Why Traditional Industries Struggle to Embrace AI
With the help of MLOPs, AI can also be well-used in traditional industries. (PHOTO: VCG)
By YU Haoyuan
Over the past few years, artificial intelligence (AI) has developed in leaps and bounds. However, despite it is now being used in every business from food delivery to mobile payments, traditional industries are still hard to benefit from the value addition of AI.
Preconditions of AI application
"The application of AI is determined by three factors, which are data，computation capacity and algorithm," said Zhu Pengfei, vice president of the College of Intelligence and Computing at Tianjin University. According to Zhu, the data as the basis of AI application must reach a certain volume. In addition, the computing capacity must support large -scale model training, and then the algorithm needs to achieve a certain accuracy, and the client computing power must have a certain reasoning ability.
"Accurate Content Push (to provide precise content for internet users) mainly counts on improving algorithm accuracy, and the improvement of algorithm accuracy is inseparable from massive data as the basis," Zhu explained. He thinks that in this single scenario, the algorithm model is required to make the continuous adjustment and repeated upgrade for a higher degree of accuracy and formation of a positive sequence.
Why AI is seldom used in traditional industries
Recently, Andrew Ng, a well-known computer scientist and technology entrepreneur, said in adopting AI, industries other than the consumer Internet industry are facing three major challenges in the development progress, which involves "Small datasets," "Cost of customization," and "Gap between proof of concept and production."
"For the traditional manufacturing industries, in their process of transforming to smart manufacturing, data is a very prominent issue," said Zhu. He said data assessment is facing difficulties because many traditional enterprises have neither built data pick-ups nor data centers. As a result, the data of these traditional companies are fragmented and seriously lack quantity and quality, which is challenging to obtain.
Moreover, data from these kinds of industries have business value, and as a result, they are always kept confidential, preventing data from being circulated and being shared. This further creates the effect of an information island, which affects the optimization of the AI algorithm model.
"When we are developing an AI algorithm model, the data we get is often 'processed without privacy issues,' which also seriously affects our judgment because of the confidentiality of the data. In addition, with a lack of technical personnel who can develop AI algorithm models in traditional industries, it may cause trouble in the development process," said Zhu.
He believes conventional manufacturing industries must upgrade their production equipment to be information capable and intelligent if they want to gain high-quality data. Such an upgrade requires enterprises to invest a huge amount of time and energy and increase production costs. It thus has become a barrier to the application of AI in the conventional manufacturing industry, said Zhu.
Solution to getting AI involved in traditional industries
Is there still hope for AI to be used in traditional industries? The answer is yes. The shift toward data - centric AI development has great potential, but what we are doing in the Industry of Consumer Internet is not suitable for traditional industries.
According to Ng, given the current complexity of AI, the bottleneck in the application usually lies in the quality and matching of the data. Traditional industries should adopt a "data -centric" model to obtain better quality and higher matching data.
Ng listed three things for enterprises to work on immediately to transform:
"Firstly, instead of merely focusing on the quantity of data collected, also consider the quality, make sure it clearly illustrates the concepts needed for the AI to learn.
Secondly, make sure the team considers taking a data -centric approach rather than a software-centric approach.
Thirdly，for any AI project intended for production, plan the deployment process and provide MLOps (Machine Learning Operations) tools to support it."
"Some good application cases have also emerged in traditional industries. For example, the image recognition AI system in the medical field can help doctors read CT images, identify tumors and other lesions, and assist doctors in making judgments," said Zhu.
In terms of AI technology application, the data should be given priority, and application should not be taken into account until high-quality data is realized. Without good data, it is hard to give birth to a really good application, he added.