About the talk
Kai-Fu Lee discusses the rapid progress of artificial intelligence in China and the implications this has on the present and the future.
01:16 Wave 1 of artificial intelligence
04:13 Wave 2 of artificial intelligence
09:13 Wave 3 of artificial intelligence
10:18 Wave 4 of artificial intelligence
11:35 Five premises of artificial intelligence
11:46 Artificial intelligence in China
18:55 Chinese unicorns in AI
19:24 Predictions for artificial intelligence era
20:03 Half the world will be powered by Chinese tech
20:56 Sinovation ventures
Hello and good afternoon. My name is Kai Polly. I'm here to talk to you about Ai and China. I'm going to have someone Super Famous tell you what AI is all about. It's a great thing to Build a Better World with artificial intelligence and wait a sec. So you only thought his granddaughter could speak Chinese but that was not President Trump talking. That was a Chinese speech synthesis system built on deep learning that can speak multiple languages and deep running is the core technology that has a huge breakthrough in the last 10 years. And the way it works is think of it as a mathematical black
box. If you pick one single domain, ideally one single task because it can't do lots of domains and tasks throw in a lot of data and give it a lot of labels and tell it what to optimize then it will automatically learn at a superhuman accuracy. So that is a huge breakthrough and that's at the very very core of their artificial intelligence Revolution and give you a little bit more color on what types of applications there are. Sometimes people think of AI as a speech and vision sometimes think about this Google and Facebook sometimes you
See the flying cars it is all of the above because underneath all those applications is the same algorithm deep learning that is doing what is described single-domain lots of the label and the with a knob to tweak and what's called an object to function and we see that AI is coming in four ways there waves not faces. So they come all together some earlier some later all four waves are happening now, obviously with the internet being the first wave. So why is that because AI works better with more data, it is really not rocket science anymore.
It's pretty well understood by a lot of really smart people and if you have a ton of data that technology just works better. So every time you put 10 more 10 times more data is just jump to the next next level of performance. So now you understand why that leaders and they are all internet companies because the internet has a ton of data, so In the u.s. Google Amazon Facebook Microsoft in China tencent Alibaba Baidu and others are the ones with the most data because every day we use the internet we are contributing data, but more than contributing data. We are labeling data.
How are we labeling data everything you buy on Amazon every story you click on Facebook. He is a strong piece of data to that company that signals you or someone like you at a time like this may want to read this or may like it or may buy it and everyone else likes clicks and buys is Rocket Fuel that makes these companies stronger. So we've been contributing hugely to these near trillion dollar companies. Think about it. If you are a business executive and someone gave you a black box. And it says use your pick pick your
favorite business metric and tweak that metric automatically and money or whatever just comes out. So Facebook May tweak more minutes per user Amazon my mate week more Revenue per visit and or more total revenue and Google make a tweet in the more appclix. I'm not saying they do that but that is an amazing capability that as the CEO that you now have to tweak the data how to tweak the way I so that is optimized is your business objective. So that's why AI is
like printing money for these internet companies. So that's Phase 1 Wave 1 there will be more internet companies with internet companies. You got together user get a day. Then you're on to something now. Let's talk about way too because internet companies aren't the only companies with the data. Who else has Businesses right in there enter enterprise software in the day the warehouse what type of business is the ideal business for AI in business is financial because there is no friction. There's no Warehouse Logistics delivery. All you've got is a human fabricated a
game of numbers where accounts receivables Investments insurance policy. They're all numbers they come in and then the CEO says tweet something minimize credit card fraud minimize loan default maximize investment return and so on and Magic happens, of course, they're not quite there yet because these companies have both the older DNA takes time to transform but in due time all of our financial institutions will be fully in infused with artificial intelligence with much greater efficiency and profitability. Let me jump to want to give you
one example, we invested in the company now. But an app to borrow money. So this app will you download it and you can just say I want to borrow $300 and in less than one second that money goes to your phone in China mobile payments. Okay, and of course it may reject you but if it except you instantly so how many of you would take $300,000 for your money go outside in the Los Angeles and find a thousand people who you think will repay your loan and just give them $300 each. What kind of default rate would you have 80% maybe a little bit
lower in Los Angeles, but but really high right but this app has a default rate to really strangers know people that never met at 3% and now charging credit card like interest rate you can you do your math in your mind and see how wildly profitable this business is. And how is it done? Its AI what does the info today I of course you have to What's a lot of things in the form your name address income and so on but also there are three thousand other attributes that you agree to send from your phone that are part of the Dai and it's those 3000
after abuse that are really really for the cult. You're probably thinking I was sending all my secret photos and every know it's sending nothing more than what an Android and iPhone with permit an app developer to get same as Facebook. No more no less. So that information goes in and now all of a sudden the AI engine and consider not just your application. But also what type of phone do you have? What apps do you have installed? Do you have games? Do you have gambling apps? Do you have knowledge based apps and it also know some when you bought your phone how much memory it
has and it is also knows what day of the week what day of the month is it? Is it before your payday or after your payday, of course Barney? I need before payday is reasonable after payday, maybe not so reasonable. So and also how long did it take for you to type your address too fast or too slow May indicate fraud. So all of these things go in and just out of curiosity. We went in and asked Joe least show us the least relevant feature but still useful and then they pulled it out number 3000 on the featureless was the battery level of your phone that
contributed to whether to give you the long or not. Why is that because if you are a note have OCD plug in your phone all the time. It's probably a little bit correlated with someone who repaid the loan and if your battery if you keep letting your battery run out while maybe it's a little bit correlated to the contrary. Now, this is this is not giving you a terrible Credit One loan decision, but you can imagine why is drawing correlation from 3000 features against the pool of people who repay the loan and the ones who defaulted. So
how did it ever train the system while with RVC money, of course, so when we invest in this company we gave in 10 million dollars and then half of your later they came back as well. We spent it all what did you spend it? They will help we made a lot of bad loans and now we have data. And that is why we've gone from you know, 18% default down to 8% Give us another 50 million. So we get another 50 million came back later spent it all lost it all again. But our default rate is now down to 5% week and I'll break even let you alone. So this is time to really grow our
business. So that's an example that hopefully illustrate the power of the eye of why humans can never do this kind of thing. Okay ways to also goes to obviously clinics governments and cities and so on is perception AI that some speech recognition computer vision and autonomous stores and things like that. So taking giving you an example of All the computer vision app, there's a face recognition company in China. Mostly you're worried about you no surveillance apps, but I'll give you an example of how this can be used to make our environment safer. So recently
a very famous singer in China Jacky Cheung gave for concerts in China 40000 people came and as a result of these concerts 23 most wanted criminals were arrested because in the entry each person, there's a camera and the cameras connected to the Most Wanted database at 23 people were arrested. So this no human whether you agree with that use or not, whether you think it's a privacy intrusion or not. Think about can he just think about a technical level can a human policeman ever have in his or her head a million names and faces and then watch all the people entering
absolutely no way. So that's why I is amazing. Then the fourth level is way for that's where robots robots could do. Things like mmm agriculturally water and the fertilized plants pick fruits if can be used in factories to for visual inspection and simple assembly you can be used for delivery with seeing all of all of that and we're invested in an autonomous fast food and it's basically like McDonald's accept a new humans you're going in and just stopped either with your face or with your WeChat. You tell it what scan in and
tell him what you want to eat tell the assistant what you want to eat and a bowl of beef Chinese noodles is a dollar 75. So this is 1/4 the price of McDonald's and tastes a lot better. So this kind of thing will become popular and I think a lot of convenience stores and fast food will turn over into a I of course high end restaurants will still have humans, but most restaurant Santa coming in stores. We don't really need that social experience. And of course, there's autonomous vehicle. Those flying cars and so on. So these are the four ways in our mind over the next 15 years will
continuously have new applications of providing huge value to humanity and the five premises of making you at work cuz I said it's a single domain lots of data that has to be tagging lots of computer power and some experts to make things work. And now let's talk about China you probably think that u.s. Is way ahead in AI technologies that is true in terms of inventing a deep learning AI us least trying to sleep the all the all the inventors and the thinkers were American. However, why does China have a
Chance in This Global competition for leadership in the eye because there's a few observations that need to be made. The first is that AI is not rocket science. AI doesn't have a lot of breakthroughs there's only one big breakthrough how deep learning deep learning is getting easier and easier to use a really good computer science student can use the running and apply it to all kinds of problems and and there are not a lot of breakthrough is coming out every year. So if there were breakthroughs every year then you asked would be ahead of the rest
of the world. But if there aren't that many breakthroughs and frankly a lot of the top university professors and a I have been poached by Google Facebook and Microsoft so for Academia to make progress anywhere is hard so my argument and he is really that we've moved from the pain year ago five year ago early adopter phase where the very few people know deep learning and they demand the premium and they can do things. No one else can today the Technologies are getting easier and easier and the dishwasher favor not he or she who What's the best papers thinks the
deepest thoughts but who is good in implementation? And a lot of platforms are making this really lowering the berry then Tree Top computer science students can do this. So China really has six advantages. I'll just quickly talk about them. Number one. Lots of great Engineers Chinese University in engineering programs are really great and out. All the engineers want to be a i Engineers now, so we get to pick the best. We as in entrepreneurs NBC's and tech companies number two is that China has a great VC ecosystem that is
equal in my opinion to Silicon Valley equal but different to Silicon Valley China's approach to entrepreneurship is incredible hard work ethic winner-take-all and the CEO makes all the decisions you can argue that's not the Silicon Valley way, but at that actually gets companies going fast. Turn faster delivery results focus on a tension and the execution number three is that the Chinese Innovations are really matching the US it's hard to imagine but 10 years ago. China was largely still copycat, but because the market is so big and
entrepreneurs are so tenacious and hard-working in the execution oriented and the market kept growing and capital kept infusing into China many of those copycats became good entrepreneurs and some of those entrepreneurs became became innovators. So today I can list probably 10 apps in China that you've never heard of here as an example and financial definitely not PayPal you can look into it. If you want to look into Tik Tok it is actually out doing Instagram and Snapchat in terms of video social network. Then there are many
many other examples. So China is able to innovate and that's the third. Reason the fourth reason is that China has a ton of data. So in the ear of it a i data is the new oil and China is done near Mopac. And what is the status of this data isn't government collected a lot of people confuse this it's still corporate collected by this day that comes from a large number much larger user group four times that of the US and actually each person has more deaths in usage. So it's breath and number of people and theft and usage as an example.
The payment data is arguably the most valuable data because you know, what you click on may or may not indicate your intent, but what you buy is a really strong indicator. So who has all that data Visa has a MasterCard has some they don't have any way up really figured out what to do with it. But in China Mobile payment, if you ever go to China you'll find there is really no credit card or cash anymore and that Mobile payment has the following effect on the China economy. First no longer is there credit card taking away two or three percent tax at the economy who is much
faster to put on your phone and your thumb imagine your favorite social app Facebook or whatever has a payment in it. Then you can pay any imagine. You can pay anybody any friends you have on Facebook any Merchant any person by just scanning my wife was out shopping a month ago and she told me she saw a beggar holding up a sign says I am hungry scamming and it only takes 2 seconds to scan and pay somebody. I was just with one of the attendees trying to add each other on WhatsApp and it took us 3 minutes. So this is really efficient. It will also because of the
expediency of payment it will turn China from a savings account of me to a spending economy and most importantly this is an incredibly important data for training Ai and Is the reason is actually more money is going to Chinese AI that USA I measure by Venture Capital spending and it finally at Chinese government is providing a lot of help subsidies and most importantly infrastructural support as an example. Autonomous vehicle u.s. Is clearly had to China by about two to two-and-a-half years. However in the US there a number of barriers such as truckers union
lobbying and potential liability and lawsuits and and then so on that Mesa and also media headlines that may slow things down but in China what's happening is the Chinese government ask a question while Rai companies aren't as good as you ask but if what if we change the road what if we built this city so a city is being built with two layers one end at CityCenter one layer is for pedestrians one layer for cars and that would completely avoid the kind of accident that we saw in Phoenix with Uber so that will help the Chinese. Autonomous vehicle companies to launch sooner safer collect
more data and potentially ketchup or even when in terms of technology and the cost to build a new city while you figure it's the giant City the size of Chicago with a whole downtown has two layers. So those are the kind of infrastructural spend that I think are very very smart and arguably similar to President Eisenhower when he failed the interstate highway many decades ago. So the markets are also his people here are some examples of China now has 18 unicorns Nai. We are proud investors in five of them their value that
21 billion dollars. I don't really know the u.s. Numbers but I suspect Chinese AI companies are probably worth a little more than us right now measure the area by area China currently has the world's most valuable drone company computer speech recognition company translation company and Provision Company What's the who's really a had? My belief is if we look at research u.s. Is clearly ahead ignoring research. Just look at implementation user base monetization revenue and market cap. I think this is what we're looking at China going from nothing a few years ago
going into slightly ahead of the US for 5 years from now. Again, this is purely commercial. This light is not intended to show a zero-sum game because Chinese tech companies AI companies sell only in China and in the future they may sell to Sunbelt Road countries. So their success would not come at the expense of an American company, but people ask a lot. So that's my answer and you see that this is emergent. So was this put pushing forward my belief is that just one piece of the technology? You can look at all the things on your phone. It's got your your apps your eCommerce social
entertainment. The two makes your phone and then you got the server. Play Store in a 5G and the cloud and the Theta and the AI connecting cloud and phone. So this is my prediction in 5 years. If we just did a survey of every human being and every server in the world. I think China will power about half of the world. This is not saying it's a good thing or bad thing nor is the saying a parallel universe is a good thing. But this is just I think an inevitable outcome that will be accelerated by the current geopolitical tensions and the rest country's been paying countries are the
ones will be more Chinese and blue countries will be more American so lastly a few words about sanitation Ventures. We are actually very much built as a venture capital firm on this premise that China will have a lot because of us power Top China Powers 1/2 u.s. Is more or less status quo. China is all growth a lot of people questioned whether Chinese software is any good because only Chinese use it, but I think I've shown you a lot. Examples in Mobile and I and others that they found with technology is good. If it's good, it will glow go global global. So that is the premise
of salivation Ventures. My company is a VC firm. These are our investments in the last 7 years in all kinds of AI companies and our fund has done very well. Alright, I find this up about 56% net, irr manually and the work that we do is not just in BC because one of the really unique things about China is the large population and lots of Engineers lots of engineering population. So so we're impatient to wait for the engineers to meet the entrepreneurs and and come up with a business plan. So we created
a very unique entity called the AI Institute where we hire Engineers because we've done while on our funds. We just that we can hire Engineers directly with her about 300 and the spin-off companies. Not very many maybe when the When we really see a business opportunity that may be entrepreneurs don't yet see but this is not an incubator. It's driven by our investment team based on feces that we see in the way that doesn't compete with a portfolio companies and a very unique thing that we also do is we train AI Engineers since I told you this is not rocket science.
Therefore it means you can train them University's generally do a very poor job training I Engineers so we actually bring our portfolio companies in to train with us. These students will graduate within the year and then in just in less than 2 months they become very good AI engineers and part of the magic is forget the accademia just train them on real-world problems with real data with a large amount of real day. That's what they really know. What they eyes about you have to have a large population of students to select from this year. We will get fifteen thousand applicants
from which we will take a 1000 and we will train them and then we will have Select Portfolio companies and are not a i companies who have a hard time hearing a i and the rest can come to a Eye Institute, and we will incubate for them. So, I think this is a really exciting opportunity from a technology point of view on China and us are both leaders and that we look forward to changing the world in a i with you. Thank you.
Buy this talk
Access to all the recordings of the event
Buy this video
With ConferenceCast.tv, you get access to our library of the world's best conference talks.