Keynote: The Era of AI: The Rise of China and the Future of Work – Kaifu Lee | SVIEF
10:40PM Sep 29, 2018
Okay, ladies and gentlemen, our next speaker is so well known that also doesn't need any introduction. He's a thought leader in the innovation community, one of the world's leading AI experts, a venture capitalist and a writer. Now, his latest book, AI Superpowers hit Amazon's best selling book list. And ladies and gentlemen, please join me in welcoming Dr. Kai Fu Lee, Chairman and CEO of Sinovation Ventures and president of Sinovation Ventures AI Institute. Ladies and gentlemen, welcome Dr. Kai Fu Lee.
Thank you. Thank you very much. It's great to be back at Silicon Valley. So I live in China most of the time, so many of you probably don't remember me. Then I'm going to introduce someone else that you are more familiar with to tell you about AI.
Hello, is the sound down? Let's back up and try again.
And wait for this.
So that's not our work. This is the work of iFlytek. But nevertheless, it shows you with the power of deep learning the quality of speech synthesis of President Trump speaking in a foreign language is actually very, very realistic. And that is the biggest in technological improvement over the last 52 years, 62 years in the history of artificial intelligence. It's deep learning.
Deep learning is a network of highly connected neurons in thousands of layers that can, in a single domain, take a huge amount of data and train to recognize, predict and decide and synthesize at a much higher accuracy than humans.
Now, on the one hand, it is not human intelligence, it cannot think cross domain. It cannot play plan in very deep sense. It has no strategic or creative thinking capabilities. But in a single domain, again, with a huge amount of data, it is beating humans in almost every task imaginable. We have seen AlphaGo. We've seen speech recognition, face recognition. There are beginnings now of diagnosis of how to read how to read MRI and doing a better job than radiologists. So this amount of improvement is leading to what I call four waves of artificial intelligence.
So in this picture, what you see is what I believe four waves of business opportunities that deep learning and associated technologies will bring about to this world.
The first wave naturally is the internet wave, because internet has more data than any other domain. With so much data, it powers Amazon to predict what you might want to buy. It powers Facebook to predict what you might want to read. And similarly, all the American and Chinese companies, all the great AI companies of today are all internet companies, because they have the most labeled data. And we every day, are their lab rats labeling data for them. So a lot of great companies are near trillion dollar companies have been created, the seven giants of AI, all in wave one, and that will continue.
But wave two has, of course, already begun several years ago, because you can ask the question, who else has a lot of data? And it is businesses, banks, insurance companies, hospitals, they have a lot of data in the past, they viewed data as a call center, as a legal requirement to archive. But now data has become a goldmine. So think about the bank that has all the transactions from you. And now it can use that data from you and all the other users to understand how you invest, how to improve your asset allocation, to give you a better return, how to sell products to you and make money from you how to recognize whether a credit card use is legitimate by you, or potentially fraudulent, and whether to approve a loan that you would like to apply or that you are that you might default. So these are the uses that could be done in a bank. And you can extrapolate for insurance companies, hospitals, and so on. And that's going to be a big wave that's running across many segments of business and government.
The third wave is what I call a perception AI, that is you're adding to the AI the ability to see and to hear. So it's computer vision, speech recognition, speech synthesis, understanding combined together, and it's also can be viewed as digitizing the physical world.
Whereas the first two waves are taking big data that's already stored. The third wave is transient data that usually just disappears, but now they're being captured. And they're being used to power new usages and applications. So what can you do? Well, Siri, Amazon Echo are examples of speech recognition. The demonstration, you see, there is another another one live live transcription. And in terms of computer vision, I think the applications are even bigger, you can have autonomous stores, you can have face recognition, you all know this at airports. Now, there are almost no humans taking care of our immigration because face recognition and OCR are taking over. And these are all examples of perception AI.
And then the fourth wave is when AI becomes autonomous in its ability to move around and manipulate sort of like having hands and feet. So that is in the area of autonomous vehicles, and robotics. So in autonomous vehicles, that will bring about a huge transformation, because it isn't just about a but
it isn't just about the button that you press to have the car drive for you. But it's going to be a synthesis of electrical cars, shared economy, and also autonomous driving. So no longer do we have to own cars. Did you know The car you own is the worst investment you've made 96% of the time it's sitting, idle, depreciating. So why would we own cars for that 4% of the time that we drive. But if there were Uber that can come to you in 30 seconds, anytime you need it. And it's, um, it's actually always has a perfectly courteous driver, because it's autonomous, and it has no emissions because he's electrical, and this perfect route planning, it saves us time doesn't pollute the air, less traffic congestion. No longer do we need our parking garages. So it's going to bring huge changes to our lives.
Furthermore, when car autonomous driving, when they collect more data, they will get better and better. When it's first launched, it might be a little bit safer than human. But in five years, it will be a lot safer. In 10 years, it will be much, much safer with many, many lives saved. And after that as car start to talk to each other, they will be able to do things that humans cannot do, they can transmit to each other and say, I'm going to have a flat tire. Get out of my way. Or they might talk to each other and basically miss each other by one centimeter, which human humans cannot do. And when they get that good, the only danger on the road to human lives is ourselves.
And the natural next step is we won't be allowed to drive anymore. Just Just like you can't ride a horse to an interstate highway, right? It's the same thing. But for those of you who love to drive, not to worry, this is probably 30 years away. And when it happens, even today, we can go to a ranch and ride the horse. So there will be something for you at that time.
In terms of robotics, there are so many innovations happening, it's moving a little more slowly, because it requires mechanical aspects. And not just software. But that will definitely come about we are invested in autonomous robots that move items in the warehouse that put packages together, we're invested in autonomous robots that deliver to homes and apartments, we've invested in robots that can pick fruits that can wash dishes, and all of these will come about over time. And the price will come down as volumes increase.
For example, the dishwashing robots is a company we invested in. Would you like one? You can just throw everything in and it cleans the dishes and, uh, puts all the refuse garbage into recycling bins, saving a lot of time, now, you all want it, right. So it's, I'll take a check for $300,000 each at the end of the session if you want one.
But don't don't laugh. Because if you have a restaurant with two five people, washing dishes, it pays for itself in about a year and a half. So that's how this will begin selling. And of course, over time, it will get better and better, cheaper and cheaper. One day, we can each buy one for our homes for $500.
So this kind of robotics and autonomous driving will be the fourth wave these full waves together I think will bring about huge changes to our future economy, perhaps each wave can be 10% more to the global GDP over time, perhaps it might also take away 10% of our jobs.
So to make AI work, we need the following five things, you need a lot of data that is tagged within a single domain, a lot of compute power and some experts to work on it. So AI is not perfect, you can't make a do perfectly unsupervised learning. You can't make it learn on very little data. You can't do AI with very little compute or no data. But once you have these AI can be applied.
So about US and China who's ahead. I'm first going to answer the question in very detailed way and then tell you that it actually doesn't matter. So I will try to answer clearly in research, US is way ahead. These are the top researchers in the US. Not only is US way ahead in research, US has been at the center of every technological revolution in the past 30 years.
But something magical has happened in the past 10 years that is a gigantic market in China has emerged. And a lot of money, capital investment went into China with smart VCs helping smart entrepreneurs build products and companies. And those products actually are so attractive they brought more Chinese users on the internet. And this loop kept going and going for the last 10 years taking China from 150 million users to about 800 million users by far the largest user base in the world. And this loop has created something that we never thought possible that a system that parallels the Silicon Valley.
So today there are actually two separate systems. It is true that Chinese companies began by copycatting a lot of American innovations and at that time, it's understandable because when Google was started US internet penetration was 30%. In China it was 0.2%. So who's a what's a Chinese entrepreneur going to do other then look for inspiration. This is not IP theft, this is just inspiration okay. And but today, China has because this cycle I described evolved into a completely different way of entrepreneurship. I don't have the time to go into a lot of details.
So I'll give you one example. Whereas in Silicon valley, it is most respectable to have a light tech company, a company like Instagram, 11 people billion dollars, a company like Yelp just does reviews and it's $4 billion valuation or great companies, very respectable, all tech only and not touching the offline ugly stuff. The Chinese way is completely different because in China, there are many other entrepreneurs who can get funding and compete with you and even copy you. The only way to succeed in China is to find a business model that is impregnable. In other words, build a business that's uncomfortable. So how do you do that? I'll give you an example. I mentioned Yelp. Let's look at Mei Mei China's a coming his company that just went public and they dare to change the way Chinese people eat. And that is
how can we make people eat cook half as much how can we make people eat out half as much instead rely on home food delivery of food to the home that is not purely a technological problem that is largely a huge operational excellence problem. It is about organizing 600,000 people on electrical mopeds delivering to the home and of course the price is very important because VCs will only burn so much money. So one she needed to get the cost per delivery down to about 70 cents to break even that sounds almost impossible to people in the US. But even accounting for urban density and wage differences. Even if you say is four times I doubt companies in the US can deliver to each home for $2 or 250, right? So how does it get to 70 cents? Well, he hired 600,000 people, he came up with innovative algorithms to motivate them. And he paste them at modest wages with encouragement, sort of the reverse of the
surge pricing on Uber to encourage them to ride when the supply and demand are out of whack. And he he, he found the cheapest way to deliver electrical mopeds which don't have enough batteries to last very long. So he has to solve that problem for them. I could go on, but I think you get the idea. The way may turn one was by building an incredible operational excellence system that could change the way people eat and, and
an under at 70 cents, 30 minutes delivery to your home. And underlying that is a tenacious entrepreneur who's not afraid to take on the very, very hard and ugly and operationally difficult problem and completely, continuously grind away at it using huge amounts of capital to interactively drop a few cents every few months, eventually getting to his goal. And by the time he reaches his goal, if a competitor wants to build an equivalent, Lee efficient system for food delivery, they'd have to hire 600 and manage 600,000 people basically add minimum wages and still manage a good level of service to the customers as well as deal with best route finding encouragement and burn the few billion dollars. So that is how he makes his business impregnable. Unlike other lighter businesses, that of Instagram, or Groupon or Yelp, that are a lot more easily attacked by other companies. Silicon Valley tends to respect others and not attack them. But in China, if you had for companies like let's say grub hub, Yelp and
Open Table and sort of Groupon, Groupon, Yelp, Open Table and grub hub, these four companies cannot possibly all survive in China, there would be only one left, but that one is the one that comes up with the most resilient business model. So, the point I'm making here is that China didn't get to where it did because of copycatting that was a good important part in the beginning. But China got to work got to by developing a completely different heavy and exciting way of building impregnable businesses.
So as a result, Chinese companies kept improving going from copying from the US to inspired by us and then leapfrogging we chat is better than WhatsApp way boy is better than Twitter, our investment to who is much more profitable than Cora. And these are all examples. But even more exciting is the third ladder where Chinese companies are brand new innovation, this Chinese model of building impregnable businesses have reached new heights, so that these brand new companies are being built. And also being copied now from China, these companies, for those of you who are Chinese, you'd recognize them, those of you who are not, I'm afraid I don't have time to explain them, because each one would take five minutes because they don't exist, there doesn't exist another correlate in the US.
So the now these models are being copied to a lot of areas worldwide. And the reason that may turn was successful, and DD and many other companies is that these entrepreneurs may have begun as copycat but they didn't copy forever. Copying was just their way of learning. And then they became great entrepreneurs with a Chinese model of entrepreneurship. And that becomes a Chinese advantage in AI. Because these are the same entrepreneurs who are going to build AI applications.
The third advantage for China to overcome its talent deficit is a large amount of capital invested. Last year, China had 48% of the world's AI investments, and us only 38%, just in sinovation ventures, we have five unicorns in AI. These are all pure core AI companies with a total valuation of 21 billion US dollars. So these are examples where China's already use the capital advantage to push forward and also AI itself, the deep learning was invented 10 years ago, there have been incremental improvements. But there are no other big breakthroughs. So we're at an age where we're no longer thinking about inventing the next deep learning. Of course, that's possible. But But today, I think we have moved from the early adopter phase where very few people understood deep learning so so a
deep PhD from us has a big advantage, we've reached a new phase that I call widespread application phase, where what matters is the application the value and creating the value with great engineers, so that you actually need the business person, a solutions person to work with the CTO. And when we're in the right hand phase is advantage China because China has all these entrepreneurs who are battle proven and who work incredibly hard and who are tenacious about finding opportunities. And that's what matters in the second phase. The
fifth advantage is that China has a lot of data with AI, the more data, the better it works, you can see on the right hand side, no matter which algorithm you'll pick with more data, the model becomes more accurate. So in an age of AI, where data is king
data is the if the data is the new oil, and China is the new Saudi Arabia. And
and this shows you that China leads in data not only in the number of users by a factor of three, but in the use of food delivery by a factor of 10, in the case of mobile payment by a factor of 50, bicycle shared bicycle rides affair, a factor of 300, you might say, hey, shared bicycle rides. That's not data for AI. But it is every time you write the mobike six sensors are transmitting all kinds of information. So that mobike extracts 20 terabytes of data every day. And that can be used for AI. So this is all data in a breath away because they're more users in a way because Chinese people use more online and offline services that have the data captured. And then finally, the sixth advantage Chinese government Chinese government assists AI in three important ways. And there are different from what you read in the medium.
Most of Chinese Chinese AI was basically built by private capital. The government came in about a year ago on the SS on the left side with a State Council development path plan in 2017 that legitimize AI as a national priority that had and the effects of
setting the tone with that document banking customers are much more open to buying products from our AI companies. And we saw that but it wasn't injecting money into the industry. The second thing in China is on the rights it's about building infrastructure. I think the Chinese government lets the private enterprise fund and build companies, but there are things private enterprises can't do, such as public infrastructure infrastructure equivalent to the electrical grid and to the 5g network.
So what is that equivalent in AI? Well, I'll give you a few examples. China is building a new city show on and that will have the size of Chicago and that will have autonomous vehicle roads built in. So Joe is building a two layer road one layer for autonomous one layer for not show is building a building a two level wrote the downtown one level for pedestrians and pets and bicycles, another for vehicles, these will all make autonomous driving safer, Georgetown province is building new highways with sensors that talk to autonomous vehicles.
So these are the kinds of advantages that I think China got Chinese government helps the AI. And the last thing I think, is that the Chinese policy in general, is what I call technical utilitarian, that is, when the new technology comes out, let's give it a try. And if it looks good, then keep it going. And then if there are the issues regulated, Western governments tend to basically that that
all the possibilities and then decide on regulation before the technology goes out. So Case in point is how mobile payment took over China. In the last three years, there are no almost no cash, no credit cards in China. Because the Chinese government permitted Tencent and Alibaba to move forward in a Western government, they might have been stopped due to lobbying and concerns raised by credit card companies and banks that software companies, companies can't reliably protect your money. A hypothesis proven to be false, not known by the Chinese government gave them a chance and Alibaba and Tencent prove themselves worthy.
So where do we stand with China and us when the for waves I think generally us is a little bit ahead today. But China will probably be ahead in four and five years. This is not about research. This is about implementation us will continue to be ahead in research for the next 10 or 15 years, because that is very hard to overcome. But having given you this picture, I also want to point out this is not a zero sum game. Us VCs fund US companies who develop products for us customers, Chinese VCs fun Chinese companies who develop products for Chinese customers, they are not going after the same market.
When Chinese company wins US company does not lose when US company wins, Chinese company does not lose. So I think the sentiment behind the current some of the current rhetoric is not correct. This is truly not a zero sum game. This is merely a keeping score of how far ahead each technology might get. So with China and us both pushing forward AI, I think AI will make a lot more progress than internet and mobile because those only had one engine the US pushing forward. And there are a bunch of other reasons such as the seven giants hiring people, training people with large amounts of data VCs being devoted to AI
and also AI platforms emerging that makes it easy for people to develop applications pw see estimates by 2030, the whole world's AI industry will add about $15.7 trillion to the global GDP. And if you're not, you're not sure how much money that is. That's about today's China GDP plus India GDP. So A is incremental value is huge. But also AI brings a lot of challenges in the US, you hear a lot about privacy, security bias from data, those are all issues and have to be worked out. And I will also can potentially help a company become very strong, because AI reinforces strong companies and monopolies. And I can increase some wealth in the quality. And finally,
AI can potentially lead to job displacement. We talked about how AI can within one task do a phenomenal job better than people. Well, how many people really primarily, repeatedly do the same task over and over again. Now, people in this room do not. But if you look across the world, a lot of people are in fact doing jobs that are purely repetitive and routine. And if you look at the jobs from left to right, it is anticipated the jobs on the left will gradually be better done by AI in the next 510 and 15 years. Some jobs are reasonably safe, because they're complex and creative. So some of us feel okay, but the AI
machinery is moving forward on both blue collar jobs and white collar jobs. And there's some examples here, there are burger flippers that are replacing humans are burger flipping robots replacing humans in the middle you see a pastry shop cashier auto checkout that we funded that's replacing cashiers on the left you see a Chinese fast food restaurant that's purely robotic. We also found that that and if that gets more market share than McDonald's and KFC will inevitably have to do layoffs. And the right hand side you see city has issued a warning about automation of jobs. And then we all know Facebook newsfeed and hotel are getting more eyeballs, which means traditional media may face some squeeze.
So the job displacement doesn't just come from one to one replacement, it may come from industry wide disruptions, if so many people out of jobs, are these enough jobs to have they can, can they all be scientists and artists? Of course not. But if we think about what AI cannot do, one is it cannot create the other is that it doesn't have compassion, it doesn't have the human touch that many jobs require the service jobs that have a strong, empathetic, compassionate human component.
So actually, this graph should be two dimensional, one dimension to show you how creative a job is. And another dimension to show you how much social and compassion for job is. And if we replied, the picture will see that while the lower left side our jobs that face a lot of trouble, there are actually a lot of additional jobs that can have more high can hire a lot more people jobs like teachers, nannies, nurses jobs like elderly caretaking, that's a job that there are a million openings in the US and it's only not being filled because it's not paying enough. So
just as we saw a big agriculture to manufacturing transition in the last two centuries, I predict we will see a routing to social social service job transition over over the next next 20 or 30 years. So in terms of AI and humans so some people say it's awesome biotic Some people say it's all displacement. This graph shows you there are at least that's not right, there are at least four different outcomes.
The lower left side what purely be done by AI, lower right side will be AI tool to amplify human creativity. The upper left will be AI doing the analysis such as doctor's diagnosis, but the human doctor wraps warmth around delivering the good news to the patient and giving patient that confidence and comfort. And then of course, the upper right is where human compassion and creativity will shine. So there you have it a blueprint for human AI coexistence.
So going forward, I think AI is electricity in the next 20 years, there will be huge opportunities and challenges. But I want to take us a moment into 50 years. When we look back ignore for the moment all these job displacement opportunities, I like to leave you with two thoughts. The first thought is that
AI is serendipity. It is here to take away the routine jobs so we can really spend time on what we love and what human beings are on this earth for. And secondly, for those worry about AI causing problems. Just keep in mind AI is just a tool it possesses no creativity. We are a eyes Master, we are the ones that have free will. And it's going to be up to us humans who will write the ending to the story of artificial intelligence. Thank you.
By the way, if you want the slides, send AI to text ai to 345345.
Thank you so much, Dr. Lee for this wonderful speech shedding light on this interesting topic. Well, one thing we are sure I hope that we all gonna keep our jobs in the race with AI, but Dr. Lee said exactly right. This is not a zero sum game. If China and the US can join hands, and instead of adversaries, the world will be a much better place