Andrew Ng on China’s Ambitions in Artificial Intelligence
GGV Capital’s Hans Tung and Zara Zhang interview Andrew Ng, one of the world’s leading experts on artificial intelligence and deep learning who was the chief scientist at Baidu from 2014 to 2017. Andrew is also the co-founder of Coursera and the Google Brain Deep Learning Project. In this episode, Andrew discusses how the startup ecosystem in China compares that of the US, his work at Baidu, and what he is most excited about in the field of artificial intelligence.
Update: We recorded this episode in late 2017. As of January 2018, Andrew has announced his new project, Landing.ai, that will help bring AI to the manufacturing industry. Landing.ai’s first strategic partner is Foxconn, the Chinese manufacturing giant that builds iPhones, among many other products.
HANS TUNG: Hi there. Welcome to the 996 podcast, brought to you by GGV Capital and co-produced by the Sinica Podcast. On this show, we interview movers and shakers of China’s tech industry, as well as tech leaders who have a U.S.-China cross-border perspective. My name’s Hans Tung. I am the managing partner at GGV Capital, and have been working at startups and investing in them in both the U.S. and China for the past 20 years.
ZARA ZHANG: My name is Zara Zhang. I’m the investment analyst at GGV Capital and a former journalist. Why is this show called 996? 9-9-6 is the work schedule that many Chinese founders have organically adopted. That is, 9 a.m. to 9 p.m., six days a week.
HANS TUNG: To us, 996 captures the intensity, drive and speed of Chinese Internet companies, many of which are moving faster than even their American counterparts.
Today we’re joined by Andrew Ng, one of the world’s top experts and evangelists on AI, artificial intelligence and deep learning. He was the chief scientist at Baidu from 2014 until this March. Before Baidu, he cofounded Coursera in 2012. From 2011 to 2012, he worked at Google where he cofounded and led the Google Brain Deep Learning Project. And before that, he was and still is a professor at Stanford.
Wired Magazine describes Andrew as someone who spreads the gospel of AI. Thank you, Andrew, for being here with us today.
ANDREW NG: Thanks for having me.
HANS TUNG: Let’s talk about what you are up to. There are constant discussions about what is your latest since you left Baidu. Can you share with us what excites you these days?
ANDREW NG: The thing that gets me out of bed in the morning is the idea of building an AI-powered society. I think that AI technology has advanced enough that we see a relatively clear path to making self-driving cars a reality for everyone. Maybe someday giving every child a personalized tutor, maybe also giving everyone higher quality, lower cost access to healthcare.
But really, I think every segment of society, every major industry will be transformed by AI over the next decade. I think that no one company, not even a large tech company can do all the work to let AI be used in all the ways it can be to improve human lives. So quite recently I announced, one of the things I’m doing is announcing a new set of courses on Coursera on deep learning to try to train up maybe millions of people, maybe even millions of your listeners, that could benefit from learning these AI tools to use them in all sorts of creative ways that I myself would never have dreamed of.
ZARA ZHANG: You recently started a project called DeepLearning.ai. Could you share with us more on what that’s about, and how that effort has been going?
ANDREW NG: So DeepLearning.ai has been focused on creating this new educational content, really courses on deep learning hosted on Coursera that I hope will help educate, hopefully millions of people, with the latest deep learning tools so that they can go and invent all the wonderful things that I think can be invented with deep learning.
HANS TUNG: It is going to be more of a 2C model or a 2E model? Who will be the audience for this?
ANDREW NG: You know today, we have individual learners coming to Coursera to take courses on deep learning, and there are also over 500 companies and governments working directly with Coursera to try to educate their workforces or the citizens of these different countries. So I think that Coursera has been a great channel, a great educational tool to reach lots of individuals as well as companies and governments.
ZARA ZHANG: So Andrew, given that you last worked at a Chinese tech company, I wonder how China figures in your vision for AI, especially now that we’re seeing a lot more research and experts coming out of China. Are you seeing an increased interest from China in your courses and your projects?
ANDREW NG: Yes, especially the newly announced government plan in China on AI, the interest in AI in China has always been high, but I think it has become even higher in the last few months. So DeepLearning.ai worked with NetEase or WangYi (网易) in Chinese to post a lot of our educational content on our websites in China to make it more easily accessible, but I think the China AI ecosystem is just incredible. I think globally the two top centers of innovation for AI today are Silicon Valley and Beijing. After Silicon Valley and Beijing, I think Canada or London, a few other places, maybe New York, are also very good but I’ve not yet seen any other centers with the intensity and talent and the rapid velocity of Silicon Valley and Beijing.
HANS TUNG: So we see on Twitter you’re actively recruiting people to join a startup. How’s that going so far and where are the talent coming from?
ANDREW NG: You know, since announcing DeepLearning.ai, I was surprised at the number of people that contacted us even before we posted any job positions to try to volunteer or work with us in various ways. We’re building up a team of folks to help with some of the things we’re doing with DeepLearning.ai, as well as with some new projects that we haven’t announced yet. So I hope that through DeepLearning.ai we can train up many people that will do wonderful things, and maybe someday someone in a village will figure out how to use machine learning to optimize their water supply and help nature locally and improve cleanliness of water. That’s great.
There are some AI vertical applications I’m excited about as well, so we’re just building a team to go after some of those things now.
ZARA ZHANG: Have you seen a lot of applicants coming from China and a lot of Chinese students for your online AI courses?
ANDREW NG: Actually, I’m seeing a lot of interest definitely in the U.S. and in China to work with us and to pursue these AI opportunities. I think AI, frankly, is global. I am seeing a lot of interest in Silicon Valley and Beijing, and also a lot of interest in India and in a few places in Europe.
HANS TUNG: Right, that makes sense.
ANDREW NG: The recent rise of deep learning, the recent rise of AI technology gives everyone a chance to build the next generation of companies. One of the things I learned watching the rise of the Internet, this was maybe 20, 25 years ago, was that if you take a shopping mall and build a website for it, that does not turn the shopping mall into an Internet company.
In fact, what makes a truly Internet company is not just whether or not you run a website. It’s much deeper than that. It is whether or not you do progressive A/B testing, whether or not your have short iteration times that ship a product every day, rather than every six months.
With the rise of AI, I think that we’re all starting to figure out what is it that truly defines an AI company. How do you architect your company to leverage AI capabilities? I think this gives rich opportunities both to incumbents such as Google and Baidu to transform themselves into AI companies, as well as a lot of opportunities for new cities and countries and new companies to rise to be maybe the future, leading AI organizations. I think we’re still in the early days of this exciting wave of development.
HANS TUNG: Well at least for the first class you’re offering on Coursera, you tweeted out last week that over 1,000 people have signed up in the first 24 hours already.
ANDREW NG: That is my on-campus Stanford class that has over 1,000 people. The Coursera class is much larger than 1,000.
HANS TUNG: What is the number now?
ANDREW NG: The first machine learning course I offered on Coursera had over 1.8 million enrollments.
HANS TUNG: Over what period?
ANDREW NG: Over the last several years, over the last five, six years. So the interest and the number of people we’re bringing into the machine learning community I think is very large.
HANS TUNG: When you look at these sort of inbound interests, have you ever done any analysis to break down by geography as to where it is coming from? We know it’s global, but is it concentrated around a few countries or areas?
ANDREW NG: So I think the majority of learners in machine learning, a lot of them come from the U.S., China or India, and also a lot from the UK and I think a lot from Germany if I remember correctly. That is really quite global. The U.S., China and India are the top three.
HANS TUNG: In the area of startups, have you seen or through your student network hear about the type of startups that are being funded today or being started? What type of opportunities are they going after? Are there any that excite you?
ANDREW NG: You know, I’m excited about AI’s application into a lot of different vertical industries. So a couple of industries I have spent time looking at include healthcare and education. You see in a lot of industries there is first the IT revolution, the digitization revolution, and that trace data which then enables the AI revolution to come in and eat the data and create value.
So you know maybe 20 years ago in a lot of industries, say healthcare, a lot of the activity was recorded on a piece of paper that might be filed in some filing cabinet over in Ohio. But over time, thanks to the rise of electronic medical records in the United States and globally, now instead of pieces of paper floating around in filing cabinets all over the world, a lot of that is now stored in databases. It is much more digital, and that gives rich opportunities for AI to come in and learn from data, make predictions and create value.
Just as a long time ago your x-ray would have been a physical piece of film, today it is much more likely to be a digital image, and so the digitization of healthcare creates a lot of opportunities for learning algorithms for AI.
And I think for education as well. You know, 10 years ago a lot of university education was a professor saying a bunch of words into the air, and then those words kind of disappear into the ether, or they disappear in the students’ brains. There was, for the most part, no digital trace of that.
But thanks to the rise of online education such as Coursera, but also Khan Academy and others, there’s now much more digital data about what was previously analog activities. Your students’ graded homework used to be sheets of paper floating all around the world, but now some fraction of that is done in the digital realm, and now there’s a digital record of what students answer which test questions correctly and incorrectly. All that data means that AI now, much more than several years ago, can now come in to understand patterns of student behavior, who’s learning what and how to really help students.
I think both of these are examples of industries where the IT revolution is well underway, and this creates the data for the AI revolution to also gain momentum.
HANS TUNG: I get that data has been created and that’s needed to make the AI revolution possible. But do you see startups taking advantage of that, or is it more big companies like a Baidu or a Google? Because if you are a healthcare service provider, you have a lot of data. You’re going to be guarding that very closely, so you will only shared that data with someone who you would trust. Do you see that the big companies have an advantage over startups in this space, and how should the startups do something differently to compete?
ANDREW NG: You know, I think that data is very verticalized. So one thing that makes a large web search engine’s business defensible is they have a data asset that tells you if you search for a certain term, you’re more likely to click on this URL rather than that URL. And so even though, I’ve led phenomenal AI teams in both Google and Baidu, so I have a good sense of how web search technology works, but without access to that data asset, I would have no idea how to lead a small team to build something competitive.
But having said that, all the web search click data is actually questionable as to whether or not it is useful for some other business. For example, how is web search click data useful for reading x-ray radiology images? It is actually not quite as useful for that.
And so I think that in other vertical applications where there isn’t already a big incumbent that is dominating the market, I think that there are a lot of new vertical applications where a small team, either through scrappiness or just hard work and sweat or a few partnerships, could scrape together enough data to start getting momentum.
ZARA ZHANG: So Andrew, given that we’re in the business of venture capital, I’m wondering, if you were a VC investing in AI startups, what would be the qualities that you would look for? What would get you excited about AI startups?
HANS TUNG: What gets you excited to want to work with a startup?
ANDREW NG: You know, in the end, whether or not a startup is going to work is a judgment decision that takes in a lot of things. Maybe one things I used to used to tell all of my teams is that in the end, our boss is the customer. Your ultimate boss is actually not the CEO. Satisfying the CEO is usually much easier than satisfying the customer, and there’s not one customer, but there are these millions or hundreds of millions of people that will vote with their feet whether or not to use your product.
So in the end, I think my boss is usually the customer and it’s a very complex piece of judgment, I guess. Whether or not you have the right team, the right product, the technology, the channel, as well as the business model to serve the customer in a way that makes sense.
I’m curious as to what your answer to that is.
HANS TUNG: We can ask Jenny. She’s the one to talk about this.
I heard you speak about three years ago on the different paths to do machine learning, that deep learning is only one of the paths to get there. Deep learning is catching a lot the wave today, and hype. Do you think it’s the right usage of the R&D dollars going into the area? How does it impact other paths that machine learning could take place?
ANDREW NG: I think that AI technology has many components, including deep learning, how it runs on your networks, but also graphical models and knowledge representation and planning and so on. There are actually a lot of different buckets, a lot of subdisciplines of AI.
Having said that, I think the thing that has advanced the most rapidly recently is deep learning, driven to a large part by the rise of data and the rise of computational power and GPU computing, but also real algorithmic innovation.
Now, because deep learning has advanced so rapidly in the last two years say, this creates a very rich opportunity set. There’s just a lot of stuff we can now do with deep learning, like accurate speech recognition or accurate face recognition that we just didn’t know how to do two years ago.
Today, computers can look at pictures and figure out what’s in the picture much better than ever before, be it recognizing faces, recognizing animals, figuring out what are the objects in the picture or video. So where does that create value?
I think speech recognition is working far more accurately than ever before. This is what enables products like Amazon’s Alexa and Baidu’s Duer and Google’s OK Google, and so on. There are additional use cases of speech to be invented yet that companies are just starting to look at.
I think that computers are starting to read medical records more accurately than before. but really, the pattern is whenever you have a large amount of data, technically it is labeled data, then deep learning algorithms are letting us make predictions or recognize the types of data or label data far more accurately than ever before.
But I think there is still a lot of work to be done to find creative business use cases and fit this into the appropriate business or the appropriate product context.
I think a lot of the reason that so much excitement centers on deep learning is because this is the piece of the AI that has advanced the most rapidly in the last couple of years, and this rapid progress creates rich set of opportunities that frankly no one, or very few people, are trying to tap yet.
In contrast, I think other pieces of AI technology, and some of the buzzwords that you don’t need to understand will be things like knowledge representation in planning, or draft models, all those other tools, I find them useful in a variety of projects, but they haven’t gone through this step change the way that deep learning has in the last couple of years. So even though there are opportunities there, maybe a set of untapped opportunities there might not be as large as the set of untapped opportunities enabled by the new deep learning algorithms.
ZARA ZHANG: Recently there have been reports of potential security issues surrounding AI. So there was a study that found that voice-controlled assistants could potentially be hijacked by ultrasound commands that humans cannot hear. Are you worried about these security issues at all?
ANDREW NG: Fortunately, there are many deep learning researchers aware of the security issues, and so this time around, a lot of us are developing the deep learning algorithms together with keeping in mind the security implications. I think as we trust face recognition systems more and more, researchers there are definitely working on defenses against the attacks.
There was a fascinating paper at Carnegie Mellon University that showed that if you wear a certain turquoise shell patterned glasses, that that could spoof a face recognition system into thinking you’re actually someone else. But fortunately, we’ve seen enough computer security that there are now actually top computer security researchers and top deep learning researchers working together to try to address this.
I would say the solutions are definitely not fully developed. We actually don’t know what those solutions are, but there’s enough researcher attention on this that I feel pretty good about the path forward.
HANS TUNG: On a lighter note, I remember 2015 when the DeepMind contest, playing Go against the top player from South Korea, it was five matches before the first win happened. Most people, especially in China, thought that there’s no way a machine can win. What did you think before that first match? Were you surprised at all by the results from those five matches?
ANDREW NG: I think DeepMind has had amazing breakthroughs in getting computers to play games, ranging from their work on Atari to the various games, I think they’ve done a fantastic job showing that reinforcement learning is really good at playing video games, as well as playing games like Go.
HANS TUNG: Do you think this is more, beyond that?
ANDREW NG: In terms of reinforcement learning technology, which is a type of machine learning technology, I am really excited by the work done by a few groups at University of California Berkeley and elsewhere on applying these types of technology to robotics. I personally do not work on having computers play video games. I think it’s great that others work on it, I just don’t do it myself.
But I think that one of the most exciting applications of this type of technology is using it to control robots much better than ever before. I’ve been very impressed by some of the work that originated out of UC Berkeley, because of Pieter Abbeel and Sergey Levine’s research groups, research teams on using robotics.
HANS TUNG: Have you seen that kind of study or efforts elsewhere in other labs, beyond UC Berkeley?
ANDREW NG: I think that more and more research groups around the world are adopting reinforcement learning for robotics, or deep reinforcement learning for robotics. Some groups in Canada, some groups in Germany, a few groups in the U.S. Also a few large corporations are now doing this, but I think maybe a lot of the fundamental breakthroughs I felt took place in Berkeley, but it’s definitely becoming more popular.
In terms of applications of these technologies —
HANS TUNG: But nobody at Stanford is doing it?
ANDREW NG: Actually, sorry, there are a few people at Stanford working on this, yes. I think it’s really become quite global by now. I still have an affiliation with Stanford, I’m an adjunct faculty and really privileged that Stanford remains, has always been and remains one of the top centers of AI technology. Well really, multiple universities and multiple corporate labs have had a huge impact on the rise of deep learning.
But if I hadn’t been doing some of the work that I was involved in at Stanford, there’s no way I would have started and lead the Google Brain team, for example, which really builds on the early work that happened at Stanford. It is fortunate to still be spending time teaching there.
ZARA ZHANG: One huge application of AI is in self-driving cars. Within the self-driving car ecosystem, what do you think are some of the holes in terms of areas that people have not devoted enough research or energy to?
ANDREW NG: I am on the board of Drive.ai, and Jenny Lee is a board observer.
HANS TUNG: That’s right, GGV is an investor in it.
ANDREW NG: So clearly I think Drive.ai is a very exciting company with a lot of potential. I think that autonomous vehicles will become a reality, probably sooner than most people think. But it’s not something to worry about either, because I think once we put these things on the road, people will also adapt to it faster than most people currently think they will adapt to it.
So that’s one piece of the AI future I’m excited about. I remember when I was an undergrad student, I started to see videos of autonomous driving work done at Carnegie Mellon University, and then over the years I helped out peripherally with Stanford’s participation in the DOT Urban Challenge and the DARPA Grand Challenge and saw really my friend Sebastian lead a lot of that work at Stanford.
I think that technology timing is important and in hindsight, I think ten years ago around the time of the DARPA challenges, the surrounding ecosystem of technology just wasn’t ready to support truly safe and scalable autonomous driving.
HANS TUNG: Correct.
ANDREW NG: I think it is only in the last couple of years that I feel the technology has turned the corner and the path to making this not just demoware but making this real is actually much clearer now than ever before. I think most of it, frankly, was driven by the rise of deep learning and to a lesser extent the available TO sensors, but I think really much more deep learning.
HANS TUNG: Google obviously has spent a lot of money investing in this area. The Waymo project is becoming bigger and bigger. Do you think that there’s still a lot of opportunity left for any startup to try to do that here in the U.S. against Google?
ANDREW NG: I think that Drive.ai has a lot of very promising technologies and approaches. There might be enough room for others in addition to Drive.ai, but we’ll see.
HANS TUNG: Well, you put them ahead of Google, that’s impressive.
ZARA ZHANG: What kind of advice do you give to Drive.ai?
ANDREW NG: You know, honestly, so Drive.ai was founded by a fantastic group of my students, an old friend and also my spouse, so I help them out a bit whenever they need me, but frankly it is a phenomenal team.
HANS TUNG: Carol is just simply amazing. We spend a lot of time with her, and she is just incredible.
ANDREW NG: Yeah, I’m married to her so I’m very biased, but I have to agree with you.
HANS TUNG: Now we’ve verified it, she is very good.
ANDREW NG: Yeah, and I help them out when they need me, but frankly, I think the team is doing great. There is a lot of phenomenal talent up and down, so I think they do well.
HANS TUNG: Yeah, it is a great team. Also, we do see opportunities worldwide beyond just the U.S. market for sure. As you travel around the world, where are some of the other markets you see progress or innovation being made in the autonomous driving space?
ANDREW NG: Yeah, you know because we actually have conversations with specific governments, that makes it awkward for me to name specific governments. I think that regulations will play a big role in the rise of autonomous driving, and without naming specific governments, I think that some governments are moving faster than others. The countries that are able to get the right regulations in place will have a non-trivial advantage, will have a big advantage in the rise of autonomous driving.
HANS TUNG: Do you have any advice for the U.S. government at any level, whether it is the state level or municipal level or national level, to do more? Because it sounds like innovations are happening more based on an ecosystem now, not just on the startup anymore, and the ecosystem involves government increasingly on sectors that require regulations.
ANDREW NG: I think we do have a lot of smart people in the U.S. government, although I think that perhaps the U.S. government could move even faster. Maybe I just say that a lot of governments at various levels in the United States, as well as internationally, have reached out to Drive.ai. So maybe I will just leave it at that.
ZARA ZHANG: So would you advise startups in this area to work with big companies or compete against them?
ANDREW NG: I think the question of whether or not to partner or go it alone is a complex, strategic calculation for any startup. I think there are great opportunities for startups to go it alone, but then there are also some things where partnering with a large company could enable a startup to go much faster. So I think it really depends on the details.
HANS TUNG: If we take Drive.ai out of the equation and just look at what other big companies are doing, there is also Google with Waymo and then there is Tesla with their sort of open project. So when you look at different big companies trying to do this, some are building and collecting data on their own, some are working with other partners. Do you see one system working out better versus the other for big companies?
ANDREW NG: You know, it’s early days. I think we will give it a few years and see how the competition plays out. Some of these things, Drive.ai has its own unique perspective on where the market is going, but give it a few years and we’ll see.
HANS TUNG: Let’s talk about China. You were born in the UK, and then you moved to Hong Kong and then Singapore. You came to the U.S. and taught at Stanford, and you end up going to Baidu and spend a few years in China. When you look at the kind of innovations happening in Asia, how does that impact your career decisions and the kind of things that you’re working on these days, just having that Asian background and experience and exposure?
ANDREW NG: The China technical or Internet community ecosystem is almost a totally different world than the U.S./global Internet ecosystem. And so I find it really important, I find it very helpful to see what’s happening in both ecosystems and to learn from both ecosystems.
So I think today, more and more people realize that there are a lot of products that are invented in the U.S. that then later go to China, as well as vice versa. There are a lot of ideas that first take off in China and then only quite a bit later make their way over to the U.S.
I think that individuals that are able to really monitor both ecosystems and learn from both will just learn faster and see the best ideas much sooner.
ZARA ZHANG: Within the field of AI, what do you think are some areas in which the U.S. is ahead of China? What are some areas where China has a surprising lead over the U.S.?
ANDREW NG: I think that historically, as well as today, the U.S. has been ahead of China in terms of basic algorithmic innovation. So you look at a lot of the detailed, very important technology inventions, I see more of those invented in the U.S. than in China, although China is really picking up. I have been really impressed by some of the basic technology inventions in computer vision for example, in China.
HANS TUNG: And on those, do you see it out of the big companies or a government lab or university labs? Where do you see those basic technology innovations happening in China?
ANDREW NG: I think that in China, a lot of the basic research is done by the large corporations, but then also some of the universities are doing a pretty good job. Somehow I feel like there are more startups in the U.S. that feel like R&D labs and fewer startups in China that feel like basic research organizations, but maybe I am biased.
HANS TUNG: I would think, based on our experience, that seems right as well. A lot of innovations we’re seeing is being done by the big companies in China.
ANDREW NG: Yeah, and just a few big companies have the capital to pay for the large research organizations, but some of them are doing a really nice job. Definitely Baidu also, but also more and more other companies as well.
HANS TUNG: We see that out of Huawei, maybe even also Xiaomi and Alibaba and Tencent and so forth.
ANDREW NG: Yeah. I feel like Baidu does have a huge AI lead in China, but other countries are definitely working hard at it.
One thing that is taking off, because China as a society moves so quickly, I feel like consumer products seem to have the ability to just go from zero to 100 much faster in China than in the U.S.
HANS TUNG: That’s right.
ANDREW NG: And there are a lot of reasons for this, I guess, ranging from China being mobile first and maybe it being a somewhat more homogenous society, so there are fewer market segments you have to fight for one at a time. It feels like the U.S. is like 100 market segments or something, and to get adoption there you probably have to fight one at a time for adoption, whereas China maybe has ten instead of 100 market segments — I am making all of these numbers up — and so things seem able to grow faster in China.
But I see that new products and new business models sometimes take off faster in China. And the fact that China is more of a mobile-first society, really for a lot of people in China their cell phone was their first computational device, rather than in the U.S. where most of us use a laptop or a desktop and then had to be retrained to use a cell phone. So that also enables mobile products to take off really quickly.
So I think it’s really valuable to watch some of the trends in China because some of them, not all of them, some of the things you see in China might be a harbinger or might be an early signal for what might happen in the U.S. later as well.
ZARA ZHANG: So the trends that you just mentioned like large population base and fast adoption, I wonder if these also give rise to a lot of low-hanging fruit in tech in China. After China goes past this first phase of adoption, how can I go beyond that and actually lead the world in innovation?
ANDREW NG: I feel like in terms of AI-powered products there’s just low-hanging fruit all over the place. You almost can’t walk around without stumbling on all this fruit. It is not even low hanging, it has already fallen on the ground.
HANS TUNG: Can you give us some examples?
ANDREW NG: Well I mentioned healthcare and education as two verticals, mind you it is a lot of work to build these healthcare products and education products, but I think it’s very clear that there are healthcare and educational experiences that we could deliver that just were not possible three years ago, because no one had the technology.
I think today, the number of great opportunities still vastly exceeds the number of teams that are qualified to pursue these opportunities. So I think teams with these AI and business skills just have many opportunities they could go after, that might not see any competition, at least for some time because there just isn’t that much AI talent around.
HANS TUNG: So as you worked at both Google and Baidu, how do you compare or contrast the culture, the approach to work and how they innovate between these two companies?
ANDREW NG: I think both Google and Baidu are great. It’s very difficult to compare the two, because they have very different products. Some similar, but also a lot of very different products, very different company cultures that operate in different countries. I love the super high quality engineering, frankly at both companies, and just the wonderful people I got to work with at both companies.
There is one thing I like especially about China, which is just the speed of decision making. I felt like our teams at Baidu, I led about a 1,300-person team at Baidu.
HANS TUNG: Thirteen hundred?
ANDREW NG: Thirteen hundred. We were making decisions at a speed that feels more like a very small U.S. startup, so that’s one thing that I think the China Internet ecosystem is very good at, because the pace of competition is so fast that it drives teams to naturally make decisions very fast as well. Large and small companies in the U.S., it feels like the pace of decision making is not as aggressive, and there’s pros and cons. When you can take a bit more time to make decisions, maybe the decisions are slightly higher quality but then also it is not as fast.
HANS TUNG: It’s a trade-off.
ANDREW NG: It’s a trade-off.
HANS TUNG: How was your lifestyle different when you were at Baidu versus Google?
ANDREW NG: Were you working much longer hours at Baidu?
ANDREW NG: I think I’ve personally worked long hours pretty much all my life, going back to when I was a student, so I don’t know that that’s very different. But again, in China the Chinese internet companies, not just Baidu, but globally are famous for people working very hard.
HANS TUNG: Work 9-9-6 a lot, 9 a.m. to 9 p.m., sometimes often six days a week.
ANDREW NG: Yeah,that’s part of the PR, I guess.
Actually let me say one other thing that might be useful. I think that so far a lot of excitement about AI has focused on the tech companies like Google and Baidu, because they had a very high technology base to build on as well as a lot of data, and so this why Google and Baidu have some of the most phenomenal AI teams in the world.
Having said that, I think a lot of the coming impact on AI will be in other industries as well. I mean frankly, I’d be a bit sad if the ultimate impact of AI is to transform a bunch of tech companies. What AI needs to do is go transform healthcare.
HANS TUNG: A lot of offline, traditional companies.
ANDREW NG: Education and construction and really all those other industries.
ZARA ZHANG: Are there specific non-tech companies in AI that you’re excited about?
ANDREW NG: Yeah, I see a lot of startups emerging, entering different business verticals. Some of the large companies, really it tends to be companies with a large set of data that can make the best use of AI. For example, GE has been doing a lot of work in AI, and other companies really with large digital assets, even companies that weren’t traditionally considered tech companies, if you have a large data asset, there’s probably a chance that AI could create a lot of value for you. Fintech clearly has been adopting AI very rapidly.
It will be the industry sectors, the companies with the data assets that I think will be best positioned to do this.
HANS TUNG: We did an industry study a while ago, comparing the top 10 market cap, most valuable companies that are listed around the world 10 years ago versus today, and 10 years ago you have a lot more traditional economy industrial companies like GE and the banks, telecom carriers and so forth.
Fast forward to today, of the top 10, seven are tech or Internet related, and of the seven, two are from China. But a question that will be interesting over the next 10 years, how many of these industrial, offline economy companies can leverage AI and other technology to mount a comeback, and grow and become a lot more efficient over the next 10 years?
ANDREW NG: You know, one of the interesting things that’s happened in tech, not just in AI, is in the tech world all of us are used to needing to learn new things every three years or every five years, because new technologies are invented all the time. So we went through the Internet era to cloud to mobile and now AI. All of us are used to the ground shifting under our feet every five years, and thus needed to shift.
What’s happened in the last decade is now tech has infected pretty much every major industry, and so as tech continues to shift every five years, now everyone needs a shift every five years. So the tech world, we are kind of used to that, we have gotten used to this way of life. I think the other industries, because their pace of change is now driven by this say five-year cycle that tech has, they will need to get used to this pace of change as well.
HANS TUNG: Right, because whoever has data, data has replaced oil and credit as the most valuable commodity, and how you utilize data determines how how much innovation and value creation you can generate.
ANDREW NG: Yes, and also even broader which is that when someone invents mobile, which happened just 10 years ago with the iPhone, or when someone invents these new AI technologies, now these fundamental shifts in the tech landscape, because all these other industries are using tech, this means that this is a competition, these things also change. So when the basis of competition on which you built your big business changes every five years, then you kind of need a new way of operating.
This is one of the effects of tech infecting these other industries means that these other industries also need to change every few years. So, it’s exciting times.
HANS TUNG: For us, one of the interesting examples is Domino’s Pizza. They were the most forward-thinking in the adoption of internet and mobile technology within a traditional offline economy business. Their revenue growth and their stock price growth over the last 10 years is on par with what Amazon has done. So it’s mind-boggling to see how a traditional economy company that is willing to adopt technology can have such a big impact in their own business, and vis-a-vis their competitors as well.
ANDREW NG: That’s great. I think these things do move people’s market caps, right. It’s very clear there.
ZARA ZHANG: You obviously care a lot about education. Jack Ma said recently that we should teach our kids things that computers cannot do better. One thing about AI is that because it moves so fast, when you train someone on AI, maybe the skills they’ll learn this year will become obsolete in three years. So how would you advise young people to equip themselves to succeed in the long term? How should we approach education in this age of AI?
ANDREW NG: I think the most valuable skill in the future might be the ability to keep learning, because so far, I think that AI displacing jobs will be significant, but it may also be somewhat overhyped.
So a typical U.S. report you might read is maybe in the next 20 years, 30 percent of jobs in the U.S. are at risk of automation and that sounds really scary. Like 30 percent of jobs, oh wow.
But the flipside of that is maybe as many as 70 percent of our jobs are not at risk of displacement, and in fact we can’t find enough people to do a lot of those jobs. We can’t find enough teachers, we can’t find enough healthcare workers. For some reason, we also can’t find enough wind turbine technicians. So there are a lot of these jobs that we just can’t find enough people to do. So it’s much more of a reskilling, retraining issue than humanity has no work left to do kind of problem.
Having said that, now that tech has infected all these other industries, I think that most other industries, whatever industry you’re in, there is a bigger challenge that you need to learn something new every few years, as the technology your business is built on continues to change every few years.
HANS TUNG: Looking back, how did you sort of discover that AI was going to be something big, major, and play a role not only in your life but in the lives of billions of people?
ANDREW NG: You know, people outside AI feel like it suddenly took off, in the last three years or something. As an AI insider, I feel like AI has been making steady progress, exponentially real fast progress, but steady progress year after year for the last maybe 20 years.
HANS TUNG: So about 20 years ago, how did you decide that AI was something that is very interesting and decided to do this? Because I remember back then at Stanford, I was freshman year in college and thought AI, oh, that has been around for a while, because there are a lot of things happening. Obviously I wasn’t smart enough to decide that was the one. How did you recognize that 20 years ago?
ANDREW NG: You know, when I was living in Singapore, my father who is a doctor, way back was looking into doing eyewear work for automated diagnosis. I learned a little bit of AI from him.
The other defining experience in my career was way back in Singapore. I once had an internship where I just remembered having to do a lot of photocopying, and I was definitely thinking —
HANS TUNG: Like my job at Merrill Lynch as a financial analyst, yes.
ANDREW NG: Yeah, and I was definitely thinking, “Gee, if only we could automate all this photocopying I was doing, maybe I could spend my time doing something else.” And then when I started to understand that AI can help us automate a lot of things we didn’t know how to automate before, then I wound up really pursuing a career in AI, pretty much most of my professional life.
ZARA ZHANG: You’ve moved to many places growing up. What do you think were some of the challenging things about going to new environments, especially that move from Asia to the U.S.?
ANDREW NG: So I moved to the U.S. to attend Carnegie Mellon University in Pittsburgh. When I came here when I was maybe 17 years old, at that time I actually thought I would return to Singapore when I got my bachelor’s degree. I was really scared as a 17-year-old at that time, because if I knew that my move to the U.S. was going to be quite final, I don’t know.
I think I really enjoy and I think people really benefit from seeing multiple organizations. I really enjoyed spending time at Carnegie Mellon and and MIT and Berkeley and Stanford, and getting to see a few different companies. You know, there’s so much knowledge there’s created in the U.S. and so much knowledge that is created in China, and most people underestimate how much there is in the other community.
Although no, maybe that’s not totally true. I think enough people in China read the English language media. I think China has a much better sense of what’s going on in the U.S. than vice versa. But still, I think each of these ecosystems has so much to learn from the other, it is easy to underestimate how different and how accretive both of them are.
HANS TUNG: So of all the places you have been to, where do you enjoy being the most and what kind of food do you like?
ANDREW NG: I like whatever. I don’t know, yeah.
HANS TUNG: You don’t eat? Or you just don’t have time?
ANDREW NG: I still love Singaporean food, and my wife Carol and I, we both eat a lot of sushi.
HANS TUNG: That is good.
ZARA ZHANG: At what point did you decide that you want to stay in the U.S. and pursue a career here instead of what you originally thought?
ANDREW NG: After I graduated from Carnegie Mellon, I realized I wanted to pursue a Ph.D. and then there were the great U.S. universities. So I wanted to go into MIT and then work for my advisor, Michael Jordan at Berkeley. And then after that, I thought you know, taking on the position at Stanford was a great place to keep on working on and trying to advance AI, so I wound up sticking around through all of that.
HANS TUNG: Do you see that at different universities on the two coasts have a different approach to innovation or encouraging professors to work with startups?
ANDREW NG: Stanford University is a very special place, and if you want to get into the fashion industry in the U.S., you pretty much have to move to New York. If you want to break into the film business, you pretty much have to move to L.A. And if you want to work in tech, frankly there’s like one place that is much better than all of the others in the U.S. So I think Stanford University is a very special place.
And maybe actually, here is one example. Many years ago, my students started working on GPU computing for deep learning and very early. We influenced a lot of other groups to use GPUs for deep learning. And the reason we wound up writing that paper, I don’t think we were that clever. I mean, we kind of saw the trend, but we just saw it earlier than others because we had friends in NVIDIA.
HANS TUNG: NVIDIA played a role.
HANS TUNG: And we just saw it coming, and we were just lucky to have the right friends and hear the right signals and so we said, “Hey, let’s do some research on GPU computing with deep learning.” If we had been living somewhere else other than Silicon Valley.
HANS TUNG: You wouldn’t have had access to the influence.
ANDREW NG: We wouldn’t have seen it that early and been able to do the work to help influence our community. So any long examples are down in Silicon Valley.
ZARA ZHANG: Speaking of the right friends, who are some people you think really helped or mentored you throughout your career, or that you enjoyed working with? So-called the “right” friends?
ANDREW NG: Oh boy, honestly there are too many to name. I am always afraid of naming a few people.
HANS TUNG: And offend the rest.
ANDREW NG: Yeah. But I have been really fortunate, fantastic mentors as an undergrad, people like Andrew Moore, Michael Kearns, Michael Jordan, Stuart Russell, and then at Stanford, all of the Stanford CS faculty were great. I don’t want to name them all, but they were all great.
And then building up Coursera, some of the VCs we worked with were fantastic. Maybe they are your competitors so I won’t name them, but you can look up online who they were. And then really at Baidu, Robin was fantastic. I think actually Robin Li was really one of the best technology and business leaders in the world. I think that a lot of people have learned a lot from Robin.
ZARA ZHANG: So I’m curious, how did you first meet Robin Li?
ANDREW NG: I had a friend way back who was working at Baidu, and when I told them I want to look at doing work in AI, he said I should consider Baidu, so he wound up introducing Robin and me. So I flew to China, had a long lunch with Robin. We chatted for three hours about what building up AI at Baidu could look like, and then it was shortly after that that I wound up joining the company.
I think Robin is actually a very clear strategic thinker and quite a few times he kind of sees further than a lot of other people. So he’s really one of these executives I very much admire.
HANS TUNG: We talked about comparing different universities. It’s at least impressive to us when we visit CMU in January, to see Professor Moore left CMU to go to Google, and after 10 years there to go back to CMU. Having that kind of exchange of experiences and ideas makes it a lot more likely to foster the next wave of innovations.
At Stanford, do you see a lot of people moving around and different people coming together to collaborate on different things?
ANDREW NG: I think that Stanford University has long had a deep connection to Silicon Valley. In fact, frankly Stanford University played a huge role in creating the Silicon Valley that we all now know today. So I think that the transfer of knowledge is often a two-way street, and the whole community benefits, everyone in the community benefits from other pieces of their community, ranging from basic research innovations at Stanford being used by companies, to companies sharing with researchers what are the most important problems or the key datasets that then in turn helps drive the research for it.
I don’t know, I think there are a lot of examples. I feel like there are at least 100 or maybe 1,000 examples of exciting things happening in AI today. But it’s just AI has turned into this huge engine of opportunity and of potential value creation.
HANS TUNG: And part of one the reasons we like to do this podcast is to share some of the best practices that the best minds in each field are working on, and also the way they work. We’re hoping that more universities will have collaboration and allow talent and ideas and even patents to be used, exchanged, so that more innovation can come out in collaboration with universities and industry.
For the next section of the podcast, we are going to ask you a series of rapid-fire questions. So an entrepreneur that you admire the most and why? You mentioned Robin earlier. Do you have anyone else?
ANDREW NG: Another executive I really admire is Jeff Bezos. I think he’s really one of the best CEOs in America, and I often listen to his talks, sometimes live, sometimes just on YouTube. I feel like I’ve actually learned a lot from his strategic thinking.
I think one of the amazing things about Amazon under Jeff Bezos is that they’ve somehow built a model that lets them keep on building up new businesses even within this giant company. And historically, that’s something that has been very difficult to do, and Amazon is one of the best countries in the world at doing this.
HANS TUNG: Yes.
ZARA ZHANG: When you were at Baidu, you made it mandatory for senior executives to join a book club. Can you share some titles you guys read that left the most impression on you?
ANDREW NG: Sure. Some of my favorites, The Lean Startup or Running Lean, Talking to Humans. Our team read a lot of books on product as well as on business strategy. I also loved a lot of Clay Christensen’s writings ranging from disruptive innovation to his new book, Competing Against Luck. There is actually a list of like 26 folks that I share about with some of my teams.
HANS TUNG: How about Malcolm Gladwell?
ANDREW NG: Malcolm Gladwell I have loved reading. His book is not on my list currently. I really enjoy reading Malcolm Gladwell.
HANS TUNG: For fun.
ANDREW NG: Yeah, for fun.
HANS TUNG: It was more related to VC, less related to building products. One app that you use most often that people may or may not know, besides Google or Baidu.
ANDREW NG: I read a lot, so there are a few magazines that I read digitally.
HANS TUNG: Like what?
ANDREW NG: Actually some magazines I enjoy include Harvard Business Review, The Economist, Bloomberg Businessweek. I also read the Wall Street Journal, New York Times. Actually I subscribe to quite a lot of the things.
HANS TUNG: You mentioned YouTube earlier, and so what podcasts or videos?
ANDREW NG: I’m not a frequent user of YouTube, just occasionally.
HANS TUNG: Too much junk.
ZARA ZHANG: You have mentioned that when you decide how to spend your time you have two criteria: one is whether what you’re doing can change the world, and the other is how much you’re learning. So what’s one thing you did that you think may change the world, and what’s one thing you did that taught you the most?
ANDREW NG: You know, I think that by offering a set of courses on Coursera, it’s not so much me changing the world, it is me giving you or me giving others the power to change the world, so I am very excited about that.
And I think in terms of learning new things, I think that as I tried to apply AI to healthcare, I find myself learning a lot about healthcare, or AI and education, I find myself continuing to learn a lot about education, and so I find that exciting.
Oh and also, I was quite happy I got to spend most of this past weekend reading research papers, which I hadn’t had time to do for a while but actually, it is very surprising if you spend 20 hours over a weekend reading research papers, you can actually read a lot of research papers on weekends.
ZARA ZHANG: So that is something you do for fun?
ANDREW NG: I enjoyed it. I can’t do it all the time, since I have other things I need to do, but I really enjoyed this past weekend.
ZARA ZHANG: So it’s a luxury for me.
ANDREW NG: Yeah.
HANS TUNG: So the last question, what keeps you up at night and what makes you feel excited and get out of bed in the morning?
ANDREW NG: Actually, some days it has literally happened, it has happened some days when I woke up at 5 a.m. and I couldn’t get back to sleep because I was so excited about work that I just want to get to it. That’s actually happened to me a few times.
HANS TUNG: And that’s AI, usually?
ANDREW NG: Yeah, oh definitely. And the opportunity to change a few different vertical industries.
HANS TUNG: With that, thank you so much for coming over and spending time with us today.
ZARA ZHANG: Thank you, Andrew.
ANDREW NG: Thanks, Hans. Thanks, Zara.
HANS TUNG: Thanks for listening to this episode of 996. By the way, we also produces a weekly email newsletter in English, also called 996, which has a roundup of the week’s most important happenings in tech, in China. Subscribers have told us it is informative and fun to read. The newsletter also features original content and analysis from Zara and me. Subscribe at 996.GGVC.com.
GGV Capital is a multi-stage venture capital firm based in Silicon Valley, Shanghai and Beijing. We have been partnering with leading technology entrepreneurs for the last 17 years from seed to pre-IPO, with $3.8 billion in capital under management across eight funds. GGV invests in globally-minded entrepreneurs in consumer Internet, e-commerce, frontier tech and enterprise. GGV has invested in over 200 companies including Airbnb, Alibaba, Ctrip, Didi, Hellobike, Hellobike, Houzz, Keep, Slack, Square, Toutiao, Wish, Xiaohongshu, YY and others with 29 IPOs and 17 unicorns to date. Find out more at GGVC.com.
HANS TUNG: If you have any feedback on this podcast or would like to recommend a guest, please email us at 996@GGVC.com. This podcast is co-produced by our friend and business partner Kaiser Kuo, the host of the wonderful Sinica Podcast. It covers China’s economic, political and cultural issues.