Why is it so difficult for traditional industries to get AI blessings
◎Reporter Chen Xi
Recently, well-known artificial intelligence scholar Wu Enda published an article explaining his understanding of the slow application of artificial intelligence in traditional industries.
Whether it’s personalized recommendations for short videos, time-consuming estimates for takeaway delivery, or face recognition for mobile payments, AI technology represented by algorithms is “handy” in the consumer Internet industry.
However, when it comes to traditional industries, it is difficult for people to quickly think of very mature typical cases of applying artificial intelligence.
Why is the application speed and scope of AI technology in traditional industries far inferior to industries such as consumer Internet?
The application of AI in the consumer Internet industry has more advantages
“The application of AI technology mainly depends on data, computing power, and algorithms.” Zhu Pengfei, associate professor of the Department of Intelligence and Computing at Tianjin University, introduced that data must first reach a certain size, which is the basis of the application. In addition, computing power must be able to support large-scale Model training, and then the algorithm needs to achieve a certain accuracy, and the end-to-side computing power must also have a certain reasoning ability.
The reason why only consumer Internet companies are currently applying AI technology on a large scale is that consumer Internet companies have more advantages in these three aspects.
In the past few years, short videos were not as popular as they are now. For example, Taobao at the early stage of development did not have strong user stickiness.
As the push becomes more and more accurate, the user experience has also been greatly improved, and finally showed a blowout user growth.
"Accurate push mainly relies on the improvement of algorithm accuracy, and the improvement of algorithm accuracy is inseparable from massive data as the basis." Zhu Pengfei explained that in this single scenario, the algorithm model needs to evolve continuously and learn for life.
Since it is not a closed data environment, new data is always added, and the algorithm model needs to be continuously adjusted and iteratively upgraded through learning to make it more and more accurate, forming a virtuous circle.
"At the same time, although the accuracy of algorithms in the consumer Internet industry has risen to a certain level, compared with the application scenarios of some traditional industries, the consumer Internet industry has a relatively low acceptance threshold for AI algorithm accuracy. For example, short videos and Taobao. Preference recommendation and Baidu hot search keywords only need to achieve the purpose of user stickiness, as long as there is a certain accuracy, users can accept it.” Zhu Pengfei said that in contrast, in many traditional industries, the requirements for technical precision are only Much higher.
For example, in the application of vision-based AI technology in face recognition, identity verification at high-speed railway stations and airports, 1:1 comparison accuracy must be as high as 99.99% or even higher before it can be applied.
In terms of computing power, the current cloud computing power can already support large-scale model training and inference, such as short videos and Taobao recommendations.
However, in a large number of traditional industry application scenarios, the end-to-side computing power on smart terminals cannot meet the real-time and accuracy requirements of reasoning.
"Compared with social networks and e-commerce systems, the closed ecosystem of traditional industry application scenarios makes cloud computing power unable to be effectively used." Zhu Pengfei, for example, took intelligent unmanned system inspections as an example, power inspections, pipeline inspections , Traffic inspections, river inspections, and photovoltaic inspections require the computing power carried on drones and robots to meet the requirements of real-time inspections. Due to the high complexity of the video analysis model, the end-side is often unable to achieve accurate and efficient real-time Inference, the lightweight network loses recognition accuracy while satisfying real-time performance.
Because the accuracy of the algorithm does not meet the requirements for use, the application of AI technology cannot be implemented in many scenarios.
The application of AI in traditional industries faces three major challenges
Wu Enda believes that in terms of AI applications, industries other than the consumer Internet industry are facing three major challenges: data sets are small; customization costs are high; and the process from validating ideas to deploying production is very long.
In this regard, Zhu Pengfei also feels deeply. He analyzed the traditional manufacturing industry as an example.
"Data is a very prominent issue in the process of traditional manufacturing enterprises transforming from manufacturing to smart manufacturing." Zhu Pengfei said, first of all, there is a certain degree of difficulty in obtaining data.
The data of traditional manufacturing companies is closed, because many traditional companies are not new information equipment, do not have sensors to collect real-time data, and do not have a data center. Therefore, the data is scattered and the lack is serious. High-quality data.
Secondly, many of the data of various factories in the industry have commercial value, so the factories are strictly confidential, which leads to the lack of data circulation and no way to share, which in turn forms the data island effect and affects the optimization of the AI algorithm model.
"When we are developing an AI algorithm model, because of the confidentiality of the data, the data we get is often'desensitized', which also seriously affects our judgment. However, companies in traditional industries lack Technical personnel with the ability to develop AI algorithm models, so the two sides also have high barriers in the process of cooperative research and development." Zhu Pengfei said.
In addition, data sources in traditional industries do not come from a single scenario like the consumer Internet field. Complex business scenarios result in data that is often “dirty” and must be “cleaned” to remove a large amount of invalid information, so that AI algorithm models can be learned efficiently. To improve accuracy.
"It's like we teach children knowledge. Only by talking about knowledge points, children can learn fast. If there is a lot of useless information in the knowledge points, the children will not be able to distinguish, and the learning efficiency will definitely decrease." Zhu Pengfei introduced, and labeled the data. The work of "knowledge points" is huge and cumbersome. It requires a special person from the enterprise to do it, and it takes a lot of time and energy.
“In order for traditional manufacturing to obtain high-quality data, it must carry out informatization and intelligent transformation of production equipment.” Zhu Pengfei said, this transformation requires enterprises to invest a lot of time and energy, and it will also increase production costs. Become a barrier to the application of AI in traditional manufacturing.
High-quality data is a prerequisite for application
In the past 10 years, most of the research and development and application of AI have been "software-centric" driven.
With the support of massive data, the software and algorithms are continuously optimized to obtain higher algorithm accuracy.
In the case that traditional industries cannot improve the quality and quantity of data, Wu Enda believes that traditional industries should adopt a "data-centric" model and focus on obtaining better quality and higher matching data.
"Under this kind of thinking, some good application cases have also emerged in traditional industries. For example, the image recognition AI system in the medical field can help doctors'see' CT images, identify tumors and other lesions, and assist doctors in making judgments. "Zhu Pengfei introduced that because a lot of data is marked on the image by professional radiologists, the data is relatively accurate, and the AI algorithm model has made rapid progress in the learning process.
At present, the accuracy of many image recognition systems can reach more than 90%. Because they are assisting doctors, doctors are required to make medical decisions at the end, but this level of accuracy greatly reduces the work intensity of doctors.
"Although there are some successful cases of AI technology in traditional industries, if you want to better integrate with AI, you have to work hard to improve data quality." Zhu Pengfei suggested that first of all, for traditional industries that have accumulated massive amounts of data, Under the premise of ensuring data security, actively open data.
There will be a lot of room for development when mining the value hidden in the data and linking it with the demand.
Secondly, for emerging industries, such as new energy vehicles, when building smart factory plans, factors such as data acquisition and intelligence are taken into consideration.
However, Zhu Pengfei emphasized that while making good use of AI technology in traditional industries, do not abuse AI technology. Make an assessment before applying it. If production efficiency cannot be improved and the industry as a whole can be improved, then blindly using AI technology is a waste of resources. Waste.
"For example, some application scenarios require AI algorithms to achieve an accuracy of more than 99% before they can be used. Through evaluation, the existing model algorithms can only achieve an accuracy of 90%, so there is no need to force AI technology in this scenario."
"All in all, for the application of AI technology, data must go first, and high-quality data will be discussed before application. It is difficult to have good applications without good data." Zhu Pengfei said.