In the past 2021, we have witnessed the ups and downs of the sub-sector of artificial intelligence. Some companies have suffered losses for a long time or are on the verge of bankruptcy. Some companies have successfully obtained financing or successfully listed. However, looking at the current road to industrialization of artificial intelligence in China, there is always a realistic question that cannot be avoided, that is, "How far is artificial intelligence enterprise from making a profit?" It is true that research and development in the field of artificial intelligence requires huge investment, but all the The real commercial value of the R&D investment of AI can only be realized in the industrialized landing scene. It is separated from the actual industrial needs, and artificial intelligence can only stay in the technology itself.
At present, there are three paths for the industrialization of artificial intelligence: (1) AI+ industry, that is, after an artificial intelligence technology company has mastered a certain artificial intelligence technology, it will be implemented in specific scenarios of industrialization. For example, well-known artificial intelligence companies such as SenseTime, Yuntian Lifei, and Questyle Technology all take this path. (2) Industry + AI, that is, companies in a sub-industry, especially leading large enterprises, as the leading force, actively introduce artificial intelligence technology to complete the upgrade. For example, the industrial upgrading of large enterprises in sub-sectors such as Ping An Insurance, SF Express, etc. (3) The transformation of industry-university-research achievements, that is, the transformation of scientific research achievements led by universities and scientific research institutions and actively facing the market. In recent years, major top universities have established artificial intelligence research institutes, and cities such as Beijing, Shanghai, Shenzhen and other cities also have many government-backed artificial intelligence research and achievement transformation platforms.
Since 2020, the author has continued to visit hundreds of artificial intelligence companies and scientific research institutions. Based on what I have seen and heard in the industry, combined with my own ideas, I will talk about my thinking on the road to industrialization of artificial intelligence.
1. The road of "AI + industry" has entered a platform period
The model of "AI + industry" mainly refers to artificial intelligence technology companies through technology first, and then find suitable business scenarios to achieve business value. This path can be learned from the development history of the mobile Internet. Internet companies such as Taobao and Didi have discovered a new 0-1 industry through technological and business model innovation. We used to think that artificial intelligence technology companies can use 0-1 technological breakthroughs and learn from the experience of the mobile Internet to cover a wide range of sub-scenarios in all walks of life. But apart from a few scenarios such as face recognition, artificial intelligence technology companies have not replicated the success of their predecessors in the mobile Internet.
There are many reasons for this. We cannot simply blame it on the market, capital or the team itself. The author believes that the root cause is that artificial intelligence technology itself has entered a relatively slow stage of progress. We take the three core elements of artificial intelligence: Computing power, algorithms and data to analyze accordingly.
Let's talk about computing power first. According to the analysis of the "China Computing Power Development Index White Paper" by the China Academy of Information and Communications Technology in 2021, although basic computing power, intelligent computing power and super computing power have all increased to a large extent in recent years In the next five years, the global growth rate will even exceed 50%, but there is still a large gap in terms of increasingly complex algorithm models and fast-growing real needs. At the same time, the integrated storage and computing architecture, quantum computing, photonic computing and brain-like computing chips are still in the research and development stage in the laboratory, and there is still a long time before large-scale commercialization. Although, leading companies such as SenseTime and Huawei have adopted the establishment of artificial intelligence computing centers (AIDC) to meet the rapid growth of future intelligent computing needs; the development of three E-class supercomputing systems in my country, Shenwei, Tianhe and Dawning The work is also gradually advancing, and many domestic hardware companies have begun to replace the localization of computer hardware. But in the short term, computing power will be a real difficulty that restricts the development of artificial intelligence technology.
Let’s talk about algorithms again. Algorithms are computer technology on the surface, but abstracting and analyzing them in essence is a mathematical problem. In recent years, there has been a lot of development in the field of mathematics, such as infinite function calculation, but the development speed in the field of computer is not so fast. As far as the development of this specific field of algorithms is concerned, there is currently no generation gap between the top algorithms in China and the United States. Although the layer algorithm needs to invest a lot of money in research and development, as far as the application layer is concerned, enterprises can download the open source code of Gitub or OpenAI by themselves, or use the existing technical solutions of Internet giants such as Baidu, Ali, and Tencent, thereby greatly reducing the cost. threshold for technology application. At the level of market competition, artificial intelligence technology companies do not necessarily have more advantages than traditional Internet companies, or even traditional enterprises undergoing digital transformation.
Another key element is data. my country has gradually tightened data security-related management since 2020. The Personal Information Protection Law, the Data Security Law, and the Guiding Opinions of Nine Ministries and Commissions on Strengthening the Comprehensive Governance of Internet Information Service Algorithms The successive introduction of ” makes it more and more difficult for artificial intelligence technology companies to obtain data. Unless they can go deep into the segmented scenarios of the business, it is difficult to obtain large-scale data for training algorithm models as in the past. And most of the data that "feeds" the algorithm model is in the hands of companies in the industry, especially large companies. Whether it is for commercial purposes or for their own business security, it is almost difficult for these large companies to cooperate with artificial intelligence technology companies, which also causes artificial intelligence technology companies to face difficulties on the road to industrialization.
2. "Industry + AI" and opportunities for the transformation of industry-university-research achievements
The path of "industry + AI" belongs to the process of spontaneous upgrading of enterprises in the industry, and we can summarize it into the process of digital transformation of enterprises. In order to adapt to market competition, enterprises in the industry will actively seek cooperation with technology-based companies or research institutions in the field of artificial intelligence, and even establish their own teams to complete R&D work. For companies in most industries, they are not facing a 0-1 new market, but are often competing in the existing Red Sea market. This long-term struggle experience in the industry has made them in the field of artificial intelligence. Industrialization has the following two unique advantages:
1. Master a large number of professional knowledge and data in specific production scenarios: we generally call it industry knowhow, such as the formulation of chemical materials or a special production process. This industry knowhow is often the core secret of an enterprise. In some areas where data collection is closed and production processes are kept secret, only a few companies can obtain enough expertise and data to train artificial intelligence models. Therefore, when companies in many industries are looking for technical partners, they will reject more aggressive technology-based companies, and often require technology-based companies to submit algorithm source code to avoid cultivating potential competitors.
2. Understand real transactions and application scenarios: such as how to establish a reliable supply chain, how to analyze market intelligence information, how to establish a new business model and profit model, etc. These contents seem to belong to business-related categories, but they are the pain points of technology-based companies. Almost 90% of artificial intelligence companies have died on the road of polishing business models and finding application scenarios. But for companies in the industry, it is their innate natural ability to keenly capture market opportunities and make money from the industry. All companies that do not have this ability have been eliminated in the past market competition.
With the lowering of the entry threshold of artificial intelligence technology, it will be more convenient for a large number of traditional enterprises to adapt to artificial intelligence technology. In the future, every enterprise has the potential to become an "artificial intelligence + company". It is believed that with the continuous deepening of the country's new infrastructure and digital transformation, "industry + AI" star enterprises will appear in all walks of life.
On the road to the industrialization of artificial intelligence, the participation of universities and scientific research institutions is indispensable. For enterprises in the industry, universities and scientific research institutions can well complement their own research and development capabilities. At present, the transformation of industry-university-research achievements in my country is not very smooth. Although the country invests a lot of scientific research funds every year, due to the huge difference between academic and scientific research, business, and the market, universities and scientific research institutions are in business judgment and market sense. It always seems less "down to earth", and more results stay in the laboratory and it is difficult to go out, facing the realistic dilemma of "the fragrance of wine is also afraid of deep alleys".
In addition, universities and scientific research institutions often lack engineering capabilities. Although it is easy to gather high-level top talents, they lack practical business operators. Precisely because it is stronger than R&D and weaker than the market, universities and scientific research institutions are often more willing to cooperate with enterprises in the industry in the form of sales, technology shares or revenue sharing, rather than developing the market themselves. Compared with artificial intelligence technology companies, universities and scientific research institutions are guaranteed by a large amount of national basic scientific research funds. For universities and scientific research institutions that are easy to gather talents, many artificial intelligence technology companies are very complex technology. , it is not difficult for universities and scientific research institutions. With the revision of the National "Science and Technology Progress Law", the way for scientific researchers to participate in the transformation of achievements will also be smoother. Once enterprises find a suitable path for the transformation of their own achievements, they can well establish "industry + technology" with universities and scientific research institutions. " of the union. It is foreseeable that in the future, various new technologies and achievement transformation platforms will continue to emerge as a bridge between technology and the market.
3. Starting from the needs of the industry and taking the industry results as the verification standard
The development and change of the artificial intelligence industry is rapid, and even practitioners in the industry cannot avoid continuous and high-intensity learning and research. When experts from all walks of life enter the field of artificial intelligence, they may need to go through a continuous process of "looking back". When the author participated in the artificial intelligence legislation of Shenzhen Special Economic Zone in 2020, the definition of "what is artificial intelligence" now seems to be insufficient in connotation and extension. In the past, we used to think that artificial intelligence was simulating human intelligence, but with the development in recent years, we found that machines have many pain points in simulating human intelligence, but they have made rapid progress in simulating the intelligence of insects and animals. Many achievements have been applied in many fields such as obstacle avoidance and behavior prediction. Therefore, we found that artificial intelligence cannot be simply defined as "simulating human intelligence", but should be "artificial intelligence". Obviously, almost all the legislative experts at that time did not have a comprehensive and forward-looking understanding of artificial intelligence.
There is a famous story of monkey climbing a tree in the field of artificial intelligence: we cannot think that based on the current technological progress in the field of artificial intelligence, they are all contributing to the arrival of general artificial intelligence; just as we cannot think that a monkey climbing a tree means that it means It's like a giant step toward landing on the moon. On the road to the industrialization of artificial intelligence, we must maintain a humble and pragmatic spirit. Everything starts from the needs of the industry, and everything is based on the actual results of the industry as the verification standard. The verification of any technology or business model has its own time window. When the dividend period of market opportunities is missed, it is very difficult to achieve rapid development of the enterprise, and it is bound to face more intense hand-to-hand combat.
Like the development of artificial intelligence technology, the road to industrialization is always easy to "look" and difficult to "do". Although we do not advocate success or failure as a hero, on the road to industrialization of artificial intelligence, being able to solve real problems and obtain objective results is the core issue that entrepreneurs need to think about. For experts in each industry, instead of adopting a "pre-judgment" argument and insisting on explaining what they know to others, it is far more convincing to make real cases in the industry. In the future, every enterprise in all walks of life will be an "artificial intelligence + company", and we are willing to grow together with colleagues in the industry and witness the development of artificial intelligence industrialization. (Peng Jiahao is the director of the Digital Governance Center of the Shanghai Institute of Artificial Intelligence)
At present, there are three paths for the industrialization of artificial intelligence: (1) AI+ industry, that is, after an artificial intelligence technology company has mastered a certain artificial intelligence technology, it will be implemented in specific scenarios of industrialization. For example, well-known artificial intelligence companies such as SenseTime, Yuntian Lifei, and Questyle Technology all take this path. (2) Industry + AI, that is, companies in a sub-industry, especially leading large enterprises, as the leading force, actively introduce artificial intelligence technology to complete the upgrade. For example, the industrial upgrading of large enterprises in sub-sectors such as Ping An Insurance, SF Express, etc. (3) The transformation of industry-university-research achievements, that is, the transformation of scientific research achievements led by universities and scientific research institutions and actively facing the market. In recent years, major top universities have established artificial intelligence research institutes, and cities such as Beijing, Shanghai, Shenzhen and other cities also have many government-backed artificial intelligence research and achievement transformation platforms.
Since 2020, the author has continued to visit hundreds of artificial intelligence companies and scientific research institutions. Based on what I have seen and heard in the industry, combined with my own ideas, I will talk about my thinking on the road to industrialization of artificial intelligence.
1. The road of "AI + industry" has entered a platform period
The model of "AI + industry" mainly refers to artificial intelligence technology companies through technology first, and then find suitable business scenarios to achieve business value. This path can be learned from the development history of the mobile Internet. Internet companies such as Taobao and Didi have discovered a new 0-1 industry through technological and business model innovation. We used to think that artificial intelligence technology companies can use 0-1 technological breakthroughs and learn from the experience of the mobile Internet to cover a wide range of sub-scenarios in all walks of life. But apart from a few scenarios such as face recognition, artificial intelligence technology companies have not replicated the success of their predecessors in the mobile Internet.
There are many reasons for this. We cannot simply blame it on the market, capital or the team itself. The author believes that the root cause is that artificial intelligence technology itself has entered a relatively slow stage of progress. We take the three core elements of artificial intelligence: Computing power, algorithms and data to analyze accordingly.
Let's talk about computing power first. According to the analysis of the "China Computing Power Development Index White Paper" by the China Academy of Information and Communications Technology in 2021, although basic computing power, intelligent computing power and super computing power have all increased to a large extent in recent years In the next five years, the global growth rate will even exceed 50%, but there is still a large gap in terms of increasingly complex algorithm models and fast-growing real needs. At the same time, the integrated storage and computing architecture, quantum computing, photonic computing and brain-like computing chips are still in the research and development stage in the laboratory, and there is still a long time before large-scale commercialization. Although, leading companies such as SenseTime and Huawei have adopted the establishment of artificial intelligence computing centers (AIDC) to meet the rapid growth of future intelligent computing needs; the development of three E-class supercomputing systems in my country, Shenwei, Tianhe and Dawning The work is also gradually advancing, and many domestic hardware companies have begun to replace the localization of computer hardware. But in the short term, computing power will be a real difficulty that restricts the development of artificial intelligence technology.
Let’s talk about algorithms again. Algorithms are computer technology on the surface, but abstracting and analyzing them in essence is a mathematical problem. In recent years, there has been a lot of development in the field of mathematics, such as infinite function calculation, but the development speed in the field of computer is not so fast. As far as the development of this specific field of algorithms is concerned, there is currently no generation gap between the top algorithms in China and the United States. Although the layer algorithm needs to invest a lot of money in research and development, as far as the application layer is concerned, enterprises can download the open source code of Gitub or OpenAI by themselves, or use the existing technical solutions of Internet giants such as Baidu, Ali, and Tencent, thereby greatly reducing the cost. threshold for technology application. At the level of market competition, artificial intelligence technology companies do not necessarily have more advantages than traditional Internet companies, or even traditional enterprises undergoing digital transformation.
Another key element is data. my country has gradually tightened data security-related management since 2020. The Personal Information Protection Law, the Data Security Law, and the Guiding Opinions of Nine Ministries and Commissions on Strengthening the Comprehensive Governance of Internet Information Service Algorithms The successive introduction of ” makes it more and more difficult for artificial intelligence technology companies to obtain data. Unless they can go deep into the segmented scenarios of the business, it is difficult to obtain large-scale data for training algorithm models as in the past. And most of the data that "feeds" the algorithm model is in the hands of companies in the industry, especially large companies. Whether it is for commercial purposes or for their own business security, it is almost difficult for these large companies to cooperate with artificial intelligence technology companies, which also causes artificial intelligence technology companies to face difficulties on the road to industrialization.
2. "Industry + AI" and opportunities for the transformation of industry-university-research achievements
The path of "industry + AI" belongs to the process of spontaneous upgrading of enterprises in the industry, and we can summarize it into the process of digital transformation of enterprises. In order to adapt to market competition, enterprises in the industry will actively seek cooperation with technology-based companies or research institutions in the field of artificial intelligence, and even establish their own teams to complete R&D work. For companies in most industries, they are not facing a 0-1 new market, but are often competing in the existing Red Sea market. This long-term struggle experience in the industry has made them in the field of artificial intelligence. Industrialization has the following two unique advantages:
1. Master a large number of professional knowledge and data in specific production scenarios: we generally call it industry knowhow, such as the formulation of chemical materials or a special production process. This industry knowhow is often the core secret of an enterprise. In some areas where data collection is closed and production processes are kept secret, only a few companies can obtain enough expertise and data to train artificial intelligence models. Therefore, when companies in many industries are looking for technical partners, they will reject more aggressive technology-based companies, and often require technology-based companies to submit algorithm source code to avoid cultivating potential competitors.
2. Understand real transactions and application scenarios: such as how to establish a reliable supply chain, how to analyze market intelligence information, how to establish a new business model and profit model, etc. These contents seem to belong to business-related categories, but they are the pain points of technology-based companies. Almost 90% of artificial intelligence companies have died on the road of polishing business models and finding application scenarios. But for companies in the industry, it is their innate natural ability to keenly capture market opportunities and make money from the industry. All companies that do not have this ability have been eliminated in the past market competition.
With the lowering of the entry threshold of artificial intelligence technology, it will be more convenient for a large number of traditional enterprises to adapt to artificial intelligence technology. In the future, every enterprise has the potential to become an "artificial intelligence + company". It is believed that with the continuous deepening of the country's new infrastructure and digital transformation, "industry + AI" star enterprises will appear in all walks of life.
On the road to the industrialization of artificial intelligence, the participation of universities and scientific research institutions is indispensable. For enterprises in the industry, universities and scientific research institutions can well complement their own research and development capabilities. At present, the transformation of industry-university-research achievements in my country is not very smooth. Although the country invests a lot of scientific research funds every year, due to the huge difference between academic and scientific research, business, and the market, universities and scientific research institutions are in business judgment and market sense. It always seems less "down to earth", and more results stay in the laboratory and it is difficult to go out, facing the realistic dilemma of "the fragrance of wine is also afraid of deep alleys".
In addition, universities and scientific research institutions often lack engineering capabilities. Although it is easy to gather high-level top talents, they lack practical business operators. Precisely because it is stronger than R&D and weaker than the market, universities and scientific research institutions are often more willing to cooperate with enterprises in the industry in the form of sales, technology shares or revenue sharing, rather than developing the market themselves. Compared with artificial intelligence technology companies, universities and scientific research institutions are guaranteed by a large amount of national basic scientific research funds. For universities and scientific research institutions that are easy to gather talents, many artificial intelligence technology companies are very complex technology. , it is not difficult for universities and scientific research institutions. With the revision of the National "Science and Technology Progress Law", the way for scientific researchers to participate in the transformation of achievements will also be smoother. Once enterprises find a suitable path for the transformation of their own achievements, they can well establish "industry + technology" with universities and scientific research institutions. " of the union. It is foreseeable that in the future, various new technologies and achievement transformation platforms will continue to emerge as a bridge between technology and the market.
3. Starting from the needs of the industry and taking the industry results as the verification standard
The development and change of the artificial intelligence industry is rapid, and even practitioners in the industry cannot avoid continuous and high-intensity learning and research. When experts from all walks of life enter the field of artificial intelligence, they may need to go through a continuous process of "looking back". When the author participated in the artificial intelligence legislation of Shenzhen Special Economic Zone in 2020, the definition of "what is artificial intelligence" now seems to be insufficient in connotation and extension. In the past, we used to think that artificial intelligence was simulating human intelligence, but with the development in recent years, we found that machines have many pain points in simulating human intelligence, but they have made rapid progress in simulating the intelligence of insects and animals. Many achievements have been applied in many fields such as obstacle avoidance and behavior prediction. Therefore, we found that artificial intelligence cannot be simply defined as "simulating human intelligence", but should be "artificial intelligence". Obviously, almost all the legislative experts at that time did not have a comprehensive and forward-looking understanding of artificial intelligence.
There is a famous story of monkey climbing a tree in the field of artificial intelligence: we cannot think that based on the current technological progress in the field of artificial intelligence, they are all contributing to the arrival of general artificial intelligence; just as we cannot think that a monkey climbing a tree means that it means It's like a giant step toward landing on the moon. On the road to the industrialization of artificial intelligence, we must maintain a humble and pragmatic spirit. Everything starts from the needs of the industry, and everything is based on the actual results of the industry as the verification standard. The verification of any technology or business model has its own time window. When the dividend period of market opportunities is missed, it is very difficult to achieve rapid development of the enterprise, and it is bound to face more intense hand-to-hand combat.
Like the development of artificial intelligence technology, the road to industrialization is always easy to "look" and difficult to "do". Although we do not advocate success or failure as a hero, on the road to industrialization of artificial intelligence, being able to solve real problems and obtain objective results is the core issue that entrepreneurs need to think about. For experts in each industry, instead of adopting a "pre-judgment" argument and insisting on explaining what they know to others, it is far more convincing to make real cases in the industry. In the future, every enterprise in all walks of life will be an "artificial intelligence + company", and we are willing to grow together with colleagues in the industry and witness the development of artificial intelligence industrialization. (Peng Jiahao is the director of the Digital Governance Center of the Shanghai Institute of Artificial Intelligence)
Comments