan. But I do think that some of the overall trends that that is pointing to, which is we’ve had no major interstate war since the advent of nuclear weapons where the benefits from conquest and territorial conquest are much lower now than they were for world war one, world war II, just because I think some of the impulses for Germany’s aggressive behavior was to secure more natural resource supplies, whereas now we have an expanded trading system and more interdependence in the world economy. So I do think that there are good system trends that point towards peace, at least from major conflicts.
Robert Wiblin: Do you know if there’s been any reactions in China to stories about Russian bots posting on social media to influence elections in other countries or just in general engaging in disinformation. I suppose there’s two ways you could see it. One would be they might worry about being a victim of similar disinformation campaigns from overseas or, on the other hand, they might be interested in potentially using similar methods to Russia in future when they have conflict with other countries.
Jeffrey Ding: Yes. So there has been one reported alleged case of China purchasing Russian AI bots to attack the Twitter feed of, I believe his name is Guo, last name is Guo. He’s like this tycoon who left China and claimed that he has a dossier on the scandals of high profile Chinese leaders. And after he posted about that, there was a massive Twitter campaign where a lot of people commented and tried to take down his Twitter account. So that was one alleged case of using Russian AI bots. I think the disinformation tactics are different between China and Russia. So Russia has a very experienced, specialized group. I think it’s the internet research agency and they had very sophisticated tactics to target specific voters in the US. For example, African Americans with memes and Instagram posts that were meant to troll and also claim that the white media was wrong about any different issue. So that’s a very specific targeted campaign that’s meant to sew chaos and they wanted chaos in a US presidency which they got with Donald Trump. For China, I think the interests aren’t as specific. I don’t know if they would prefer Trump versus another president. Whereas for Russia, there is a clear preference for influencing and sewing chaos. Whereas you could argue that China actually wants more stability so the tactics are different and the interests may be different.
Jeffrey Ding: Yeah, that’s a really good point. I mean there’s a sense in which I’m not sure why it is so great for Russia to have chaos and weakness in the United States. It seems like that potentially can create problems for Russia as well, to make things more unstable, and it does seem like definitely China probably would have preferred to just have more of a status quo, just like more stable government and more continuation of the policy that had been for 20 years, which I guess is one kind of protection against China trying to mess with the US too much. But I have wondered, it seems like Trump is taking a pretty aggressive stance towards China compared to previous presidents. Could China try to hack emails or like try to release his tax returns if they wanted to basically play the same game that Russia did, but to get rid of Trump because he’s like a dispreferred candidate.
Jeffrey Ding: Yeah, I think it’s hard to say how much of this trade war and conflict is a product of Trump specifically, if you’re thinking about it from the Chinese perspective, or if it’s about structural and systematic changes. And then if you’re being even more Machiavellian, I think the assessment is that Trump has been overall bad for the US as a whole. So if China wants to gain vis-à-vis the US, that Machiavellian instinct would say keep Trump in power.
Robert Wiblin: Interesting, yeah. Okay, so they might not like it because of the trade war aspect, but then maybe they like the fact that Trump is just weakening the US in general, or like weakening respect for the United States among other countries, which creates a vacuum that China can then fill.
Jeffrey Ding: Yeah, that could be one case. Trump has definitely not been positive for the US system of alliances.
Robert Wiblin: Yeah, and I guess maybe he’s also more isolationist, so maybe less likely to intervene in some ways.
Robert Wiblin: Okay. So a year or two ago, 80,000 Hours published this article, “A new recommended career path for effective altruists: China specialist”, which got a lot of people interested in using their careers to try to improve relations between China and the West or just find ways to contribute to solving problems in China itself. Having gotten that far, I think some of those people then struggled to figure out exactly what they should do to try to become experts in some aspect of China and I think it’s kind of a failing of 80,000 Hours that we haven’t been able to give more, more concrete suggestions to people. Do you have any suggestions for someone who wants to become a China specialist? Like where might they begin and I guess, in a sense, China specialist is almost too broad a category. I feel a little bit bad that we’re talking about it at such a level of abstraction. Maybe we should break it down into more specifics.
Jeffrey Ding: Yeah, no I think you’re gesturing at a good point here which is for me, I see being a China specialist as you also have to learn about Russia, you also have to learn about Southeast Asia. You have to also know about the US. You can’t just study China in a vacuum. I think the first priority to at least consider is whether you want to invest in learning the language or not. And I’ve talked to people who have gone both ways, some who have gone all in terms of going to Tsinghua, taking intensive language courses and really trying to learn the language. And then others who have more tried to get a feel of landscape. Maybe they’ve gotten to China but they haven’t invested all their time in learning the language and they are just really good at reading a bunch of the top English language Chinese analysis of China and then synthesizing from there, which I think both approaches can work.
Jeffrey Ding: And this also is taking place in the backdrop of advances in neural machine translation, which just keeps getting better and better day by day. So my advice would be if you have some of the language already or you have a strong passion to learn the language or you like learning languages, to invest in learning Mandarin as much as you can, especially if you want to learn about tech in China. Because for me, I read probably 60 to 40 of my stuff in terms of learning about China’s AI development is Mandarin to English. So I think some of the best stuff is still in English, but there’s just so much more stuff coming out every day in Mandarin about what is happening in China’s AI scene. It’s like if you’re a German, you’re not going to read Der Spiegel about what’s happening in terms of Silicon Valley. Even if Der Spiegel has a correspondent in Silicon Valley, sure, they may have like three people covering Silicon Valley. In the US, you’re going to read MIT tech review, you’re going to read Bloomberg news, you’re going to read The New York Times. So if you want to learn about what’s happening in China’s AI scene, sure those New York Times correspondents, Financial Times correspondents across China. They’re doing great work and they’re getting stories that you can’t get in some of China’s censored media, but for a massive scale effect, you should be reading Chinese language news or translations of Chinese language news.
Robert Wiblin: Yeah, but it sounded like you might be saying if you don’t particularly like learning languages and you haven’t started, then just wait for the AI translation.
Jeffrey Ding: Honestly, the neural machine translation is getting really, really good and it can augment translation capabilities to the point where you can build pipelines, and I think the Center for Security and Emerging Technology, one of the organizations I work with, is trying to build a more automated industrial scale pipeline for translation of these materials. I’m sure these exist across other institutions and I think it can get to the point where you can be someone who knows none of the language, but you have read all the best English language work on this and then you can say, “I want a translation of this thing” and then use that. Feed it through the pipeline and then work from there. Yeah, so I’m a little bit agnostic between the choices, but just know that there is a trade off.
Robert Wiblin: Interesting. So you’re doing a whole bunch of translation every week for the ChinAI newsletter?
Jeffrey Ding: Yeah, every week a translation usually just of a piece that I found interesting, or working with contributors for pieces that they pitch, or working together on a joint translation.
Robert Wiblin: Yeah. How much can you just rely on the machine translation and do you notice that it’s getting better and better? Just like literally over the period of a year?
Jeffrey Ding: It got really good in the summer of 2016, because I remember being in the embassy in Dakar, Senegal. I was working for the State Department, and I was doing a lot of French-English translation because a lot of the operating language there is in French. And I just saw a big jump in terms of the French and then I was also reading some work in Chinese and there was a big jump and that was when Google translate implemented neural machine translation. I don’t know if it’s gotten that much better since then, but it’s to the point where every translation I start with plugging it into Google translate and then correcting each line from there.
Robert Wiblin: Interesting. Is Chinese easier or harder for machine learning translation? I guess like some languages have more ambiguity right in like one word can mean many different things which might slip up the algorithms. I wonder, does Chinese have more or less of that than English?
Jeffrey Ding: So the first cut answer to that is it’s much easier to do neural machine translation with high resource languages like Chinese and English because you just have more data on that versus some language, which we call it low resource languages that are much more obscure or there’s not as much text. But the deeper cut is that there are some difficulties in terms of idioms and phrases.
Robert Wiblin: I guess what does it do when it encounters an idiom that doesn’t exist in English? There’s no correlate.
Jeffrey Ding: Yeah, so idioms you have to correct line by line and I just use a dictionary Pleco and I just plug it in. So, I think there’s an interesting phenomenon where now there’s more loanwords in Chinese from English. So “fantasy” is now like just there’s some Chinese word that basically just spells out fantasy in Mandarin. So the sense is that as you have more people working across both spaces, and as languages intermingle more, it might just become easier to translate across.
Robert Wiblin: So you’ve kind of specialized in China and AI. I guess that’s an area that’s like big enough in itself. So maybe if people are going to specialize, become China specialists, they have to pick a speciality within China specialists. Like maybe they could work on farming issues within China or it’s like there’s specific relations between like China and one other country. I guess because there’s so much going on, it’s like you can’t possibly specialize in China as a whole.
Jeffrey Ding: Yeah. So I don’t know how scalable this advice would be, but for me, I found that AI is nice because it’s so broad, but then at least it marks it off so that you’re not just studying everything about China. So I actually stayed away from a lot of the Hong Kong debate stuff, because I used to work in Hong Kong for the Hong Kong legislative council. It’s like very interesting, the Occupy Central movement was actually marching with people in the summer of 2014 but I’ve just stayed away because in the space, you have to try to limit what you’re covering, otherwise it just becomes completely untenable. So I do like half of my work specifically on China AI, the other half I’m just lucky that I’ve been able to look at broader technology and great power competition issues, so a lot of times those broader themes are influencing what’s happening specifically. But then also, sometimes, a specific case in China and AI is also influencing my broader conception of trends.
Jeffrey Ding: So an example of that is, I have gotten really interested in the Chinese attempts to build an industrial internet. Just trying to connect a bunch of manufacturing devices and have them talk to each other. And that is a nice model of improvements in a capital goods industry at making machine tools better. And that connects to historical accounts of the American system of manufacturers and machine tools and how that was actually really key in our manufacturing revolution back in the late 19th century. So I think another piece of advice would be, don’t just study China, read the general academic literature in whatever field or specialty within China that you’re interested in.
Robert Wiblin: Yeah. So one of the challenges with trying to become a China specialist is that it’s not an established career path with like a clear progression where it’s like you study this and then you go into this internship and so on. So I’m kind of curious as to what was the progression for you? How did you get into it? Because maybe that can be an example that other people can follow, at least in some parts.
Jeffrey Ding: Well I got really interested in this from high school policy debate where we’d have resolutions like, “The US Federal Government should substantially increase its space development and our exploration”. And it would be a year long topic like that. And then almost every debate would involve some interaction with China, like US-China competition and what is China doing in space? And so I got really hooked from there and then just being Chinese American, it seemed natural to look at US-China relations. Thought I would have more of something to contribute there. And then I got interested in Chinese foreign policy specifically in Africa. So that’s what I studied for the MPhil at Oxford. It just so happened that there’s a student group that ran an AI lab that mashed together machine learning PhD students with people with no technical background like myself and put us together in groups to tackle AI for social good problems. And I got hooked on the technical stuff and learning more about the technology from there. And then, so I don’t know how applicable this is because it’s just a series of like random lucky opportunities, it just so happened that the Future of Humanity Institute had a call out for interns and they had Mandarin expertise preferred or China expertise preferred. So I threw my hat in the ring.
Robert Wiblin: How much of a benefit do you think it is to have parents who grew up in China? Does that give you a big advantage over people who just have like no particular connection to China?
Jeffrey Ding: Well, maybe on some credibility level because you look Chinese or you’re of Chinese descent, people think you can be like a China whisperer of sorts. On some level it’s a disadvantage because when you brief generals, they’re like, “Oh, it’s the Chinese influence campaign come to us in person!”.
Robert Wiblin: That’s quite dark.
Jeffrey Ding: Yeah, so I think it’s a double edged sword, but I think it is helpful in the sense that for me a lot of my commentary also draws from the fact of my identity as a Chinese American. So I think that I have more of a fuller understanding of what it means to be Chinese or just, I’m more willing to think… A lot of my work on at least Chinese user terms privacy or Chinese views on AI ethics consists of me saying “Chinese people are humans too”. And that helps when you are also a Chinese person to like think of yourself as a human. So I’m being a little bit playful with this, but I do think there is something meaningful there of being of Chinese descent, being either Chinese American, British Chinese and having that potential of understanding both sides.
Robert Wilbin: Yeah. What organizations out there are doing work on China and AI, other than the group that you’re involved with in Oxford? I imagine it’s a rapidly growing space?
Robert Wiblin: Yeah. A lot of core organizations are doing work here. New America has a DigiChina initiative led by Graham Webster, and they do translations of Chinese government policies and technical standards frameworks. So I’ve done some translations for them. There’s also the Center for Security and Emerging Technology. I’m affiliated with them and they’re also looking to scale up analysis about China’s AI development. I think there is a range of think tanks in DC and then a lot of companies are probably hiring in the space as well. I don’t think I’m doing anything particularly exceptional with the newsletter, but just from the demand for the newsletter shows me that I think there’s a lot of demand in this space.
Robert Wiblin: Yeah. There seems like there’s been an explosion of newsletters to do with AI and effective altruism and maybe China as well. Is that still a good way to advance your career? To create a newsletter that adds some value for people or maybe is there too many of them now?
Jeffrey Ding: Yeah, we might be reaching–
Robert Wiblin: A newsletter saturation.
Jeffrey Ding: Yeah, I think people were saying that about podcasts a couple years ago. I mean you’re doing amazing and like there’s more and more podcasts popping up. So I think it’s the same thing with newsletters. I work with Substack and they make it really easy to just do your own newsletter and I think I’ve always been interested in blogging, so I kept a blog since first year of undergrad. So I see the newsletter as basically the new model for a blog. Just to get your blog to an audience who’s super interested in your work rather than put your blog out there and then say go there.
Robert Wiblin: Yeah, that makes sense. I guess podcasts have just gotten more and more specialized perhaps over time. It’s like you have to carve out a new niche, but that’s actually really great because it means you can potentially reach like the thousand people who are most interested in your very specific, say, policy area and that’s like the conversation that you want to be part of.
Jeffrey Ding: Yeah. I experimented with one podcast for ChinAI and our pitch was, “This is for like the 50 government bureaucrats who’re really interested in getting in depth on AI policy. If there are 50 people who think you have something interesting to say, maybe invest some time and figure out how to say that well.
Robert Wiblin: Yeah, I suppose if there’s only 50 people you need to reach, then it’s very easy to do your promotion. You just email them every week. Just subscribe to the show.
Jeffrey Ding: Yeah, friend them on Facebook.
Robert Wiblin: Yeah. Are there any university courses that are worth highlighting for people who want to become China specialists and maybe who don’t want to go to China to study for a long time, perhaps because they have concerns about, say, security clearance or just for whatever practical reason?
Jeffrey Ding: Yeah, so I think Johns Hopkins SAIS school has a good course and it’s one year in DC and then you can also do a year in Nanjing as an optional course that they have like an arrangement. That’s a good start I think. I mean there’s a whole field of Chinese studies. There’s some of the UC schools, University of California schools are good on this. In the UK, you have the School of Oriental and Asian Atudies that has probably one of the best libraries of materials on China. So here, Oxford also has a China center. So I think there’s going to be hopefully more of these institutions popping up.
Robert Wiblin: Yeah. Do you think it would be useful for relations between China and the US to have more people moving between the two countries? I guess especially maybe having Westerners go to China and living there for a long time, which is a bit less common perhaps than the reverse.
Jeffrey Ding: Yeah, I think it’s an interesting question because they’ve done studies of the effect of short term study abroads on people’s perceptions of nationalism and the country that they study abroad in, and there’s a study by Jones in the International Studies Quarterly. I think it’s a 2014 study that was published. And this was an experiment with students who’re selected for study abroad before and after their study abroad in China, I believe. And the finding was that the students coming back actually became more nationalistic, so more proud to be American because they saw the differences with another country, but it’s what he calls an enlightened nationalism. So they were more nationalistic and proud to be from the US but for, I think the question was, how tolerant are you of a conflict between your home country and your host country? They’re much less likely to be tolerant of a conflict. So it’s what he calls an enlightened nationalism effect. But this is just a short term study. I don’t know what the effects of long term moving would be.
Robert Wiblin: Yeah. I’d be very interested to see. Is that a published paper? I need to look at that.
Jeffrey Ding: Yeah. I think if you just search ‘Enlightened Nationalism, Calvert Jones’. I think they’re at the University of Maryland.
Robert Wiblin: Yeah, that’s interesting. I guess having lived overseas, I do appreciate a little bit more about what I like about Australian culture and what’s distinctive about it. But I guess it also makes sense that it becomes much harder to imagine going to war with another country that you’ve been living in where you have friends in. Yeah. Are there any conferences that people who want to become China specialists should be attending? I guess possibly there are hundreds of them, but any ones you’d like to highlight?
Jeffrey Ding: So there’s Chinese studies conferences. There’s also, I think within the effective altruism space, there’s an effective altruism Chinese group. I’m not like a big conference person, so I don’t really know that many conferences off the top of my head.
Robert Wiblin: Yeah. What about any conferences around China and AI? Is there a thing for that yet or is it yet to be developed?
Jeffrey Ding: DigiChina’s hosting a conference on AI in China. I don’t know if registrations are closed yet. I think it’s happening recently. It’s happening soon and I think they’re going to host more of them.
Robert Wiblin: Okay, great. Yeah, we will try to provide a link to that and stick it in the show notes. Are there any areas of knowledge about China that you think are particularly lacking in the United States where someone could just like go and learn about this and then fill an important niche?
Jeffrey Ding: Yeah, I think the first cut is that the China AI model, which is just translating the best research and analysis in China on a subject every week, or on a regular basis, can be applied to basically every domain. So I’ve thought about applying it to basically reading the best international relations research coming out of China. And it’s happening on journals that are in Chinese language. It’s also happening in blogs and different WeChat public accounts that I look at. You could just do that model for Chinese foreign policy stuff on the Belt and Road Initiative, cybersecurity. So I think the model scales. In terms of subjects that I think are really important, but I think are undercovered is… I’m more focused on the technology space, so maybe like technology and society. So we talk a lot about technology on a nation state level and that’s actually what I focus on a lot in my research. But there’s a lot of interesting stuff on the relationship between technology and individuals. So we throw around terms like techno-nationalism, which is often about nation states competing over technology. But there’s something interesting within techno-nationalism about how the Chinese public thinks about technology and that’s often lost. And I think there’s a lot of things that could be unique, could be similar across countries, but understanding how the Chinese public thinks about technology, I think, is another cool space.
Robert Wiblin: Are there any books on China that you’d particularly recommend that listeners check out? I’m halfway through this ‘Deng Xiaoping and the Transformation of China’ book and really enjoying it, although that’s maybe a bit more historical than it is about China today.
Jeffrey Ding: Yeah, that’s cool. I haven’t read that one. I am reading Matt Sheehan, who’s a good friend of mine at MacroPolo. We’ve done some work together in this space. He came out with a book called ‘The Transpacific Experiment’ and it talks about how China and the US cooperate and collaborate across a wide variety of domains. But there’s a chapter on AI and technological innovation. I got halfway through ‘The Party’ by Richard McGregor, who’s a long time foreign correspondent in China and he dives deep into what the role of the Chinese party actually is. The separate role for the government versus the party structure.
Robert Wiblin: Yeah, my colleague Niel Bowerman was telling me about that book and recommending it. I was very sad to find out that there’s no audio book version. So I’d have to find the patience to sit down and read a real book on paper. Alright, so you may not be a conference person, but we’re going to have to get you back to EA global so you can give your presentation in a minute. Actually, what’s the presentation you’re giving about?
Jeffrey Ding: The presentation is going to be about Redeciphering China’s AI dream. So two years on, what are the new research questions we should be asking and how was I wrong and basically what to do to move the field forward in my opinion in terms of having dealing with problems like the technology abstraction problem, the China abstraction problem, techno-nationalist assumptions. A lot of things that we talked about.
Robert Wiblin: Nice. Yeah. If the video for that is up, we’ll stick a link to it up there in the show notes so people can check out what you’re about to present. I guess our final question is… You’re reasonably active on Twitter and I saw that you seem to post almost every day about what work you’ve done that day. Is that right? Is this some kind of commitment contract you have to like if you’re being public about what you’ve accomplished or not accomplished, it encourages you to get more work done or have I misunderstood?
Jeffrey Ding: No, this was just like a week long fad where I read something, well a friend sent me something about how a big part of the millennial burnout culture is because we don’t think about how we’re influencing other people’s burnout. So this week I basically just tried to post things about how I’m actually not being productive and not publishing anything. Wasting a lot of time to like have people give themselves a little bit of grace that on Twitter you usually see people just killing it, like publishing 17 different books in one day, but for me, I talk about how I wasted a bunch of time refreshing my email and it’s okay for you to do that as well.
Robert Wiblin: Yeah. People don’t post about how they got up really late. Listeners, I got up at 1030 this morning. I’m certainly not all that productive. Got up just in time to make this interview. Alright. My guest today has been Jeff Ding. Thanks so much for coming on the podcast, Jeff.
Jeffrey Ding: Thanks for having me, Rob.
Robert Wiblin: Sure there’ll be a lot more to talk about with China and AI in the coming years and decades.
Jeffrey Ding: For sure.
About the Author
This article was produced by 80,000 Hours. 80,000 Hours is a web platform that is dedicated to helping as many people as possible lead high-impact careers.They do this by providing career advice for talented young people who want to have a social impact. see more.