KAI FU LEE: AI SUPERPOWERS

good evening and welcome to tonight's program  hosted by the committee of 100 and the common walk   Club Silicon Valley my name is Dennis woo it is my  pleasure to introduce Chi Fuli chairman and CEO of   sign evasion dentures and Arthur of AI superpowers  China Silicon Valley and New World Order dr. Lee   has been at the forefront of AI development for  thirty years he was the president of Google China   and also held executive positions at Microsoft  SGI and Apple he received his bachelor's degree in   computer science from Columbia University and his  PhD from Carnegie Mellon University dr. Lee has   authored ten US patents and into 2013 was selected  as one of the 100 most influential people in the   world by Time magazine moderating this evening's  program is the Honorable Michelle Lee former   Undersecretary of Commerce and director of the  United States Patent and Trademark Office Michelle   was also the Hermann Pfleger visiting professor of  law from 2017 to 2018 at Stanford University where   she taught a course on artificial intelligence and  the novel legal policy and ethical issues posed   by such technology ladies and gentlemen please  join me in welcoming Chi Foley and Michelle Lee I'm thinking this one you're taking the other  one better Thank You Dennis so good evening   everyone and welcome I am Michelle Wie and  I have the pleasure of being your moderator   this evening I mean thank you to the Commonwealth  Club and also the committee of 100 for putting on   tonight's program titled artificial intelligence  and how it will change our future we have the good   fortune of having dr. high Fuli one of the most  respected AI experts and also author of the just   published book AI superpowers China Silicon Valley  and the New World Order we will be discussing the   contents of this book this evening but before  I begin allow me to set the table a little bit   artificial intelligence or AI the desire to  replicate human cognition has captured the   imagination of mankind since antiquity it  began with myths and stories followed later   on by books and movies of artificial beings  sometimes in the form of robots endowed with   intelligence or consciousness by their creators  the modern-day AI field traces its roots back to   1956 Dartmouth College at a workshop held there a  group of computer scientists began discussing the   possibility of building an electronic brain many  predicted that a machine as intelligent as a human   being could be created within a generation well  it's 2018 more than a generation since 1956 and   that hasn't quite yet happened that said within  the past several years there has been tremendous   excitement over the developments in artificial  intelligence and also the promise that it holds   and there's a sense that we are at the cusp of  some incredible consequential changes to our   lives brought about by artificial intelligence  in decades so with that dr. Kai Fuli I'd like   to ask you my first question what's the big deal  now versus say a couple of decades ago when you   were doing AI research at Carnegie Mellon and  I was a graduate student at the MIT AI lab and   AI was supposed to revolutionize the world what's  different now versus then okay Thank You Michelle   and thanks to a c100 and the Commonwealth Club my  pleasure to be here this time is really different   AI has had its ups and downs but this time there's  a dramatic difference that is the amount of data   that we're now generating is many orders of  magnitude more than we had not only compared   to the 50s let's just compare to my own PhD  thesis in 1988 I published the world's best   speech recognition system that paper was built  on the world's largest speech database and we   had a hundred megabytes of data it's like five  songs on your iPhone right but that's I mean we   forget but that's how much memory storage and  CPU have progressed over the last thirty years   so we literally I mean people cannot do good  speech research without at least one or two T   of data the best companies have 100 T so this is  1 million times more data and what happens is when   you have that much more data you also look at the  algorithms I use and you laugh at it and you say   well with that much data you can surely train a  much more complex system so more data encourage   people to develop new algorithms that more data  and better algorithms and that led to about 10   years ago a number of researchers invented what's  called deep learning and that is currently the   type of machine learning that can use the largest  amount of training data it can have thousands of   layers of neurons with connections trained on  hundred T of data and that combination really   made AI just work so much better but keep in mind  this is not what we originally talked about what   Michelle talked about fit in 1956 John McCarthy  and others dreamed and also what I dreamed when   I started AI that we would build another human  brain the AI today is very interesting it is   like a precocious super brain but it only does  one thing it's train a huge amounts of data but   each AI algorithm can only do one thing such  as play go decide whether to give you a long   target ads maximize users time on news fees and  things like that so we now have many companies   each of which doing one AI which is really super  smart and better than people but it has no where   the human capacity of common-sense reasoning  planning strategy and creativity let's look   into the future look through your crystal ball  and give us a sense of the changes that AI will   bring to our lives in your book kai foo you give  a very good example and you paint for us what my   grocery shopping experience will look like in  the future with AI you also discuss China's AI   city share with the audience what lies ahead ok  I I can generally divide a I progress based on   the steep learning technology in four waves and  those waves are overlapping some are already at   maturity others are just beginning I would say  the first wave is clearly the Internet wave that   is the Amazon Google Facebook accumulating so  much Internet data and what you may not realize   is you're labeling and tagging data that trains  the AI system right you never went in and tagged   the data but by clicking or not clicking buying or  not buying you are helping the AI make inferences   what you want and don't want and it's able to use  those inferences all the users so the internet   data is the by far the largest and the world's  leaders in Internet AI would be those companies   from the US that I mentioned plus China's Baidu  Alibaba and ten cents and more data makes more   powerful AI makes more money gets more users and  buys more machines and the cycle continues so   that will continue there will be more internet  AI companies that first gather a lot of usage   and then use that to help it make money the  second wave are the businesses I see many of   you may be owners of businesses or executives at  businesses today the companies that you have the   what you used to keep basically your cost center  right that Bank keeps its data Hospital insurance   company they're actually goldmine that can now  be used into your business process to make make   money and save money for example a bank kept all  the transactions of a customer and that used to   be just an archive but today that data can be  used to estimate the customers net worth what   kind of products might want to buy and also how  weather when that they apply for a loan whether to   approve or not whether a credit card transaction  might be fraudulent or not based on training from   all that data so suddenly a lot of data that was  archived and collected can be used in a big data   sense that's the second class a second wave the  third wave is when AI actually has eyes in years   that it can see and hear and actually understand  to some extent and the best examples in the u.s.   are Amazon echo that you can talk to in your  speech and speech interaction will be more and   more popular in more and more places and also  Amazon go is the other example a store in which   there are no people it's an autonomous store that  has cameras and sensors that watches what people   do so when you take things and put it in your a  skit or even your purse or pocket it recognizes   that and charges your account accordingly so that  kind of automation will hit many many areas I can   imagine there will be autonomous schools not the  teacher will not be there but AI can help great   papers exams figure out what each student  may need help just like Google understands   you from all the products it has in the future  the students will also have a profile just like   user has a Google profile and that profile will  know everything about the student what you might   need to improve what areas you didn't understand  what kind of teacher you prefer whether you need   tutoring how you did on your last exam we're  what are your strengths and weaknesses how   do you improve and in fact we just did a segment  with 60 minutes hopefully you'll see it in about   a month it will air and show the types of AI  that companies we invest in are injecting into   education so the third wave will be basically  using it's basically digitizing the physical   world and capturing all kinds of information and  using that to create applications that weren't   before possible then the fourth wave is when AI  becomes autonomous that means it can move it has   equivalent of our legs arms hands it can move  and manipulate and that will come in the form   of robots from industrial robots that basically  replace people on the assembly line to commercial   robots that might be security or receptionists  or a courier to also in in cases like restaurants   dishwashing robots in agriculture most of what's  done in agriculture like picking fruits or spreading insecticides or fertilizing those can  all be done with robots and eventually in our   home with education and other robots that will  take care of our our chores so each of these   four waves I think overtime will generate on  the order of 10% increase to the world's GDP   and it will also probably displace on the order  of 10% of human jobs so that's the positive and   negatives that come together 10% both ways  roughly speaking yes right let's step back   kai food to the Internet AI you talked about the  key to developing powerful internet AI algorithms   or solutions is the data doesn't that mean that  large companies with huge databases of data have   a competitive advantage and what about the small  start-up or the late entrant to the game how can   they compete in that space right I think the large  companies definitely have an advantage because of   the virtuous cycle that I described so in the  past monopolies are maintained because you have   a unique resource brand or customer affinity  or perhaps you have a great technology or you   have a big barrier of entry for your competitors  but today AI and data becomes another monopoly   maintenance tool I mean large a tech companies  don't want to talk about it but this is quite   powerful imagine a new company that has even the  resources capabilities technologies to do what   Google Amazon or Facebook does but they cannot  enter the market and beat the incumbent because   the incumbent has accumulated all that data that  has fed it's AI and build its product to be more   accurate and better monetized so new entrants  will have a hard time in particular with respect   to internet businesses although as Michelle said  the good news there's good news good news is the   Internet giant's can't cover everything there's  no real advantage for the Internet giant's to   rebuild a retail our schools or hospitals because  all of these things as I mentioned to you as AI   has eyes ears and hands legs and can move those  are brand new areas and commercial applications   or agriculture or the future of hospital or  autonomous vehicles even though Google happens   to be ahead in autonomous vehicles it has no real  edge in terms of launching those vehicles because   it doesn't have a uber like business units or  a General Motors like business unit so because   the internet business is very larger valuable  but the only a slice of the pie of the entire   economy so the startups that we fund and the  startups that are in Silicon Valley still have   great opportunities to target those areas that  are non non internets so it's not as terrible   as the first part sounds not too late so if AI  development were a baseball game are we at the   first inning the ninth inning or the fourth inning  I think we're in the first inning but sometimes   it feels like we're in the fifth inning because  there's too much hype and bubble and expectations   but I think we're just that there it begins so  it's not too late for those companies out there   no to jump in let's change subjects for a moment  now let's talk about the United States and China   and in particular the title of the book is AI  superpowers China Silicon Valley and the new   world order what is the state of AI technology  in the United States versus China and which of   these two superpowers is the greater superpower  now and in the future right I bet that's what's   on everybody's mind well you did title the book  as you did I did and I think president Trump's   speech in the UN exacerbates the yeah the issue  today I'm gonna give a two-part answer first I   will answer it and then I will describe why that's  not the right way to look at it today US is ahead   clearly in research the top schools in US are way  ahead of China and China has no chance catching   up in 15-20 years that's the state of research  however in the state of implementation I would   argue China has already roughly caught up with the  US and is probably going to leap ahead in the next   five years and and that has actually just really  happened in the last two years so this would sound   incredible to you that such research depth why  does it not lead to implementation advantage and   the reason is really twofold one is that Chinese  entrepreneurs and engineers are incredible they   may not be as deep in writing papers thinking  creative thoughts but they're very fast learning   new ideas finding every opportunity to make money  and building operational excellence and that's a   lot of more detail in the book later maybe I can  give an example and the other part is that China   has so much data and as I mentioned earlier AI  works better with more data and a lot better   with a lot more data so in the age of AI data is  the new oil and China is this new Saudi Arabia and   and that is a little bit of an unfair advantage  and those the two things combined together makes   China better in implementation but I do want to  refute myself for a moment because I really don't   think it is a zero-sum game this this competition  of who's winning who's losing it's not like in the   old days of war where there's one piece of land  to grab or one won a finite amount of oil to win   u.s. wants better lives for American people  China wants better life for Chinese people   these two are not at the zero-sum game both can  succeed and win without hurting the other also   also the to the two China and US pillars don't  even intersect I described in my book their   parallel universes with Chinese VC spun Chinese  companies build products for Chinese people and   America the same the two parallel universes don't  cross see all the media is asking me what Google   succeed in China will they be allowed in are  their censorship or regulation issues those   are not the issue the issue is at this point the  two parallel universes are are so completed that   companies can't cross over to the other parallel  universe so whether or not those regulations are   or are not important the parallel universes cannot  be crossed Alibaba cannot be successful in the US   and Facebook cannot be successful in China even if  laws and regulations were a non-issue user habits   are very much fixed similarly chai companies  would be the same and then going beyond the   US and China obviously US companies will lead  in all developed countries China will probably   catch up in other parts of the world including  Southeast Asia Islamic countries because those   are two regions u.s. largely ignores and of course  Africa so I think we'll end up actually with two   the world divided into two parts each using  different technologies so they don't really   compete in a zero-sum game the winning of one one  set of companies is not the loss of another so I   don't think this war metaphor is ideal interesting  let's turn to now the difference of opinions in   advances in AI and where it will take us there  are at least two camps one view is that AI will   bring us to utopia where human beings can not work  at all or perhaps work fewer hours and yet enjoy   a heightened standard of living there's another  view supported by some prominent businesspeople   and scientists who believe that AI presents  this existential threat to humanity Elon Musk   is quoted as saying AI presents the biggest risk  that our civilization faces and Stephen Hawking   disagrees which is it utopia dystopia and how  can we guide it in the direction of utopia okay   a very big question I think both camps have some  valid points but I do want the first dispel a few   incorrect some incorrect parts there are many  dystopian and some utopians who project that   singularity will happen probably you've heard  it right singularity super intelligence AI will   exceed humans and perhaps become uncontrollable  and that's the existential risk in I strongly   believe there's absolutely no basis for that  and as someone who's worked on the engineering   product aspect I know how limited our algorithms  are and there's no no reason to project that kind   of capability and what that may bring about  whether it's utopia or dystopia that's not   to say it can never happen Never Say Never right  you know a thousand years ago people wouldn't have   projected that we could put people on space or  have airplanes that fly but those inventions were   made but I'm just saying at this point we have  no engineering evidence to make those projections   so it's basically a waste of time however AI may  bring some exists tential risks in the following   ways suppose half the jobs are wiped out and  people fall into depression suppose many countries   have no AI and no technologies and all the jobs  are gone and they're so poor and they go into   this either a very rogue state that tragically  endanger others and that joblessness can cause   these things also carelessness and malicious  programming of AI can also hurt humanity for   example autonomous weapons one could imagine how  dangerous they could be autonomous vehicles let's   say they became pervasive and someone hacked  it it could be turned around not to miss human   but to hit humans and one country or one set of  terrorists can completely not just disabled and   almost destroy a country because all the vehicles  might be hacked so those kinds of issues I think   are real and we have to be cautious on how to  get there I am hopeful that this utopian world   is one day possible probably more for two three  generations from now we if we get over those   existential issues in the next one or one and  half generations I do think AI can eventually do   not only the routine tasks but more and I think  they will become more and more capable and they   will increase the profitability reduce the costs  and and basically inject a huge amount of value   into our global GDP and and that we will one day  look back and see this removal of routine jobs   not as a cruelty but as as something that frees us  from having to do routine tasks and really spend   time what we love to do with our loved ones on the  hobbies that we love inventing new things thinking   about the meaning of life so I am I'm certainly  more with the utopians but I think we've got to   solve a lot of the problems that emerge well thank  you on that and do you think that there should be   limits on the use of artificial intelligence you  mentioned the use of AI in terms of autonomous   weapons where are Thomas weapons make decisions  about targets and who to kill and maybe even in   the area of say criminal sentencing you have an AI  algorithm that determines how long of a sentence   and whether or not to parole an individual should  decisions about life and liberty be allowed to be   made by computers well I'm pretty sure we'll get  a unanimous unanimous answer from this audience   and I would tend to agree with that there are many  limits that we have to set but it's actually very   very tricky because I think I I think all of us  can agree we should seriously regulate autonomous   weapons we should be extremely cautious and  perhaps not in our lifetimes see machines   sentencing people however it's kind of a gray  area because you certainly don't want machine   to send in someone to death but what about the  machine to settle a small claims dispute or maybe   a traffic ticket and then you go into a slippery  slope where do you stop right so I think that's   not a black-and-white issue and also I want  to emphasize that different cultures will be a   little bit different in the ways they view this  I would say China I think is generally speaking   more utilitarian that is looking at the greater  good and the ultimate answer and not as focused   on individuals for example there is already a  sentencing assistance that's active in Chinese   courts AI is telling the judge these this is  the kind of sentencing that would probably be   appropriate the judge makes the decision but you  got to believe the tool is influencing the judge   right right and then in autonomous vehicles  people in the US especially Europe love to   debate the trolley problem right would you hit  three children or throw the owner over a cliff   right though those kinds of issues and some car  manufacturers already said are saying we protect   the driver or the owner so there's a business good  for business but it's going to be very challenging   debate but in China I think those will not be so  much debated my guess is the government will say   the greater good is for fewer lives to be lost  therefore if an AI is launched that is proven   to be better in general than people then let's  let the algorithm run and then with more data   it will get even better and eventually so many  lives will be saved then the early lives lost   may be worth it it's not additional lives lost  keep in mind however I think people in the United   States may have a lot of trouble with let's say  you know a hundred thousand peoples are killed   today we have a new AI algorithm that would only  through traffic accidents we have a new algorithm   that would only cause ninety thousand to be  killed but it's a different ninety thousand right of course it is so if you're one of the if  you're wondering the 90,000 you're gonna be angry   with all kinds of lawsuits and if you're one of  the hundred thousand you would never know because   you because the machines have taken over so in  a system where legal disputes and a fight for   individual liberty is so strong how do you ever  get the system going from a hundred thousand to   ninety thousand because in the future it will be  fifty thousand and ten thousand so I think we can   all agree ten thousand it's going to be worth it  but if you don't accept it at ninety thousand you   may never get to ten thousand that's a paradox for  for this society well it's it's interesting you   mentioned in your discussions how in sentencing  or in medicine there will be an AI algorithm and   maybe the human might interface with the machine  and provide the final decision the final authority   do you worry that if I'm the human being and  this AI computer has pored through all the data   and returns a result that this is the diagnosis or  this is the sentence that as a human being due to   liability concerns I'm not very likely to jump in  and circumvent the recommendation of the computer   for fear of liability for fear of you know the  computer looked at all the data surely it must   be right so in essence there is no intervention  that's right i but you're the expert on this issue   well so I guess you're AI systems just have to  be better is what you're saying well the thing   is they're going they're going to be eventually  better by far but initially maybe only better by   a little so the problem is merely transitory but  but getting over that hump is going to be tough   okay let's let's turn to another topic I got  a question here about jobs you had mentioned   10% loss of jobs you also mention that within 15  years in the United States forty to 50 percent of   the jobs in the United States will be automated by  some form of AI right what I said is technically   replaceable technically and economically replace  okay it may not happen due to you know the CEO   feeling oil tea to the employees or government  regulations or labor unions so it's hard for   me to estimate those things but technically  and economically yes so how do you see that   unfolding who will be the winners and losers in  terms of jobs job gains and job losses and what   if anything do you think the government should do  to help soften the blow right so I think the job   loss will be the routine jobs so that includes  some examples I've given assembly line workers   fruit pickers dishwashers but also white-collar  jobs for example customer service representatives   and telemarketing telesales and and also expanding  to some pretty high-end jobs like radiologists but   but more detail in the book but basically these  are the routing aspects of those jobs will be   replaced those are the ones lost and in aggregate  it's probably not hard to convince yourselves   that's at least 50% of human jobs are scripted  repetitions of similar things and especially those   quantitatively oriented very easily replaced what  jobs will be gained well clearly AI will create a   bunch of jobs and we probably don't even know what  those jobs are and that is probably the best the   best news and that's what a lot of people would  argue every technology revolution always ends up   creating more jobs than destroying them but I do  think there's a difference in this case because AI   is coming too fast and also it's run with dual  engine US and China and also the replacement   aren't always one-to-one but sometimes through  disruptions by inventing a robotic restaurants   McDonald's will lose jobs right by having rope by  by having Facebook newsfeed traditional media will   lose editors not because they hire the robot  to do the editor but because the traditional   industry got squeezed so so I think the losses  will be substantial and then the AI creation   of jobs we have no way of telling how many there  will be I actually believe in human wisdom given   enough time maybe three generations I think  I will create more jobs than it destroys but   AI will happen so fast that this may not be the  case so what what can be done I would call on a   different dimension if you want to make creative  jobs AI jobs you know robot repair AI programmer   data scientists it's very hard to train for them  educate for them not to mention taking someone in   the routine job making them a data scientist  that just isn't practical but what what may   be practical is another dimension which is jobs  that require human touch human compassion human   empathy jobs like nurses elderly care tourist  guide concierge jobs like the future of doctors   the nurse nannies so these are the types of  jobs that can be large enough in number and   also require modest amount of retraining so that  the Reese killing could happen taken things as   an example elderly care I understand there are a  million positions open in the US but they're not   filled because the pay is too low so what can  government's do to give a very simple example   a lot of people talk about universal basic income  which says basically tax the ultra-rich and give   money to everybody I would argue that's a very  terrible idea it fixes the income redistribution   problem but it doesn't fix the individuals loss of  meaning problem when you lose your job you don't   just want to get Social Welfare and treat you as  the so called useless class that some author has   chosen to use which i think is terrible but you  really need to I think people today need to find   meaning in having a job and and also why don't  someone who lost a job customer service job why   don't they want to be an elderly care person it's  because the job doesn't pay enough so some policy   that that basically adjusts the elderly care pay  would be a much better way to spend the tax from   the ultra-rich than to give money to everybody now  what is the policy again you're the expert I don't   know what policy that should be but there must  be some way to inject money into care reeducation   social value it's kind of questioning the social  contract we have today I mean our social contract   is that we are paid based on the economic value  and economic value only that we contribute to the   society and what this is saying is maybe there  are other elements to consider maybe we can be   paid based on some combination of economic and  social benefit that we bring about so technical   engineering and social engineering needed yes both  alright let's turn to privacy okay our privacy is   increasingly under siege these days there seems to  be a trade-off on the one hand between privacy and   the power of an AI system the more comprehensive  the data collected the more powerful and the   more intelligent the AI system but also the  more exposed our privacy so if you look at a   spectrum we've got Europe on the one side with  the GDP our general data protection regulations   providing some of the strongest protections in the  world and I think it's fair to say that China is   at the other end with the u.s. somewhere in  between possibly closer to China but having   a lot of angst as to where is the proper place  to play along that spectrum do you think China's   less restrictive privacy environment gives it a  competitive advantage in terms of the power of its   systems yeah I think this is sometimes a little  bit misunderstood what you say is not incorrect   but it's a little bit misunderstood because every  time someone says Chinese people don't care about   privacy they get blasted in Chinese social media  so I don't dare make that statement anymore so   certainly the Chinese people think they care a  lot about privacy I think the issue however I   think they have not been educated in this kind of  environment to know exactly what rights to fight   for so I think over time the increasing privacy  will happen in China is just that that hasn't   historically been something people have been aware  of so III think it is for the user standpoint it   is a trade-off between privacy and security and  convenience so by giving up some privacy you let   yourself be monitored and protected against you  know bad people is that something you would do   on the other hand there's privacy for convenience  by letting the big companies have more of our data   your user experience is better you save money  is that worth it so I think Europe's GD P R is   an attempt to give people back a little bit of  control in those two dials and I think GD P R is   is quite clumsy in the way it's designed now but  I applaud Europe for doing it because at least   it shows something should be done China actually  has done something and what's done is basically   it preventing a company who's collected a lot of  data from selling it or giving it to someone else   with extreme penalties similar to GDP are also  extreme penalties and so if you look at that well   the what's the what's the biggest privacy issue  that's upsetting Americans they must have been   the Facebook Cambridge analytical issue right  and that was the case of one entity giving or   giving data to another entity without the consent  of the individuals so actually that case would   have been either prevented in China or punished  very severely with imprisonment imprisonment so   I think each country is trying something different  and I I do believe so finally the last point is I   do agree that because more day that makes better  AI if privacy gives users the rights to not give   the data then the AI will not be as good so a  society or government has to measure the impact   of the economic impact of creating very strict  laws that makes the AI work less well let me take   a question here do you foresee a future where the  United States and China are collaborating on AI   advancement maybe that's already occurring and if  so how it is occurring at academic levels you know   when I did my PhD thesis on speech recognition  that data was all shared by a National Institute   of Standards and technology's and Carnegie Mellon  decided to and I decided to open-source our code   and that kind of began AI becoming a very open  and sharing community people are publishing in   real-time and also sharing data in real time and  often sharing source code so when all that sharing   happens it helps move the whole field forward  so so someone who published a great result may   be challenged and then knowing that he or she is  likely to be more honest ai is a science where   the same day that going in should yield the same  results unlike many of the other natural sciences   so these natural aspects made the whole world  very much connected if you go visit one of the   big conferences like cvpr or nips or ICAST you  will see every country researchers from every   country very openly sharing discussing talking  there are no national boundaries no competition   of superpowers so that sharing is going on whether  or not the other competition goes on so that I see   I also see I also hope the governments will work  together on things like restricting autonomous   weapons on things like agreeing on general  principles about security and privacy maybe   not agreeing on the final outcome but general  principles and I also hope come countries will   share best practices on how to adjust our social  contract because that is such a hard task that I   think there are great things various countries  do like in for example in Korea the gifted   and talented education may be something to be  shared in Japan and Switzerland the the types   of craftsmanship and that being a great career  path that may be something that could be shared   in Canada and the Netherlands the volunteerism  maybe that's something that could be shared so   I do hope more countries will share and in terms  of sharing best practices it's not limited in   superpowers every country has some good good  ideas to share so you can foresee international   effort to define for example AI values AI FX and  so forth I hope so there are some institutions   like partnership for AI is trying to do that you  and may be doing something but I think in practice   even without the current push to nationalism each  country I think is going to be more independent   at how it makes decisions so the cross country  bodies will have really only limited powers okay   I have an interesting question for you Caillou  how will a eye affect the fields involved with   human sensories such as taste food and smell for  example wines will a computer be able to select   your favorite wine from California where where  Avicii so we there may be a start up for you yet   yeah okay I imagine some sensor can probably do  something sure but I just don't know how far it   can get interesting okay fair enough you said that  one company like Google won't be able to acquire   all data of the society but how about a nation  like China this comes up a lot I think I think   there is a belief that Chinese government collects  all the data and uses it it is absolutely untrue   it's impossible to collect and it's and actually  no government at this point AI expertise is   definitely in the private sector so that the task  of doing that is just too large so I think that is   perhaps based on paranoia and fear and concerns  about about China rather than based on anything   that's realistically likely to happen fair enough  in terms of artificial general intelligence do   we need a paradigm shift from deep neural nets  to achieve a true general intelligence I think   there's general agreement that we would that deep  learning is very well understood and engineered   and ready for applications and it's going to  continue to build us those wonderful narrow   prodigies and that it may be a part of the future  AGI but it's certainly not the only part but again   I want to say that AGI may be impossible so AGI  I know that a lot of researchers are optimistic   that AGI is possible and many have projected 10  or 15 years on the realization of AGI but when   you drill down they can't give you a roadmap so  AGI meaning the ability of a computer or machine   to perform all the functions that a human being  that's right can perform in terms of intelligence   that's right and and sometimes they forget that  all the things we perform includes our love our   soul our emotions all right and and no one has  any idea how to do that and even if you throw that   away and just say build an analytical human but  then there's strategy and planning and creativity   and inventions we still don't know how to do that  there are some very basic things like common sense   that AI doesn't even have no it really doesn't has  no no clue so I would say we're farther away than   10 or 15 years I think a lot of researchers tend  to be just a little bit optimistic okay why did   you decide to write this book I decided to write  it because I found many of the other books to be   inaccurate and misleading not completely each book  of course has its merits but I found some books   have incorrect technical understanding especially  the ones that project singularity AGI and super   intelligence I feel that's dangerous to believe  I also think some books inject a strong sense of   negativity by calling the future of 90 percent of  our people to be in useless class I think it's an   unfair insult it's it's insulting to our humanity  and equality and I think it injects a negativity   that can become self-fulfilling prophecy and also  those were my thoughts when that approach to my   book agent and published or and they approached  me but they said that's not gonna sell a book he   says he says you know historians and philosophers  write better than you do and and that's probably   true but but he said here's what you can do if  you can put China in the book that's gonna sell so so I thought there was a color there was  an angle because he said you're the only China   expert no one can write like you you're AIG expert  China expert put them in the book does it fit so   so if the book reads like two books I deeply  apologize okay fair enough you write a lot in   your book Chi foo about how China had a Sputnik  moment and how the government took very concerted   steps to develop and promote the advancement  of technological innovation and in particular   artificial intelligence share with the audience  a little bit about that and the question the   reason I'm asking that question is this particular  audience member says how can europe reach the same   level of AI knowledge and economy like china and  us so those two questions might be related okay so   the first question is on the the role of Chinese  government's first I want to say that a I pretty   much took off in the private sector the judge the  government noticed it about two years ago and that   Sputnik moment was when alphago beat léa Seydoux  the korean expert and that shocked chinese because   all the AI experts including me have been saying  in a goal is a game that we don't expect machines   to be people for another 15 years and suddenly it  happened also go is an ancient Chinese game that   Chinese invented and is believed to be more than  intelligence but there's Zen and you know just   and you know human insight brilliance philosophy  it's like it's one of the things that that that   were represented the Chinese art and and and and  all that and suddenly this US UK company jumped   in and beat the best and while most of America  didn't even see those matches because you most   of you don't play go but in China 100 million  people watched and that really woke people up a   little bit like the Sputnik moment did for America  when USSR launched it so so the Chinese government   woke up and made a lot of very decisive and I  think smart steps so the point I want to make   is that it really began in the private sector  government is secondary however having said   that the good the secondary government aspects  are very very strong and very helpful and very   effective and very worthy of studying by US  and Europe now the government support for AI   is not what you read in media especially the  anti-china media that basically says okay the   Chinese government blocks foreign companies from  entering injects money into Chinese companies so   that with the lower cost they can beat American  companies I'm not saying those things have not   been true in the past in certain cases but  that is not how Chinese government basically   super scaled AI here's what Chinese government  did there are three things that they did that   were very instructive and maybe a little hard to  replicate but but you need to know what they are   I think first is a is infrastructure building  and and basically let the private enterprise   do the investment entrepreneurship products  customers and all that but there are things   that private sector cannot do such as building  roads and water roads for AI these are autonomous   vehicle roads these are new highways that have  sensors to help cars be be safer and building   a new city China is building a new city called  Shaolin it is the size of Chicago and it will   have autonomous vehicle built in it that Chinese  cities are also very entrepreneurial a city called   pseudo is building a 10-kilometer piece of land  where what one layer is purely autonomous vehicle   another is human vehicle and another example is  where two layers of downtown are being built the   top layer is for people bicycles and pets and  the bottom is for vehicles including autonomous   vehicle but with controlled lighting so that the  Tesla acts accidents cannot happen it was it was   really a lightening in color and things like that  that confused the car into thinking the truck was   sky right so all these is it's a noun trip renewal  environment where huge amounts of money went into   infrastructure now some of you might be saying  hey that's unfair that's like indirectly helping   but when President Eisenhower built the u.s.  interstate highway it had the same same effect   right for a different age so I think that's what  governments should do is infrastructure building   leave everything to private else to private  that's number one the number two is a general   Technol utilitarian policy this may be hard to  emulate and that's and may not be right but it   for every case techne utilitarian policy in China  means when the new technology comes around just   to give it a shot don't regulate and block it and  watch it and if it's something it goes wrong let's   regulate it if it goes really terribly let's stop  it that whereas us in Europe will tend to want   to debate all the other all the possible outcomes  and regulate it before it comes out an example is   Chinese mobile payment in a short period of three  years china now has no cash and no credit cards I   say no I mean very small percentage I don't carry  cash or credit cards when I go out it's all mobile   pay and mobile pay today is 700 million people  paying who can pay each other instantly with   no Commission not nope and you know as you know  credit card charges two or three percent at the   merchant side this is no Commission at either  the merchant or the user side there is a 1.1   percent transaction fee for taking money out of  your bank but that's a very much smaller amount   and and it's micro payment capable so you can give  somebody 15 cents so that came about because you   know Tenzin Alibaba did some mobile payment and  the government obviously had concerns obviously   credit card companies and banks don't like it but  the government basis is it let's just go ahead and   see if it goes well can they do it securely and  they did now it's taken over by storm there were   other things that worked less well for example  p2p lending that caused certain amount of fraud   and the government now is regulating it after  letting it go out and [ __ ] the currency that   led to a huge amount of money laundering and the  government didn't know what to do to regulate it   so they stopped it altogether so it's not that  China is letting all technology go loose they're   just letting it go out a little bit try it and  then regulate it only on as needed as needed   basis so these are the types of things that the  Chinese government does and and there's one other   thing I should say it's setting and that relates  to the form of government it's basically setting   a tone so when the central government issues a  plan actually you've probably all read this or   heard about this manufacturing 2025 there's also  an AI to thousands are very ambitious and target   driven but actually the central government doesn't  have all the say and the budget it's about setting   a tone because it is run top-down as a government  when they set a tone when the AI document came   out our AI investment we have an AI investment  building AI for banking they were having trouble   opening doors to banks but when the AI document  came out the bank said we better do AI so suddenly   we're theory we're able to sell the software into  the banks so the setting at owned by the central   government is important but it is not like what  you'd read in the media it is not like the central   government has six hundred billion dollars and  is giving it to Chinese companies our companies   basically got nearly nothing from the government  I mean we have we currently have five unicorns   valued at about twenty three billion dollars that  we invested in and none of them were funded by the   government all of them were privately funded that  generated revenues and rose to to their success by   themselves maybe after success government would  give them some free rent or something but it's   inconsequential interesting so AI will effect  the tax revenue since robots are not currently   taxed and currently human beings who provide labor  earn income which is taxed in the form of income   tax Bill Gates says that we should tax the robots  others say that attacks on automation is a tax on   innovation and progress and we didn't do so in  the industrial revolution so why should we do   so now your thoughts on how our governments can  keep the lights on and provide the services that   they need if there is a decrease in income tax a  major source of revenue for the government right I   can't believe that government former government  official keeps asking he policy questions and   I'm an engineer and investor but nevertheless I  will try I I certainly think redistribution is   absolutely needed if you just look at even without  a I for the last 15 or 20 years the American   disparity between the top 1% and the bottom 50%  have not only crossed each other but continued   to increase the gap is widening the wealth held  by the top 1% and the wealth held by the bottom   50% and that is very much a result of the ultra  rich from tech in the past maybe internet mobile   and so on in the future ái companies and in the  AI is going to be double jeopardy because job   displacements will will make the bottom 50 even  more challenged and the routine jobs are largely   correlated with the bottom 50% so that rapidly  increasingly widening gap must be closed and I   wish we we had some magical formula other than tax  to close them but but had a dire time like this I   think we got it use approaches that are proven  for centuries to work and taxes the only one   but rather than taxing specifically AI companies  or robot companies I don't think it's possible   to do right I'm a board member of Foxconn where  the robots are and the Foxconn margins are this   much compared to apples margins so please tax  Apple not Foxconn you know and and I think also   you know targeting AI companies who's at our AI  company I think in five AI is like the Internet   today there are a bunch of AR companies I think  in five years there will be no more air companies   every company will have it's only I it's just  like internet so I think I don't think you can   do it by classes and and I think you know again  let's not invent something when we really need to   do this income redistribution and taxes based on  I think taxing the ultra-rich is the easiest way   I mean occasionally there will be a collateral  damage of a very rich person who had no AI but   will take their money never anyway so I see we've  got the signal for the last question because we're   running down to the to the last few minutes  here and as moderator I'm going to exercise my   moderator prerogative and ask you the following  question on the very last page the very last   paragraph of your book Caillou you write luckily  as human beings we possess free will to choose   our own goals that AI still lacks we can choose to  come together working across class boundaries and   national borders to write our own ending to the  AI story let us choose to let machines be machines   and let humans be humans let us choose to simply  use our machines and more importantly to love   one another any further thoughts you can wear the  ending well they have to read all of it in order   to get how you doing how you get kidding yes I  I think I feels very strongly about that because   I think as you've heard I I really don't like  some of the books that are so dystopian and and   I believe a thought leader has a responsibility of  leading people towards good outcomes because what   thought leaders say may become self-fulfilling  prophecy so if if someone writes a book and says   most of you are useless class is hopeless a is  gonna take over what's going to happen people   are going to work less they'll be depressed and  even though this may not be the end this will be   the end because peak if people believe that kind  of saying and and if we if we if we believed and   in in my book I clearly point out there are  ways in which if that the a very interesting   side effect or maybe the actual outcome is that  our maker or our collective consciousness if you   don't believe in it religious has in mind we have  been trapped in this industrial revolution driven   work equals meaning of life and we're just  repetitively routinely doing work thinking   that if I work harder even if it's repetitive  work it will give my people give meaning to my   life and give my better living to my family while  that is understandable that cannot be why humans   were placed on this earth and I think if there  were a maker he would be disappointed looking at   humanity that after you know thousands of years  all we do is repeat routine work and thinking   that's the meaning of our lives and he has not  been able to make us wiser so he brought in AI so AI takes away all the routine jobs okay now  you humans can you please figure out why you exist I hope you enjoyed this evening's program brought  to you by the committee of 100 in the Commonwealth   Club Silicon Valley again we would like to thank  CAI Fuli Arthur of AI superpowers China Silicon   Valley and the New World Order the Honorable  Michelle Lee former Undersecretary of Commerce   and director of the United States Patent and  Trade Office our audience here in Silicon   Valley and those of you who joining us on the  radio and web and now this meeting is adjourned

Channel: Commonwealth Club of California

Views: 12,735

Rating: 4.8534031 out of 5

Keywords: KAI FU LEE, AI SUPERPOWERS, Google China, Microsoft, SGI, Apple, Commonwealth Club, San Francisco

Id: clbSkL3BiRg

Length: 64min 3sec (3843 seconds)

Published: Mon Oct 08 2018