In this episode Byron and Dr. Kai-Fu Lee talk about the potential of AI to disrupt job markets, the comparison of AI research and implementation in the U.S. and China, as well as other facets of Dr. Lee's book "AI Superpowers".
Dr. Kai-Fu Lee, previously president of Google China, is now the CEO of Sinovation Ventures.
Byron Reese: This is Voices in AI, brought to you by GigaOm. I'm Byron Reese. Today I am so excited… my guest is Dr. Kai-Fu Lee. He is, of course, an AI expert. He is the CEO of Sinovation Ventures. He is the former President of Google China. And he is the author of a fantastic new book called "AI Superpowers." Welcome to the show, Dr. Lee.
Kai-Fu Lee: Thank you Byron.
I love to begin by saying, AI is one of those things that can mean so many things. And so, for the purpose of this conversation, what are we talking about when we talk about AI?
We're talking about the advances in machine learning… in particular Deep Learning and related technologies as it applies to artificial narrow intelligence, with a lot of opportunities for implementation, application and value extraction. We're not talking about artificial general intelligence, which I think is still a long way out.
So, confining ourselves to narrow intelligence, if someone were to ask you worldwide, not even getting into all the political issues, what is the state of the art right now? How would you describe where we are as a planet with narrow artificial intelligence?
I think we're at the point of readiness for application. I think the greatest opportunity is application of what's already known. If we look around us, we see very few of the companies, enterprises and industries using AI when they all really should be. Internet companies use AI a lot, but it's really just beginning to enter financial, manufacturing, retail, hospitals, healthcare, schools, education and so on. It should impact everything, and it has not.
So, I think what's been invented and how it gets applied/implemented/monetized… value creation, that is a very clear 100% certain opportunity we should embrace. Now, there can be more innovations, inventions, breakthroughs… but even without those I think we've got so much on our hands that's not yet been fully valued and implemented into industry.
I tweeted once – and it's a really arbitrary tweet… I said that if all R&D related to AI just stopped today, it might take us 10 or 15 years just to fully implement what we know everywhere it can be applied. Again, that can't ever be proven one way or the other, but does that feel like a reasonably accurate statement to you?
Absolutely. It sounds almost exactly like what I would say. So, I guess, I fully agree.
If I went back 25 years to 1993, that's when the Mosaic browser came out, and in the 25 years that we've had the Internet – which, let's be clear, that's a technology that only allows computers to talk to each other. They don't think or anything, they just talk to each other – that's created something on the order of $25 trillion in wealth, a million new businesses. If you were to say: Okay, what will 25 years of AI get us? With that data point as a reference, is that going to be massively bigger than what the Internet gave us? Or is it like another Internet or two? Or do you have any gut level of how big of a thing we're talking about?
I do. I think it'll be a multiple of what Internet has brought us. In fact, both McKinsey [Global Institute] and PwC have done studies that in about 12 years AI will bring about $12-17 trillion in value. So, if we add that to 25 years, it ought to be a few times [that] of the Internet if we believe in the PwC and McKinsey studies. And the gut feel is right because not everyone has to embrace the Internet. I know almost everyone we know does, but the whole world is not fully penetrated. But AI truly could go into the tech businesses as well as very traditional businesses – including agriculture, manufacturing and so on.
I think the reach is bigger in terms of industries. But also I think Internet was a great thing. So I think I'll say a few times [that] of the internet, based on the technologies that are known and natural extensions thereof. If there are bigger breakthroughs, let's say if AGI actually gets invented, I think it'll be much much larger than that, but I don't place high odds on that.
Machine learning is really a simple idea. It says, let's take a bunch of data about the past, let's study it and make projections into the future. When you say it like that, it sure doesn't sound like such a big deal, but of course it is kind of a big deal. But where do you think that methodology really works well and where does it not? And I asked the question because the central assumption behind machine learning is that the future is like the past. That works for identifying a cat, where a cat tomorrow looks like a cat today. But it may not work for other kinds of problems. It may not, for instance, be able to predict what I'm going to say next or it may not be able to come up with anything that's creative, or anything like that. So where in your mind does that simple idea of machine learning work well and where does it not really work well?
Well, it has a number of constraints. It works well when it's in a single domain and a large amount of data can be gathered, and that [data] can also be tagged and labeled with the proper prediction, decision or outcome… and the new data that emerges can fit into the same categories of data that we've seen. So, that would basically say that a lot of the quantitative work that was fabricated by humanity – banking, insurance, stock market, investment – those numeric tasks obviously do fit. Internet obviously doesn't.
I also think a lot of the jobs that are done routinely to process numbers or data or text also fits. As it extends into robotics with the abilities of computer vision, computer speech, movement and manipulation, I think a lot of the routine jobs that are blue collar in nature will also fit. So it fits a pretty large number of things that happen in the world. Probably more than half the tasks that we do are quantitative-based, objective decision-based and based on relatively routine, repeatable, predictable kinds of outcomes and forecasts.
The things that don't fit would be things that are more abstract – things like creativity, strategy, concepts, cross-domain thinking, common sense, intuition, self-awareness, emotion… Those are the things that don't fit.
Well, I really would love to explore your 50% number, because I confess to you I see things a little differently than that. So let's try to get at the bottom of that. I'll set the problem up by saying: There was a fantastic study that came out of Oxford a few years ago – Frey and Osborne [The Future of Employment: How Susceptible are Jobs to Computerisation by Carl Benedikt Frey and Michael A. Osborne, Sept. 17, 2013]. They came to a conclusion, which I think was routinely mis-represented, but it was reported as 47% of jobs can be automated away.
If you read what they wrote in that report, they go out of their way to say: We are not making any predictions about how many jobs are going to be lost. What they actually said was, 47% of things people do in their jobs can be automated. That's not particularly interesting, I mean that's not particularly surprising. My father sold insurance for 30 years and over the course of his 30-year career a whole lot more than 47% of what he did got automated. But he still had a job.
The OECB did a study [that asked]: What percentage of actual jobs can be eliminated by this technology? They came up with 9%. I have a big survey on my website where I ask people to score jobs on these different markers, and I have a hard time finding any jobs… I don't want to say none, but I find so few… because when you start going through them, and you ask: Okay, what about a plumber? Well, a plumber's got to come into your house and look around… or an electrician… or you come up with any job that requires motion or mobility. You come up with any job that requires empathy, any job that requires creativity, any job two people would do differently from each other. The range of jobs like this one simple technology can do seems really small to me, so I would love to give you the microphone and have you really kind of teach us what you're thinking about with that 50% number.
Certainly. I think at the roots it comes back to the Oxford study. Other people have challenged the study by saying it's not the job itself but the tasks [that face] displacement. And the tasks also come to around 50%, if not higher. That's in the McKinsey study. So the question now is: If the tasks can be automated, would they lead to an equivalent number of job displacements? In my book I'm also cautious to say that I'm talking about technical capability as opposed to actual employment because there are actually a lot of things that depend not only the items you mentioned but also things like the employer relationship, employee loyalty, pension, labor unions, stock regulations, Social Security and so on and so forth.
But let's just come to how many jobs might be impacted. I would still stay with the higher number for the following set of reasons: First, if you displace half of the tasks in the job, arguably the pool of workers ought to be shrunk so that the people would do the jobs that machines can't do. So…
But let me let me put a pin in that right now. That historically hasn't been the case, has it. I mean like when we came out with the ATM machine, which did a lot of what a teller did, you would assume there would need to be fewer tellers. There actually need to be more tellers because lowering the cost of that to zero opened more bank branches that need more tellers.
Google Translate can translate a document as well as a human. The need for translators is actually exploding because when you lower the cost of translating an email to zero you still need a real translator to do a contract, to do a face-to-face meeting, to do the rest. So isn't it the case that throughout history when technology lowers the cost of some aspect of a job to zero, that increases demand for that and increases the number of people who need to do that?
Well, for the case of Translate, I think is a temporary phenomenon because technology is currently imperfect. I would predict in five years technologies would be so good that the number of people would come down again. The same applies for truckers. There are more truckers needed now because of the booming e-commerce. But when autonomous vehicles take over that will come down. Tellers is an interesting one. I would bet tellers are largely gone in ten years because while there are more branches, I think people are now seeing [that] banks are extracting a reasonable fee, and the services they provide are better provided by fully automated systems.
But let me give you an exact example. We fund a company that does loans by apps. This is an app you download and instantly get about a thousand dollars directly to your phone. This is in China with mobile payment. The app does not do the job of the teller. It’s not a one-to-one displacement. I'm sorry, in this case it’s more like a loan officer. But to the extent that such apps become pervasive, the industry gets displaced and there would be no more loan officers left to be displaced.
Another example is that we're investing in autonomous fast food restaurants. While McDonald's is not hiring a robot to displace each cashier or hamburger flipper, these new autonomous fast food [restaurants] will probably offer food at one third to one half the price. To the extent that these autonomous fast food restaurants take over, then the jobs of McDonald's and Kentucky Fried Chicken will still shrink.
I think my first argument was that when a large number of tasks are replaced within the pool, the number of people needed to do the job will shrink… especially because the percentage will go up over time.
If you take a paralegal, maybe it's 10% doable by AI today, maybe 50% in 10 years, maybe 80% in 50 years… and as AI takes more and more it would shrink the pool of workers needed for that job.
The second point is that disruption is not a one-to-one displacement but a whole industry with a whole brand new way of doing things causing all of the old jobs to go away.
The third is that AI is specialized at doing routine jobs, and we actually have a large number of routine jobs in the economy. A plumber, I would argue, is an exception because it requires mobility, human interaction and dealing with all kinds of different types of plumbing data, whereas more of the blue collar workers are in a single place – like an assembly line worker, dishwasher, fruit picker that requires a lot less of the other skills. And I think we should do an inventory: Are there more plumbers or are there more assembly line worker types of people? I think we’ll find the latter will be larger.
The last reason I will give is that this technology revolution is a dual engine… that is, both the U.S. and China are making amazing progress. Whereas, before, the whole world depended on the U.S. coming up with solutions for Internet and PC and software and mobile... Now there are two engines, so there should be faster progress.
But having said all that, I'm not predicting a 50% unemployment. I'm just saying 50% of the tasks will clearly be doable for much less money, and many people have jobs that a large percentage [of which] would be displaced. That plus the other reasons ought to lead to a larger percentage of job displacement than 9%. Will it be 40 or 50%? I think, you know, time will tell. There are so many factors. I'm not an economist, so I can’t go there.
So, you of course make a compelling case. But just to look at some of what you said… The number of translators, I would argue, even as it gets better it will continue to go up because if you take the premise that the technology is going to enable international commerce across language, then all of a sudden you need localization. You need people who are actually in that country fulltime. You need people who are conversant in the customs of that country. I mean, there are so many things that happen that need translators suddenly.
If you look at the Frey and Osborne study of the kinds of jobs that were listed, and even the jobs that you just listed… you can't possibly think we're going to build a machine to be – for instance – a dishwasher… which you mentioned. We are so far away from something that could do something as simple as picking up dishes and holding them and then putting them in a… I mean we are so far away from that.
Frey and Osborne had things like waitress, short order cook and pharmacy assistant… that was one of their 98% jobs. Like we're not going to need pharmacy assistants? No. I think what we would say is that we're going to do different things but that doesn't mean pharmacies won't have assistants.
I think Frey and Osborne, if you look in the detail, some of the higher and lower numbers are certainly debatable. I find examples I disagree with also. But actually, technology is progressing very fast. I gave you the example of a loan officer, which would be thought of generally as sort of a higher level job in society. We can go as high as a radiologist, right. I don't think Frey and Osborne, at the time they wrote the paper, deep learning hadn't come out. So 47% is probably an underestimate …because at that time they were probably thinking: “Radiologists. There's no way to displace them.” I remember two years ago I talked about radiology as something machines would do better than humans one day in the not too distant future, and many people in the medical domain disagreed with me. But this year at the Global Radiology Conference, one of its topics was when would their jobs be gone.
So this progression is a lot faster... even in the example you gave – dishwasher – we're actually an investor in a company called Dish Kraft that is building a robotic dishwasher. It's not doing… I think you're talking about building nimble fingers that can handle dishes the way a human does, but that isn't the way… to come out with a giant machine. The goal is to get the dishes clean, not to create finger-like manipulation...
So let's put a pin in that and say some number of jobs are going to vanish. Does that, in your mind, whether it's 10% or 50%… Does that in your mind imply we will have some increase in unemployment?
I think it doesn't have to be. I'm actually an optimist, but my point in my book is that: AI will not create enough jobs to take over the jobs that may be displaced. I don't think it can possibly be 10%. I would put it maybe at 20 to 45% in the next 15 years… or something like that. That is a larger rate than any technology has displaced in the past. AI will probably over time create an equal number or larger number of jobs if we give it enough time. But because this 15 years is very very short… you know, electricity, the industrial revolution, they took much longer.
I'm with most people that technology ultimately will create jobs that we haven't seen before and everything will balance out. But that's only if the technology revolution takes 25 years as Internet did or hundreds of years as electrical and industrial revolution and steam engines. So it's a lot faster. I also think, even if as I predict AI will not create enough jobs, then it's up to us humans to create those jobs. And I think those jobs actually do exist.
So I am an optimist. A lot of jobs – especially service, empathy, compassion-related jobs – currently are vacant in the society because they don't pay enough, because their social status is not high enough. And I think we need to consciously shift that… and that if we do, I do agree that eventually everything will be OK.
So let's look at the Internet as an example of this. If you had gone back 25 years… well, you were studying artificial intelligence before that. So we're going to use you as our own incredibly smart forward looking person. It's 1993. I go back to Dr. Lee and I say: Dr. Lee, in 25 years this technology is going to have billions of users. What jobs do you think will be lost? And you may have said: Well, stockbrokers and travel agents and probably newspaper jobs lost because people are reading online, and postal jobs lost because of email, and there will probably be less yellow pages because people advertise online, and there will be less retail jobs because people will buy online… and you would've been right about everything.
But of course what nobody could have seen was eBay, Etsy, Baidu, Amazon, Google, Airbnb and a million new companies… literally trillions and trillions of dollars worth of wealth. So, do you believe that as the Internet has progressed over the last 25 years and has eliminated certain jobs, has it created jobs at a slower pace or has it created jobs in excess of what it was destroying… as it did it in real time?
I actually think Internet has created an equal or maybe even larger number of jobs than it has displaced. I think the examples you gave are great. Those are exactly the categories that came down. But then there are the website designers, the people who are involved selling things on eBay, things that people started doing with e-commerce and so on and so forth. I think it probably created more jobs.
The difference between Internet and AI are twofold. The major difference is AI is basically designed to displace routine jobs at the bottom of corporate hierarchy. Internet is not. Internet is a connection tool, which will create jobs and destroy jobs as it kind of changes industry’s outlook. People aren't using Internet technologies to build entry-level routine jobs, but with AI, people are. We're investing in companies that are essentially doing customer service, telemarketing, dishwashing, fruit picking, loan officers, fast food restaurant workers, cashiers... So we're actually seeing machines or software that are exactly designed to displace jobs – which was what prompted me to write the book. I think that is the one big difference. It's kind of a natural technology for job displacement.
The other thing is that AI will come about faster because a lot of this job displacement won't require any infrastructure building. In order for eBay to take a bunch of sales off of Walmart or whatever, it takes time for people to shift their habits. It takes time for the mail services operation to improve. With AI, basically just plug it into the Cloud and the job is done. AI displacement and AI revolution will be faster.
So, basically [there are] two differences. One is larger numbers of directly targeted routine jobs. The second is faster implementation.
What I find interesting, however, is that… and I'm just talking about the United States at this point because it's the only thing I know the numbers off the top of my head… in the last 250 years in this country unemployment has been between 5 and 10%. It never has varied other than the Great Depression, which was kind of a special case unrelated to technology.
So, you have 5 to 10% unemployment for 250 years. Meanwhile, I've taken a lot of time to try to figure out the half-life of a job, and I believe the half-life of a job is about 50 years. I believe every half century half of all jobs vanish. It might be a little less; it might be 45 years. But for 250 years half the jobs go away every 50 years. And what's interesting is that we've maintained full employment.
I know you're talking about a faster pace. I understand that. But we've maintained full employment for 250 years losing half the jobs all the time. And, I point out, we've maintained full employment and rising wages against that backdrop. Now, to me what's interesting is that when a new technology comes out… and I'm not sure that AI is actually being adopted faster… and let me justify that. The United States went from generating 5% of its power with steam to 95% in 22 years. So in 22 years we changed how all power is generated. Millions of draft animals, all the infrastructure to support them, all the animal power in the world was obsolete… and all of it in this country in 22 years.
The electrification of industry, the replacement of steam with electricity, happened… depends on how you want to measure it… in maybe 7 years. What's interesting to me… and you can debate whether AI is going to be faster or slower… what's interesting to me is when you look at unemployment numbers, you cannot see those technological revolutions. You cannot see them in the unemployment numbers. I could show you the unemployment rate for 250 years and say: Guess where the replacement of animal power with steam happened. Guess where electricity happened by looking at unemployment. I would argue that no matter how fast it happens people just take new technology instantly and increase productivity, and there's never even a blip in unemployment. What do you say to that?
Well, I just gave you seven categories of jobs that we as one small VC in one country are looking at displacing, and we're seeing higher performance by the technologies. I don't think steam to electricity was aimed at displacing individual jobs. These jobs are basically displaced, and no new ones are needed. When you don't have cashiers anymore, it's not like you create the new category of super cashiers or AI cashiers. They're just gone. And when you displace customer service, they're just gone. These seven categories I gave you are probably easily 150 million jobs in the world. And it's a complete displacement. That is the scale and an approach that hasn't been seen before. Again, coming back to agreeing with your longer term view, I certainly do believe AI will create a lot of jobs, jobs that we definitely cannot name today. But I don't see that pace of job creation keeping up with the job displacement.
That's my real question. If you can't see a blip… I understand all the other ways you're saying is different… My question is: Just because you run the engine faster and you advance the technology faster and you eliminate jobs faster, don't you think that has the corollary that it creates the jobs faster as well because there's just so many more economic opportunities all of a sudden, so many more entrepreneurs start so many different kinds of companies and hire those same exact people… Why do you believe that the job disruption engine runs faster than the job creation engine in this case? Just because you're saying, when you eliminate that cashier there's not some new instant job… but there are some unintended jobs you can't even imagine that are instantly created as well… Aren't there?
There could be. I mean when you went from steam to electricity, you can almost name… for every steam operator you have an electricity operator. Even though the jobs are different, the training is different, it's sort of… some go away and some come along. You can almost imagine that even if you couldn’t name the exact job, we're at a position where I can't at all tell you what are the 150 million jobs that would come when these 150 million are gone. So that makes me concerned.
It's to be expected. We couldn't have seen eBay or Etsy or Airbnb or Baidu or Alibaba or any of those jobs 25 years ago. Right?
That's right. But the jobs were not directly targeted and displaced. In the case of the Internet, it’s weaving a new thing that caused a gradual shift of people's buying patterns. You can almost say for very non-Internet sale that took place, there's an Internet sale that takes place. Therefore, there needs to be a human behind it. Therefore, there will be jobs created. Twenty-five years ago I could give you a logical explanation why jobs will be changed but not gone. Now I can't, because jobs that have been had for 250 years… customer service reps have been around for 250 years. Telemarketing agents have been around. Fruit pickers have been around for thousands of years, and overnight they’re just all gone. So I think you can be optimistic and say more will be created – and in the longer term I believe it.
So I think we're basically agreeing to disagree on this issue. But let's just say for the moment you are right and that lots of jobs are created. And I would tell you there is a gigantic training problem because we're displacing a set of workers who have the lowest education, the least amount of training, and are most unskilled. That's what AI is targeting, and whatever jobs are created by AI, the skill sets are not going to so easily match. I think for sure that the training process will be tougher, longer and costlier – if at all possible. Even if let's say 30% of the people are displaced, and let's say you're right that 30% of the jobs are in fact instantly created… but those might be data scientist jobs, robot repair jobs, not something a fruit picker and the dishwasher and the customer service rep can do.
So, I've never found that argument terribly compelling to be candid… because I hear it like this: It says technology is great at creating awesome new jobs like geneticist – high paying, high skilled – but it destroys jobs like order taker at fast food. And then people ask: Do you really think that order taker is going to become a geneticist? The answer is Well, no, of course not. A college biology professor becomes a geneticist, and a high school biology teacher gets the college job and the substitute teacher gets hired on at the high school all the way down the line. The question is not: Can that person who's displaced do that new job? The question is: Can everybody do a job a little harder than the job they have today? And that's 250 years of what I think has happened, which is technology creates great new jobs, destroys bad jobs and everybody shifts up just one notch. So it doesn't require some massive retraining of unskilled workers to be data scientists.
Oh, I certainly disagree with that. I think you describe a utopian escalation where everybody steps up. We're talking about a world where there's already a giant increase of inequality. I mean, over the past 30 years we're seeing wealth inequality, education inequality… and the reason there's so much nationalism and unhappiness all over the world is that the “haves” and “have nots” have increased a lot. And what we're about to do is to take the “have nots” – the most “have nots” of the “have nots” – and take away their jobs. And then, I think, let's just say 25% or 30% of the “most have nots” of the “have nots” to have them enter a career ladder that's gradually shifting everybody one notch. The statistics just don't work.
All right, fair enough. I mean I would argue that any time you can increase human productivity, you're going to increase wages. It will always happen.
Well, give it time. I give us 30, 40, 50 years, it might work, but AI's coming in 10 or 15…
So you make a very compelling case, Dr. Lee.
So do you.
So I think this kind of embodies the conversation a lot of people have in Silicon Valley, and I will give you the last word on it...
I will say this is one of the best conversations I've had. I think your ideas are very thoughtful. We certainly don't agree on many things. But I think it's a very good discussion.
Well, thank you. I guess I'll say one more thing. But I really will give you the last word. Is it the case really… so inequality is going up, certainly… isn't it the case that inequality is going up because it's now easier to make a billion dollars than ever before. In other words, if you look at the number of billionaires in the world who made their own money it's higher than it's ever been. Google minted seven billionaires. Facebook minted eight. So all of a sudden it's easier for people – not me, unfortunately – to make a billion dollars than ever before, and that mathematically increases inequality. But is that really the thing people should worry about. Shouldn't we instead ask: What is the plight of the bottom 20%? Are they getting wealthier or are they getting poorer? Do these technologies help them by increasing their productivity. Or do these technologies somehow hurt them? And I guess I think when you deflect and say inequality is going up, that's different than saying poor people are worse off today than they were before. So, do you think that a bottom 20% person in China or in the United States is worse off than they were 20 years ago?
Oh, of course not. They're better off. But people's psyche is that they're not just comparing their absolutes, material wealth or how they have done in the last 20 years. They're comparing with other people. I wish people could just look at their own, but I think it's demonstrated in psychology that people are very concerned about how more inequality raises social instability, creates more unhappiness, causes things like unusual election results that we have seen. So I think there are two sets of issues that are both independently correct, but the issue still remains. And, in the end I don't want to be viewed as a total dystopian or negative viewer because if actually you've read my book you know that I have an optimistic ending.
And I think that with 25 or 30% job displacement, if efforts were put in for retraining I do believe there are enough interesting, useful, economically valuable and socially valuable jobs –
in particular jobs that relate to empathy, compassion, caring for the elders, nurses, nannies, even the future teachers and doctors – will be large enough categories to absorb the people who are displaced and if we put the training in place and make sure the new jobs are paid fairly and have a decent social status. So I do have an optimistic final outcome as well, even though I see a lot of chaos in the process.
So let's switch and talk about one of the thesis of your book. I mean it's called “AI Superpowers,” the plural for which we'll discuss in a moment. I've avoided asking you philosophical questions about artificial intelligence because what I have discerned from reading the book is you're a practical man interested in the practical ways these technologies help people and so forth.
I'm just going to ask you one though, because you mention that the AlphaGo and Lee Sedol Go match was this watershed event in China where people were like, “Whoa, that’s amazing!” So I specifically want to draw your memory back to Move 37 where that was the moment that AlphaGo made a move that the [Google] Deep Mind people said there was only a one in ten thousand chance that a human would have made that move. Even Lee Sedol later would say that move was… I don't remember his exact words, but he was so gracious and so generous… and he said that move was divine, I think was what he said. That was the moment that people talked about AlphaGo being creative. Do you believe Move 37 or AlphaGo in general is creative?
I do not, because I have a high bar on creativity. I think creativity is about… really, you know, the next Picasso or inventing the next cure for cancer or inventing the next theory in physics. AlphaGo has a human set function in an artificial domain in which it was able to smartly search the space and come up with things people haven't come up with before. So I do understand if someone has a low bar on creativity. If anything novel is called creative, then fine, you can call AlphaGo creative. But I have higher bar.
So, give everyone, if you would, an overview of the most novel aspect of your book, where you talk about how we're going to enter a world of two AI superpowers and how the Chinese view artificial intelligence differently because you're a man with a foot in both worlds. You've had education, you lived here for a long time in the United States… For obvious reasons, you're uniquely suited to be the definitive voice on the differences and similarities in these two worlds. Can you just take a moment and intrigue our listeners?
Absolutely. The U.S. remains by far ahead of China in research, in deep research, in technologies and in expertise. Whether we measure by research paper h-index, whether we measure by people who are really deep thinkers or brilliant ideas people might come up with [for] the next breakthrough… the U.S. probably has about 10 times more such people than China. So it's way ahead.
However China is a very unique country in the sense that it has the following four things. First, it has an amazing set of entrepreneurs who are incredibly tenacious, incredibly hard working… and they are the epitome of Lean Startup and pivoting. They will take a real problem that people have and use every which way to solve it, including using AI and algorithms to optimize the results to meet customer satisfaction or better revenue. So these entrepreneurs I think generally are faster to results. I'm not as innovative as American entrepreneurs but much faster to results, much more results-oriented. And that's very suitable for iterating A.I.
The second thing is China is just full of data because AI gets better with data. China has three to four times more users than the U.S., and each user actually uses the Internet probably two to three times more than the American users. Example… ordering food, sharing bicycle rides and particularly payments. All the Chinese payments are actually done by software companies. Chinese people carry no cash and no credit cards. These payments are the equivalent of Facebook and Amazon. So imagine if you will that Facebook acquires Visa and MasterCard and Amazon acquires American Express and they fully integrate everything into their engines. So… much stronger predictability, as well as they pass the AI and the data back to people who transacted through them. So, if I’m a convenience store owner of 100 stores and I use WeChat Pay then I will actually know what customer bought, essentially giving my offline store some online capabilities of being able to do predictions, product placement, sales forecasts and so on. China has much more data and that is a very very key point.
The third thing is that China actually has more capital. The total money invested in AI is larger than the U.S. starting in 2017. And that money fed into hungry, tenacious entrepreneurs will yield more results.
And finally, of course, the Chinese government is very supportive of AI. The State Council plan of 2030 [calls for China] to be a world leader in AI technologies. A lot of money has been spent, especially on infrastructure where private capital cannot fund the building of highways and cities that are suitable for autonomous driving, and that will give China some edge.
So, I think the view of AI is very different. China is very practical… going into not only Internet but finance, banking, insurance, manufacturing, education, schools, hospitals and retail… and [apply AI] in much bigger and deeper ways than in the U.S. The U.S. is leading in research and technology, but China is running faster in implementation. From a results standpoint China already has the world's most valuable companies in speech recognition and synthesis, in computer vision and in drones. It's already building products that are equally good… and getting monetization and valuation at a comparable level to the U.S.
In what sense is it useful to think of American AI and Chinese AI? I mean, isn't it really Google AI and Facebook AI and Baidu AI and Alibaba AI? What is the nationalist component, as it were… because, these companies are all multinational… they have employees everywhere. Why does it still have kind of a nationalist tint that makes sense to view it through?
I think the media and maybe the governments want to see that, but I certainly don't see it. In fact, I see it as parallel universes. Chinese companies were funded by Chinese VCs and they build products for Chinese people. So, there is no zero-sum game. Amazon is a much bigger competitor to Google than Baidu is… right? And Google is a much bigger competitor to Facebook than Tencent is. So the Chinese companies operate in the parallel universe. Their gain comes at each other, not at U.S. companies. So, I would go further than what you said. There really should not be a discussion of a zero-sum game. I think, in theory, in an abstract universe, I would love to see American researchers teamed with Chinese implementers building products that can be used for the world with half the money from the U.S. and half from China… and everybody makes money. I think that is the ideal combination, but I realize that's not very realistic given the current trade situation.
I think it's interesting that you can see how certain cultures develop a certain natural advantage in artificial intelligence. For instance, China leads the world in two things for a very specific reason. One is voice recognition because the character set is so hard to type with on a mobile device, and facial recognition because you have so many unbanked people in China. As you were saying, commerce has to be done securely. So it has to have really good facial recognition to know that’s who is holding the phone. It's really interesting to me that a particular country’s cultural situation creates an environment where certain things naturally evolve.
I've also noticed that Chinese AI researchers by and large are much more familiar with what's going on in the U.S. than vice versa. The information doesn't seem to be flowing well… U.S. researchers don't seem to know about advances in China nearly as well… or at least that's my experience. Do you see that? Or is that not true?
Well, first, very good observation on the Chinese environment creating some advantage. Another thing to add to that is: China is actually behind in many areas, which gives AI a chance because the leapfrog is bigger. Right? China didn't have a lot of credit cards, so mobile payment was a bigger leap because it was a huge advantage compared to cash and a smaller advantage compared to credit cards. It's harder to adopt here because credit cards are very heavily entrenched. Similarly, American malls were the world-leading way of doing offline commerce. Before China could copy that, all this AI comes about. So Chinese retailers are saying: Well, let's do retail with AI and have the Chinese style of malls. So, I think being behind also has certain advantages.
On your other point about the lack of symmetry of information, I think that is hurting the U.S. I agree with you. I think for academic research the information is shared. Both countries have academicians who would go to conferences; they read each other's papers; they cite them. That's reasonably connected. But you're very much right that all Chinese entrepreneurs really study both Silicon Valley and China, so they kind of have two teachers. For Chinese entrepreneurs, Jack Ma Yun is their teacher – virtually – as is Larry and Sergey [Larry Page and Sergey Brin, creators of the online search engine Google]. But for American entrepreneurs, they generally view China as copycats and not worthy of learning from. But, in fact, Chinese business models, Chinese ways in which AI is applied… I think that is very much worthy of learning. And I hope to argue the point in the book to open eyes of some American readers that it's worthwhile – even if you don't ever want to do business with China and you want to build a purely American company – it can't be bad to study another business model like this. You know, it’s like going to first grade, third grade, fifth grade… skipping second grade, fourth grade and sixth grade. That can’t be a good thing.
I think a lot of it just boils down to the percentage of Chinese who speak English is vastly more than the percentage of Americans who speak Mandarin. But, in essence, I think the language barrier can be an excuse for looking elsewhere for solutions.
That’s part of it... and machine translation won’t help you because even with good machine translation, what they translate is not very readable. So that's why people should read my book because it's written in one of my two native languages... and hopefully easier to read.
We're running out of time right now. So, I will ask you one final question, then I will invite you to close with whatever I haven’t asked that you'd like to talk about.
When you look forward 10 or 15 years, are AI techniques the property of the human race? China is going to make advances and we're going to learn from those… and they'll be promulgated, and the U.S. will make advances and then Israel and Great Britain and all the others... I mean, is it really a global community? Or is it the case that nationalism and business practices mean that people will lock their algorithms, their data behind… you know, there'll be trade secrets and they'll protect them ferociously, and we’ll never really have the technology widely shared?
I think there will be a great deal of sharing. Sharing by the AI academics is natural. I remember when I did my Ph.D. thesis, many people questioned how could the results be so good. My advisor says, “Just share your source code, share your data.” And now we see a lot of people doing that, because AI is one of the few sciences where results are replicable if you implement the algorithms faithfully. So, given that, it has led to a group of researchers who have been very open, willing, transparent, helpful in sharing. So I think that will continue. There will be some exceptions. Large companies, a Google or a Tencent, might choose to protect some things within its firewall or platform patent to protect something, which is understandable. But I think mostly it will be sharing.
Similarly, I think good practices will be shared. Like you said, if new AI-created jobs are created in China, someone in the U.S. will create those jobs. Hopefully if someone comes up with a new way to deal with AI-induced displacement, like I said, another country will learn that too. Maybe one country will come up with a good way to protect data privacy or a good way to fight data security and that will be learned somewhere too. So, I think there will be an increasing amount of sharing at the individualistic level. It's hard to predict whether the political nature of each government will choose to be more or less nationalistic. They're certainly not going in that direction. I hope that will change, but I am not an expert.
I've imposed on your time enough already. Is there a closing message you have about your book, about your work, that you would like to share with my listeners?
I think this book “A.I. Superpowers: [China, Silicon Valley and the New World Order]” is really three books in one. First, talking about why China is different from what we read in media. There are things that are scrappy and maybe frowned upon, but there are also a lot of things that are worthy of respect and learning from.
The second is that I think with the U.S. and China both driving AI forward, it creates a lot of opportunities… but also a lot of challenges, and we can overcome challenges if we work together. And the third is that I share a lot of personal stories – including my own illness, facing my own mortality… having had cancer and having overcome it… but then looking back and realizing that we need to look differently at work.
Displacement with work is not a horrible monster. Ultimately it might be something we'll be very appreciative of because there are a lot of things in life that are more precious than working.
That's a wonderful place to leave it, Dr. Lee. I want to thank you so much. This has been so informative for me, and I want to thank you for taking the time.
Thank you. Thanks a lot.
Byron explores issues around artificial intelligence and conscious computers in his new book The Fourth Age: Smart Robots, Conscious Computers, and the Future of Humanity.