This article was originally published by AI Frontier.
He studied 5,000 AI companies and said that’s what AI apps should do!

Hi, my name is Henry Shi. I am a serial entrepreneur with a PhD in ARTIFICIAL intelligence in the US, and I also focus on early stage investments in AI (MANAGING Partner of AI List Capital). Most of the companies I invest in are early-stage companies in the United States. However, some of them are now in the Chinese market and have been deployed in many cities in China, so I pay close attention to the application of AI in different industries in the United States and China. We also did a lot of research, including our team studying more than 5,000 AI companies around the world to analyze their technology and business models.

I’m also a content partner at Avaricious Technologies. We are an artificial intelligence and education company, founded in Los Angeles, USA. The Greedy Academy under Greedy Technology is China’s first adaptive learning platform with AI and big data content as its main content. The platform provides the most professional and standardized AI course system, and provides each user with a customized learning path through artificial intelligence technology. The era of artificial intelligence brought about by the Internet revolution is rapidly reshaping our understanding of the classroom, education and even human learning patterns. We believe that the combination of artificial intelligence and online education will bring about great changes in the education industry.

I am committed to providing the platform of this company with very high-quality artificial intelligence content, so that the most cutting-edge talents and some scientific and technological information from the United States can be better spread to China. Greedy Technologies is a company invested in by the AI List Capital fund. You can follow the public account “Greedy Technology” to learn more about the company, with a QR code attached at the end of the article.

No matter you are AI entrepreneurs, investors, or interested in AI technology enthusiasts or business people, I hope that what I have said can be of certain reference significance to everyone.

Today, I will focus on the following aspects:

  • The first is the basic understanding of AI, from an investment perspective, how we perceive AI;
  • Second, a brief introduction to some dimensions of AI company analysis;
  • Third, I would like to discuss the application of AI in vertical industries and the comparison between China and the US.
  • Finally, I will give you some advice on starting your own business.


Basic cognition of AI

The basic understanding of AI is very simple. We believe that the core of AI is data-driven to improve productivity and production efficiency. If it meets this condition, we basically think it has the characteristics of AN AI company, that is, as long as it obtains data and uses it to improve its overall productivity and production efficiency, we think it has AI elements.

The second thing I learned about AI is: AI is an inevitable outcome of the development of the Internet or mobile Internet, because the mobile Internet and the Internet created a large amount of data, so for now opportunities of the Internet, mobile Internet should be limited, today you want to do a the unicorn to a company in the field of the difficulties, there are a lot of opportunities, but AI I believe you also are very concerned about the AI company financing, are AI company in China for very large financing, it’s the whole valuation push up very fast, it may speed than before the Internet, mobile Internet company much faster, it give us the opportunity to created the very big, both entrepreneurs and investors.

Third, I would like to analyze two common types of AI companies. The first type is to directly use AI to solve problems. One representative here is unmanned driving. The second type of companies are intelligent after data accumulation, or the development of Internet and mobile Internet companies will naturally become AI-based. For example, LinkedIn and Facebook can use the data to make intelligent friend recommendations after they have accumulated a large amount of data. They will be more and more intelligent in the future. In the future, many Internet and mobile Internet companies will naturally become AI companies.

Here are a few examples that relate to both types of AI companies. Here most of the companies are our investment, but it is also in the subsequent round by the United States is the mainstream of the investment institutions to led, at the same time, they are good examples of AI in the different vertical industry, I think this a few companies in the vertical industry applications of AI provides you with some concrete examples.

The first company I want to talk about is called EverString, which uses AI directly to solve problems.

So what did EverString do first? In fact, after they started about five or six years ago, they got invested in Real Time fund very early, because EverString and most of them came out of Stanford, and the first thing they did was to help VC and PE identify potential investments, which was in 2013, and this is what you see here, In fact, he integrated the information of many global companies and presented it on a map in a visual way to help VC and PE find valuable targets. For example, the circles on the map may represent those companies with large size and rapid growth.

Through this kind of interesting big data visualization, the company can find some fast-growing fields, or some fields that actually have a lot of opportunities, but are not targeted by mainstream VC and PE for the time being. In general, this company does the same thing.

But it was actually doing is relatively limited, such as they later found that the market is too small, although they provide something valuable, but few VC and PE to bigger pay for him, then they put the product upgrading to a big market that is hundreds of times for dozens of times, namely AI plus the market sales, Use AI in sales. Use AI to identify potential business customers.

Let’s take this EXAMPLE of VC looking for projects and think deeper.

VC is looking for investment projects, investment projects can be considered as VC potential customers, in this sense you can think of it as a B2B company, B2B companies want to find good corporate customers, VC is only a small category of B2B companies. In fact, there are a lot of B2B companies that have to find enterprise customers. For example, this graph shows some of EverString’s current customers, including Salesforce, Oracle, IBM, and so on. These are very large enterprises. They have to find a lot of corporate clients.

EverString do thing is that when companies have a lot of potential customers, in my this picture on the left, that is the gray head, that is to say every head may be a potential corporate customers, then you may have many, many, for example the tens of thousands of potential customers, you come over to an enterprise, As a matter of fact, for an enterprise, I will talk with these customers and sell products. It is very important to select the right enterprise customers to sell products. This will improve efficiency and reduce the repeated and wasteful work of sales staff, which is equivalent to improving their efficiency and reducing their waste of a lot of time on unlikely customers.

After acquiring many potential customers, EverString moved on to the next step: matching and analyzing existing customer data in the enterprise, such as analyzing the Web Bhev of those potential customers. Take Amazon as an example. I found that a customer may be recruiting some people who are capable of doing cloud services and cloud computing on the Internet. It is likely that this company has a great demand in cloud computing and cloud platform. For example, a company may have just raised a new round of capital, and it says that it is going to make important strategic development in some areas. Then the analysis of the data can also provide this kind of screening information for potential customers.

In general by among a large number of customers to pick out some very potential customer, which is a combination of CRM and some users online behavior, finally he scores can give a lot of potential customers, is there this picture in the last step, every potential users called him a number of scores, the higher the score is, the more likely it is that Then you can have your sales people talk to those customers first.

After such a transformation and upgrading, EverString is developing very fast. They basically want to enter the unicorn industry in the United States, and now they are also a B2B company with great potential in the AI field in Silicon Valley.

To summarize the above example, the company has two attributes: first, a vertical industry application of AI plus sales; The second is that the company directly uses AI to solve problems.

Let me talk about another security company that we invested in that also uses AI to solve problems.

The company uses AI to defend itself against automated cyber attacks. What is an automated cyber attack? According to an authoritative study, in fact, more than 90% of the login page visits come from automated programs, also known as bots. What does that mean? For example, if you are an e-commerce website similar to Taobao, the scale is relatively large. In fact, a large number of login behaviors, user names and passwords are not from people, but from the machine program. What is the purpose of this login behavior of the machine program? To steal the account, of course. How?

In fact everyone a lot of time in different web sites, registered user name and password when using the same set, but many sites of it security is very poor, for example a lot of hacking programs could be to some security more bad of the BBS, website, and then put your user name password information leaking out, In addition, there are a large number of bots with these information to test on many mainstream websites, such as e-commerce, banks, airlines and so on. It is easy to steal the information. As long as the user name and password are paired successfully, it will log in, which may cause great losses to you. Shape Security is here to solve this problem.

Who are Shape’s customers? For example, taobao, the purpose of Shape is to help Taobao to analyze my Taobao login these people are people or Bot. If it is a Bot, it needs to be blocked. This Shape is also developing very fast, and it is almost reaching the rank of the unicorn. Its investors are all top VC in The United States, such as PCB, Google, Ventures, etc. This picture is a good embodiment of Shape.

Did you see the picture above, the black area is that the site visits, so before starting Shape, site although the traffic is very high, but a lot are produced by Bot, the Shape after the operation, Bot landing is big filter, the traffic is very normal, is basically comes from people, Shape did just that.

You might ask, what does this Shape AI use? The AI here is mainly to determine whether the login behavior is from human or Bot, which is not easy to judge, and now Bot is becoming more and more intelligent, it actually uses AI to counter AI, which is similar to this anti-virus software and so on, but the problem that Shape solves is that many existing firewalls. Anti-virus software and so on is not quite able to solve. This is why Shape’s customers are all the top ones in The United States, including the biggest banks, the biggest airlines, and other large customers like Starbucks. It is also a very good application of AI in the vertical field, that is, security, and it is directly using AI to solve the problem, to solve the problem that this program attacks.

The third company I want to talk about, ObEN, which was one of our early investments, also had AI to solve problems. What ObEN is doing is interesting, it’s an application of AI in the entertainment industry, and it helps you build artificial intelligence avatars.

For example, if you look at this first picture here, this is the two founders of ObEN, and they give them a picture, and it generates his ARTIFICIAL intelligence avatar on their right, and the avatar not only looks and acts like him, but also sounds like him. Maybe you are also more or less familiar with this technology, for example, in many movies and TV applications.

The core of ObEN is its ability to build this avatar in a very fast way and make it entertaining. For example, it needs two minutes of audio to model sound; It takes only a few photos to create this image of the avatar expression. Of course, the more data, the better it will be, but it can be generated quickly.

For example, here you can see most the right side of this picture, is it the company to the Indian, the founder of the virtual image is constructed, made him in a somewhat like the singing of BR, AR scene, but also let him sing Chinese songs, or the person own this tone, these are the application of it. It can also allow you to sing jay Chou’s feeling, even very like it, it can also allow you to build Jay Chou’s voice, let him sing another singer’s songs, and even let Jay talk about guo Degang’s cross talk, all of which can be done.

ObEN develops very fast, and its investors are very strong, including SoftBank, Tencent, Chinese Culture, SM Entertainment and so on. SM Entertainment is a very famous entertainment company in South Korea. Last year, ObEN and SM Entertainment jointly established a joint venture called AI Stars, which is also called Magic Stars. The core is to use AI to provide virtual image services for many Stars of SM.

Why, you might ask, did celebrities create this avatar? Actually one of the purposes is to be able to interact with the fans better, you can imagine, the future every star, he has a virtual image, you can use an APP to interact with him, for example you can interact and jay Chou, can you ask him a lot of problems, he will try to give you the answer, some problems he may answer is bad, But he probably answered it for him through his agency backstage, and then he continued to learn and interact better with people. And fans to interact with him, he can send jay Chou the voice, and jay Chou’s unique to this kind of expression, etc., and even users can point song, let him sing what song he’ll give you sing what song, this is very valuable, is willing to pay for his many fans, it should be said that created a new entertainment experience.

The whole development of ObEN is also very fast, in addition to the cooperation with SM, for example in China it is currently working with SNH48. There are a lot of apps out there, including partnerships with wechat and other companies. Therefore, ObEN is also an AI application company in the vertical industry that I mentioned. It applies in the entertainment industry and directly solves this problem with AI. And now ObEN, which is laying out the whole blockchain, has launched the world’s first distributed social AI platform, integrated with blockchain.

You can check it on the net, their news recently also is very active, it is artificial intelligence and chain blocks, a good combination, I believe many of you here may also know block chain recently a fire may be based on artificial intelligence reached last year after tuyere, and achieve the tuyere is another technology this year, There are a lot of exciting ways to combine AI and blockchain. ObEN is at the forefront of this, and I believe there will be a lot of great companies in this area soon.

The next ai company I’m going to talk about is Honey. Honey is the second type of AI company I mentioned earlier. Honey is an Internet company, but after it acquired a lot of data, it naturally became an AI company. Or it naturally does a lot of what AI does here.

What does Honey do? It’s actually an American company, and it fits the American market very well. It’s a browser add-on that automatically finds the cheapest coupons, called Coupons, for your purchases on more than 6,000 U.S. e-commerce sites. These are coupons that are being rolled out overseas in the United States and Canada and so forth. At the same time, it is a very important way for e-commerce to promote themselves. Many people will find shopping coupons when checking out, but it takes a lot of time in this process. Honey saves very little time. It only takes a few tenths of a second to find the items you buy and automatically fills them with the cheapest coupons in the entire network.

For example, on eBay Check out, Honey analyzes your purchases, finds up to a hundred dollars in savings, and says, “I’ll fill it in for you.” What it does is, the concept is very simple, it saves people a lot of time trying to find coupons, and it automatically searches for what should be the cheapest coupon on the web.

Honey was a very simple concept, it was an Internet company, even it was a browser add-on, but it grew very quickly. Its profit model is relatively simple, that is, sales share, because it is the last step of the flow, so the general e-commerce will give it three to four percent of the commision in the industry. Therefore, Honey has experienced a rapid growth in the last two years, and now it has more than six million users. And its net revenue is over $10 million a month, which is a lot of revenue, but it’s based on a very simple browser plug-in.

It’s a company we invested in very early, and it’s fully profitable now, and it’s growing very fast. It was originally an Internet company, but it got a lot of authorization. At that time, it began to make use of AI. The application of AI in this field mainly consists of two layers:

First automatically grab coupons, it is also used in some natural language processing, because coupon it may often in a different form, have a plenty of structured data, some unstructured data, just will appear, so Honey need to analyze these data, equal to the coupon out to its database, This is a very simple application of AI, and maybe you don’t even call it AI, you call it an intelligent crawler;

So the second piece that it’s building now is called the Smart Shopping Assistant, which means Honey has a lot of data, Honey has data about what you spend on 6,000 e-commerce, Honey knows that you probably looked at something on eBay, and you ended up buying it at Target. Think of eBay and Amazon, which also have a lot of data on their users’ e-commerce spending, but it’s second only to their platforms. For example, eBay doesn’t know that some people look at my things on eBay and then buy them on Amazon. They don’t have a lot of such cross-e-commerce data, but Honey does have it. When people use Honey, it certainly needs the same Honey to get its data. Otherwise Honey won’t know exactly what you’re buying or what coupons you’re using.

Honey have a across more than six thousand user behavior data on electricity, it is more able to help you to more smart shopping, you can imagine how many electrical manufacturers now have the function of automatic goods recommended by the, but it is actually electricity as a starting point, the core is better to let the customer to pay, recommend it something better. However, Honey can be regarded as an intelligent assistant from the perspective of customers. It knows your consumption behavior in many e-commerce and what you like and what you are sensitive to. Therefore, it is also his future AI application, which I think is a big one, and it is also created by him in Gravity. Honey, as an example, combines AI and consumption. It is different from the previous companies in that it does not use AI to solve the problem directly. It automatically generates such intelligent application scenarios based on the accumulation of large amounts of data.

We’ve actually just shared four companies with you, and that’s it. Let me just give you a quick reminder:

  • EverString is an AI company that solves this problem in sales;
  • Shape Security is a combination of AI and Security;
  • ObNE is a combination of AI and entertainment;
  • Honey is an Internet company upgraded to a company with AI attributes. It uses AI to solve problems in consumption.

So these are four companies, and I give you an example, which I believe may play a more vivid role in your understanding of AI’s application in vertical industries.


Dimensions of AI company analysis

Next, I would like to share with you the possible dimensions of AI company analysis as an investor. Of course, there are many dimensions of AI company analysis. Here I would like to focus on two dimensions, which may be helpful for you to understand AI company and entrepreneurship.

The first dimension is value.

The question of what value AI plays in this is important because it affects the business model and valuation of the company. And of course from an investment point of view, an investment especially an early stage investment like ours is a value investment, where you have enough value to be worth investing in, so that we can see the potential for future growth. The value of AI in many of these companies is mainly reflected in two aspects:

The first level is to improve efficiency and reduce cost, which is called improving productivity. For example, it can reduce customer acquisition cost, maybe it can reduce customer turnover rate, or it can reduce labor cost and so on. For example, unmanned driving obviously reduces labor cost, which is a kind of value.

The other kind of value is that it might create new value, for example, ObEN, ObEN does this thing reduce any cost? In a sense, for example it reduce the cost of the star and the interaction of fans, improve the efficiency of the interaction, but from a broader perspective, the application of AI in the entertainment it tend to create new value, which means it may create a new user behavior, so as to help the company to produce a new source of income, Or it can help the company to expand the user base. That is to say, AI may sometimes not directly reduce costs and improve efficiency, but may create new values and find some new ways for enterprises to interact with users.

And the value but there is another very interesting characteristic, is that it often can give the enterprise to produce unique data, is that it is a bit like a UGC engine, users it will produce unique data, in the middle of the interaction of the data is a BAT is not possible, is not to say that you combine with a traditional industry company. This is your exclusive data. The AI company that creates new value is also a kind of company that we pay close attention to, because of its strong data barrier.

As for barriers, we think the second latitude is barriers.

Barriers matter, you have value, but do you have barriers? Is it competitive? Barriers, the first diagram we see is a triangle, and if we analyze AI companies, it can actually be divided into three categories:

1. The bottom layer is the infrastructure, such as cloud computing, chips and so on;

2. The upper layer is general technology, that is, iFLYtek, for example, is to do speech recognition;

3. The next layer is called vertical industry application, which means we think AI and industry application is the strategic highland for startups.

Why is that? Because the common technology and infrastructure is often done by large companies, or it often requires a lot of people and money, many startups do not have a great opportunity in this area, or it is difficult to enter the unique advantage now. And like general technology, after many large companies are mature, I believe they will open source, or let start-up companies use it at a very low price, forming an ecology of it.

So at present, the great opportunity for start-up companies is to apply AI to a vertical subdivision of the industry. These applications often have a certain threshold of the industry, and many large companies may not have so much energy to invest in them, but we think there is an opportunity for start-up companies. This is from a barrier perspective, which is when you’re facing a lot of competition, startups I think think a lot about vertical industry applications.

The second thing that creates a barrier is this concept called data network effects. This is very important also, the effect of data network mean, when you have more and more data, actually your company is getting stronger and stronger, and you can get more and more data, it is a positive cycle, when you have customers you will certainly get more data, when you have more data after your algorithm performance is improved, The algorithm gets better and then it helps you get more data. It’s a cycle.

Therefore, data network effect is often also a barrier for AI companies. We often say that the first-mover advantage of AI companies is very important. Only when you enter the market first and obtain data first can you continuously obtain more data and more customers. Sometimes from an investment point of view, AI’s data barriers often outweigh its technical barriers.

As many AI companies are using deep learning, or the AI chips used at the bottom are becoming more and more standardized, in many cases, this data is still the core, the difference of algorithm level, or its advantages may be in the continuous structure, not excluding specific industries, of course. Perhaps some special algorithms still have strong advantages. In particular, we think that data barriers are very important for startups doing AI and vertical industry applications, and to make full use of this data network effect to rapidly expand the volume of your data, first-mover advantage is very important. As AI entrepreneurs, our advice is: as soon as possible to cut in, as soon as possible to combine with the industry data, as soon as possible to grasp the opportunity to lead. I’ll come back to some entrepreneurial advice in a moment.

There is another dimension, which I also want to add, which is called the technical analysis dimension.

This is also when we do investment to think about, and I think for entrepreneurs is also a very critical question.

Is in this graph, here I provide two graphs, the first graph is that: any industry application it actually has certain requirements for accuracy. For example, here is the red line, for example, automatic driving, its accuracy requirements are very high. For example, when AI is used to analyze medical images, it should also have a certain accuracy, which is often compared with human. But sometimes, it is more demanding than human requirements, such as unmanned driving, human driving may have a relatively high accident rate, but people think that the application of AI, the accident rate is much lower, but how accurate can THE AI algorithm do?

In fact, it is very related to training data. When you have a small amount of training data, it is really difficult for you to meet the requirements of industry applications. What should you do at this time? Or when I’m an investor and I judge a company, I think you might not be able to be as accurate to some extent, is this a company to invest in or not to invest in? In addition to the changes in the amount of data in this place, we should also see the progress of technology, yesterday was not possible, maybe a few days, a few months may be possible.

In 2010, when ImageNet was just launched, the accuracy of this AI was very low. Its error rate reached nearly 30%. But in just a few years, Due to Deep learning, now the AI has surpassed human Performance and made great progress every year.

This is very important for entrepreneurs and investors, we want to think about what technology will be the next few years, the development of these technologies and may cause what kind of impact for the industry application, today I can’t in the very good application in AI, maybe tomorrow you can, and so on, this is everybody to want to think about.

So the concept I mentioned before can be said from one aspect, it actually has different risks in applications. Some are called high-risk applications, such as unmanned driving, such as Shape Security for Security, which has high requirements. If someone uses your system, there will be a big loss. Some are low-risk applications, such as Ever String recommending corporate clients to you. If the recommendation is wrong, the problem is not that big. It just wastes some of your sales resources. ObEN does entertainment, you say it’s not that realistic and probably not too risky; Honey does the same thing. It makes smart purchases and recommends products that aren’t exactly accurate, but at least it’s not a high-risk app. We have to figure out what kind of application we’re going to have.

Another is how to solve this problem when there is not enough training data to improve the algorithm to meet the requirements of industry applications, or when the accuracy is not high enough? It is usually solved through reasonable product design, such as ObEN. ObEN technology uses a very short sound and a small number of pictures, but it is impossible to achieve high precision artificial intelligence virtual image, movie level effect and special effects level effect, so I increase its entertainment. You can see that many of its applications can do a lot of entertainment functions, and people find it interesting. In addition, sometimes artificial assistance can be added. For example, if the AI is not 100% able to solve the problem, people can be involved in the early stage until the AI gets a lot of training data and then becomes more intelligent.

Unmanned is a good example of this, now you are not trust the unmanned these technologies, so it can be called intelligent auxiliary driving, when people trust it unceasingly, and it is the auxiliary driving after get the data, it can continue to improve performance to meet the requirements of the unmanned more high. Therefore, the third dimension is technical risk, we need to keep pace with The Times to view technology, and in a reasonable period of time to do some reasonable and clever product design is very important, start-up companies must think: how to design products? What does the customer really want? How to find a balance after considering various factors?


AI application in vertical Industries and Comparison between China and America

What we’re talking about today is the application of AI to vertical industries. I have already given you a lot of examples, today we don’t have too much time, we will discuss what kind of application in different industries one by one, the discussion of those scenarios, I actually talked for several hours in cheung Kong Graduate School of Business a while ago, is to analyze the application scenarios of AI in different industries. At this time today, we can only make some general summaries. If there is an opportunity in the future, I can still do some analysis with you.

So now AI in various industries should be said to have a lot of application opportunities, and it is really very hot application. We think there is an analogy that can better understand the effect of AI in this application, which is the difference between camera and video camera.

Before ARTIFICIAL intelligence, data acquisition was like a camera. For example, when we were thinking about the combination of AI and education, students might take tests once a month, and the teacher would know how the student was doing. It was like a camera taking pictures of you regularly. But when combined with AI, it can analyze the students the whole learning process, it can be to monitor him, he can go to higher density acquired data, like a camera, it took all day, the amount of data is different, in fact as a system can be in a shorter period of time to know the students, the usage of the user, To do some processing on it, and there’s a potential for data-driven, high-density data-driven applications.

Therefore, the application of AI in many industries conforms to this logic, that is, with a large amount of data, AI can achieve automation, personalized, adaptive and other characteristics. The more data, of course, you can achieve personalized, adaptive characteristics such as thousands of people. So that’s the idea of AI in vertical industries and I think it’s a pretty basic thinking.

To take a few examples, for example, the application of AI in medical treatment, using AI to look at medical images, which we are all familiar with, is a lot of automation, quickly learning from the experience of experts, and then doing analysis. For example, personalized diagnosis and treatment, according to your situation, according to the past historical data, it can help you make more personalized diagnosis and treatment plan, this is a kind of personalized; For example, in the application of AI in education, a very hot direction is adaptive learning, that is, to obtain your learning data more quickly and at a higher density, to model for each user, and then to provide him with adaptive learning solutions, so that everyone can have different learning paths. This is also greedy technology, which is what a company that I’ve invested in and that I’m deeply involved in is doing. We want to combine artificial intelligence with online education to better analyze the learning data of each student, build models for him, and improve his learning efficiency.

Speaking of the comparison between China and the United States, I can summarize the following points:

First, at present, we still think that the United States has a great talent advantage. The United States has more than 10 times as many basic AI talents as China, because many big companies and universities spend a lot of capital on cultivating these talents. China’s talent advantage will be weak, but it is growing. According to my knowledge in China, AI talents are in short supply. I believe geebang platform is also helping to promote the construction of AI talents, which I think is very good. In the short term, the United States still has advantages, but China’s market advantage, I think is very attractive, China’s overall market is very large, and now many companies are willing to combine innovation, this is China’s advantage. Therefore, as an AI venture capital institution, we also hope to connect excellent talents and technologies from the United States with the Chinese market, which is also the direction THAT I have been working on.

Second, China has elevated AI to the level of a national strategy. After the 19th National Congress of the Communist Party of China (CPC), AI has been promoted all over the country, which is relatively weak in the United States. Once the Chinese government promotes something, the effect will be very great. It will cause a great change in the capital market and the whole entrepreneurial ecology, and people are more willing to embrace AI. So at this point, we think there’s a great opportunity to start a business in China.

Third, China is actually in some industries with high threshold, and I think there are opportunities for the application of AI, such as energy, security, agriculture, manufacturing and so on. These industries tend to have relatively monopolized resources, but now, after the national strategy, these industries are also seeking AIization, and I think AIization is inevitable in these industries. If you do not do it, you will often lose the opportunity of such transformation. Because the threshold of these industries is relatively high and the resources are relatively monopolized, the cycle of start-up companies may be a little longer. It is not so easy to get data and do business, but I think there is a great opportunity. This aspect I think can draw lessons from the United States, because the United States on the one hand, in these areas is relatively is more market-oriented than in China, such as in energy, in terms of agricultural manufacture and so on, and even security you can see there are lots of good AI companies in the United States, I think China can draw lessons from more in this respect, if you are interested in these specific areas, We can also have a chance to communicate more.


Advice for AI entrepreneurs

The last thing I want to talk about is startup advice. Here are some suggestions that might give you some inspiration.

The first is that AI companies should now have more rational valuations and lower talent bonuses. In 2017, there are a lot of AI company valuations are very high, I believe 2018 should also is such, but relatively valuation will be slightly more rational, because a lot of things on the tuyere, it had a fever after cooling process, but AI company valuation is still relatively high, because it’s the market potential is very large, We expect it to be a little more rational. Talent dividend reduce means that before you can have one or two team AI very cow, did your company valuations will very high, by the scarcity of talent is likely to attract capital very much, so such a dividend may decrease, means that investors don’t just see that one or two AI, it still need to pay more attention to the fundamentals of the company, That’s the first suggestion.

The second piece of advice, which is closely related to the first piece of advice, is that in the first environment, I think people need to identify paying customers and monetization models much earlier. If you’re going to convince investors when you’re starting a startup, you need to show them that there are actually users who are willing to pay, and you need to have a clear monetization model. Of course very early stage companies may have a hard time identifying paying customers, so at least you can find some customers who are willing to pay, and I think that’s very important.

The third point is that more consideration can be given to introducing some strategic investment in the development process. Strategic investors often bring critical industry resources and launch channels to these startups. This is something to consider when raising money for a startup. You can also see that many of the funds raised by these AI companies are strategic investments from investors. Many AI companies pay attention to, for example, unsupervised learning and reinforcement learning. Because many things based on supervised learning have been applied in many aspects of technology, more attention can be paid to these aspects. AlphaGo Zero is a good example.

The fourth piece of advice is to grasp the trends of AI development. As an investor, I would like to share a few trends THAT I see in the next 1-2 years. In terms of barriers, AI startups are raising barriers in three ways, including 1) integrating hardware (keep an eye on CES), 2) B2C business models (which can create unique data barriers) and 3) traditional industries with higher barriers of penetration (e.g. Ng’s AI+ manufacturing company Landing. AI). Technically speaking, most AI startups use more supervised learning algorithms. Unsupervised learning and reinforcement learning should still have great application potential, such as the success of AlphaZero in 2017. In addition, the application of AI and blockchain is also attracting attention as blockchain has become a hot topic recently. Blockchain has a good combination with AI in solving data security and sharing, which will produce many valuable application scenarios.

For more content, you can follow AI Front, ID: AI-front, reply “AI”, “TF”, “big Data” to get AI Front series PDF mini-book and skill Map.