Kai-Fu Lee - Will China Supersede Silicon Valley in the Next Era of AI? - What's Now SF

well good evening I'm Lanny Cohen I'm Capgemini global chief innovation officer and it's a real honor to have everyone here tonight I can tell the demographics of the audience is a little bit different I think we've got a little heavier European mix I think we got a little heavier millennial mix here tonight how many first-timers for what's now I guess I guess kind of right oh wow okay well listen everyone welcome it's a real pleasure to have you here we've got an incredible program tonight from a very very distinguished person our industry you know I'll let Pete do kind of all that introduction one thing I want to start with on a serious note though is these wildfires in this state are just it's just unbelievable what's going on and I would just ask we're not taking any donations here tonight but but anything you can do to help out with the the relief and the recovery efforts please even if it's just rounding up you know your your uber your lift whatever you can do is really really gonna be helpful because this is gonna be a pretty big thing to overcome so anything you can do much appreciated I also want to welcome some some guests here tonight a lot of Mike Mike colleagues are here from Capgemini and we've been on them for since we started what's now like you know consumed or kept generally people come so we've got a lot of them tonight in fact we've got a lot of our automotive industry leadership here tonight so thank you guys and girls for all being here really appreciate it and a lot of our top global customers are here and for purposes of confidentiality I'm not going to say who they are but you know who you are so welcome and thanks for being here as well we'll really appreciate it and then as I always like to do I always like to introduce our applied innovation exchange team from Capgemini so when we're done for the evening I just encourage you to search them out and get to know them and get to know more about what we're doing here so our aie gang can you guys just put your hands up or stand up or do something so again when you're done just say hello to them and they'd love to tell you about what we do here and then the last thing I'm going to say is coincident with tonight in about the next 30 days or so Capgemini this is because it's our place we get to say a couple of things about ourselves but we're launching our global AI strategy into the market it's called perform AI by Capgemini it's gonna be an incredible program so any of you who not that we're hiring people I know some of you think we might be but well we can talk but but the real thing is if if you want to know more about what this what this solutions about and what these services are about again love to spend some time talking to you about that as well but you'll hear more about that through various marketing channels towards the end of the year in the beginning of next so with that where's Pete there he is so we've been doing these what's now's for I don't know pushing three years now and we've got them in New York all the time and we've got them in cities around the world now so it's a great great program and we're fortunate to be partnering with reinvent and Pete and Andrea and Brad have just been incredible to work with so let me introduce Pete Lydon and we'll get the show started Pete thank you Lani Thank You Laney I think you camp Gemma again uh who have really been the great partners for this we have built an amazing network of folks across disciplinary like you over the years and we're getting literally world-class speakers coming like Kai Fuli today and it's really been great right and we're also deep into planning for 2019 so we're gonna have some stuff coming up into the next year I'm Pete light and I'm the founder reinvent which is the media company which is you know helps produce these and also streams these three camera shoots by the way if you're gonna tweet about it use the hashtag they're opening up these to wider audiences but thirty years ago I was starting my career as a journalist as a foreign correspondent in Asia I was working with Newsweek and literally I was able to watch China cover visited and cover it actually as a young reporter literally right around the time of 10 min squared 89 90 and that kind of period in fact I was literally in China about a month after Tiananmen Square wasn't always in the big cities I was in Guangzhou I was in Chengdu I was in Tibet under martial law all kinds of different things but I had a look at what China was like thirty years ago and when I came back to the stage that started covering technology and started working for Wired and all that kind of stuff I always kept an eye on China and it's been an incredible story I mean it's literally gonna go down as a world historical story of what China has accomplished in the last thirty years in just the way they've transformed their economy and how they've mastered so many industries but one thing that over those years they haven't really or they've been let's just say the technology world they really haven't been driving that anywhere near or what it's been happening in Silicon Valley in the American kind of economy and so I've watched that over the years as they've evolved at the best you could think of it as fast following and sometimes just outright copying and but rarely kind of driving this story until now and in fact tonight our speaker Kai Foley and in his book AI superpowers really lays out a provocative argument that China is really going to probably be driving the next era of AI and he makes a very convincing argument now I will tell you I read a lot of books like this and I do a lot of speakers you have books and things I will say this is one of the best books I have literally read in the last year or two honestly and I don't always say that it's a terrific book it really is thought-provoking it's accessible highly recommend it to anyone actually and for several reasons and the one reason is even someone who's tracked China pretty well he really opened up my mind of what has happened just in the last five years or so maybe ten to five even the last five years in China the technology seemed really where they are and where they are closed to go in a way that it was really really opened it up to me and one of the reasons I think he's so good at that is he's had a foot in two worlds he actually studied here as a student and computer science he's worked Apple he worked for Microsoft he actually when Google's office in our whole operations in China so he gets what's going on in an intimate way what's happening the states but of course he's Chinese he lives there now and he has his own company but you see company called sin evasion ventures which is got its fingers in all kinds of new companies over there so he gets both cultures deeply which is a big thing but if that would be enough of a good book just to kind of explain China but actually he's one of the leading thinkers in kind of AI has been for years here and he really comes up with for an insight into four different waves of AI that are coming in various stages were kind of in him right now fascinating way to lay that out he I'm sure he we're gonna talk about that later so he kind of is his own insights on kind of where this is going and that would be good enough but he income then he has a whole other piece of the book is there really an analysis of the disruption that's gonna come to the economy to jobs in ways that I think put a different layer on it then that kind of a lot of these books that have been out there recently and he really has some profound insights into how disrupt this is gonna be and frankly he's on the high end of disruption compared to folks out there and he'll lay out some that I think today but he also comes up with a bunch of solutions including some new ones some really transformative ones ones big-picture once so there's all kinds of solutions in there and the final thing that's fascinating out this book is there's a personal side of this book he actually has an experience a personal experience which he may or may not talk about here that actually transformed kind of his own thinking around the future of AI and where we need to go on the planet here in the coming decades so for all these reasons it's just gonna be a great night now I will say we're gonna have him come up and he's gonna do about a 30 minute presentation kind of laying out the big themes of the book so get everyone on the same page but then as is our way we're gonna roll into a conversation with me a little bit to draw up some particular themes of interest in this community and then we're gonna roll into a conversation with you many of the folks here have expertise in AI we know a really interesting group of gathered here and so it'll be a great conversation and to start that all off let's welcome Kai Fuli thank you thanks to Peter for the probably the best introduction I have heard thank you and I'm not going to go into the personal story I need to leave a hook for you to buy the book so I am going to talk about the two aspects that the two other aspects that Peter talked about and when we talk about deep learning I think most people here know it but to kind of summarize it deep learning machine learning is the big advance in computer science and what that has led to is take one single domain and throw tons of training data at it and it will do superhuman performance so let let's let someone much more famous tell you more about that it's a great thing to build a better world with artificial intelligence I'm going to now jump that guy mentioned yet so so that thank you that was not president Trump talking that was a speech synthesizer using deep learning trained from President Trump speech and it could speak not only English but Chinese and that system was built in China so hopefully that gives you some background about what this is all about and I think artificial intelligence with the core built on machine learning with deep learning and related technologies like reinforcement learning again and other technologies are really starting to mature and in my book I talk about how they will permeate the society in four different waves I think sometimes people get confused they think oh Alexa that's AI or autonomous vehicle that's AI but what people don't realize is the first wave of AI has been here for ten years and you're all using it and that is internet every time you use Google Facebook or Amazon it is actually using those very same deep learning algorithms that I talked about as the autonomous vehicles so it's one set of algorithms behind all four waves okay so the first wave is the Internet because as I said deep learning is a huge amount of data one single domain so what is the best case where we have four that is clearly the internet space right think about how much data we are generating as guinea pigs for Amazon Facebook and Google and Facebook actually is able to construct a knob that says okay of all the clicks I get from all of you what should I show you in the newsfeed so asked to achieve some objective function and that function could be more users per minute more shares more sessions more minutes per day more minutes per month it's all one knob so imagine the temptation that a business person has when that magical knob comes to tweak the business however they want so if people thought Facebook a couple of months ago went a little far on other things it's not that they intent fully did it it's just that they couldn't resist the temptation to use tweak that little thing to for more money and then there were side effects that they surely did not anticipate but that happened so the same thing with Amazon it can tweak do I want more revenue durable more profit do I want you to come back more or do I want to make more money today Google can tweak it for the ads and so on and so forth so it's large amounts of data every time we click we are contributing as guinea pigs data for Amazon Google and Facebook and that is why the most powerful AI companies are all internet companies because they've had the very good luck of picking the right area where where there was enough data in a single domain to do so well but the second wave has happened the second wave is around using AI for businesses I'm sure a lot of Capgemini customers are thinking about it this is potentially usable in banks and insurance companies and in anybody who has data that is historical data in the past only used for archival reasons but now can be turned into rocket fuel for AI and the more data you have the better it performs so a bank with this customer transactions in the past was just a burden to have to store them but now they can be used for things like asset allocation targeting ads products for you and also credit card frauds and determining if you're worthy of a loan and so on and so forth and before you may be a 1 best example to illustrate how these industries well both either we'll both transform themselves and be disrupted let me give you an example of a company we invested in a second wave it does a long application and this long application is an app you download on your phone and like any other loan you would fill out your name social security number address who you work for how much you make where do you rent or buy a home and then also it asks one other question it says send data to send data from your phone to help determine the long application and the data it would send is not all your secret stuff on your phone it's the same thing that Facebook gets or that Amazon gets that you click when you download their apps and with this input it's able to build an amazing long determiner a loan officer that no human can ever possibly match so here's what it does imagine would you if you had $500,000 would you go outside the street and find the first thousand people you know and give them five hundred each and say please pay me back with such-and-such interest rate what would the default rate be I mean be phenomenal be 70 percent 80 percent and if you were a bank it would probably be still 20 percent 30 percent right so this app has it down to three percent so how does it do that well it uses all the informations through deep learning and all the samples of defaults and non defaults to learn what to do so what data does it have well it has not only all the information you entered but how long it took for you to enter if it took you a long time to enter your address or that's something fishy right and all the apps you have if you have a bunch of games or gambling apps things like that well cannabis apps and things like that well that could all go against you on the other hand if you have really wonderful apps like horror or things like that well that's really respectable then it might be in your favor and also what kind of phone you laughs do you have the fancy new iphone XR or do you have some old five year old phone that will be a matter for the determination and it goes all the way to what day of the month is it and it can guess if it's before payday or after payday before payday loans are better after payday loans are terrible right and it uses all the way it has 3000 features the lot wa and we just for fun I went to look at the last few features that it was marginally useful but still useful one of them was your battery level can you believe that does that what does it matter well I think about it if you have an OCD and plug in your phone and charge all the time you're probably a little bit more correlated with the type of people who repay loans right if you keep letting your battery run out you're probably a little bit correlated with a little bit less responsibility so I mean I'm not gonna this not about data bias or ethics this is just about efficacy right so 3,000 features goes into a deep learning and the company just iterated initially now you can say well with who label the data for them well we the VCS gave the money to lose to label the data for them right we funded them 10 million dollars and then they lost it all because bunch of people default it but they got personal the signatures of these people and many people who returned then they learned and they got the default rate from 20% down to 10% then they raised another 30 million dollars then they reduced the default rate again and at one point in time they said we're making money now we're gonna not raise any more money from VC and just borrow money and because we know that we can make more money than the borrow rate so that's an example of a disruption in the second wave but there will be many subjects disruptions so before you think about the banks will be banks insurance or companies or insurance companies so just you use a I think again there will be disruptors out there let's move on to the third wave the third wave is when there are eyes and years and see and hear and also other sensors that capture things that weren't captured before and those things are taken into account to build applications that couldn't be done before face recognition is the most famous or example an anecdote from China recently a very famous singer named Jackie Joan Hong Kong singer he went to China five cities and he gave concerts and a total of 35 criminals were arrested so how did that happen well for say for safety of the concerts they put in cameras all over the place and the cameras were connected to face recognition and face recognition was connected to most wanted criminal database and and maybe a hundred people were apprehended there were maybe 65 false false alarms and sorry we didn't mean that and then 35 arrests so Jacky Cheung is not known as the top police performance in in China that is something we again there are other considerations that many of you are thinking but in terms of efficacy crime crime rate and just using that for profit or for benefit or for good there are many imaginable things if you can recognize faces you can recognize people animals you know cucumbers and and anything actually a I think a Midwestern farmer's son wrote based on tensorflow a cucumber shorter and and that saved his family a bunch of money because then they could put the you know the the big and cling and shiny ones from the you know ugly ones broken ones and then they could ship them and put them in different grades so AI really is everywhere and that's computer vision the cucumbers water as well as the face recognizer but going further there will be not just recognition of objects there will be recognition of intense if you go into a store pick up something look at it and feel happy or depressed or unhappy or disgusted that may get captured and the store will remember that to show you less of those things if you were disgusted and also it's even more powerful offline because online you're just clicking that your system doesn't know if you were happier it discussed it but offline the cameras can capture that and also your actions right Amazon Go is a great example it's happening all over China stores where cameras are capturing people taking things and putting them into baskets or trays and get having them recognized and charged directly to their phone when they're done shopping so leading to completely autonomous stores and that is talked about in the US but already being implemented in China and we're investors in several of them then we move on to the fourth wave whereas the third wave gave the eyes and ears the fourth wave give the arms and legs to this system not physical arms and legs but the ability to move and manipulate and that is the age of robotics and autonomous vehicles now that's going to be a little bit further out because the mechanical aspects are not instantly fixable but there are many things that can be done in manufacturing inspection we have someone from and ruines group and they're doing inspection for factories to see if an iPhone and IC board has an issue there are uses in dishwater they're their dishwashers possible with any of you want an automatic dishwasher that actually takes everything from the tablecloth and just dump it all in like we have out there and it cleans everything up and separates their recycle bins would you want that so I would along with the sign books will take orders for those those are three hundred thousand dollars each okay so anyone still want them some restaurants do and this is how technologies get adopted right there are some restaurants that have 10 dishwashers people these washers and they buy one of these giant machines and they make up the cost in seven or eight months and the rest was all free so they will be the first adopters then the prices will come down as the cost to make them become less in volume so that kind of robotics and of course autonomous vehicles I think we read enough about those I won't say too much other than that you can imagine one day you know we're going to have much safer more efficient and healthier air clean energy all delivers through autonomous vehicles that not only get better with data but that they talk to each other so they will avoid incidents you know one car can tell the other I have a flat tire stay away or they could say my my Peter hired me and he's in a hurry so get out of my way I'll give you 25 cents right so these kind of communications beep on they'll be communicating to no exact location there will be high-definition maps that will cause one car to miss another by one centimeter and both cars know there is zero chance of accidents so ultimately we the people are the biggest danger as drivers so not only well autonomous vehicles be on the street but at one point maybe 30 40 years from now humans will be this allowed to drive so if you enjoy driving drive a lot it's just like you know horses aren't allowed on highways right so for the same reason cars one day won't be allowed in various types if I'm humanly driven cars won't be allowed but but not to worry if you don't want to ride horses you can still go to farms and there will be car farms for those of you who will still want to drive so that is the progress of AI in the four ways and as I said the big big issues are for AI to work it has to have a lot of data and ideally automatic labeling right the labeling of the long app I told you about didn't come from some human but they came from actually whether the person repaid or not and a single domain that is actually a huge constraint because systems have to be training one domain with one objective function they can't think out of the box and of course a lot of compute and a lot of AI experts and speaking of experts while us has all the experts these are the top Turing Award recipient AI experts these are the top deep learning inventors you don't see any Chinese so why should try and have a chance if you look at go down the list of top thousand researchers US has 68 percent China six percent again why would China have any chance well let me tell you the reason is I think we're missing three very important knowledge points and I'll go to each one of them the first one is that there has been one huge invention in AI and that invention is well understood by the Americans and Chinese and all the other people who care to study it's that and if you look at the history you know I'm buried somewhere in here with a little bit blip I started working on a a long time ago it started as 62 years ago and really I think if you measure the magnitude of impacts deep learning is just by far the single greatest breakthrough and it is now reasonably understood I don't think people understand everything every theory about it but in terms of using it for real applications it's well understood and ready to be applied also this slide is another answer to the people who believe in singularity or that there will be super intelligence coming to take over the world well if you look at the history this is 62 years one big invention eight years ago none no more after that so how many more would singularity require probably dozens so I think we're safe from singularity until we start to see a larger number of breakthroughs which is not impossible but based on history not terribly likely the second important thing based on that point is that we're going from an early adopter to a widespread application in the early stage when deep learning was understood by very few people maybe 5,000 people 2,000 people well if you were one of those people you could start the company and be the CEO and deep mine is a great example of that and you'll do things others can't but when deep learning is understood by hundreds of thousands if not millions at least in terms of applied engineers well it's completely different it all depends on who's got more engineers who's got more entrepreneurs and who's got more data and the advantage actually would shift to the countries that have more engineers and entrepreneurs and the third point is just that as I said that the barrier of entering machine learning and AI is going lower and lower these are all the open source efforts that allow anybody in any country to get started in AI so it's not the difficult high barrier super expert kind of thing that it used to be so when I talked about the super AI experts actually for most applications no there's obviously apps exceptions like autonomous vehicles but for most applications you really don't need the experts if you have a bunch of very smart young people they can do great things let me show you what great young people have done we hold a summer camp for people with no prior AI industry experience and not graduated from college these are just students who maybe taken a course in AI we give them five weeks and let me show you some examples of what they built here's an example and what the autonomous vehicle that they built this is the camera of the autonomous vehicle driving inside the Peking University I mean it's a small sized car but it's obviously recognising people bicycles skateboards and not running into them so students can now do this supervised by a senior person but students can now do this so the difficulty is not what it used to be and that's it important to understand because once you understand that you will see why China despite being behind in in the deep research has a chance because of these three things which weren't fully understood so now that we understand these three things that some deep learning is not as hard as it used to be and that we've reached the era of application and China has and also the platforms are making it easier and easier for even college students in five weeks training to build apps like that so now what's what's good for China well first while Chinese researchers are actually a lot better than what I portrayed earlier if you look at the total number of researchers who publish papers Chinese publishers are actually 42% amazing number this shows you a pyramid where the Chinese engineers are flooding in they're not yet at the master level but they're going to keep growing okay and this is the picture of the summer camp we have this is a picture of me lecturing to Chinese universities about the act so you can see how popular these things are getting also the Chinese entrepreneurs are now innovative this was one of the key points Peter talked about I don't have time to go into details I will just say here that if you look 10 15 years to 10 or 15 years ago Chinese companies were basically copycats about five to ten years ago they began to localize and innovate a little bit so that's actually pretty good because like we Chad is probably a better app than whatsapp and Twitter sorry wait boy is probably a better app than Twitter and and actually not every case better but I think on par with Silicon Valley five to ten five years ago but what's most exciting is in the last five years these Chinese companies well this one's a little older maybe seven years old but the other ones are all within five years old and they were created and they're worth a total of about three hundred and twenty billion dollars today and these companies are all innovative brand new Chinese companies with apps that don't exist in the u.s. some of them would take way too long to explain so I'll just pick pick a couple and financial that spun off Alibaba but nevertheless it is basically the ideal frictionless financial institution with payments and all the financial instruments you can imagine it's all in one quite show and quite so and the subsidiary of total I'll call tick-tock are they have 200 million daily active users not users daily active users between the two apps tick tocking question 200 million daily active users I mean in the US how many apps have that you know Facebook Google and only a few like that and what do they do they actually deliver a video based social network something tried in Silicon Valley and failed with vine if you remember but succeeded in China so hopefully gives you an example the entrepreneurs are able to find new and innovative ideas not breakthrough like iPhone like SpaceX none of those but really solid products more like facebook snapchat level of companies these Chinese companies can do and at this point we've developed to a stage where China and US are actually parallel universes so it's a picture when anarthas just drew for me you're the first group that's seeing it because this kind of demonstrated Silicon Valley a way and the Chinese way and it's entirely different and parallel and for those of you who always ask me is Google going to succeed in China I think this gives you the answer you know you're crossing parallel universes China has developed its own complete entrepreneurial ecosystem with Chinese Internet and mobile companies worth about the same as the American ones and with Chinese own innovations so I think we have reached a point where crossing the universe may be a little bit difficult and in doing so well what is the secret formula of the Chinese people that they just have a bunch of Steve Jobs no Steve Jobs is the essence of Silicon Valley right starting with dr. Turman who invented the the transistor and there Steve Jobs and others that's what makes Silicon Valley great but what made Chinese companies great are some things that may not sound so glorious but very very critical the very fact that the Chinese ecosystem was full of copycats forced the entrepreneurs to work incredibly hard in a gladiatorial environment where one wins and everybody else loses and not only do you have to win everyone else has to lose because otherwise if they're still alive they'll copy you again and again they'll copy this feature copy this and copy that the only way you can become King is if you build a product that's uncopyable so that sounds to Silicon Valley that sounds crazy you know what's uncopyable every feature is copyable well what's uncopyable is through incredible iteration and a process excellence that you build something that would be way too expensive for a competitor to build and it'd be impractical an example is a mate one which is some people call China's Yelp but it's absolutely no Yelp Yelp is worth about three billion dollars May 20 is worth about 50 billion dollars as a public company and what did they do that's different they change the way Chinese people eat right they figured out that in order to change the way Chinese people eat you do food delivery such an incredible level that you let anybody order from several hundred restaurants and delivery in 30 minutes and delivery charge is 70 cents that's what it would take so they just went and kept cutting away at that problem reducing a few cents a month and burning billions of dollars per year and eventually got their things they have to do are things like hiring six hundred thousand people at minimum wage and buying the cheapest electrical mopeds figuring out for the batteries to change but just relentlessly shooting for that goal of thirty minutes get the food to your table and cost no more than 70 cents otherwise people don't buy it and they succeeded and that Iraq's a huge barrier of anyone who wants to beat them had better spend that much money and that much effort and they may not even work because they now have the brand and essentially a monopoly in this space so that kind of winner-take-all truly winner-take-all people here accuse of Microsoft and Facebook and Google of being monopolistic but the answer is in the internet space you need to be number one and number two gets nothing and if you can find a way that your competitors can't even copy you then you have a way to be number one and you might ask was that just one case in may - I know that China's uber didi has done something similar and I won't go into details you have to buy my book for that but there are lots of examples Taobao has done stuff like that there are many many examples of the Chinese playbook which is make a product that by far number one and uncopyable and that I think will be studied in many business school books while it's not as elegant and brilliant as Steve Jobs or Fred Terman but it does work and it creates very powerful and valuable businesses and of course Chinese engineers work incredibly hard one commercial by one startup company said we have great work/life balance where only nine ninety six and that means 9:00 a.m. to 9:00 p.m. six days a week as opposed to 9:00 12:7 for many other startups now I'm not advocating that lifestyle but this is the Chinese work ethic because it they're people urgently want to become that final successful company this is actually an old picture that shows you $0.10 employees at som this is a large company not started up at 2 a.m. these taxis were lined up to pick them up so not just startups but throughout and maybe a lesser point is that more money is pouring into China more money than in the US China the u.s. in AI for the last year and Chinese AI companies are very successful these are five examples these are the five unicorns that we have invested in they're worth 21 billion dollars we were in at a series a they're about a total of 15 Chinese unicorns today and they're the most value also the most valuable speech recognition machine translation computer vision and drone companies are all Chinese so this is all a result of just relentlessly working hard and applying known algorithms to AI problems and another this is the probably the biggest point is that data is very important this is a very typical graph you see in various AI papers what this shows is with the more data you perform better more data the better you do and it's more important than what algorithm you pick as long as you get more data and if in the age of AI data is the new oil then China is the new OPEC and that used to say Saudi Arabia of it not too good to mention nowadays but anyway the China has more people and more usage per person more usage I mentioned with food delivery ten times of us 300 times shared bicycle rise and this is actually a giant IOT system don't think the share bicycles are dumb they're smart devices and most importantly more mobile payments 550 times more than the u.s. in China last year there were more mobile payments than China GDP and that's a little hard to imagine but okay next you can ask me later if you don't believe it it is if you think about a dollar of sales there are actually many dollars spent to create a dollar of sales in China people don't use credit cards and cash anymore my wife found was last month came back home and told me she saw a beggar holding up a sign that says I'm hungry scam me I would never joke about something like that this is really what you see and this is a farmers market so that kind of mobile payment I think does many things for one thing it removes the credit cards which is essentially a 2% tax on the American economy but furthermore it proves software companies are trustworthy and can do all this despite what financial institutions might say but most importantly all this is incredibly valuable data for training the AI because this is a lot more valuable to know you bought something then if you clicked on something in Amazon or Google so this is incredible for the AI companies in China and finally there is the Chinese government which is obviously very supportive of AI there have been various documents the most famous which was the July 20 2017 AI document where China says it wants to be a world leader in by 2030 in research and by 2020 leading in technology and applications and by the numbers I'm going to show you I think 2020 is clearly on track and then 2030 I think research may be a little hard but at least to be number two is and undoubtedly doable and that while the central government sets the tone the local governments come up with creative ways of how they adopt AI the city of Nanjing said we have good universities we're going to build the world's largest AI incubator and and Science Park and the city of Shaolin was built it's a brand new city size of Chicago being built with autonomous driving built-in with two layers of road in the downtown top layer for it's very green for people pets and bicycles and and so on the bottom one for vehicles so that immediately rules out the possibility of what happened in Phoenix with with the uber autonomous driving cars and pedestrians are separated by just sheer brute force costs of building a two-layer Road and those are the kinds of things infrastructure plays that I think will be very helpful to Chinese air companies think about an American autonomous driving company it may be technologically farther ahead than China but China may have more safety problems but by doing the two-layer road maybe that evens the difference then what happens is maybe China will launch first collect more data and then eventually get better because data is very important so this is kind of my projection of what will happen that's basically four or five years ago China was nowhere in AI and China woke up and is kind of behind catching up with the US and probably will catch up and lead the u.s. in whether you measure by monetization adoption usage or revenue probably certainly not in basic research that remains America's strong point so with this these factors all going with US and China both going at it and seven giants producing more talent and lots of VCS including Softbank with a hundred billion dollars investing in a space and AI becoming lower barrier what will happen is this will generate a tremendous amount of value various companies had different estimates I use a PwC here they estimate that AI will have a net incremental revenue DP of about 16 trillion dollars to the world by the year in just 11 years and with that's benefits that may eradicate or reduce hunger and poverty there are also many problems I'm gonna choose to focus on just one of them but we can discuss the other ones in the Q&A later I'm gonna focus on the job displacement issue because we talked about AI as a single domain large data super human performance and when you extrapolate that that means a lot of human jobs not not any of you in seat seated here but most of human jobs are routine jobs in the single domain quite repetitive and routine so if we kind of have plot jobs all the way from the most repetitive to the most creative we're going to see that the jobs will be replaced with only a few people just barely squeaking by the complex jobs are cannot be done by AI because as I said AI is one domain right complex job is a multi domain creative jobs can't be done because a ice has objective function which we give it and it learns to optimize it doesn't really rewrite or event that but the displacement is already happening because the routine jobs can be better done cheaper done by AI with no complaints labor union or you know unhappy workers and so on it's already happening in in the white-collar jobs as well as blue-collar jobs I won't go into details here but you know white collar jobs in general will actually come first because they're just software blue collar jobs a little more difficult but in various types of blue job up local jobs that are fixed location it actually can happen this is some one of our investments that is are just checking someone out with the pastries you bought these pastries instantly checks you out so one machine replaces one cashier about two thousand dollars one-time costs and in this case it's an autonomous fast-food restaurant they will make you a bowl of beef noodles or a fish ball soup for about a dollar fifty to two dollars so about a third the price of McDonald's more tasty at least for the Chinese and that will not directly displace our McDonald's employee but to the extent these types of restaurants eat into the shares of McDonald's and Kentucky Fried Chicken jobs will be lost so displacements are very very real so with all the potential displacements on the a I will create more job so who knows what they are but can they really do these jobs certainly not but if we really think about what AI cannot do we've covered the part that AI cannot be creative or strategic but another really important thing is that AI has no empathy or compassion that is there are many things that we really expect of a person in the job that is human connection that you only trust a human to do X if you think about that there are many jobs that are like that so instead of thinking about just creative jobs from less creative more creative we can think about another axis that shows you compassionate required compassion empathy required or not and then we will find that there are actually many jobs that can grow or be created or entrepreneurs can build them or social entrepreneurs can build them that the people here in the lower left corner if they lose their jobs they can more likely be retrained to the upper left then the right side for example elderly care that's something that has about a million jobs open right now and they're not being filled because there's they don't pay very well so rather than thinking about universal basic income why not find a way maybe through giving each person over 80 some kind of stipend or ability to to hire elderly caretakers at a higher salary than today then suddenly a million jobs will appear and those are surely needed because older people are your parents or some of the younger ones your grandparents are not gonna want to be taken care of by robots they want humans to take care of them and that is the kind of job I think will require human connection there are also many luxury jobs because there will be many millionaires and billionaires who make money from AI and those people are going to want the concierge and the luxury vacations and so on and so forth and that will create a bunch of jobs so AI will create jobs but there going to be more like data scientists kind of jobs and and I don't know if we can expect to retrain the lower-left into those jobs but fortunately we have the upper left so this is a summary of a short-form summary of what I say in the book is that this quadrant actually tells us about four ways in which we will coexist with AI the lower-left jobs will be sold totally replaced with it by AI and that's just going to happen over time it might be 15 years it might be 30 years but it won't take longer than that but the lower-right jobs the creative jobs actually are very useful because these are the you know inventors researchers who will use the tools to help them become more creative and then the upper-left are the cases where like a doctor would use a tool to analyze and diagnose the likely problems of a patient but the doctor would wrap warmth around it half the human comment connectivity and and convince the the patient that he or she has a high likelihood of recuperating and then finally the upper right are the jobs that we will Excel because we have the combination of creativity and compassion so there there is the coexistence map of humans and human and AI it's a little more complicated in that there's blue color and dexterity and a bunch of other things but this at a high level gives me hope that we we not AI can create a bunch of jobs that will help the displacement problems and in the long term AI will create a lot of jobs so the problems will go away so in this talk I have talked about the next 20 years full of promise and also full of challenges but let me now take more like a 50 year view 50 years looking back I think the next 20 years are challenging because there will be AI that makes money a lot of privacy bias security issues jobs lost jobs gained etc but when it's all said and done AI will have taken all the routine jobs away 50 years looking back to today will actually only be thinking two things number one is that we'll be very thankful that the era of AI came about because AI liberated us from all the routine jobs so we can do what we love and and find what why we're here as humans and lastly we will be very thankful and that's we are the only ones who have free will ai is merely our tool that we use as this master and that we're the only species with free will and the ending of AI is for us too right thank you fantastic fantastic folks you got a little taste of what that book is Wow thank you for that I saw many people putting their phones up and trying to like grab insight in what you just laid out there you laid up so many things so masterfully and I don't want to repeat any of them but I want to kind of draw out a few things that you didn't touch on okay they think is worth bringing out particularly this crowd okay one is you really talk about AI as a kind of a meta innovation it's it's almost like a world historical innovation that is really in a different league more like with electricity or is it talk more about how why this is such a different thing than just another kind of innovation that we've seen many many times in history yeah I I think the most the most powerful thing is that AI is Omni use that is usable for every area where every domain every industry so in that sense it's even more than what the Internet has done because it can be applied to every industry and it's increasingly better to do single tasks jobs better and better than we do and I've tried to illustrate through the four waves about how expansive the impact can be but actually I think there may be another 10 waves after the four waves so even without further giant breakthroughs just the extrapolation of what we already know I think is creates something like electricity because there's an enabler that can be put into every domain in many cases enhancing the work yes cases disrupting and making it dramatically more efficient or more profitable than before so the flipside of the disruption without going into all these is incredible productivity gains as you're saying through the AI this world historical technology and because of that at least in the current way the systems built incredible inequality and do you want to just talk a little bit about that you talked about the disruption but talk about how the system is essentially that you will concentrate it'll-it'll tend to concentrate well absolutely I think just looking at the first wave look at the winners from the first wave they occupy what ten of the Fortune twenty top twenty ten of the richest people ten of the 10 of the 20 richest people ten of the 10 wealthiest ten 20 companies or so and that's the first wave is a combination of large data and AI and disability to tweak and imagine that starts to permeate every other possibility so the first Bank that figures out the brilliant I Bank and that might be and financial for example right that's going to be a trillion dollar company the first company to figure out an e aí assisted and AI disrupted insurance that's going to be a trillion dollar company so so the more riches will be created than even the Internet and mobile era and that's on the one side and then a lot of the people who do routine jobs their jobs will be displaced they might make it as a compassionate worker they might do something or they might find something to do or they might retire early but the average per-capita income is going down for the bottom you know 40 50 percent of the population so AI will exacerbate the inequality so when you talk about solutions you kind of alluded to it but again and I want to get the conversation to open it up a little bit here quickly but I do want to touch on a few of these which is you incredible wealth incredible productivity but lopsided wealth there seems to be some mechanisms that have to move that wealth differently through society and you threw out in the book you didn't really touch it I'm here some kind of ways to think about that do you want to just have a few kind of thoughts on that just a reflection I'm sure think about that sure so first I don't like you be a universal basic income it is part of the solution perhaps in the sense that it does the income redistribution there are two parts of it one is income distribute redistribution and the income district redistribution were likely to have to be done through one of the two things we can't avoid tax right other than the government taxing the ultra-rich it isn't clear how this money will come up with a lot of Silicon Valley Giants are proposing ubi and I think they know behind that is that they will pay higher taxes and I think people are willing but somehow the government needs to get away to tax the ultra rich Bill Gates has talked about the robot tax I think what it means is people who make a lot of money from their having the robots have to pay taxes I'm on the board the Foxconn so if you purely tax on robots were in big trouble Foxconn doesn't make that much money Apple makes more the money but anyway I'm the I think the ultra-rich or the people who make huge amount of income or benefit greatly from AI are the ones who can pay for that distribution but that distribution of income is not even half of the problem I think the main issue that will hit the people who may face this job displacement isn't the loss of income but the loss of meaning because many people have associated job with meaning and there's also very strong research that shows people with sustained unemployment face a much higher percentage of likelihood of depression suicide and substance abuse so when we think about how to deal with the potential displacements just giving them money isn't the isn't the only answer and but what what can we do to accelerate then the upper left of the quadrant right I pointed one example is maybe elderly care could be something that could be subsidized that if the government said anyone over 80 is entitled to 15 hours of a week of elderly care and by the way that job has a minimum wage of 25 dollars an hour then suddenly I think the the well-being of the elderly as well as the the people who shift jobs over I think will both increase and also the amount of time it takes to train an elderly caretaker is not so outrageous it's not like training a data scientist and would be too hard if at all possible there are a lot of other things that could be done Amazon this April announced that they would put 12 thousand dollars of training per employee they will reimburse for up to four years and and they actually listed the courses they would reimburse and when I look through it I think you know Jeff clearly thought about these are the jobs that don't go away they include jobs like Arenado creep air something AI can't do for a long time nurse nursing another one and and I think it's it's very much matches you can just imagine Jeff is thinking that at Amazon the warehouse employees they're going to be replaced by robots in three four five six years I don't know but he knows and then these whole food cashiers they're gonna be replaced sooner so here is a benefit that helps them get the next job and I think that is very noble thinking that a corporate a corporation is not just responsible for using the employees to as an exchange of money for contribution but there is a responsibility for the corporation to take care of the employees even you to find another job even if it's not in the company so it's a combination of things that governments can do and corporations can do in my book I also talked about instead of ubi can there be some kind of a stipend that is given to people if people are displaced then they can get a stipend they keep learning new skills that make them employable or if they contribute in some way to these kinds of elderly care or voluntary services I think that's going to be a lot less expensive than ubi and also increase and accelerate the path from lower left to upper half in that one of the most interesting things I thought about you talk but also your book was that you kind of flip on its head things that Americans think about China is problematic you actually flip it around as potentially advantageous for example the copying of the entrepreneurs which we kind of would see from one way they and the lack of the intellectual property actually stimulated a different kind of entrepreneurial energy you laid it out well but one that's actually really interesting one right now is when you look about the disruption of the AI coming and also with climate change frankly these our governments are very different and given that you're familiar with both you know kind of the Democratic way we do it here and the kind of Chinese way of doing things how would you handicap both societies in terms of how we're gonna deal with what's coming up here in terms of the moves we have to make and how how we might fare any reflections on that well I think China is really taking advantage of the form of government and the opportunity in AI because I think all the steps that I have listed are they're already doing all the right things I think on the unit you on the US side I think it's important first to not feel us is the only way to lead forward I think is to accept that there are two ways of building products and innovation and to study both ways otherwise I think American entrepreneurs will be missing half of the materials in the textbook and I think the government should think about what is the core advantage if you kind of place through all if you accept everything I've said today the biggest chance for us to become a big leader again in AI is the next breakthrough right I think that is what the government ought to do is to fund dramatically increase the funding in a our research and find ways to protect them through patents and bring the world's smartest people to us to study I mean that is the magical weapon right because it's it's um most countries have brilliant researchers roughly proportional or ought to have brilliant researchers roughly proportional to their population u.s. is the single exception because most of our rules researchers want to come here or our students want to come here to study and once they study here a decent percentage stay so that's why US has so many I mean plus all the other research institutes and universities and I think I think immigration needs to be kept open for students I think visas should be granted for people who do excellently in in schools and you know every PhD in AI maybe should get a green card automatically you know if we want to be a little extreme I think these are the metrics I think basically fundamentally is betting US can make the next big breakthrough and then regain the leadership I think that is plausible but the current policies don't seem to maximize that chance final question before we can so think about folks really my own view thinking about ideas their questions um it's not just the US and China you actually have a global perspective of how the two countries actually interact with the rest of the world in this field we have a lot of Europeans here kept Gemini's rooted in Europe you give a little reflection on how the world kind of the large the broader world actually fits in this future and the opportunities there for example or whatever and maybe I'll reserve one last question of that but to talk a little bit about that yeah well I think Europe is a is very challenged in this in this situation because if you think about what made Silicon Valley great right it's the visionary technology top universities breakthrough and the strong venture capital ecosystem as serial entrepreneur right if you think about what may China very successful is the ten tenacity speed execution of the entrepreneurs a strong VC ecosystem and coming up with a number of playbooks that help companies build the business models that are uncopyable now Europe doesn't really have most of these almost all of these attributes so in theory Europe should have the wealth and population as these two as the third big entity but it doesn't seem to have happened and I think the challenge is how to I think I think of all these things there are many things that are cultural just lock right place at the right time and sheer size I'm sorry China also has a huge size of the market right and government policies so for Europe I think most of these things that China us possess are very hard for Europe to have maybe the only one that can be actively built is a VC ecosystem because that seems like something you can do with money and by poaching the right people you know in in you know China's VC equals is the most built by basically picking a few of China's entrepreneurs and a few of Silicon Valley's VCS and said okay you're the beginning of China VC ecosystem then they've got going so my advice to Europe would be one is to build up a VC ecosystem even if it by brute force and money and then the other is the European universities are very good but there are a little academic and not as connected to industry and applications well many US universities have the same problem but but Stanford would be the counter example right what would be the Stanford's of Europe right that would be another fair question worth asking and and also I think Europe should be cognizant that the steps that it takes while very respectable to protect privacy with measures like GDP are it is going to cause the process of building a giant AI company harder as a result so that's the price that's paid it's okay if that's done Kanchana bleah but-but-but this is just a trade-off final question without giving away the book which I encourage you all to buy but you do have a personal epiphany there about how you had this kind of workaholic yeah kind of you know life yeah but are you pivoted to a kind of a more a humanistic kind of understanding do you want to just say a few things about that oh we can give away the end yeah it's not a suspense yes okay yeah I have been working the Chinese hours for most of my life certainly once since I moved to China until about five years ago when I was diagnosed with for stage lymphoma and that was when suddenly I realized well first was the denial and all the other stuff I had to go through but once I accepted the fact that this is what I have I faced I finally look at my life and said it was all not worth it because all that hard work I don't even want to spend another minute working and that the purpose in my life couldn't possibly be just because of the working hard and the accomplishments and the fame and the wealth that was not worth anything and and it turns out that many people facing death almost all people facing death come to that same conclusion that there are many things much more valuable about our life giving love back to the love people who love us and doing things were passionate about and not being just listening to the environment and society and what we should do but do what we really want to do and and people don't do that they conform and then they work and they think this is what they want and then the epiphany I came to was that I done this all wrong and and yet it's an yet I wanted to first write a book and tell people about my death facing experience and which I did in Chinese and it was the worst selling of it it did sell okay it's so the 200,000 copies in China still did okay but it was the worst selling of my books because and it turns out the people who generally buy them to give to their parents or grandparents who have cancer and say well read this read this this will give you hope oh it would it didn't deliver so I think I then realized okay maybe China is just in a phase of development where there's not yet a large enough middle class there's still too many people who have not streaming to be the haves there are too many people who've been poor for 10 20 generations with high expectations they have to Excel that's continuing to preach that message my might not might not work but I did try anyway but when I came back to work and I saw that the once upon a time the know one of the loves of my life AI has come back to life and deep learning was working and then of course I started investing in these companies but I was thinking well maybe I'll give it one more chance that maybe or maybe this is not maybe I'm any one person near-death experience just can't be transferred to another person it seems plausible that can't be transferred but maybe it is our maker or our collective consciousness if you're not religious that really got fed up with this world that's just consumed by working hard and work being the meaning of life that it's that that the our makers through AI at the world and said okay I'm gonna take away all your routine stuff so now you have no excuse right you have no time you have time you have time now right because maybe in the future there is going to be three day work weeks maybe we can do what we love maybe there will be stipends given to us to pursue hobbies maybe we can now rethink that work isn't the only Center in our lives that there's more to it than life and that's that by AI taking away the routine work we can finally slow down a little bit and think about what it is that we love to do and why what does it mean to be human what away all right questions let's change to some questions here's one in the back here yeah let's give it right there Gregg okay get him a mic by the way just introduce who you are and wait for the mic and and let's say go ahead thank you very much my name is Gregg DeLong I'm with the urban innovation exchange looking at green and clean technologies for cities thank you very much for the humanity you're bringing to this and and your thoughts on displacement of jobs etc we're facing an existential crisis with the climate wildfires sea level rise you could throw in population explosion all that if it's not too much of a red herring do you see how this AI revolution and what the next 30 to 50 years look like how what's the relationship or is there a relationship between AI and a solution set for climate change in our environment okay thank you it's not an area that I've studied very much but this seems to fit in the lower right quadrant right scientists who are looking for four causes and solutions and correlations can look to AI to help become a tool that helps them discover what it is they need to discover I think these tools are a little bit nebulous when I talked about them because they don't exist yet but it's sort of like today as a writer you couldn't do without Microsoft Word as a photographer you couldn't do without Photoshop I think in the future there will be tools developed for people who study the climate people who want to gather insight from satellite images people who want to draw correlation between the forest fires and other factors so I think the scientists really need to be trained on using AI not becoming AI expressed just like you know writers don't have to know how to write a Microsoft Word they just have not how to use one I recently gave a talk at the recipients of current and past projects through price and that was yuri milner Sergey and Mark Zuckerberg and others donating a lot of money to the top scientist in the world and and winners are selected every year and I gave a talk to them and and actually to my surprise very few of them understood what AI was and could do and almost none of them I asked for a show of hands were using AI on their day to day job yet they were brilliant but they haven't thought about how to use a AI think that kind of connection and advocacy for scientists to use AI is probably the best way for us to gain more insight on that other questions here there's a swimming back here okay let's get one there and I'm just looking at put the hands up folks who wanna okay good all right hi so my name is soft brown I work for the British government as part of our science and innovation network I've also had stay for cancer and it was interesting to you kind of hear your take on all that experience left you with I'm interested that it doesn't seem to have left you with an impulse to kind of press pause on some of the ramifications that AI might have it seems to me that it would make sense for humankind to kind of stop and Porras and get together and consider at an international level what it is we could do to ensure that a chart that AI results in more social goods that it might take away okay sorry you said that poor the social hasn't got me to do more of what paws things all paws yeah oh of course it did that's why I wrote the book I could have spent the time investing in more unicorns I think part of the book is to articulate what I think some of the dangers are and I am a part of actually I'm a part of the number of task forces looking at how to address the dangers of AI one of them is World Economic Forum's AI counsel I just agreed to to co-chair that council and I'm close to people like partnership for AI I actually know all three members of your UK AI Council and we've we've exchanged email tweets and phone calls so I I do try to helped some of the problems are are really big problems and I think all of us have to jump in and help and I think there's actually a lot of misunderstanding outside the AI community on some of the dangers what's perceived as the largest danger may not be as large as perceived what's perceived as maybe small issues may actually be bigger so I think it's incumbent on AI scientists to really share what we know and and and actually and actually try to find ways to to address them but at the same time we have to recognize cultures are different so there may or may not be a single solution probably there isn't a single solution for all countries there are probably some lowest common denominator values across all humanity that's that's that can be collected but there are aspects of Europe UK US China Japan that may lead to two different ways of of working on those problems right here but by the way I'll just say again if I need another plug for the book the book is extremely accessible to anyone and I would HIGHLY encourage anyone to kind of who wants to wrap your head around a good way to start go ahead my name's Jenny Jenny Oh Jung and I'm with github so even the head of Google China and of course with your own company sin Ovation ensures you obviously have a lot of experience both in US and China what advice would you give American companies that are trying to expand into the Chinese market given its protectionist ik kind of government policy as well as like the cutthroat kind of competitive market environment mm-hmm okay well give hub is different I'm a big fan of github as you may know I've helped you a great deal about seven or eight years ago in China but we can talk about that later but well that I can't say that I can't say that the environment hasn't hasn't been hard for American companies but I think rather than saying it's mostly protection is mystic today we'd really have to parallel universe's so that's why for you know Facebook to expect to enter China it's not simply a matter of getting the permits and just launching Facebook simply is by far not good enough for Chinese users compared to WeChat I mean you may you may have used both to give you the counter example you know we just saw Alibaba did thirty billion dollars in one day right eleven eleven sale so they must have a great product right that's how they got all the sales so let's say Ali Baba launched their e-commerce in the US are you gonna use it or will you keep using Amazon of course you use the Amazon so I think the parallel universes have been formed so it's really difficult to to to displace them and the the difficulty is probably more with how to compete locally and how to change user habits I think generally it's going to be really hard for a company that already has a leader in the equivalent space in China to enter it won't be successful regardless so Google search Facebook's you know a social product will be very hard but new products that don't exist locally may have a chance so github I think has a chance there is a local C Sdn but it's not nearly as good as github you just need to do a few things right to to get in I think I've exchanged email with your CEO a long time ago but we can talk more later okay we get right here do you want to stand up in energy sure yeah hey yo my name is Shawn Ron I work with a trader I thank you so much for going through all the things that you showed us and so I'm a small question I'm gonna run through a few slides that you showed so you said in one of your slides that China is catching up with the United States with dd0 Suber and WeChat is eco so what's I even though you could claim that doors are better than what we have right now and and the other slides you said China is leapfrog frogging in the implementation of AI while us is more on the research side and one of the boxes said China has more data and then you said most of the Chinese apps don't even exist in anywhere except China like it's not open to the world and you can't download most of the apps here and there so my question is what is China doing that they have more data compared to others because everything that we have is running all over the world like the rules of Facebook Instagram it's open to the world while most of the Chinese app is just open to China so how is that they have more data and that is what so basically that is what the whole idea is that since they have more data they are implementing it and they're leading in this whole process yeah so how is that so what are they doing to have that more data okay very good question yeah you're right if you compare just US and China China has three times more users but if you say Facebook covers the world so that should be more data than the WeChat has just purely in China I have a couple of answers to that I think the first answer is the Chinese dude users have more depth so as an example Chinese mobile payments that's a kind of depth right so the data that $0.10 and Alibaba have on me is much more than Amazon and Google even if I use the products you know comparably because I pay everything through them so they have all that record and then I talked about food delivery and the shared bicycles so the Chinese people use new things more and then those get recorded and become data that becomes rocket fuel so that's one part of the answer I think another part of the answer is that homogeneous data actually works better so if you just mix the data from you know Indonesia India and Nigeria and us together they're not quite the same right because the you know local people click click on local links for Facebook and so on so the the if Facebook had you know a billion people users and we Chad had 500 million it isn't clear that had billion I'm making up the numbers right but it's not clear that billion data across 100 countries is better than that 500 million so those are some reasons that explain they don't completely address the issue I think there is an underlying question that how can China ever claim to be leader in the world when the software is not used anywhere outside China which is a fair question then people ask me all the time china is exporting this software if you think about why china has an export let's just think about consumer internet software so let's think about people like us there are probably about 250 million people like us in China who use internet roughly in the same ways social you know ecommerce and so on and for China to export software in those areas ecommerce social search to other parts of the country unfortunately that's no longer possible because they missed the window because US took over right specially the developed countries however if you look at the when I gave you a list of the 300 billion US are china new innovation like social video video social networks those are actually largely in China used not by the 250 million people like us but the and not the other you know 600 million users these are young people from small towns and and sometimes it's frivolous just for fun and and those um those usages are much more similar to the types of demographics that we see in Southeast Asia possibly in India possibly in South America and eventually in Africa so Chinese companies who are not who are doing innovative software software that doesn't exist in the US software that's targeting smaller cities less educated less wealthy people actually may have what it takes to go to the belt one belt one road countries over time and the way in which China is doing that is not necessarily through a expansion of a single platform if you think about the Silicon Valley mentality the world revolved around Silicon Valley everybody used Silicon Valley products so it feels like that's the only way to go one brand taken to the world but China having been technological colony of the American technologies and having seen the inadequacies of Windows and Google and Yahoo and so on and build his own products is actually quite empathetic to other countries so when I said DD is very successful in a global expansion it's actually partnering and making deals with many local companies in share right share hailing right hailing and supplying not just the capital but the technologies including AI to help them compete against uber so that's the interesting direction so it's a more a little more collaborative partnering approach and we are seeing more Chinese software being exported in the innovative classes so I would guess in five years we will find that developed countries continue to purely use u.s. software almost purely maybe 95 percent but developing countries there will be a significant penetration by the Chinese software that's a prediction hasn't happened yet yeah you get the mic there yeah and there just a few more questions and we're gonna have drinks and more food and signing books my name is Carlo Bellini I'm a physician professionally working in AI having two questions a personal and a technical one for you with the interesting life journey and insights you've had with your illness I'm wondering what makes you most fulfilled in your life and secondly regarding artificial general intelligence do you believe it's achievable and in what time frame okay well most fulfilling would have to be with the people I love and that's by far the most important thing to me and that's the revelation that after my illness that I realized that I had not paid the right kind of attention to them and that I was able to get a passing grade by them it was all through a clever AI algorithm in my head you know placating their loved ones while focusing on my work but now I put a priority what my loved ones want and they're very reasonable they don't want me to spend you know 50 hours a week with them you know it is and they have those things that they I know that they need and want I want to be there before without them having to ask me and then I still have lots of time left for work which which I do and a GI artificial general intelligence well I think it's a long time from now we we saw the picture of the last 62 years there's only one breakthrough I think getting to AGI is going to require maybe 15 more breakthroughs and based on that chart it's hard to draw a line of whether the 15 breakthroughs will come in the next decade or two or five or ten I would guess if if 62 years we've only seen one breaks through and that was eight years ago it's probably not going to come all of a sudden and I've talked to a lot of people who believe in AGI and they haven't shown me any engineering evidence that there is progress towards that so I'm going to say not in the next 20 or 30 years beyond that it's really hard to tell but I'm also going to leave me myself with a little way out in case we start to see some breakthroughs I will be the first to tell you through my next book although I will just add in the book you make reference to potentially even centuries I mean it might be way off from what you thought that's right yeah actually it might be centuries might be never right I think for those of you who believe the sanctity of the human soul or believe there is a God then AG I can't be done right all right we got a question here maybe yeah I am Gerald Harrison president of the quantum planning group we do scenario analysis so my question really is around the future of AI you seem to be very much a supporter of it but as we all know one of the problems with AI is garbage in garbage out terms of the data right so what happens when the underlying meaning of that data changes or there's some Black Swan event and you have some AI systems still feeding on that data and it does damage in the real world who's responsible for that how how do we get compensated for that who do we hold accountable for that we have cases in this country where AI has been used against arrogance because it really did not didn't recognize our faces okay so I have a real question about your optimism here in the lack of you saying who's gonna be responsible for the mistakes that are made okay very good actually from the earlier question from the lady from UK similar question I think there needs to be these ethical questions discussed and consensus that come up with and actually among the first things that are discussed and I was just at a this was a bloomberg group that of gathering of AI experts and the first thing that we agreed on is figuring out the accountability because I think that's something we can do and with the accountability that other things will fall into place if we go back to the time when credit cards first came out in the United States what happens when your credit card is stolen is the accountability for the user the merchant the bank issuing the card or Visa and MasterCard it was somehow decided it was Visa MasterCard and then all the things fell into place because they had to make sure they charged enough money put enough safeguards put in the fraud and then what figure out the insurance systems so I think accountability is something totally that needs to happen first autonomous vehicle is a good example people are debating that now and I would really encourage people who debate the autonomous vehicle to talk less about the trolley problem more about the accountability problem so I'm totally with you on that point on the parts about the balance of data garbage in garbage out I think what's important is that a well there's natural incentive for AI developers to build AI that works well and for that to happen there has to be very good high quality data that's why the loan company I told you about was losing tens of millions of dollars until they got real data of people who paid back and didn't pay back they didn't go for some garbage data of some randomly labeled thing that would have been no good so I think the diligence and then the natural you know commercial benefits will cause most cases to be okay but if we come to data and how to make sure that discrimination doesn't happen I think it comes through several things first I think we want the data as much as possible to be trained on real outcomes not on some human labeling because then the human bias will come through secondly I think we want the data to be balanced you met your example with African Americans examples were simply an unbalanced database that needs to be balanced and I think you can trace back and say therefore maybe a future rule of AI building is that your data becomes balanced before you launch at scale right and I think the other thing you have to make sure so a high quality label balanced data sets and the other thing you can do is if it's really important you felt that a certain minority group or gender has been discriminated against for centuries and we need to purposefully move it out just in case there's any data you can remove gender or race or some such things from the training data so I think there are many things you can do is it perfect of course not can you infer someone's gender from other data or race from zip code and things like that of course it's possible but there are steps that we can conscionable e take that will make what make things better and and also I think rather than assuming AI is heavily biased and mistaken I think we should also take a step back and think about the human bias I would argue the human bias is much more serious and in fact not fixable whereas I gave you three ways to make AI bias better and I hope there were are ways to force developers to be trained on it and do it and check against it and you know just like today software developers have to make sure there's their code is as bug free as possible as security hole free as possible there should be some AI check in the future as well but I my point is that at least AI and data can be checked I think human bias unfortunately is very serious in my book I gave an example of study in Israel where the judges before lunch gave much tougher sentences than I've for lunch right and that I mean that bias is an even discrimination based just I'm hungry I'm gonna be me so I think the human controllability is a lot tougher so we should do everything we can to control the AI but but also not assume that a human represents some kind of a golden benchmark okay what okay you got the mic all right go ahead last question Henry at school it's international triple helix Institute University industry government interaction one thing that's predictable is that if you ask the question late in the event it's likely they've already been at least partially answered and that's the case here but the question anyway is as you're aware a I was invented at least in its modern form at a conference of academics in the 1950s at Dartmouth College some of those inventors are still around today but they expected that the problems of AI were going to be solved relatively quickly translation would just take a summer give the problem to a graduate student and that was it well so you had the fourth and fifth waves are you on the optimistic side that these problems we solved in short course or will they tend to be further out than we think okay very good question and what you say is exactly right that people have been over optimistic about AI two times before already and this is the third time are we over optimistic I think we're definitely not and the reason I say that is we've crossed the human capability barrier right before their optimism were purely based on passion intuition and things like that and and and extrapolation not based on reason but today that this is really based on solid facts I mean AI recognizes speech better than humans do AI recognize faces a lot better than humans do AI place go better than humans do and and the list goes on so the four waves I'm I would say I'm highly optimistic that the first and second wave in the next five to eight years will make huge disruptions in the society the third wave is a little trickier because there privacy issues about putting cameras everywhere in countries that choose to do them there will be disruptions as well but then it's not just cameras I mean if you don't want cameras there are new sensors that don't maybe maybe we feel okay about the privacy so I think still third wave will be significant even in countries that don't put cameras everywhere the fourth wave is is is is probably the trickiest I think the conservative even the most conservative computer scientists would place about 30 years as the time when autonomous vehicles and pervasive robotics are around us the most aggressive ones would be saying like three to five years I think there I would tend to be a little more conservative because what it takes to adopt these things are not just technology being good enough but also issues that have to do with liability accountability security hacking and things like lobbying of various groups and laws and regulations and explainable AI all these things will need to come to place at a level in each country that feels is acceptable so the fourth one I'm going to be a little more conservative but even then we're going to see I'm talked about dishwashing robots those are going to happen right I talked about in examinations of icy boards by an assembly lines those are going to happen agriculture AI that's going to be relatively easy compared to to manufacturing to building an iPhone so they will happen over time but the biggest ones right the ones where there are no more blue-collar work and there's no more drivers that's easily twenty thirty years old and what a place to say let's give a big applause for tyfa league and incredible he's there's books here he'll be here to sign him for a little bit there's more drinks a little more food and I will say one thing unusually we're gonna have in the next two weeks from now we're doing our next one with Hal Harvey on a whole kind of a slew of ideas around climate change policy another it's a very different field well worth doing but just it's coming up fast will be the next two weeks and that'll end the year and then we'll see you next year but anyhow for now let's have another drink great thank you thanks a lot was straight

Channel: Reinvent Futures

Views: 34,606

Keywords:

Id: i7Aex-MWC_k

Length: 95min 16sec (5716 seconds)

Published: Thu Nov 15 2018