Dr. Zhang Hongjiang: AI is the Next Big Opportunity for Chinese Start-ups

“The world is experiencing the third wave of artificial intelligence (AI), and this AI wave is very different from the Internet wave. The focus of the Internet wave was driven by business models, while this AI wave is driven by technology.” 

At the 2017 New Business Summit, Dr. Hongjiang Zhang, Venture Partner at Source Code Capital, shared his thoughts on the essence of this AI wave, as well as the challenges and opportunities that it will bring.

Dr. Zhang believes that the machine learning, which forms the core of AI, has already shifted from traditional mathematical modeling to relying on big data. This wave of AI will bring more possibilities. By leveraging troves of big data from various industries and the reserves of AI talent, it is possible for China to lead the world in AI technology, which is an opportunity for Chinese start-ups.

As for how AI will affect people’s lives, Dr. Zhang believes that we should be ready for AI to not only assist humans in the future, but also replace or surpass some functions, as some new AI algorithms are already equipped with “god’s view.” AI has already surpassed humans in the capabilities relating to perception, such as speech and image recognition. In cognitive capabilities, however, AI still has a long way to go.

In his speech, Dr. Zhang also reminded founders and investors that as any company can brand itself as an AI company these days, it presents the risk of creating a bubble. If the company only has algorithms and a few experts, but does not own, or has difficulties obtaining data, and lacks application scenarios, then it will not grow in scale.

The full transcript of Dr. Zhang’s speech is as follows:

Mr. Feng, distinguished guests, good morning. It is my pleasure to be here at the Forum organized by 36Kr. I would like to spend some time today talking about the essence of the current wave of artificial intelligence, as well as the challenges and opportunities it will bring.

1. Data as the new religion

If we look at the development of AI over the past 60 years, the core of the AI we talk about today is really a branch of AI – machine learning. More specifically, it is the technology of using neural networks to conduct machine learning, which is the technology of deep learning.

What are the differences between this wave of AI and previous ones? I believe that the AI wave hitting China today is more aggressive than the ones hitting anywhere else in the world. The reason may be that AlphaGo has defeated humans for the first time and that has led us to gain a newfound respect for AI.

The second wave of AI was from the mid-1980s to the early 1990s, but it disappeared so quickly that for a long time, people like me were embarrassed to say that we studied artificial intelligence. An important reason behind its quick disappearance is that although its neural network was very popular, it also died down quickly. There were two reasons: one was that the neural networks back then did not have the support of big data, and two was the lack of support for computational resources.

Today, we not only have new deep learning algorithms, but more importantly, we are equipped with high quality, categorized big data and extremely strong computational resources. From a technology perspective, today’s deep learning is fundamentally different from previous AI methods, especially when comparing to the expert system methods.

Today’s algorithms are in fact driven by data, and not solely reliant on the rules of thumb.

Let us take a look at the two drivers of the current AI wave. First, Computational resources have grown by leaps and bounds in the past 30 years. In the development over the past few decades, we saw exponential growth in the performance of supercomputers and an exponential reduction in their unit price. Second, data, another important support for AI, has begun to explode.

According to research by IDC, the data created by the entire human race will increase tenfold between 2013 and 2020, equivalent to an annual growth rate of around 40%. The data we generate every day has exceeded tens to the 19th power (or ten quintillion bytes).

A large amount of today’s data is created by the Internet. This data is not only large in quantity, but has also been labeled. For example, your phone will record the time, location, and other information about a photo that you took. It is thanks to the explosive growth of this type of data that we can provide better training data for artificial intelligence.

I want to share some of my own experiences. Today’s smartphones can take photos, recognize faces, and not only beautify these faces, but also tell you how many people are in the photo and who they are. That was my dream 20 years ago – to have a mobile device that can recognize the people in a photo.

After 20 years of development, this was finally realized in smartphones. Why it took so long to achieve this is mainly due to the two drivers previously mentioned: computational resources and big data, plus the newest machine learning algorithms. When we first started to work on this problem in the 1990s, we only had a few hundred photos and a little over 100 people in our entire database.

Only when industry giants like Google, Facebook, and Microsoft started to use hundreds of millions of photos to train deep neural networks, and use the method of deep learning to conduct the recognition (around five years ago), did we see an increase in the recognition rate and become able to recognize people in hundreds of millions of photos. So what I want to stress is, when the scale of your training data begins to increase, the accuracy of your learning should start to experience great growth as well. The more resources and data used in training, the more linear the growth to be achieved in the accuracy of the training.

The products made by today’s AI companies are no longer supported by tens of millions of training data, but hundreds of millions. They have wide coverage, from different scenarios, to different environments, to different profiles. That is why a company like Megvii Technology is a facial recognition technology giant not only in China, but also globally.

I want to summarize here that the core of the current AI technology is machine learning, and machine learning has shifted from the classic theoretical modeling to being driven by big data.

There have always been prophets in the history of technological development. Today, I want to introduce a distinguished scholar who previously worked at Microsoft Research, Jim Gray. Ten years ago, he proposed the concept of “four paradigms of human scientific research”: starting from pure observations, to using Newton’s mathematical theory to describe the entire world, to using computational methods to simulate the world starting 50 or 60 years ago, to using data as the fourth paradigm to drive all research today.

It is precisely due to the development in technology and the popularity of big data that we see a lot of companies around the world, including IT and traditional manufacturing companies, already starting to build their businesses on the basis of big data.

Data has become a new religion. As data is now the new fuel, it explains why blue-chip companies such as Intel have been acquiring AI companies.

2. A view from heaven

We just mentioned the two forces driving this wave of AI development, computational resources, and data. Next, I’d like to share my thoughts on the influence of AI on our industries and on our lives in the future.

The first point is that AI will be able to do what humans can do, and do it faster and on a larger scale, like how humans will never be able to play a million games like AlphaGo.

The second is whether humans can be like Tesla, which learns the data derived from the tens of thousands of test vehicles running on the road every day. Another is whether humans can compare facial data captured by all the cameras in an instant. The conclusion is that whether we are looking at speed, scale, or the ability to learn from a group, we humans cannot be compared to these machines.

We should not doubt that AI machines will not only be able to assist humans, but also could be able to replace or surpass humans.

We used to say that machines are logical and that they might be able to surpass and replace humans in mathematical applications. In the areas that cannot be expressed by mathematical equations, but can be done by humans, such as how to drive, how to ride a bike, and how to draw, we can see from the performance of AlphaGo that machines have started to surpass humans even in these areas.

In reality, whether it is in autonomous driving or drawing, machines have already showcased exceptional capabilities. After Ke Jie faced AlphaGo, he said: “we only saw the tip of the iceberg after thousands of years, but AlphaGo has gained the heavens’ perspective.” After many years of training, the peaks climbed by AlphaGo are much higher than what we have obtained. Thus, it sees a much fuller picture than we can, as we humans are only able to see a small segment. It is as if AlphaGo is equipped with the “view from the gods.”

Now, if we think about the future as represented by AlphaGo, we can see that intelligence systems and intelligent machines will surpass and replace humans in the future. In reality, looking at the development of human history or of the Earth’s history, we surpassed many organisms in abilities when we evolved from primates to homo sapiens. That is why many years ago, Alan Turing said that after God created mankind, many animals grieved their fate.

As we see AI surpassing humans, maybe we can think about how these intelligent animals felt after God created man. We have already arrived at this turning point today.

Evidently, it is likely that many industries will be replaced by AI in the future, replacing jobs such as translators, news reporters, and banking analysts. The U.S. government has already assembled a group to cope with the impact that the replacement of truck drivers by autonomous vehicles will have on the American unemployment rate in the next ten years. There are more than 8 million jobs in the U.S. that are related to the trucking industry, while the entire working population of the country is only 120 million. The impact can be colossal.

We hope that in the future, AI will become intelligent assistants, but wanting a machine that can help us but not replace us is just our wishful thinking. In reality, we see that AI will replace us in many, many scenarios in the future.

If you have read the book “Sapiens: A Brief History of Humankind” and liked the author, then you should read his other book, “Homo Deus: A Brief History of Tomorrow.” In it, he proposes that the future development of AI could produce two types of people: gods and idlers.

What is scary is that possibly only 1% of the population will be gods, while 99% will be idlers. When 90% of the work in the world can be replaced by computers, a series of social issues will manifest.

Some people also say that there are three groups of people who can withstand the impact of AI. One is the capitalist. Thus, all the investors and VCs sitting here today, as long as you can get funding and invest in the right projects, then you will have nothing to worry about. The other two are celebrities and craftsmen. The population of the three groups, however, will not exceed 1% of the population. Thus, the acceleration of technological advancement is not only impacting our jobs, but is on a deeper level, impacting cultures, values, and ethics, which is hard to predict today.

Another problem with deep learning is that the decisions made by deep learning will not be explainable as its capabilities continuously increase, especially with the drive of big data. This might be a reality that we will have to accept.

We need to consciously understand that today, we are talking about intelligent machines and not the intelligence of machines, or that the intelligence of machines is still far behind the intelligence of humans. We see today that the Chinese market is filled with AI fantasies, and there exists many bubbles in this area of investment. Fundamentally, people have not clearly understood the strengths and weaknesses of AI today.

Nevertheless, I want to tell everyone that in most situations, the AI algorithms of today have already exceeded humans in the areas of perception, such as voice and image recognition. In the areas of cognition, however, AI still has a long way to go. Thus, if anyone tells you that he/she has created an AI system that can model human cognition, then you should probably think twice about what he/she said.

3. Bubbles and Opportunities

Since this wave of AI is different from previous ones, then as industry professionals and investors, how should we make our judgment on this AI opportunity and its related investment opportunities?

There was one great point made by Mary Meeker in her report: in the agrarian society before the 18th Century, people’s survival depended on planting, harvesting, and handicraft. By the Industrial Revolution of the 19th and 20th Centuries, people relied on machines and industries. The people of the 21st Century will rely on the unification of computational power and human power.

When we invest, we need to think about which areas of AI will better allow humans to realize their potential. In the past few technological waves, companies that provided platforms have always emerged.

We need to look at which companies have large scale applications and can also generate large amounts of data. When you have data, and when the data is also large in size, AI can be introduced naturally to improve overall efficiency. It can further improve productivity, create new applications, and thus create sizeable companies.

We must be clear on one thing when we evaluate an AI-related investment: this wave of AI is significantly different from the previous Internet wave. The focus of the Internet wave was driven by business models, but this AI wave is driven by technology. AI starts with overturning existing industries and improving the efficiency of these industries, not making them disappear. Therefore, it must be closely integrated with vertical industry applications.

All investors must know that any company can label itself as an AI company these days, presenting the risk of creating a huge bubble. We have to be clear on a few things: if this company only has algorithms and a few experts, but does not have data or application scenarios, or it will be difficult for them to obtain data in the future, then these companies will not grow in scale, nor will they last for long.

When you look at these AI companies, you need to understand clearly if they currently hold data, if they can continuously generate, extract and control data, and if they can possess more data than others. These are the basics. Only when technology is combined with the ability to extract data will there be a solid barrier.

Let me give an example. We are all aware of the fast growth of Toutiao in the past five years, but what is the reason behind it? The reason is that it solved a basic human need and connected people to information. Thousands of years ago, our ancestors used knots on ropes to record events. After the creation of paper and printing, the distribution of information became even more convenient. Now in the PC Internet era, all the information is knitted together and presented in the form of the Internet.

As we arrived in the smartphone and mobile era, the information device is now always in our hands. It is much more convenient for people to acquire information, but this has also created information overload. The speed of information generation is much faster than information consumption. The search engines of the PC Internet era will no longer satisfy people’s needs, because it is not enough to ask people to search for information anymore – you should push it to the users when they need it instead. This is exactly Toutiao’s business model. Its core is technology. Toutiao uses its AI algorithms, big data algorithms, and the data generated by your computer to understand what exact information you need, and when you need it.

Toutiao’s capabilities in the areas of big data are quite scary, too. Currently, it has over 200 million monthly active users, over 100 million daily active users, and the daily app usage time is over 76 minutes per user. Through technology and data, Toutiao has built itself a natural barrier. In the world of mobile applications, only one app has a longer average time spent per user per day, and that is WeChat at 90 minutes.

Once we realize the potential of this AI wave, once we understand the value that can be created by this wave of AI, these are the three points to remember when starting up a company. First of all, without a doubt, AI is the core competitiveness of the future. Second, there are currently three AI business models: to develop one’s own products, to sell technology, and to offer AI services.

Currently, most companies fall into the second category. They have some technology and a few talents, but they lack data and application scenarios. Thus, they are only providing consulting services, which is not scalable.

Last but not least, AI is an opportunity for Chinese start-ups and an opportunity for China to lead in the world in this area. In the mobile Internet era, China’s WeChat and Toutiao are already world leaders. In the field of AI, there are two barriers to entry: data and talent, neither of which China lacks. In all of the academic journals around the world, the number Chinese authors writing about machine learning, specifically in the areas of deep learning and neural networks, has already surpassed American academics as of 2014. Moreover, the amount of labeled data in China is already the world’s largest. We have a large base and a large amount of data.

Thank you, everyone.