Kai-Fu Lee: I attended 1,400 video conferences this year and didn’t make a single friend | MEET2021

The editorial department compiled from the MEET 2021
Quantum Bit Report | Public Account QbitAI

What is the biggest impact of the epidemic on the AI ​​industry?

What are the most exciting new AI technologies and applications this year?

Even in the next 10 or 20 years, what will be the direction of AI development?

…

Standing at the end of 2020, such an extraordinary year, the AI ​​industry has too many questions like the above that urgently need to be answered.

At the MEET2021 Smart Future Conference, Kai-Fu Lee, Chairman and CEO of Innovation Works, gave an in-depth interpretation of the "restart, reshaping and reconstruction" that is taking place in the AI ​​industry in a conversation with Li Gen, editor-in-chief of Quantum位.

In order to fully present Mr. Kai-Fu Lee’s latest thoughts and ideas, Quantum位 edited and sorted out the content of this conversation without changing the original meaning. I hope it can bring you more inspiration and thinking.

About MEET Smart Future Conference: MEET Conference is a top business summit in the field of smart technology hosted by QuantumBit, dedicated to exploring the implementation and industry application of cutting-edge technologies. More than 20 top industry experts including Kai-Fu Lee shared their views at the conference, more than 500 industry audiences participated, and more than 1.5 million netizens watched the live broadcast online. Dozens of mainstream media including Xinhua News Agency, Sohu Technology, The Paper, and Cover News reported on the event, with a total online exposure of more than 20 million.

Key Points

• The biggest impact of the epidemic is that it has accelerated dataization, onlineization and AI.

• The epidemic is a huge disaster, but it has also greatly promoted the further implementation of AI. This year may be equivalent to three years of progress.

• The most exciting black technologies are AlphaFold and GPT-3, the latter of which has greater potential to disrupt the industry.

• The greatest value that AI brings to society is definitely not black technology.

• Embracing AI is everyone’s biggest opportunity, and the sooner you embrace AI, the greater competitiveness you will have.

• The new infrastructure will further promote the new AI economy. This will be a new era and a new Renaissance.

• AI may be able to solve the problem of finding a partner for the next generation better.

• AI practitioners need to remember: with great power comes great responsibility.

(The following is the full text of the conversation between Kai-Fu Lee and Gen Lee)

The epidemic has accelerated the development of AI, with one year equivalent to three years

Li Gen : Thank you again, Professor Kai-Fu, for helping us summarize this year. What is the biggest difference this year compared to previous years?

Kai-Fu Lee : The pace of work has changed this year. I have participated in a total of 1,400 online meetings. It was very efficient and I held a lot of meetings and speeches.

But after 1,400 such meetings, I didn't make a single friend.

So I think that although online meetings have brought efficiency, we still need to return to the essence of face-to-face communication .

Li Gen : But this process affected by the epidemic is also a process in which onlineization is accelerated?

Kai-Fu Lee : This is the biggest impact of the epidemic - it has accelerated the development of data, onlineization and AI.

The most important core of AI is massive data, which can catalyze more new applications.

For example, in online meetings, it is not always easy to distinguish whether the person in the video conference is real or not. So sometimes when there is an online lecture or the network is unstable, we will choose to record it, and once recorded, it can be used many times.

By extension, in the future, when you are doing sales, customer service, or even TV shows online, you can actually use virtual people to replace them.

We are not yet able to create a 3D virtual human that you cannot tell apart, but it is certainly possible in the video.

DeepFake is not an application for good, but if its technology is applied in a constructive and beneficial direction, I think it can bring a lot of applications.

And with data, all industries can evaluate which tasks or processes can be further digitized.

After digitalization, repetitive work can be replaced by AI.

In addition to the shortening of social distance, we can find that robots are appearing more and more frequently. For example, when we go to a restaurant, robots will deliver food to the guests.

Many robot products are widely used, whether in warehouses, factories, or even in the apartment where I live, robots are used for every final delivery.

The situation is similar in the United States. There was no Alipay or Meituan in the United States before, but now they are both available.

So to sum up, I think the COVID-19 pandemic is a huge disaster, but at the same time, it has also greatly promoted the further implementation of AI, bringing about digitalization, IT, and AI. Perhaps this year is equivalent to the past three years.

The greatest economic value of AI is not "black technology"

Li Gen : Are there still exciting technological advances this year?

Kai-Fu Lee : The most exciting technologies this year are DeepMind's AlphaFold and OpenAI's GPT-3.

Among them, I think GPT-3 may have more potential to disrupt the industry.

It doesn't necessarily mean using it to generate novels or something, but rather using extremely large amounts of natural language data to train a pre-trained model. This model can be applied to many new fields, such as writing poetry, having conversations, or writing novels, etc.

Moreover, once this new technology was launched, other companies quickly adopted it, such as search engines, because this technology has been able to improve the accuracy of several points; in addition, OCR is improving, customer service is also improving, and Microsoft Xiaoice is also continuing to improve the accuracy of conversations.

So I think this kind of pre-trained language model may enable us to train a huge model at one time and then quickly generate AI applications.

Of course, there are many problems:

• It costs 12 million US dollars to make a model. What should ordinary startups do? What should professors and scholars do?

• In addition, training a model consumes a lot of energy. What will happen to our energy-saving future?

• Also, will the giants create a monopoly? Only giant companies can make such a huge model.
  

Of course, GPT-3 itself has not yet verified enough commercial value, but there is no doubt that it is the biggest technological innovation.

However, I want to emphasize that the greatest value or economic value that AI brings to society is definitely not this kind of black technology .

There will be a lot of black technologies, and we are also continuing to invest in projects such as robots.

However, I think AI technology has matured today and can be embraced by traditional companies.

For example, AInnovation, which has just become an AI unicorn, can see that AI can bring immediate and huge effects to traditional industries, whether it is manufacturing, health, medical care, retail, finance, insurance, logistics, etc.

So today I am here to appeal that embracing AI is everyone’s biggest opportunity, and the sooner you embrace AI, the greater competitiveness you will have ; the later you embrace AI, the more your living space may be compressed.

Li Gen : How can we get everyone to embrace AI?

Kai-Fu Lee : We still need data. I have said so many encouraging words, but in a pragmatic way, will traditional companies bring value if they embrace AI this year? Most of them will not.

Because AI consumes data, companies are required to have massive, structured, labeled, and accurate desensitized data that can achieve business goals. It is best if only you have it and no one else does. Then your opportunity will come.

What kind of companies have it? For example, banks and insurance companies naturally have it, and data in the financial field must meet the requirements.

Some areas will be a little more difficult, such as industrial manufacturing. You need to first replace your industrial manufacturing with robotic arms or unmanned forklifts. You may need to place a lot of sensors to understand what is wrong with the machine.

So it is easy for some companies, but difficult for others. If you have data today, think about how to apply it; if you don’t have data, think about how to accumulate it.

New infrastructure promotes the new AI economy

Li Gen : We were just talking about this year, and the new infrastructure is believed to lay the foundation for the next decade of development?

Kai-Fu Lee : Looking back at the miraculous rise of artificial intelligence in China, we can find that entrepreneurs, mobile Internet, data, and high-tech talents cultivated by China's excellent universities are the most important.

But in fact, there are very wise policies. This is not an easy thing to do.

This has been discussed in many countries. For example, in the United States, the Obama administration strongly supported a new energy company, but it failed to succeed and was criticized a lot.

China is in stark contrast. China's overall science and technology policy should allow the private sector and industry to do what they are good at and compete through the market economy, while the government helps them do what the private sector cannot do.

For example, the rise of our AI is largely due to the support of government-guided funds in various places, which allows people to selectively become LPs and help the early AI industry to be established.

In addition, the implementation of the "new infrastructure" policy is also a great benefit. In addition to 5G and big data centers, the new infrastructure also includes the implementation of IoT, artificial intelligence, smart cities and other aspects. The favorable policies will further drive the transformation and upgrading of China's economy towards digitalization and intelligence.

So, when the government does the above things and forms a platform that startups and large companies can use, I think this is very wise.

Of course, how big is the scope of AI in the new infrastructure and whether some new technologies have the opportunity to be included? It may be worth further discussion.

For example, new technologies like GPT-3 require a lot of resources to train in order to further promote their potential. We have found that AI training is becoming more and more expensive, and it is no longer affordable for a team of professors or startups.

It is worth looking forward to whether there will be an opportunity to promote such technologies with such potential for explosive growth from a higher dimension.

Li Gen : You have recently put forward the "AI new economy", so is it time to use this new concept to define this era?

Kai-Fu Lee : Yes, although it is a very positive description, it is time. I recently wrote a column in the New York Times, in which I talked about how the epidemic mentioned earlier has promoted dataization, IT, and AI.

The epidemic has promoted the digitization of various industries. In fact, a hidden concern is that "work has been digitized" and can be outsourced or replaced by AI.

When we have so many robots, they will also replace many blue-collar jobs. The same is true for white-collar jobs. Many back-end and BPO jobs, document processing, or customer service jobs will be challenged.

Some people would say that we should develop technology more slowly, but this is absolutely unrealistic, because no technology will slow down .

On the one hand, we need to embrace these technological advances in the new AI economy and make them reach every corner as soon as possible. But at the same time, we need to take targeted measures to deal with the unemployment problems it brings.

So my article mainly talks about how to quickly make society aware that complicated, unskilled and repetitive work will disappear, and then how to retrain this group of people to help them quickly find new directions to do more valuable and better things.

At the same time, we can spend our precious time on things that are more valuable and that we love more, such as art, charity and other creative and decision-making work, or spending more time with family and friends, making new friends, etc.

This will bring out the brilliance of humanity. I call it a new Renaissance , where people can find meaning again - why do people exist?

Globalization is still the inevitable path, and Chinese software will account for half of the world

Li Gen : The last Renaissance was accompanied by the beginning of the first globalization, but the process of globalization seems to be hindered in recent years? Chinese entrepreneurs want to do business with 8 billion people, but it turns out to be difficult.

Kai-Fu Lee : Globalization is the inevitable path. Many "America first" people say that China has no innovation and that only Chinese people use Chinese software. But everyone here knows that China has undergone tremendous changes in the past 10 years. This statement is completely wrong today.

Because China's innovation ranges from TikTok to Huawei, to Xiaomi, to Meituan, and there are many other innovative companies, including our four AI dragons, which are not available in the United States.

So we should be very proud that China's innovation today has the opportunity to overtake others at least in the fields of IT and AI. The United States is stronger and has more advanced technology in some areas, but we are faster in commercialization. Under this premise, we will naturally go global.

How do we go global? Here I have to be more pragmatic. In fact, there are huge differences in usage habits, language and culture in many countries.

If we want to push all the good products from China to Europe and the United States today, this is actually unrealistic because the usage habits in Europe and the United States have become solidified.

It is difficult to change the system they use, whether it is To C or To B, but I think the users in other countries and regions, such as those from Southeast Asia to all countries and regions along the “Belt and Road”, including the Middle East, Africa, South America, etc., are actually more like our young users.

Moreover, their habit of using American products has not yet been solidified. At the same time, American companies do not pay much attention to these markets. However, we believe that the economies of these countries will rise in the future.

So I think what you just said about “moving towards 8 billion” is unrealistic, but moving towards 6 billion is possible, which should be 1/4 of the world's GDP and 3/4 of the population.

Our entrepreneurs are willing to go global, so I think going overseas will become very important.

Our software exports, technology exports, and artificial intelligence exports will definitely happen.

When my book "AI: The Future" was published, I said that within 10 years, China will account for half of the world's software.

At that time, many people thought this was a very optimistic prediction, but now it seems that we are moving in this direction, and we had not seen products like TikTok at that time.

But we have to be cautious. It’s not that countries like Europe and the United States are still quite powerful, but TikTok’s success in Europe and the United States is not so easy to replicate.

“+AI” will bear fruit in various industries

Li Gen : Professor Kai-Fu Lee just mentioned that it is becoming increasingly difficult to start a technology business. Will you still invest in technology companies?

Kai-Fu Lee : Of course we will invest in technology companies. Everything I just said is our investment strategy. I have said so much about being optimistic about robots, and I have said that I am particularly optimistic about large language models. I will continue to invest.

It’s just that we can’t expect as many as in the past. In the past four years, Innovation Works has produced a total of 7 AI unicorns in the field of AI. Our small goal for the next four years is to set 3 AI unicorn companies first.

Li Gen : But is it a trend for AI to penetrate into traditional industries? Recently, Sinovation Ventures has also invested in companies in the "agriculture" field, which was rarely seen before.

Kai-Fu Lee : You mean XAG? But XAG cannot be defined by agriculture. XAG's products are high-tech drones and autonomous driving, and the scenario is to industrialize agriculture.

Industrial manufacturing is very different from agricultural production. For example, if you want to do smart manufacturing, use AI to make robots, or make an iPhone, clothes, or shoes, the requirements in different scenarios are completely different.

However, agricultural sowing or fertilization scenes are easier to standardize. For example, cotton fields and rice fields are easier to batch and large-scale operations.

This is true not only in China, but also abroad. At present, the penetration rate of AI in the agricultural market is only about 5%. We not only highly recognize XAG itself, but also think it is a very good international product. This is because agricultural scenarios, such as sowing seeds, are almost universal around the world, without too many language barriers.

Li Gen : It can solve the food problem for 8 billion people.

Kai-Fu Lee : Yes. (Laughs)

Li Gen : We just talked about the topic of "+AI" in traditional fields. Let's expand on this topic. Could you please talk about how to "+AI" in the following fields? For example, "online education" has attracted the most attention this year.

Kai-Fu Lee : China is doing very well in online education, which is very interactive and children really like it. But for some courses, especially for children, using some cartoon AI characters may be more effective.

We can imagine that in the future, there will be some cute cartoon characters as children's teachers, and some classmates may even be virtual classmates. It will create a good atmosphere and be more efficient.

In this case, not only will your costs be lower, but the children will also get higher test scores and have a higher interest in learning. Therefore, the human-computer interaction model after "+AI" can not only reduce costs and improve efficiency, but also increase children's interest.

Li Gen : This field doesn’t sound very challenging for Professor Kai-Fu. Let’s ask a difficult one. How can we add AI to a cocktail?

Kai-Fu Lee : In fact, machines can definitely make cocktails better than humans, because cocktails have different ingredients and machines can calculate more accurately.

However, we have to consider that if we go to a bar to drink, the main purpose may not be the wine, but to have a very understanding bartender to chat with you.

So I think in the future the bartender will just need to chat with you, and the drinks will still have to be made by machines. Ultimately, machine bartenders cannot replace the communication and companionship between people.

Finding a date can also be "AI-enabled", matchmaker AI knows who is the best match for you

Li Gen : I wonder if the problem of finding a partner in a bar can be solved by using the “+AI” method?

Kai-Fu Lee : This may come true in the next generation - but not my next generation, because my next generation already has a partner.

The way your next generation will find a partner may be completely different from today's traditional way. Just like today Meituan knows what you want to eat better than you do, Ctrip knows where you want to go better than you do, and Toutiao knows what you want to watch better than you do, in the future your matchmaker AI will definitely know who is the best match for you.

When everyone looks for a partner in their lifetime, they probably look among the people they know. No matter how you count, the number will be tens of thousands, or at most hundreds of thousands.

However, in the future, AI may understand you better, understand what kind of person you are, and understand what kind of person you would like. The people it recommends to you will definitely be more reliable and more suitable than those you find yourself or those recommended to you by your friends.

In other words, the range of your choice may not be from among tens of thousands of people, but from among 8 billion people.

Of course, the final choice is in your hands. Today, you rely on your friends to recommend five candidates for you and you choose one; in the future, the five candidates recommended by AI will come from 8 billion people and will be closer to your needs.

So your children's future will definitely be happier. (Laughs)

AI practitioners must also take the "Hippocratic Oath"

Li Gen : Thank you, Professor Kai-Fu. Time is almost up today. Finally, could you please give us some advice from the perspective of tomorrow? For example, from the perspective of 20 years from now (2041), what responsibilities and missions should we have?

Kai-Fu Lee : For AI workers, we all underestimate the impact of AI ethics today . We may see online discussions about privacy and fairness issues, or that large companies have too much data and influence our thinking.

But as big companies gain control of AI data, these problems are becoming more and more serious. How do we deal with them?

I think we must be AI practitioners, just like medical practitioners have a Hippocratic Oath , which states that I will regard human life as the most sacred thing throughout my life and I must defend it.

Being an AI worker is not actually the same as being an engineer. You have to consider that every algorithm you make, and every contribution you make may result in unfair treatment due to a certain race or gender; it may lead to some people being unable to be hired by the company; it may cause some people to be unable to climb up in their lifetime; it may cause some medical misdiagnoses that harm people, or cause self-driving cars to hit people.

Therefore, every engineer should have a sacred oath and should use various tools seriously to ensure that the software they produce brings progress to mankind rather than various adverse consequences.

I think we should take our responsibilities very seriously. With great power, comes great responsibility. I strongly agree with the saying “With great power, comes great responsibility.”

In fact, AI engineers and practitioners have enormous power and must take their responsibilities seriously. At the same time, ensure that future users receive help and benefits, rather than being harmed.

One More Thing

Finally, there is a small segment, and we will have this year's quick question and answer session.

Li Gen : What is the most successful personal stock investment this year?

Kai-Fu Lee : Zoom.

Li Gen : What book or movie do you want to recommend most this year?

Kai-Fu Lee : The Queen's Gambit .

Li Gen : What was the most inspiring exchange this year?

Kai-Fu Lee : One-on-one conversation with Yuval Noah Harari, author of “Sapiens: A Brief History of Tomorrow”

Li Gen : Which is more reliable, Michelin or Black Pearl?

Kai-Fu Lee : China's black pearls are reliable, but foreign countries still have to rely on Michelin.

Li Gen : Why was I not optimistic about Musk’s million RoboTaxis in 2020 from the beginning?

Kai-Fu Lee : Musk has a famous quote that I think best describes his personality.

He said that all his predictions will come true, but don't listen to him about the time. So I believe he will succeed, but it may take 5 or 10 years. (Laughs)

(Finally, if you want to watch the entire conference, please click to read the original text)