Kai-Fu Lee talks about artificial intelligence: In the future, repetitive jobs will be replaced by machines


AsiaIndustrial NetNews: At the “Searching for Chinese Makers Season 3 Kick-off Summit” held this morning (March 30), Kai-Fu Lee, Chairman and CEO of Innovation Works, gave a keynote speech on artificial intelligence, covering the Development, core technology and outlook for the future. The following content is organized according to the shorthand of Kai-fu Lee’s live speech.

Zhang Quanling: You know that the heroine of “Twilight” recently wrote a paper on artificial intelligence and published it on the paper sharing platform of Cornell University. This article caused great backlash, so I think it crosses the border. Maybe useful. The next talk in our Yantai can be said to be a worthy venture capital mentor in the artificial intelligence field, please Mr. Kai-Fu Lee. The topic of the speech was “I’m not Kai-Fu Lee, I’m Artificial Intelligence”, and the title made me suspect that it was a fake Kai-Fu Lee who came up.

Kai-Fu Lee: Since I joked about this sentence, I found that I was pulled into any WeChat group, and someone asked me if you were Kai-Fu Lee or an artificial intelligence. I would take a selfie and send a voice, but no one was there. Believe.There is always someone saying “UnitrobotNo red envelopes will be given.”

What I want to talk about today is that I am an artificial intelligence. I am a person called artificial intelligence. Mr. Kai-Fu Lee is busy today and cannot come here. I am going to Shenzhen for a meeting. As an artificial intelligence, I will first introduce the history of my growth to you. I was actually born in 1956 and was raised in the US. My growth process was very bumpy. Every time it seemed that I was going to destroy humanity, someone called me a liar. This happened twice. But I always want to be the best, so in 2009 I read the book “Because I’m Different” by Kai-Fu Lee, and I found my life direction.

The core is the technology of deep learning. Judging from everyone’s appearance, the biggest thing about this technology is that at this time last year, AlphaGo defeated Li Shishi. Everyone thinks that Go is a machine that cannot be better than a human. There are thousands of people. Human wisdom, how could a robot beat us and the best chess player in the world? This matter subverts everyone’s thinking. I don’t think AlphaGo will change the world in the world. But it brings people’s awe and recognition of artificial intelligence, which must be a milestone event.

Beating chess, Go, Texas hold’em, that doesn’t mean anything at all, these are games. The thing that real human thinks to do well is computer vision, and what Sun Jian has done beats human vision. In the lower right corner is Hearing, which was done by Huang Xuedong, a researcher at Microsoft, and Sun Jian in the Face++ team that we invested in. They are both Chinese and have built two systems that really surpass human beings. Where is the value beyond human beings? To put it simply, when you can recognize more faces than people, the security guards don’t need it, and the functions of the police department don’t need it. When you can surpass human hearing, customer service doesn’t need it, and many people don’t use speech to do things. Needed, so this is definitely a job-replacement thing.

Kai-Fu Lee talks about artificial intelligence: In the future, repetitive jobs will be replaced by machines

Therefore, I boldly predict that in the next ten years, not only because artificial intelligence can hear and see, but because artificial intelligence can understand, judge, classify, and predict, in any field with big data as the core, when a person’s Work that can be judged within 10 seconds, or doing repetitive work, their work will be replaced by a machine, but it will be a complete replacement, not a machine replacing a person, but a machine. A machine can replace thousands of people. It’s not that machines are as good as humans, but machines are far superior to humans. For example, the Face++ we invested in can recognize 500,000 faces at any time, and the human memory is only 5,000 people. We have surpassed human beings in the number of face recognition by 100 times, so this is really not one-to-one. In comparison, these computers only need to spend electricity, very cheap, not tired, not on strike, working 24 hours a day, 7 days a week, so these jobs are basically useless in such an environment, I don’t have time today to talk about human beings what should be done. I have a book published next month, please read it.

What I want to talk about today is the enormous value generated by these jobs, and how we as makers should grasp it. And humans are definitely not talking about high-end Robots shaking hands with you, or driving unmanned vehicles, which are all important. But the core of artificial intelligence is the deep learning algorithm, which is now in your mobile phone, and you use it every day. Every time Didi takes a taxi to help you get on the bus, when Toutiao decides which article is the first one, when Baidu gives you search results, and when Taobao recommends products to you, it is full of depth. Learning, what is deep learning? I have limited time today, so I will explain in half a minute.

1 2 3 4 Next > page

We all know Excel, enter a batch of numbers, press a twist to calculate it, and put all kinds of finances in this month. Deep learning is an astronomical Excel. Input the data of various faces, and when you press it, it will be Yu Minhong or Wang Chaoyong. Input all kinds of Go data, and it will tell you where to play. Enter astronomical numbers, and it will tell you what product you are most likely to buy at this moment on Taobao. It is a magician, trained with astronomical numbers, and that is such a thing.

What value does this have? First of all, of course, it is on the internet. The data on the Internet is already huge, and it is used for artificial intelligence. Secondly, there are various fields. I don’t have time to talk about too many details here. I just talked about identification, BI, data already existing on the Internet, and data outside the Internet. Whoever has big data can do artificial intelligence .

There are basically four prerequisites for artificial intelligence. If you have it, you can do it. The first is the massive amount of big data, at least tens of millions or more. Who would say that 10,000 data is useless. The second is that these data need to be labeled, and the labeling is not necessarily done by people. There are also labels that many people don’t realize. When you buy on Taobao and don’t buy it, it’s labeling data. The third requirement requires a single field, not cross-field. Artificial intelligence is smart, but stupid. It only understands one field at a time. Those who understand catering do not understand travel, and those who understand buying stocks do not understand insurance. That’s why it’s so dumb, because only one domain can be understood at a time. The fourth requirement is that top scientists know how to use the method of alchemy. Deep learning is not a platform, and it is not possible to recruit a few people to do deep learning. It must be accumulated and experienced. After these four conditions are met, artificial intelligence can be done, so BAT is most suitable for artificial intelligence.

The seven black holes in the world are Google, Facebook, Microsoft, Amazon, and BAT, because they have huge black holes, they can only go in and not share. They have a lot of technology. Therefore, we have certain challenges in doing artificial intelligence outside, but in any case, if the four conditions just now are established, we can also do artificial intelligence. When your app had 10 million daily active users, it must have artificial intelligence, or it will be boring. For example, Zhihu and Mobike, which we invested in, all say that we have a large volume now and need artificial intelligence. Otherwise, how do I know if the class is satisfied or not, or how do I know which article to read on Zhihu? Therefore, in any field, we must make the product bigger, and when the product is bigger, artificial intelligence is needed.

In addition, there are many other applications, financial applications. Finance is a fictional game of human beings. There is no manufacturing, no distribution, and no logistics. If we can make AI better, how much money will be made for every extra point. If I set the insurance more accurately, whether it should be insured or not, I will charge him as much as I should, and the money will come immediately. So there is no doubt that finance is the first huge milestone for artificial intelligence.

Yesterday I expressed this point of view at the meeting of Zhirong Group, which we invested in, that is, the financial field is the best. They have made a product called Qianbao. After 18 months of establishment, how many loans can be issued each month? Disbursed 1.2 million loans a month. You can imagine which bank in China has issued 1.2 million loans, and a startup company has surpassed the loans that banks can provide in 18 months. What is the concept? His core is artificial intelligence to judge whether this person should borrow money, and his bad debt rate is definitely lower than that of the bank. So you want artificial intelligence to judge bad debts more accurately than people, and to borrow money more accurately, and then he does not need to charge interest, he can borrow 1,000 yuan with Qianbao, because the marginal cost of the machine is zero, so he can expand unlimitedly, can To 10 million orders, as long as you buy a few more servers, you still need people.

In addition to finance, there are medicine, education, etc., not much to say, so there are actually many development processes. From game, perception, decision-making, and reaction, we are constantly making breakthroughs. I also talked about a few important areas just now, and it is here that we see the greatest value, and I will not repeat it here.

Self-driving is a particularly great opportunity for the future, but it is very difficult to do today. Self-driving it can assist us, remind us, and help us make fewer mistakes. But below it it’s going to be 100% self-driving, because there’s no example, driverless I, as a coach, correct it, this example doesn’t exist. Think about it, if you buy a car and come back to the mall and say that you will occasionally make mistakes, you have to help it brake, and you will not buy it. This has been verified. Googel made a system. After taking it home manually, he went to the back seat to drink coffee and read newspapers. After Googel saw this decision, the driverless operation was to do L4, which was completely automatic.

Of course some companies are “not so responsible”. Like Tesla, the L2 technology that is obviously not done is allowed to be driven by others, and the result is dead. So unmanned driving is a very big challenge, but once it reaches L4, this is not just a matter of driving the next car you buy, it is the subversion of the entire human travel and distribution, the car will be there when you need it. Here, you only need to pay for use, and the car only needs to be loaded with one person, because most of the time we are not buying, the car can be smaller, the road can be wider, the traffic congestion is gone, and the parking lot needs to be closed. So all kinds of effects.


Don’t believe in the inevitability of this era, just say how long it will take for driverless cars to come, some say 10 years, some say 15 years. Its speed definitely does not come from the maturity of technology. Technology can mature within five to seven years. The trouble is human values, morality, and whether those who have immediate interests will cause trouble, as well as government laws and regulations. These It is the unpredictability that makes the rollout of driverless cars. It can be solved purely by technology in five to seven years.

What exactly is such a process? Basically it can be divided into three stages. In the first stage, our entire mobile Internet and Internet, we are helping artificial intelligence to do the foundation, and we have uploaded all the data. We do web pages and content, transactions start to take place on the Internet, bank transactions, stock transactions, these have become, and it’s marked. In every example I mentioned just now, Meituan is helping to label people’s eating habits, Didi is labeling people’s travel habits, Baidu is labeling people’s search habits, and so on, so all these habits have been captured. is the first stage, and we are at this stage now.

In the second stage, with the gradual promotion of intelligent hardware, unmanned driving, and robots, and as the price of sensors drops, all kinds of intelligent things in our hands are also capturing these data. These data are definitely not the era of IoT that everyone hears. Behind all this, there must be artificial intelligence to have IoT. Why does IoT become popular after a while? Because artificial intelligence has not come up, and because big data has not been integrated, its value after integration is huge. When your actions and questions are captured, big data health can definitely solve your problems.

The third stage, a bit closer to a sci-fi world, where everything goes online, into a full-scaleautomationIn this era, our life will be very cool and very happy. We have unlimited time, and all things will be done by computers for us. Of course, we will have no privacy.

So to a certain extent artificial intelligence is going to take over the world. This taking over the world does not mean enslaving us, controlling us, because artificial intelligence still has limitations, it has no self-awareness, no emotion, and it will not enslave human beings, it is our tool. But to a certain extent, this tool is so large that it manages everything we humans have. The biggest challenge for human beings, the most important thing for us to keep us human is our spirit of self-realization, but when all we can do is done by machines, whether this spirit exists or not, this is the challenge of human beings.

So such a challenge is far more than ten times, it will be in the economy, education, society, these details will not be said. Just to give an example of education, continue to educate children in this way, let each child do repetitive work, that is, let them wait to be replaced by artificial intelligence, they can only do one thing, which is another technology VR. In a few years, they will mature. These children can’t find a job. He can only wear a helmet at home, drink milk tea, and play games while lying down. This is the worst possible challenge and destiny for human beings. These details are not much to say, just buy my book and you will know.

The following is the unmanned driving has been talked about, L1 to L5, not much to say. Here comes an interesting paradox, which company is the greatest? Evergreen, a company with good values ​​and low-key hard work will become a great company. But this may not be the case in the age of artificial intelligence. Google is the kindest company. He said that the car can’t surpass the safety of people. I will not launch it. Tesla is a savage company. It throws out what it doesn’t have and lets everyone drive it, causing life danger. Which of the two will win in the end? It’s hard to decide, Googel technology is better, Tesla has more data. The first point of the success of artificial intelligence is the largest amount of data, and it must be the most savage company to get the largest data, regardless of the safety and privacy of users. How do we as humans manage this? Don’t let the most savage company become the company that rules the world, that’s the biggest challenge.

Of course, today’s major companies are paying attention to artificial intelligence. IBM has made a transformation, Google has made a transformation, and has not made a transformation, so this is an inevitable state. IBM has done the medical aspect, but for those who are worried about using artificial intelligence to brainwash people, you might as well search for the word perspective. He was very successful in helping Trump’s election and brainwashing others. It is a super investment in artificial intelligence. Brainwashing you to believe his point of view is a very shocking and frightening thing. What is the basis for this data? Every time you like a like on the Internet or a like on WeChat, how much is a like worth? When you click 70 likes, this technology understands you more than your friends. If you have 150 likes, you will understand you better than your spouse, and if you have 300 likes, you will understand you better than yourself.

Therefore, artificial intelligence is an investment outlet for people today. The world is growing very well, and China is also facing a particularly good opportunity. 43% of the world’s artificial intelligence papers are written by Chinese. So we’re going to pull it in and join us in investing in AI. What is the reason? Because Chinese people are good at mathematics, the core of artificial intelligence is not computers, but mathematics. So if my friends are studying mathematics, or have children studying, let them do artificial intelligence as soon as possible, and the programming level is good, and mathematics is better. So these guys can jump into the field.

The rest of the full text page

Second, Chinese young people love new fields, especially those fields that change the world and can make a lot of money. Artificial intelligence is like this. A doctoral student from Stanford University received an annual salary of US$2 million after graduation. This is very attractive, and traditional Chinese companies like banks are very weak. Artificial intelligence can easily defeat them, and now the Internet is large, users are large, and data And the protection and privacy of data are not as entangled as in the United States, so it is easy to do well. So you want very large data, a single field.

Having said so much about artificial intelligence, I would like to mention a few things that are worrying. Because the investment and valuation of artificial intelligence is definitely in a bubble today, we will be very cautious about investing in those companies. Because artificial intelligence must be added to every business plan, I met an entrepreneur a while ago. He said that he wanted to do artificial intelligence for underwear. There are various ways to identify whether this entrepreneurial project is artificial intelligence. I will write articles on the Internet. Tell everyone. For example, home robots are a big trend. Many of them have eyes, ears, hands and feet. Don’t be a human-shaped home robot, because it can’t be done, because people have too many expectations, your housewife, the child will be Disappointing because it’s not as cute as in Star Wars.

Is AI a platform? Must be made by a big company. Computer vision to solve unmanned driving is ridiculous. It is absolutely impossible for a few computer vision scientists to come out to be an unmanned team. How many parts do you think a car has, and how complex and diverse the unmanned driving will be? It may even be one of the most important, but it may be is one of ten technologies. Don’t think that getting the first place in computer vision means you can do autonomous driving. We are very proud to invest in face recognition, but is there such a big market for four unicorns? Everyone has misunderstood speech recognition. Speech recognition is recognition, not understanding. Hearing your words, it doesn’t make sense unless you are in a small area of ​​specialization. So when I see various words about these things, everyone follows suit, which is very scary. AI is great, but please don’t be misled by what it can do.

In addition, it is very, very difficult for scientists to start a business, and many companies are very selfish. I think the seven black holes in the world are a great harm to artificial intelligence. We need a more open platform, if more entrepreneurs can use artificial intelligence. We need access to data, good algorithms, and good calculations to help you make up for various shortcomings. This is a particularly big demand. The seven black holes are so selfish, are they shameless? In fact, it is not. Of course, each company must do it for its own interests. Otherwise, it is impossible for all seven companies to be closed. We are open, and our innovation workshop is about to open our platform to embrace everyone to do artificial intelligence entrepreneurship. There are also artificial intelligence valuations that are too high and cost too much.

The last point is that artificial intelligence is a toB business. I just talked about using Qianbao to make artificial intelligence a toC. It is impossible to hope that it will bring in tens of millions next year. We are one company to solve one problem, one case, one to get money. This kind of growth is very slow, so when you invest, don’t speculate your valuations too high, and if you speculate, I will not vote.

Artificial intelligence is like a product manager-type entrepreneurship in the mobile Internet era. Let’s talk about the era of returnees in the Internet era, because only they have seen the Internet. In the app era, the industry of product managers, because they use Lean Startup to polish. The O2O era is a start-up that understands back-end technical cooperation. The AI ​​era is actually a collaboration between engineers and toB. Therefore, “Lean Startup” is a collection of startups, but it is not suitable for AI startups.

The innovation workshop is two entities, the right is our VC, the left is the early incubation of artificial intelligence, we provide the platform, provide data, provide technology, come to us to start a business code can be taken away, our entire investment team is you If you are a technical person, you can also create a business consultant, and we will help you match a CEO. These are some of the characteristics of innovation workshops.

The Links:   3HAC3766-1 3HNA004958-001 IGBT