14
2020
-
05
Unveiling the Secrets: How to 'Train' Artificial Intelligence
The meaning of artificial intelligence is quite broad; the intelligence of industrial robots is also called "artificial intelligence." When industrial robots are endowed with intelligence, their potential will be limitless.
The Ministry of Human Resources and Social Security, the State Administration for Market Regulation, and the National Bureau of Statistics of China recently jointly released 16 new occupations, including intelligent manufacturing engineering technicians, with AI trainers among them. Regarding the work content and role of AI trainers, the Sputnik News Agency interviewed Xu Canhao, associate professor at the School of Computer Science and Technology, Soochow University.
Xu Canhao said: "For example, the AI used in autonomous driving needs to recognize traffic signs on the road. A certain traffic sign is an inverted triangle, which represents yielding. However, this inverted triangle sign will have differences in different images; sometimes there will be some color difference in color or contrast, or some differences in angle. At this time, we need thousands or even tens of thousands of photos, and the AI trainer needs to tell the AI that although these photos show inverted triangle signs with different colors, angles, brightness, and contrast, they all represent yielding."
According to the notice issued by the Ministry of Human Resources and Social Security of China, an AI trainer is defined as personnel who use intelligent training software to perform database management, algorithm parameter setting, human-computer interaction design, performance test tracking, and other auxiliary operations in the actual use of AI products.
The main tasks of an AI trainer are to annotate and process the original data of images, text, and voice; analyze and extract professional field characteristics, train and evaluate the algorithms, functions, and performance of AI products; design the interactive processes and application solutions of AI products; and monitor, analyze, and manage AI product application data. As for why this work of training AI cannot be done by the engineers who develop AI, Xu Canhao explained that the work handled by AI trainers is to make the algorithms designed by AI engineers more accurate and closer to people's conception and needs for this AI through training.

He said: "For example, when people were fighting the COVID-19 epidemic, they knew that some characteristics of CT images represented infection with COVID-19. So we extracted these characteristics and told the AI what characteristics are representative. If we improve it from the algorithm's perspective, even if our current algorithm has some self-learning capabilities, it is still some distance from maturity."
Alibaba Group's survey data shows that by 2022, the number of AI trainers worldwide is expected to reach 5 million, and a large talent gap needs to be filled. This shows that the employment prospects for AI trainers are very promising.
Next page
Related News