139: Kai-Fu Lee | What Every Human Being Should Know about AI Superpowers

Transcript for Kai-Fu Lee - What Every Human Being Should Know about AI Superpowers (Episode 139)

Jordan Harbinger: [00:00:00] Welcome to the show. I'm Jordan Harbinger. Of course, I'm here with my producer, Jason DeFillippo. There's a lot of talk about artificial intelligence these days from whether it'll take all of our jobs and leave someone unemployed, or whether it will just murder all of us in some particularly brutal fashion. Now, while watching experts in science fiction authors debate this endlessly online, I came across this book by Kai-Fu Lee, former President of Google China, discussing the rise of AI in China and what this means for AI for the rest of the world. We’ll learn just how close or how far we are from the different types of artificial intelligence and how AI will begin to change the world and our position in it. We'll also discover why AI is as important as the industrial revolution was and yet it will happen a lot faster. And of course, what this means for us as mere humans.

[00:00:48] Kai-Fu Lee is a venture capitalist, technology executive, writer and an AI expert and this was a really fascinating conversation, and I hope you enjoy it as well. If you want to learn how I book some of these amazing guests, well, I'm teaching you how I do outreach and manage my network over at jordanharbinger.com/course. That Six-Minute Networking course free. All right, here's Kai-Fu Lee.

[00:01:10] We have all these kindergarten questions about AI things like will it replace us? How quickly is this going to happen? Are we going to end up with robot overlords? My phone already bosses me around, so how long until something actually is forcing me to take action. And I think people, aside from Elon Musk, people like yourself are more excited and less scared when it comes to AI. How far away are we from generalized AI that can do all kinds of -- what we as sci-fi readers expect AI to be able to do?

Kai-Fu Lee: [00:01:45] Well, I think the really important answer to the question is no one can possibly know because there are still probably 10 to 20 breakthroughs needed to get there and no one can predict the speed of breakthroughs. But if history is any guidance over the last 62 years, we've had one breakthrough and that was nine years ago, and no more since then. So if you want to be super optimistic and say “We can have 10 breakthroughs in 20 years.” I would say “That's optimistic but unlikely.” If you want to say “In 100 years, I think it's possible, but who knows.” So we can't answer it. And anyone who tells you and gives you a timeline is simply trying to sell a book or being too optimistic or just doesn't understand it, yeah.

Jordan Harbinger: [00:02:28] Gotcha. Okay, we did have a Sputnik moment, at least China did for AI. Tell us about what the AlphaGo victory -- well, victory over that poor human told China and the world about artificial intelligence.

Kai-Fu Lee: [00:02:43] Yeah. So actually, a very small number of people have been working on AI. We've been in investing in China, AI for five years. China has had a number of institutes like Microsoft research, which I established about 20 years ago that has been working AI. However though the popular understanding and knowledge of AI really began two and half years ago when AlphaGo beat Lee Sedol, a Korean Grandmaster when a former World Champion, and I think that was a big movie moment for China for a couple of reasons. One is Go is believed to be a game that requires not only intelligence but also wisdom and zen and the ability or the humanity involved in the game and strategy. And there was a Chinese pride because it was invented in China. Plus all the AI experts have been saying Go is at least 20 years away. And suddenly this US, UK engine called Deep Mind beat one of the world's best players by far. And I think the Chinese people many of whom play Go, many of them whom love Go and thought there was the last stronghold of humanity, which it isn't, but that's what people felt. And hearing AI experts saying, “It's 20 years away.” Suddenly it is now. so it feels like a fast forward moment that AI is suddenly here and that woke up the Chinese people perhaps due to national pride, perhaps due to misunderstanding, perhaps due to a surprise, perhaps due to a chance to make money, perhaps due to is China behind, all of these reasons suddenly ushered in and push for a huge amount of focus, investment, engineering, people wanting to study it, people wanting to invest it and the government wants to help it. It just happened in the last two and a half years.

Jordan Harbinger: [00:04:42] It's kind of incredible to see the progress China's made in the last two and a half years with AI and just AI progress in general. And for those of you who don't know what Go is, this is like, and I'm going to butcher this, but it's ubiquitous like chess in China, it looks like Othello, but as much more complex and everyone thought there's no way we can teach IBM to beat people in chess. It's simpler, it's got sort of an algorithmic, if you do this, there's only a few sets of correct moves. Go had hundreds of thousands or more combinations potentially, and you just can't memorize it. It has to be strategic and this computer beat this amazing champion. People thought it wouldn't happen and it beat multiple games in a row and the poor guy starts crying on television, and I think everyone kind of felt for him and felt for humanity in this moment.

Kai-Fu Lee: [00:05:35] Yeah. That was actually the year and a half ago [indiscernible][00:05:38] game against But yes, the game against Lee Sedol was four to one, and then against was five to zero by far, and yes, started crying and to put numbers down, since this is a very -- you have a very engineering savvy crowd. Chess is on order 40 to the 20th power in terms of search space. Go is on the order of 400 to the, let's say a 100th power. So just think about the magnitude of these numbers. And that's why even when a Kasparov lost to Deep Blue, people were saying, “Oh, the order of magnitude is so much larger for Go and it's at least 20, 30 years, and that's how it came about.

Jordan Harbinger: [00:06:26] Right. So that was my next question really, is how was this different than IBM's Deep Blue beating Garry Kasparov? Because Gary's been on the show. He's a very intelligent guy. He was the chess champion, Deep Blue beat him in chess. And everyone went, “Oh my gosh, this is the end of our era.” And then kind of nothing really happened as a result of that. This however, is different. This is the Sputnik moment where America, or in this case, China wakes up and says, “Oh, wait a minute, the alarm bells are going, there's a thing in space from the Soviets. We got to get on this.”

Kai-Fu Lee: [00:07:01] Yeah, I think in chess, people could sort of extrapolate it. Finally, you mentioned Othello. I was the one who wrote the Othello player that be the World Champion in ‘85. I wrote in ‘85, I think it'd be the champion ‘87.

Jordan Harbinger: [00:07:15] Oh wow.

Kai-Fu Lee: [00:07:15] And then chess is after that. But one could extrapolate from Othello to chess because Othello was on the order of -- 10th to the 15th. And if chess is 40 to the 20th, you can sort of say, “Okay, if we build a machine that's a lot faster, it could do the trick, which is exactly what IBM did. Deep Blue was just a super-fast machine that ran something not too different from our algorithms for Othello. It was a hardware breakthrough in the sense that it made compute so much faster, but you can only make compute maybe five, 10 orders of magnitude faster. But we're talking with Deep Blue here, much, much bigger number. So it had to come from not just a speed of breakthrough, it was an algorithmic breakthrough using deep learning. It was a data breakthrough using huge amount of data, originally human against human, later machine against machine data, and that's the essence of what happened in the last 15 years.

[00:08:16] That nine years ago, Jeff Hinton and his students invented deep learning and it became used in more and more scenarios. And deep learning was something that did something nothing ever it before could do. It’s a learning algorithm with a massive amount of parameters that could only be trained on large amount of data and the works best when you have huge amount of data in one single domain that is a properly labeled winning or losing, making money, not making money, clicked not clicked, kind of a simple objective labeling, and Go fits every aspect of those requirements and it became the first step poster child for deep learning for the masses.

Jordan Harbinger: [00:09:01] This makes sense, so there's different -- there's three different approaches to problem solving and they get more complex. So you have this rule based where if I'm playing checkers, because I'm pretty basic in my game knowledge, so forgive me here. If I'm playing checkers and I move the this way and I get King. There's a very finite set of rules that you can program into almost calculator level computer that could say, “I can beat this person in checkers.” I've got the rules, I know how this works. Those might get progressively more complex and any of the expert systems and now we have neural networks. Can you tell us why neural networks are different than just programming a bunch of rules in? Because when I think of AI originally, before I read the book, of course, I just thought, “Wow! They programmed in a million rules.” It's doable, but that's not really what's going on here.

Kai-Fu Lee: [00:09:50] No, no, actually the expert systems really almost never worked for any domain. Even my Othello system and IBM's Deep Blue were built on machine learning. They were just not running on as powerful computers with such advanced algorithms trained down as much data. So the way a rule based system would work as a human would write all the rules down about if I under check, then I need to search for a way to get out of the check, and here are the ways to get out of the check to move your piece away, to block it with another piece and so on. But you can't really enumerate everything and people spraying actually it doesn't quite work that way. It's not completely rule based even though we think it is, but we don't know how the brain works. So using rules to approximate what we do is a very brittle, it falls down in a lot of tail conditions.

[00:10:51] So the way deep learning and machine learning in general take care of the problem is actually use a very minimal amount of human knowledge. Just tell it the rules of the game and feed it a lot of data, and the day that would be here is a game where a white one, here's a game where black one, then learn that the positions and moves in the game which ones were the ones that lead to the white to win and in the future move more like that. And you train with millions or even billions of positions so that it covers the tail. That is the key point is you have so much data. In some sense, it's dumber than people because we don't need to read a billion positions to learn to play Go. We don't need to watch a billion clicks to know what the person might like. But the system with deep learning after seeing a billion samples actually does a better job than the human, that we can sometimes work with five samples, 10 samples that's where we're really smart. Machines, deep learning can't do that, but once you have a billion samples and a single domain, there's almost no way we can come close to a deep learning performance.

Jordan Harbinger: [00:12:07] This makes sense. Okay, so rule-based, we tell the machine there's a thousand or 10,000 combinations, but with neural networks and machine learning, we don't necessarily tell the computer that there's any combinations. We just say you can't move left, you can only move right, and it says, “Okay.” And then we say, “This is a success and this is a fail. And we input as much data as many successes and fails as we can and the computer reverse engineers at and probably pulls out some insights that we would never think of like “Oh, when it's cloudy black wins more. We have no idea why. Nobody can explain it. But the computer has done enough statistical analysis to realize “Hey, if it's raining, choose the black stones because for some reason you have an advantage.

Kai-Fu Lee: [00:12:52] Yeah.

Jordan Harbinger: [00:12:53] It's probably not weather based but that's an example.

Kai-Fu Lee: [00:12:55] The raining is a weird one but--

Jordan Harbinger: [00:12:57] Yeah, it's a weird one.

Kai-Fu Lee: [00:12:58] Maybe when the room is dark.

Jordan Harbinger: [00:13:00] Sure.

Kai-Fu Lee: [00:13:00] You have an advantage playing white.

Jordan Harbinger: [00:13:02] Sure.

Kai-Fu Lee: [00:13:02] Because you might not -- you have a tiny little lower probability of not seeing a black piece or something that like that.

Jordan Harbinger: [00:13:08] Something like that.

Kai-Fu Lee: [00:13:09] Yeah.

Jordan Harbinger: [00:13:10] Exactly. And I would never figure this out with our 10 samples. Even if we played for 50 years and we had hundreds of thousands of samples, but the computer can figure out what's going on. It just can't necessarily explain why, which is fine because we're just looking for results, right?

Kai-Fu Lee: [00:13:22] It’s fine and it's not fine. If you're just looking for results like Go or Monetary decisions, making investments, you only care in the end, you make the most money. But in some cases, like I'm deciding if someone's guilty of our time, right?

Jordan Harbinger: [00:13:36] Oh, yeah.

Kai-Fu Lee: [00:13:37] Or deciding where a car should go, each of which might hurt someone or diagnosing cancer, those kinds of things people do want an explanation.

Jordan Harbinger: [00:13:47] True.

Kai-Fu Lee: [00:13:48] And it is tricky because what the machines have are gigantic neural networks with numbers and to convert those into a language humans understand, it's not easy. That's why we begin with problems where you don't need explanation and then it takes time -- and then as a research topic of how to do explainable, yeah.

Jordan Harbinger: [00:14:06] Sure, yeah. Because I don't think we would want our self-driving cars to find that the quickest way to our house is straight through a pedestrian mall. It may be correct, but there might be an objection here along the way that the computer doesn't care about that we might want to correct for.

Jason DeFillippo: [00:14:19] Your listening to the Jordan Harbinger Show with our guest Kai-Fu Lee. We'll be right back.

Jordan Harbinger: [00:14:26] This episode is sponsored in part by Calm. I love this app. This is a meditation app. I know there are many, but I like this one. The woman in most of the meditations has this really soothing voice. It's not annoying soothing, if you know what I mean, and starting your day with meditation can really change relationships, focus, anxiety, stress, patience, things like that. When I go through really tough times, I often will wake up in the morning and start a meditation routine or do more meditation as well. And New Year is coming up, it's that magical time when you set some ambitious goals only to give up after two weeks, but if your goal is to slow down and be a little bit more mindful well, you can accomplish more and Calm, we'll help you do that. If you had to calm.com/jordan you'll get 25 percent off Calm premium subscription, which includes hundreds of hours of programs. And what I love about Calm as well is it's not just talking you through meditation. They have anti-anxiety stress, they got focused stuff. They've got the daily Calm. There's sleep stories, which are basically bedtime stories designed to help you relax before you doze off. They've got meditations for different types of slowing down in different types of mindfulness, which I really dig.

Jason DeFillippo: [00:15:30] I've been trying this as well, and I really do love the sleep stories. The new ones from Stephen Fry. I've listened to that over and over again. It just puts me right to sleep. His voice is just so amazing and it's read so well. I love this. So get unlimited access to all of Calms content today at calm.com/jordan, at C-A-L-M.com/jordan, give yourself the gift of Calm and start 2019 off right. Seriously, it might change your life.

Jordan Harbinger: [00:15:55] This episode is sponsored in part by HostGator, one of my faves. These guys are so nice. Everybody who works there is a gem. I'm serious. It's ridiculous. It's like how did you find just the nicest people and get them all in one place? Everybody's a brand nowadays. On social media, it's getting harder and more expensive to get yourself notice and since those social media sites come and go or get vacuumed up by Cambridge Analytica. You got to have your own stable place on the internet that you can call home and that's why we recommend HostGator's Website builder. You can easily create a professional looking feature packed website. The best part, no coding, no figuring out why something is unaligned because you had slashed somewhere or a period somewhere and you got to spend five hours finding it. You can choose from over a hundred mobile friendly templates, so your site's going to look great on any device, smart phone, tablet, desktop. They also have a ton of add-ons, SEO stuff, PayPal integration, stuff that lets people buy items that you create directly from your website and 99.9 percent guaranteed uptime. That means it's not going to go down and they're like, “Oh yeah, we're on it a see you on Monday.” Their support team is there to help with any issues you have 24/7, 365, and of course, they're given our peeps up to 62 percent off all their packages for new users. So go to hostgator.com/jordan right now to sign up. That's hostgator.com/jordan.

Jason DeFillippo: [00:17:15] Don't forget we have a worksheet for today's episode so you can make sure you solidify your understanding of the key takeaways from Kai-Fu Lee. That link is in the show notes at jordanharbinger.com/podcast. Thanks for listening and supporting the show. To learn more about our sponsors and get links to all the great discounts you just heard, visit jordanharbinger.com/deals. If you'd like some tips on how to subscribe to the show, just go to jordanharbinger.com/subscribe, and now back to our show with Kai-Fu Lee.

Jordan Harbinger: [00:17:43] So deep learning, we seeded a bunch of data. We use this in businesses. I think we get preoccupied. We as civilians who are not familiar with AI, we get preoccupied with what's going to happen, what's going to happen to our jobs, what do we do when AI comes along? We're not really thinking about, “Hey, China might get there first.” So before we figure out what's happening with jobs, I'd love to discuss what AI needs to thrive? If this is a plant in a garden, China has healthier soil and a better environment. Why?

Kai-Fu Lee: [00:18:14] Well, that's what my whole book is about--

Jordan Harbinger: [00:18:17] It is indeed.

Kai-Fu Lee: [00:18:17] AI Super Powers.

Jordan Harbinger: [00:18:18] Yes.

Kai-Fu Lee: [00:18:19] But well half of the book, the other half is on the other questions we're going to talk about next. So the China, China has some disadvantages and some advantages. So that the US advantage is clearly us invented deep learning, invented AI. US has probably about 11 times more brilliant AI scientists than China. So this is a huge edge for America it would seem. However, what's missing to a lot of the people is what has already been invented in AI. Namely deep learning and related algorithms is already good to make about $16 trillion of value worldwide. So we're kind of in a phase of implementation, investment monetization. So in terms of using this is like electricity or the Internet. It's already been invented. So you could try to invent electricity 2.0 which never happened. Internet 2.0, which did happen. But you could do that. It may or may not happen, but the rush now is who can build the best businesses, make the most money create the biggest opportunities and make the next big giant unicorns using AI either as a core technology or as an enabler. So in that rush, China is potentially ahead. China has caught up in the mirror 2.5 years as we discussed just in the last two and a half years through lots of entrepreneurs working incredibly hard, every engineer wanting to study AI and China has a lot of engineers and more venture capital going into AI in China than here say in 2017, and most importantly more data in AI, because there are more people, people use the Internet more so in at least a multiplicative effect of data, and as we mentioned, the more data, the better the AI works.

[00:20:15] So in implementation of let's say an Internet app, e-commerce app or a bank loan application, or using AI to predict what video you might like. China just has a ton more data than any other country, including the US. So we have already got to a point where today in computer, speech recognition, computer vision, drones, machine translation, China already has the most valuable companies in the world. And we as a venture capital firms and elevation ventures, we've already helped create five AI core pure AI unicorns that are worth $21 billion.

Jordan Harbinger: [00:20:58] Wow.

Kai-Fu Lee: [00:20:59] So China is already rapidly catching up in just two and a half years. And if we project this further, we're not anywhere close to being done. AI has not at all penetrated even the banking insurance, stock areas should be easy and hasn't even begun on retail, manufacturing, healthcare, agriculture, and so on. So as AI moves to computer vision, speech recognition, autonomous robots, autonomous vehicles, I think China's going to move faster. So US is ahead in research and technologies. So to the extent there's a breakthrough, US could regained the leadership to the extent there's not research breakthrough China is all caught up with the US and may lead to the US in five years from now. In terms of commercial value creation.

Jordan Harbinger: [00:21:53] The key bottleneck it sounds like is the amount of data China has so much coming from apps like WeChat, which is essentially there's really no analog in the United States. If people haven't seen WeChat, imagine Facebook, Twitter, YouTube, probably podcasts, any sort of messenger you have, text messaging, web browser, PayPal, Yelp. I mean I can just continue.

Kai-Fu Lee: [00:22:15] Visa and MasterCard.

Jordan Harbinger: [00:22:16] Yeah, Visa, I mean--

Kai-Fu Lee: [00:22:17] Because PayPal is not pervasive.

Jordan Harbinger: [00:22:18] That's right. I just mean everything put together. I could probably list every app on people's phones and it still would say, “Well, WeChat also has this.” I mean it's just everything in one. All that data is free. One thing I didn't see in the book that I'm curious about is the United States and Europe seem to have these really strong privacy laws where, “Hey, you can't send that app data back to the developer, you can only anonymize it and you've got to strip out this and that.” Just trying to have those protections because if not, they have an advantage there too.

Kai-Fu Lee: [00:22:46] The rules are basically similar. It's ends up being EULA, right? The end-user agreement and China has those two. So probably both Americans and Chinese tend to agree to send up data when asked and maybe China, Chinese users agree a little bit more.

Jordan Harbinger: [00:23:04] Maybe.

Kai-Fu Lee: [00:23:04] I don't think that's the critical issue, but probably there is a greater willingness to trade privacy for convenience or security. Chinese airports are full of cameras, train stations, and people are okay with it. Even though you lose your privacy, you get a greater a degree of security because potential hijackers and a most wanted list people would be recognized and that taken off the plane or the train.

Jordan Harbinger: [00:23:35] Yeah. Even the facial recognition technology that we see coming from China is to me, super interesting. I think for many people, really scary. I mean, there was one special that I saw. I think this is on VICE where if you jaywalk it finds who you are and puts your identification photo up on a billboard right next to the crosswalk. So essentially it's like, “Hey, yesterday Kai-Fu Lee, he jaywalked here, he didn't wait for the light and it's a little embarrassing to have your face up on the sign.

Kai-Fu Lee: [00:24:02] I haven't seen that one. I have seen the bill being sent to the user for the breaking the law of jaywalk.

Jordan Harbinger: [00:24:09] Oh wow!

Kai-Fu Lee: [00:24:09] Yes. Face recognition is used. Recently, there was a concert by a famous musician, Jackie Chan and then I think about 20 people were arrested during the concerts because they thought these are most wanted criminals. They thought going to a concert would be safe. There are 10,000 people and no one would recognize them. But face recognition did.

Jordan Harbinger: [00:24:29] Yeah. That's incredible. And it kind of makes you think which artists have the higher percentage of criminals and their fan base.

Kai-Fu Lee: [00:24:37] One could think that, yeah.

Jordan Harbinger: [00:24:38] Hard to say, but it's a great way to catch criminals having to ever big event that they're interested in. I know that this is going to create -- AI will create a bunch of value in the economy. Can you give us a prediction for what that'll look like and who's going to take the lion's share of this? What kind of value are we talking about?

Kai-Fu Lee: [00:24:54] Yeah, I think the Internet companies have already taken a ton of value, right? Basically every Internet company is the perfect example of single domain, huge amount of data. The fact that Google, Facebook, Amazon in the US and then Alibaba 10 cent in China are the ones who have already created hundreds of millions of dollars of market cap because AI has become a little knob that they can say, I want to you to automatically give me more users, more minutes, more revenue, more profit. You can tweak your business model now much better. And then I think this will then move to the businesses, banks, insurance companies, investment companies, because those are merely numbers, games and AI will clearly do better, then there are a little bit tougher because you'll move into healthcare retail, maybe the use of computer vision sensors, speech become important in some of those scenarios. And then you'll move into the area of robots and autonomous vehicle which will basically replace the people who are blue collar jobs as well as drivers in that case.

[00:26:03] So I think the gainers will be -- well first the AI companies who sell AI to enable others and charge money for it, but probably even more money will be made by people who naturally have a ton of data and can use a map to make more money. The Google Facebook are perfect examples, banks, insurance companies are others, but also disruptors. Imagine a new company coming up with an app that allows you to get along within one second. We invest in one such company and that I would argue is likely over time to take, basically eat the lunch of all the bank loans, so they'll make lots of money. And then basically because anyone who has lots of data and the knob or who can create a lot of data and the knob and the knob will say, “Make me more money, get me more minutes, get me more users. And when you can run your business like that it essentially it becomes a cash printing machine.

Jordan Harbinger: [00:27:00] Sure. You just turn the knob up to 11 and you kicked back ideally.

Kai-Fu Lee: [00:27:04] Well, there are consequences when Facebook turned the knob and say maximize use in minutes. It also had certain effects which people are now complaining about.

Jordan Harbinger: [00:27:13] Yeah, I was just talking with a friend of mine, Mark Manson is a really popular author and he said that he's stepping a little bit away from Facebook because he's convinced that they've accidentally or deliberately optimize for outrage because the negative comments will bubble up to the top. And I'm not, I'm not sure if Facebook optimized for outrage or if we just are optimized for outreach.

Kai-Fu Lee: [00:27:34] I’m certain Facebook would never optimize for outrage, but they would optimize for more minutes and may be outrageous correlated with more minutes.

Jordan Harbinger: [00:27:44] Exactly, exactly.

Kai-Fu Lee: [00:27:45] That's the unintended consequence, but somehow we have to deal with that.

Jordan Harbinger: [00:27:49] Right, whereas I would normally just say, forget it. I don't need to look at more cat photos or another story or status update. I might stay for a few minutes just to finish my tight rate about this neighbor of mine that thinks they're right about everything, and then those get bubbled up to the top.

Kai-Fu Lee: [00:28:04] There you go.

Jordan Harbinger: [00:28:05] I know that China has essentially Zhongguancun, which is this Silicon Valley of China, if we can call it that, and Silicon Valley formed over decades. You've got all these HP and Atari, and then Apple and Microsoft and things like that taking off here. Zhongguancun essentially it was brute force. It was kind of like, we're going to have our own Silicon Valley bulldoze this and put everybody in the same street.

Kai-Fu Lee: [00:28:27] Yeah.

Jordan Harbinger: [00:28:27] This seems to be almost a microcosm of how China has adopted technology and AI. There's mass innovation, mass entrepreneurship, a lot of political will in government pushes. What are some of, aside from more data to the advantages that China has? You mentioned there's a lot more engineers and subsidies. How are we leading here in the United States and China has more engineers? Are we talking about a lower level of engineer and then that's what's more important than just a few elites? It seems like there's a break there.

Kai-Fu Lee: [00:28:58] Yeah. I think in the era of -- because we're in the second phase of AI, the first phase was the era of discovery inventions and that's where the big shop scientists matter more. The second phase is implementation. How do I get this algorithm for bank loans to work better and tweak the data and import it from the customer database and then normalize the data and then tweak the parameters? Well, that's where you don't need a super scientist. So since we have moved to the era of implementation, that's where China's mass number of junior, but hardworking and capable AI engineers can help make a difference on your first question about the role of the government. I think that general sort of the way of thinking in the US is governments should not become pickers of winners because governments aren't trained to be VCs.

Jordan Harbinger: [00:29:56] That's for sure. And maybe governments shouldn't even pick technologies because they might not have the right skillset. The first part is certainly true. I don't think in China government picks winners. But I do think of government's picking important areas, does make sense. I mean, in some sense US funded on DARPA, and DARPA picked a speech recognition which I worked in and computer vision and autonomous vehicles as some of the grand challenges. And these were very instrumental to the launching of the research. I don't think there's anything wrong for the government to say, “Okay, well now that the early technologies are proven in universities, let's fund the infrastructure to help make them become commercialized.” I think somehow the quote unquote American thinking that governments should not get involved in building up a infrastructure is a two clouded by maybe success of Silicon Valley.

[00:30:57] Silicon Valley has succeeded through semiconductor, PC, Internet, mobile, without any assistance or much assistance from the government. Therefore, it kind of feels we don't need you. But I don't think that's true. If you think about the American economy, much of the success is built on President Eisenhower's decision to build the interstate highways. That is the infrastructure push that made the commerce and delivery possible. And the Chinese government is saying, “Well, we need more entrepreneurs, so let's go buid -- go get all the good ones on the same street and give them subsidies and then let them do the work.

[00:31:38] So government's not stepping in to do the work of a DC, it's making the VC's job easier by giving them rental, subsidies, and proximity to the entrepreneurs. Similarly in AI, the Chinese government is not picking, I'm going to invest money in this company or that, but they are maybe giving some investments to the top VCs who are good at AI. They are building infrastructure such as new cities and new highways that enable autonomous vehicles. So those I think are more akin to what president Eisenhower did and I would say these have been wise decisions. Obviously no government, including the Chinese government picks every winner right?

Jordan Harbinger: [00:32:18] Sure.

Kai-Fu Lee: [00:32:18] Solar energy was one that maybe didn't pan out for China, but it's kind of like a VC. The government picks technologies and if it's a well-informed picks, then it'll be right, maybe five times out of six, which is good enough.

Jordan Harbinger: [00:32:34] Good enough, yeah.

Kai-Fu Lee: [00:32:34] Yeah, exactly.

Jordan Harbinger: [00:32:35] So instead of planning the tomatoes themselves, the government is building the garden for those tomatoes to grow in the first place.

Kai-Fu Lee: [00:32:41] That's right.

Jordan Harbinger: [00:32:42] Whereas the United States kind of says, “Hey, if you want a garden, build your own garden we're not going to mess with this, other than DARPA, which is essentially the defense department's technological investment arm, if you can call it that. Which is why we have things like internet for example, in the first place.

Kai-Fu Lee: [00:32:57] Yeah. And I think the stock market capitalism, VCs, and Silicon Valley have done well without government help. But that doesn't mean with the government help, they can't do even better.

Jordan Harbinger: [00:33:09] Right.

Kai-Fu Lee: [00:33:09] Okay.

Jordan Harbinger: [00:33:10] What are the four waves of AI that we're looking at? I found this fascinating. Amazon already knows what I want to buy a lot of the time, or at least it's really good at convincing me I need things. Netflix kind of knows what I want to watch. We have this Internet AI. I see business AI advertised a lot. Hey, put your customer data in here and we'll tell you what your customers want to see more of. What else are we looking at phase wise?

Kai-Fu Lee: [00: 33:34] Yeah. So first that's already happened is the Internet AI. So any Internet company with a million daily active users can mine that data and have their knobs to do a great job of getting more users, getting more business, getting more minutes, getting new revenue. The second phase, as you said is business AI. That is any company that has a repository of data such as a bank with the customer transactions or an insurance company or know investment company or a large hospital. They all have data, and either it could be historical data or new data that they gather and that data is actually a gold mine on which they can predict user behavior, increased monetization, just like the Internet companies at a smaller scale, and there are AI companies selling tools and some of these companies are now using clouds from Google or Amazon. Others are developing their own, each of these can work, but it's all about creating efficiency may help you make money, save money based on data as a fuel for AI in any business that has some traditional data.

[00:34:47] The third phase is when basically it's about perception. It's about seeing and hearing, speech and video, and it's about turning transient data, which are lost. Like you're having a conversation with someone or this show or can all be captured and become training data to be used for something else? For example, the day that you collect from me can be used for better speech recognition or face recognition. And the actual uses go from Apple, Siri, Amazon Alexa, all the way up to a face recognition we talked about and autonomous stores. And you can imagine elderly homes where sensors and cameras are trying to make sure that elderly people don't fall down or danger of preventive maintenance of signals. So sensors, cameras, microphones giving the eyes and years and the sensors in the world to capture that, to do apps that weren't before possible.

[00:35:55] The final phase is autonomous AI. That's where AI is able to move and manipulate. And that's when you can imagine an AI being capable of doing manufacturing robotics to make something, to see something, to look for defects. You can see agricultural AI in terms of selectively watering and planting and fertilizing and then picking the fruits. So agriculture will almost be fully autonomous. You can imagine commercial AI, AI that can cook and wash. We fund an autonomous fast food store that has no humans.

Jordan Harbinger: [00:36:34] Where is this?

Kai-Fu Lee: [00:36:35] Is about a hundred restaurants in the Southern China.

Jordan Harbinger: [00:36:39] Oh my gosh.

Kai-Fu Lee: [00:36:39] Guangzhou has the flagship store is always a line behind it, and you can basically get a bowl of beef noodles for about two dollars.

Jordan Harbinger: [00:36:49] Oh wow.

Kai-Fu Lee: [00:36:50] Compared to five dollars at McDonald's. Over time, I think fast food is not one that determines on human, depends on human interaction, right?

Jordan Harbinger: [00:36:58] Right.

Kai-Fu Lee: [00:36:59] You're just as easily get it from a machine if it's half the price. So that's going to basically I think changing the whole world, but that's just one app.

Jordan Harbinger: [00:37:08] Sure.

Kai-Fu Lee: [00:37:08] I think all stores in the future will go into becoming two tiers. All the average to below price range stores will become autonomous. There's no point because we don't look to interact with people there. Only the luxury stores or the service oriented stores like a massage or something requires a human continuity, and then of course there's automo -- in this final phase of the autonomous AI, there's autonomous driving, which will completely change the way we basically move ourselves from place to place as well as the whole logistics and delivery of goods, and that will of course displace a lot of jobs of people who drive for a living.

Jordan Harbinger: [00:37:54] I want to get into the jobs thing in a second. But the perception AI wave is very fascinating to me because when I think of a computer seeing something, it's hard for me to wrap my mind around that because if I feed the computer a regular computer, my iPhone, for example, 700 pictures of my cats a couple of years ago, it's just a bunch of pixels that are in different colors. It doesn't know that those are my cat unless I say this is a cat. I have noticed recently in new iOS updates that it literally says cat, and there’s Momo. All these photos, it can now figure out or has figured out or been told how to figure out what combination of pixels and certain colors, angles, lighting, whatever it is, make up what looks to be a cat. And it's pretty darn accurate which is -- and this is happening on a server somewhere of course, but it's happening near instantaneously on my phone when I upload a new picture, which I have way too many of my, of my cat.

[00:38:51] So that to me is really interesting. And I, for one thing, an app I've always dreamed of is uploading a picture of myself to say Facebook and finding out everyone in the world who looks within five to 10 percent just like me, because I must have theoretically there's thousands or even tens of thousands if not more people that look so much like me that my friends wouldn't be able to tell at first glance. I kind of want to know who those people are in a weird way and this is all possible now.

Kai-Fu Lee: [00:39:18] Yeah, yeah, that's right. Well, it's just like before we couldn't find people with names like hours. S

Jordan Harbinger: [00:39:25] Sure.

Kai-Fu Lee: [00:39:25] Now we can have people who talk like us, look like us and so on and so forth. There are a lot of things you can do with it and it will be not just the face. It'll be able to recognize your gestures, your gait, how you walk about, your habits, and also it can infer your intentions because imagine an autonomous store that saw you move in. First it can separate you from someone who looks like you because the way you walk, because you might look alike but you walk differently. Then they can look at what products you pick up, basically your gesture, that the system will understand. You're picking up this you know this book and then you're flipping through it. That means you shows interest is like a click on the webpage and then you're smiling which means you like it or you're frowning or you feel disgusted, and all of those emotions are connected to feedback regarding this book and it's matched to you and your inclination to buy this book or other books like it. So it's very, very powerful.

Jordan Harbinger: [00:40:26] That would be interesting. So I pick this up, I look through it, I smile and then maybe I show my wife where I looked through it for an extended period of time and then someone comes up and says, “If you're interested in this, you'll be interested in all of these different things." And it doesn't have to be a person, it could be a screen next to the book.

Kai-Fu Lee: [00:40:41] That’s right.

Jordan Harbinger: [00:40:42] It says, not interested in AI Superpowers, I already read it. Here's three other books on a similar subject or different subjects.

Kai-Fu Lee: [00:40:48] That’s right.

Jordan Harbinger: [00:40:49] That is similar or similar in the way that it's written or explained.

Kai-Fu Lee: [00:40:52] Yeah. And of course it already happens on Amazon, but you can now extrapolate that to things that you would actually see in the store, but not Amazon. Let's say if you like romaine lettuce, it can suggest other types of lettuces, which might be similar. If you like ranch dressing, it's out of ranch dressing but then my suggest other dressings or how to make ranch dressing, so the combinations are really tremendous.

Jordan Harbinger: [00:41:16] And of course, this goes really deep. And again, the amount of data that we feed it, it'll come up with insights that we haven't seen and that humans can't recommend. If you pick this up, and I see you pick this up, I might ask you if you like this and recommend three other books, but if I have every book you've picked up in the last 10 years along with everything you've ever watched on Netflix along with everything you've ever seen on Amazon. I might say, “How about this wine?” And you'll go, “Sure.” And this isn't based on me thinking you'll like it or this is based on AI saying, based on people who like these books and these documentaries and this type of food who travel here and here. This has been a big seller and gets high ratings.

Kai-Fu Lee: [00:41:54] Well, there's no doubt that AI will know better than we do while we want to read and eat and vacations to go to. Now AI won't make all the decisions for us. It'll give us some choices and then we'll pick among those choices. But those would be more informed than accurate choices then reading a book review or asking a friend or just guessing yourself in the bookstore.

Jordan Harbinger: [00:42:16] Fascinating. And I think the idea that AI is going to make our lives a lot easier, or at least we won't have to think as much for people in our shoes is really exciting. I think what people are scared of aside from a machine deciding to kill all of us because we're killing the planet or some other reason is the more immediate threat that, okay, if this computer can pick all these recommendations and it can pick fruit or whatever in the near future, we're going to have a whole lot of people who, I think you've all, Noah Harari coined the term the useless class which is pretty harsh, but there's going to be a class of people that are never able to produce enough economic value to support themselves and that curve of approaching that might be coming up pretty quick.

Kai-Fu Lee: [00:42:59] Yeah. And well, in Professor Harari’s new book, I think he is become more optimistic. He talks about the same thing I talk about that may be people have to go back on their human instincts, their compassion, things that machines can't do because we're talking about people who can't do routine jobs better than the machine, the job of a telesales or customer service or fruit picker, dishwasher, driver, etc., these jobs will be gone, and they won't be able to find other routine jobs because those will also be gone, but what will not be gone are jobs of compassion, jobs of human to human interaction. So I think he has come around and certainly I say strongly in my book that AI cannot do a good job of emulating a human in the humans true sincerity and desire to help others. So whether it's an elderly care and nanny, nurse, doctor, teacher, concierge, these are jobs that our first AI probably cannot do for the next 30 years, maybe longer. Secondly, even if AI approximated them to some degree, people don't want that because “Oh, elderly person doesn't want the robot to take care of him or her, right?

Jordan Harbinger: [00:44:18] Right.

Kai-Fu Lee: [00:44:18] So I think those jobs will grow and we'll have to figure out which jobs those are and find a way to retrain and transition that displaced people onto those jobs and only then can we have a society that is not with a lot of resentment and depression.

Jordan Harbinger: [00:44:36] I think a lot of white collar workers, and I'm a former attorney. I guess I'm still technically a lawyer, I wouldn't hire me though. We think, “Oh, well it must be a bummer for people who clean houses or work on assembly lines or pick fruit.” But the AI is coming for us too. It's coming for the attorneys. It's coming for the doctors in some respect.

Kai-Fu Lee: [00:44:54] In some respect, I think is coming for the paralegals. It's coming for the part of the attorney's work that is looking up information and checking, cross checking. It is not replacement for their attorney who appealed to the jury for their support. It is not replacing in court argumentative parts. It is not replacement for the lawyer's job to calm down the accused and help find a strategically a way to either prove innocence or minimized sentencing or whatever, those things are not replaceable. Similarly for a doctor, AI may be better at diagnosing many or even all diseases, but it takes a good doctor to tease out all the conditions, family history from a patient, and gain the patient's trust and help the patient feel better and more confident that he or she might recover and that will actually increase the chance of recuperation. So the role of a doctor and the lawyer, especially a doctor will change, but there may be more lawyers and doctors not less, but it's just that the jobs and the salaries maybe different from what they are today.

Jordan Harbinger: [00:46:11] So in a strange circumspect way, AI makes us more human because instead of spending eight out of the 10 hours of my legal work day, going through LexisNexis and finding case law and putting it all together and making sure I'm not wrong, that part can be completed much faster by AI, and then I can spend time speaking with clients about how their case is going to work. Instead of saying, talk to this person over here, I'm busy, I'm neck deep in paperwork, we can actually make our jobs more human. Doctors can spend more time with patients instead of looking at radiology graphs and X-rays.

Kai-Fu Lee: [00:46:41] Yes. Either you can be more attentive to your customers so they feel high degree of satisfaction and pay you more money, or you can have more customers and make more money that way but your hourly pay may come down. Because well, but then it depends on whether the AI doing the work is counted in your billable hours or not.

Jordan Harbinger: [00:47:03] I know that you almost missed your own daughter's birth because you had a meeting, so you're no stranger to very intense work is workaholic. Is that fair to say at that point?

Kai-Fu Lee: [00:47:13] Oh yes. I have been an absolute workaholic, yes.

Jordan Harbinger: [00:47:16] What is studying and learning about AI and in fact being a thought leader in this field taught you about being human because I know you had a health scare that kind of had you turn a corner a little bit.

Kai-Fu Lee: [00:47:27] Yeah. Well it was actually the health scare that woke me up because when I was diagnosed with a fourth stage lymphoma and faced with the possibility I may only have months to live, it suddenly woke me up that that workaholic ways that I used to have were really dumb because if I had only months to live, I wouldn't want to spend one minute working. I would want to spend all of my time with my loved ones doing the things I love, and then you go back and say, “Well, why would I not live that way if I had years or decades to live?” There's no difference. It's just that somehow it takes a shocking life-threatening experience to wake me up and that experience has changed me, but it's not transferable on other people. So when I talked to my entrepreneurs in China, don't work so hard, they'll still like, “You're crazy. I'm in the prime of my life about the change the world, create an unicorn. What do you mean I need to spend time with my family?”

[00:48:27] So I think ultimately one's personal experiences are very personal. They are not easily transferrable to a lot of people. So I still share my story, but I don't expect a whole lot of impact. However, I think AI coming to take the routine jobs may be are collective will that we as humanity have become too much work alcoholics and that we can't seem to get out and that it takes some divine or collective will act to send AI to us to remove all the routine jobs. Now, we've got nothing routine to do so we can either spend time doing creative things or things that we love or spend time with people we love or build a company in the direction we love. So it brings out our passion, our creativity, and our humanity and our compassion.

Jordan Harbinger: [00:49:23] So rather than AI removing our sense of purpose and leaving us all wandering around in the streets looking for something to do, it might actually force us to shift our purpose from plowing through those 10 hours of email to spending time with our loved ones and family. So it actually could be live -- instead of our robot overlords being tyrants. They could actually be quite liberating.

Kai-Fu Lee: [00:49:43] Well, that choice is ours to make both scenarios are possible and one of the reasons I -- the main reason I wrote the book is so that people know that the optimistic choice exists and it's up to us to make that choice.

Jordan Harbinger: [00:49:55] Thank you very much.

Kai-Fu Lee: [00:49:56] Thank you.

Jordan Harbinger: [00:49:59] This really was a fun one, Jason. I did this in person with him and he's just a very intelligent guy, very sharp, and it's a different perspective than we're used to hearing. It's not just “Duck! The robots are coming to kill us,” and “It's also not great, nobody has to work anymore.” It's a very balanced and sort of nuanced look at what AI will do and in the order in which it will happen, how it will develop in our world and in our economy. So I thought it was quite a good read and this interview hopefully gave us a bit of an overview of the as well. The book title is AI Superpowers. Of course, that's available everywhere and if you want to know how I managed to book all these great guests, well I manage my relationships using systems. I've got tiny habits that I do every day to make sure that I'm reaching out to my network and I've got a whole course on this which is free over at jordanharbinger.com/course, and the drills are designed to just take a few minutes per day. It's the stuff I wish I knew a decade and a half ago. It's not fluff. It's crucial whether you're a professional and you have your own business. All that is at jordanharbinger.com/course.

[00:50:59] Speaking to building relationships. Tell me your number one takeaway here from Kai-Fu Lee. I'm @JordanHarbinger on both Twitter and Instagram. And this show of course is produced in association with PodcastOne and this episode was co-produced by Jason “The Robot is My Jam” DeFillippo and Jen Harbinger. Show notes are by Robert Fogarty. Worksheets by Caleb Bacon, and I'm your host, Jordan Harbinger. The fee for the show is that you share it with friends when you find something useful, which should be in every episode. So please, since you're such a generous person, share the show with those you love and even those you don't. It's the holidays after all, we've got a lot more great stuff in the pipeline. I am very excited to bring this to you. And in the meantime, do your best to apply what you hear on the show so you can live what you listen, and we'll see you next time.