In the view of many people, AI, which is called the new generation of information technology together with the Internet of things, blockchain and cloud computing, is a new technology in the 21st century or in the 1990s. In fact, AI is an "ancient" technology relative to computer history. As early as 1956, there was the concept of artificial intelligence, but due to many reasons such as computational power and algorithm, the development of AI has stagnated for several times. In March 2016, alphago (alpha dog) and Li Shishi, the world champion of go and professional Nine Segment chess player, fought a man-machine war of go and won with a total score of 4-1. Since then, AI has entered the vision of the Chinese public. According to institutional data, by the end of 2021, there were 7362 pan AI companies in China. Eight years have passed so far. Why can't alphago artificial intelligence be seen in work and life? In the next decade, AI will continue to live on application scenarios In the AI circle, we often hear some terms, such as deep learning, computer vision, neural networks, algorithms, parameters, models, etc., which make people feel "confused" (that is, they don't understand but feel powerful). AI is also very popular in the capital market. Institutional data show that in 2021, the financing amount in the field of AI is nearly 400 billion yuan. Let's look at the background of the founding team of AI manufacturers, the academic background of world famous universities and the middle and high-level working background of top large factories. An AI manufacturer recruited an academic bull, published papers in top journals, and financed XX billion yuan... We often see such information, but we don't seem to have seen any AI manufacturer "show off" in terms of revenue and profit. People can't help asking: how many AI manufacturers are profitable? How many AI manufacturers are living by financing? Is there a market in the world that has been living on financing? There are two groups of data to share with you. One group is the trading time and amount trend of AI investment and financing in China over the years, and the other group is the average amount of single financing in China over the years. Chart 1 trading time and amount trend of AI investment and financing in China over the years Data source: it orange, chart 2 average amount of single financing of AI in China over the years (RMB 100 million) Data source: it orange On the surface, these two sets of data reflect that the capital market is more and more optimistic about AI and has increased investment. Therefore, they are good for the AI market. Then, with this idea, let's look at another set of data: the establishment time distribution of Chinese AI companies. Figure 3 distribution of establishment time of Chinese AI companies Data source: it orange These three sets of data describe a phenomenon: the number of newly established AI manufacturers has decreased sharply, but the overall financing amount and single financing amount are rising sharply. Look at the essence through the phenomenon: the higher the amount of financing, the higher the rounds and the wave of listing have led to this phenomenon. The essence is that the AI market may be experiencing the final carnival. With an annual investment of 400 billion yuan, we can't help asking: how much return can we meet the appetite of the capital market? Is the current market performance worthy of this 400 billion yuan? If they can't meet the appetite of the capital market, how will their next investment strategy be adjusted? AI manufacturers who are "burning money" in the air, it's time to "land" to make money. In the next 10 years, if AI manufacturers want to live, they must find a profit model, that is, develop application scenarios. Application scenarios are not only the application-oriented AI manufacturers who focus on Intelligent Transportation, intelligent agriculture and intelligent finance, but also the platform or basic manufacturers who make AI chips, data platforms, algorithms and models. According to the data of Haibi Research Institute, "looking for application scenarios" has become a top priority in the eyes of AI manufacturers. Figure 4 distribution of strategic objectives of AI manufacturers (mention rate) Source: Haibi Research Institute Many AI manufacturers realize that "finding application scenarios" is imminent, but limited by four mountains, AI landing is difficult: Numeracy: poverty limits imagination. Taking alphago as an example, some data show that the TPU (an algorithm chip similar to GPU) used by alphago is converted into the common consumption grade 1080ti on the market, about 12000 pieces, at least tens of millions of dollars. Facing the high cost of computing power, can enterprise users afford it? Algorithm: addicted to open source, unable to extricate themselves. Professor Kong Dexing, director of the Institute of Applied Mathematics of Zhejiang University, said: the innovation ability of China's artificial intelligence industry is not as strong as the legend. The fact is that the industrial development relies too much on open source code and existing mathematical models, and there are not many things that really belong to China. Open source code can be used, but it is not professional and targeted enough, and the effect often can not meet the actual requirements of specific tasks. Taking image recognition as an example, the AI developed with open source code can accurately recognize human faces, but it is difficult to meet the clinical requirements in the recognition of medical images. For example, for the identification of liver lesions, due to the difficulties of fuzzy boundary, low contrast, organ adhesion and even overlap, it is difficult to accurately identify with open source code. In terms of three-dimensional reconstruction and visualization, it is difficult to accurately reflect the real anatomical information, and even misleading, which is "fatal" in medical application. Data: a clever woman can't make bricks without rice. Alphago is trained by massive high-quality chess score resources. When solving application scenario problems, Where are so many "high-quality chess scores"? In terms of quantity, China's data trading market is still in the early stage of exploration, and there are a lot of problems to be solved in terms of data ownership, trading mechanism and pricing method; in terms of quality, China has no mature data marking industry standard, and a lot of data are "rough" "It can't be used. Privacy computing, which has sprung up in the past two years, can solve some problems in concept, but the actual effect has not been clearly reflected. Can AI manufacturers afford it? Talent: lack of high-level talents and compound talents. The main flow of high-level talents is high paid Internet factories and established scientific research institutes, which are not attractive to general AI manufacturers. As for the compound talents who understand both AI and industry, they lack the training mechanism and are naturally rare. At the foot of these four mountains, many models of AI manufacturers cannot land. This is an anxious situation: the value of the model will be lost over time. After all, who will apply the AI model ten years ago to the current scene? According to the research of Haibi Research Institute, AI has solutions in many application scenarios. The reason why it is not popularized is mainly because manufacturers do not learn from each other, but prefer to make independent exploration with twice the effort. Through the reasons behind deep excavation, Haibi Research Institute summarizes three reasons: I don't want to learn. AI manufacturers hope to independently explore some areas with rare human traces and demonstrate their leadership. I can't. Without strong information acquisition ability, I want to learn but can't. I can't learn. Without powerful information extraction ability, I want to learn but can't. Haibi Research Institute believes that, in the final analysis, the lack of successful experience in application scenarios is an important obstacle to the vitality of AI. For example, nearly 100 RPA manufacturers have "rolled up" in the field of finance, taxation and banking, but most of them do not know that some manufacturers have found new application scenarios and are seizing blue ocean dividends. Have OCR manufacturers questioned whether this technology can only be applied to "character recognition and extraction scenarios", and whether any friends have found new scenarios? 2、 How long can the four dragons be popular when the big test comes? When it comes to AI, many people first think of these six manufacturers: Baidu, iFLYTEK and AI four dragons. Their situations are different. Next, analyze them one by one. Baidu should be an idol, not an example. AllinAI's Baidu, which is admirable, has developed PaddlePaddle, which integrates deep learning core framework, tool components and service platform. Its profound technology is very popular, and its strategic vision in driverless is commendable. But how many AI manufacturers do the market need? Baidu can be worshipped as an idol, not as an example. Baidu has many mature businesses such as advertising and cloud computing. These businesses can provide stable cash flow to support Baidu's investment in AI. Can other AI manufacturers? Baidu can make AI technology embody its own business, which is equivalent to its own scenes. (from the name, it is remarkable that the name of Baidu intelligent cloud is unique among the cloud manufacturers of Ali cloud, HUAWEI cloud, Tencent cloud, Jingdong cloud and other AI manufacturers). IFLYTEK is an excellent example. IFLYTEK is in a leading position in natural language processing. Based on this, it has found a voice scene, occupied a leading position in voice input, translation, transcription and other aspects, and formed a stable income and profit. IFLYTEK is an excellent example of successfully looking for application scenes. AI Four Dragons - four manufacturers under the big test. "Four little dragons" said that the market is very optimistic about the potential of these four manufacturers, as if people are used to giving some intelligent children the title of "child prodigy", but no one can hold the title of "child prodigy" for a lifetime. People will eventually grow up, It's possible to grow up and become a talent or "hurt Zhongyong". If you don't grow from "Bruce Lee" to "real dragon" within a certain period of time ", will also be coerced by the torrent of the times. The AI four dragons and similar manufacturers are collectively sprinting to the market and helping them through the key stage of commercial application exploration through a large amount of listing financing. This is a big test for them, and time is very precious. After all, the high expenses every day will not stop for a moment. In the field of AI, the ideal situation is to develop products and realize rapid reuse with low marginal cost. Unfortunately, many AI manufacturers can only receive those project-based lists. Deeply customized project-based lists will lead to three direct consequences: High labor cost. The project-based list is a great challenge to the manufacturer's manpower, which will greatly increase the manpower cost and affect the profit. In the field of digital intelligence services, human cost is often the largest part of the total cost. Cash flow is unhealthy. The project-based list with long cycle and slow payment collection will make the cash flow of AI manufacturers develop in an unhealthy direction. Lack of experience accumulation mechanism. When making products, experience will precipitate to the company, and the effect will be reflected in the products; When doing a project, the experience will precipitate to the individual, and the effect will be reflected in the project. Finally, manufacturers rely heavily on old employees, which has a negative impact on the construction of the company's core competence. Based on this reality, many AI manufacturers will highly publicize that they have recruited an academic bull and published papers in top journals to "develop strengths and avoid weaknesses" in a sense. An enterprise does not emphasize the achievements at the commercial level, but blindly emphasizes the achievements at the academic level, which may be a helpless move behind it. 3、 AI manufacturers without application scenarios will disappear in the long river of history Without the nourishment of application scenarios, technology is rootless duckweed. If a concept wants to have vitality, it must be rooted in the social level. Artificial intelligence has not brought enough experience to the public and enterprise users. Their patience with artificial intelligence may be decreasing, and the time for AI manufacturers is not as sufficient as expected. Government agencies may increasingly realize that "laboratory" and "factory" are two different things, which will indirectly lead to changes in the allocation of social resources. The story of the laboratory should stay in the laboratory, and do not put it in the context of factory production, which is confusing. After all, the road from the laboratory to the factory is much longer than expected