Dengzhidong, the most popular, made suggestions on

Deng Zhidong suggested that Foshan: deeply cultivate "artificial intelligence +" subdivision application

"the accelerated development and landing application of artificial intelligence will contribute to Foshan's transformation from a large manufacturing city to a strong manufacturing city." A few days ago, Deng Zhidong, Professor of the State Key Laboratory of intelligent technology and systems at Tsinghua University and director of the expert committee of China artificial intelligence industry innovation alliance, received an exclusive interview from the south to understand and analyze the theory and practice of artificial intelligence, its significance to the manufacturing industry, and the specific path to promote the implementation

as a major manufacturing city in China, Foshan recently proposed to cultivate two trillion level industrial clusters in the field of advanced manufacturing, so as to raise the scale and quality of the manufacturing industry to a new level. To achieve this goal, artificial intelligence will be an important part. Deng Zhidong also knows more about the manufacturing industry in Foshan, and is currently cooperating with an emerging robot enterprise in Foshan in research and development. He believes that "from a national perspective, the starting point of Foshan's manufacturing industry is very high."

Deng Zhidong suggested that Foshan should speed up the construction of artificial intelligence talent highland on the basis of fully grasping the advantages and limitations of artificial intelligence, deeply cultivate the segmentation scene and big data application in manufacturing enterprises with the spirit of craftsman, and promote the quality revolution practice of manufacturing industry through "artificial intelligence +"

understanding artificial intelligence

has made breakthroughs, but it is not a "panacea"

South: artificial intelligence is more and more familiar to people, but its connotation is controversial. How should I understand it

Deng Zhidong: artificial intelligence is a comprehensive discipline that develops theories, methods, technologies and applications for simulating, extending and expanding human intelligence, generally including perceptual intelligence and cognitive intelligence. Among them, perceptual intelligence involves the simulation of human intuitive behavior, such as vision, hearing, etc., while cognitive intelligence is the simulation of human deliberate behavior, including high-level intelligent behaviors such as memory, understanding, reasoning, etc

since 2012, perceptual intelligence driven by big data has made a breakthrough, which has been able to be more deeply involved in the industrial field, setting off a new round of development climax. The value of artificial intelligence to the industry is also mainly reflected in this aspect. At the same time, the deep reinforcement learning represented by alpha dog has made more progress than human level in the decision-making of dynamic game, but it belongs to the category of cognitive intelligence, and its application in manufacturing industry is not much

South: what are the differences and connections between AI and big data, intelligent manufacturing, etc

Deng Zhidong: with the support of big data and big computing power, AI is expected to achieve audio-visual perception and other perceptual abilities close to human level in some subdivision application fields. Big data and artificial intelligence are inseparable, sometimes referred to directly as "big data artificial intelligence"

AI adds "eyes" and "ears" to the information space and physical space of intelligent manufacturing at the same time. Through the perception of data torrent and the perception of the real world, it can empower these two spaces at the same time, making new manufacturing more flexible, adaptive and autonomous, so as to meet the personalized user needs of new retail with consumption as the core

industrial interconnection, industrial IOT, 5g communication, cloud platform and data center together constitute the hardware infrastructure of big data AI applications

South China: in terms of manufacturing, what are the application advantages and limitations of artificial intelligence

Deng Zhidong: the current new generation of artificial intelligence plays a very important role in the manufacturing industry, but there are also many limitations. Its real industrial value is mainly reflected in the aspect of perceptual intelligence, which can bring visual, auditory and other perceptual capabilities to industrial interconnection, industrial IOT and various production equipment, production lines, workshops, factories, etc. in manufacturing enterprises. This perception of data and real objects can even reach the human level after a lot of learning. In this way, the "machine replacement" can bring about the improvement of production efficiency and product quality, alleviate the difficulties caused by the disappearance of demographic dividends in economically developed regions, and promote the upgrading of traditional industries and the adjustment of industrial structure. However, the acquisition of this ability requires the accumulation of massive big data, the continuous iterative learning of data, performance and experience in application practice, and the support of artificial intelligence chips. However, the current AI has no ability of memory, understanding and drawing inferences from one instance, and lacks common sense. Therefore, on the one hand, it can indeed bring industrial value, on the other hand, there is still a lot of room for development. We must overcome the misunderstanding of "panacea" and "omnipotence" of AI, and have an objective and clear understanding of the advantages and limitations of AI in manufacturing empowerment. Specifically, at present, we should focus on strengthening the industrial application of big data AI on various specific issues

enabling industrial innovation

the landing of artificial intelligence should give play to the spirit of craftsmanship

South: specifically, how does artificial intelligence play a role in the industry

Deng Zhidong: everyone knows the benefits of artificial intelligence, but only when it is really done can we know that the implementation of artificial intelligence is very difficult. To a certain extent, artificial intelligence can be understood as "artificial + intelligence". In other words, as many people as there are, there will be as much intelligence, and as high a level of artificial intelligence as there are. In fact, it is to transfer human experience to the machine in the form of labels, and then endow it with intelligence

for example, in the medical field, if early lung cancer is detected by CT, the canceration label on the medical image cannot be done by ordinary people, but by top international doctors. In this way, the level of the machine is the level of the world's top doctors, otherwise it is the level of a general doctor

the landing application of artificial intelligence cannot be general and broad, but we should start with a specific specific problem and start bit by bit. Not all scenarios can be solved by artificial intelligence. The specific application scenarios that can be processed need a good foundation for digitization and networking, and can continuously generate big data. Then, we talked about the manual cleaning and labeling of data, including the quality audit of labels

for example, for machine fault diagnosis and appearance quality inspection of some products, after high-quality label data is accumulated, artificial intelligence through automatic learning can indeed reach the level of "listening to faults" and "looking at defects" of skilled masters, so as to replace human workers

it is necessary to establish an engineering team and invest a lot of human and material resources to do data accumulation and data labeling, especially to establish high-quality data, which is a "stupid" method. However, we rely on this "stupid method" to polish artificial intelligence products and deeply cultivate application scenarios and data

basically, more than 80% of human and material resources should be invested in specific subdivisions up to 0.001 ⑴ 000mm/min to do these "subtle" things. According to the research level at this stage, there is no universal method that can "easily" solve problems in all fields

South China: what links in the manufacturing industry are suitable for the application of artificial intelligence

Deng Zhidong: these application scenarios should first meet the conditions mentioned above, such as easy data collection. On this basis, typical application scenarios can give priority to those links with poor production environment, high value or scarce personnel

at present, the technological breakthrough of artificial intelligence is mainly in the field of perceptual intelligence, so artificial intelligence technology should be more widely used in all kinds of work that requires senior skilled workers to use "eyes", "ears" and other perceptual intuition, especially in the perception and mining of data and information on Industrial interconnection and industrial IOT

no matter in which link it is applied, it should be deeply cultivated with "craftsman spirit", start from "small things", strive to achieve "machine vision", "machine hearing" and "machine touch" that are closer to the human level, further enable the product, process and service experience, and achieve a significant leap in production efficiency and product quality

at present, the technological breakthrough of artificial intelligence is mainly in the field of perceptual intelligence. Photo by Lin Dongyun

South: in addition to the production line, what fields can AI empower industrial innovation

Deng Zhidong: artificial intelligence can be applied in many scenarios, whether in physical space or virtual information space. We call it information physical space, which needs to be equipped with large punctuation elongation devices (CPS). In fact, the intelligent manufacturing and new retail we do are all in CPS, one part in the information space and the other part in the physical space. These two parts can realize intelligent automation through the empowerment of artificial intelligence at the same time, which can not only improve the efficiency of production, but also improve the quality of production products

made in Foshan

can speed up the construction of artificial intelligence talent highland

South: what is the current development status of artificial intelligence industry

Deng Zhidong: the current main track of AI industry is mainly concentrated in computer vision, speech recognition, AI chips and "Ai +". The dominant enterprises of visual AI mainly include Baidu, Shangtang technology, Kuangshi technology, etc; The advantageous enterprises of voice AI, such as iFLYTEK, Spitz, yunzhisheng, etc; AI chips include Huawei Hisilicon, Cambrian, horizon, etc. In "artificial intelligence +", China's application scenarios and scale rank first in the world, including security, consumption, autonomous driving and other vertical segmentation fields

these AI enterprises are currently most distributed in Beijing, followed by Shenzhen, Shanghai, Hangzhou, etc

in terms of industrial support for "artificial intelligence + manufacturing", domestic cities are actively exploring and starting. Although no experience worth popularizing has been accumulated, Shenzhen, Beijing, Shanghai, etc. can measure the secant elastic modulus under strain, which is in the forefront

South: what suggestions can we offer for the research of artificial intelligence and the application of manufacturing industry in Foshan

Deng Zhidong: the manufacturing industry in Foshan is very diversified, and our private enterprises are blooming everywhere and very active. There are giants like Midea, including KUKA, which it acquired, and many small and medium-sized enterprises. From a national perspective, the starting point of Foshan's manufacturing industry is very high

it is suggested that Foshan should promote the deep integration of AI with traditional manufacturing and industrial interconnection in four dimensions: application scenarios, big data, AI algorithms and computing power. With the accelerating penetration of AI into traditional manufacturing, industrial interconnection and other industries, there are serious deficiencies in the stock and increment of AI engineering and technology development talents, and the contradiction between demand and supply is becoming increasingly prominent. Therefore, it is necessary to speed up the construction of AI talent highland in Foshan

the industrial application of artificial intelligence also requires the deep integration of universities, scientific research institutions and manufacturing enterprises. On the one hand, enterprises have a thorough understanding of specific application requirements and application scenarios; On the other hand, many first-class universities and research institutes have long accumulated high-end R & D talents in AI. The combination of the two can make "Ai +" land quickly

this