It is commonly thought that the people who develop large language models (LLM) are artificial intelligence (AI) engineers. However, at a tech company in Beijing's Zhongguancun, there is also a team of young psychological trainers for AI, all of whom graduated in psychology.
What are their roles in developing LLMs?
"In the early stages, when asked what I should pay attention to when swimming, the AI will immediately give a very complete answer. It's a good answer, but it is not from the perspective of a child," Wang Jingwen, AI psychological trainer at Hanwang Technology Co., Ltd., told China Media Group when demonstrating a detection interface for English communication for primary and secondary school students.
"It behaves just like an expert, but we hope to make it into an image that is as friendly and lively as a child. So I can give the instruction that the answer needs to be understood by primary and secondary school students. This actually embodies the logic of developmental psychology, helping Al better understand what we want it to become," added Wang, who has a PhD from the Institute of Psychology at the Chinese Academy of Sciences.
The LLMs are just like children, and they need encouragement, Wang said. The AI trainers would repeat training instructions many times to allow them to gradually enhance their ability to communicate with people. Just like children, if there are more positive incentives, the large language models will perform better, she said.
The instructions given by Al psychological trainers are translated into code language by algorithm engineers. The LLM can quickly improve their ability to understand and generate human-like communication.
Representative image. /CFP
"Al trainers will provide a lot of high-quality and targeted data, which will greatly improve our model's ability," said Peng Fei, team leader in large language models.
"They also provide higher-level guidance and training methods. This is not a specific algorithm, but some theoretical ideas we can work towards," Peng added.
"As a psychology practitioner, when an LLM emerges, employment options might suddenly present you with choices beyond your imagination. Instead of interacting with people, you may interact with machines," said Wei Jingchao, another psychological trainer in the team who graduated from Fudan University.
The psychological trainer for AI bridges the gap between psychology and AI development, designing and implementing psychological principles and techniques to improve the interaction between AI systems and humans. Such cross-border integration of occupations is also emerging with the advancement of new quality productive forces.
"The development of these new professions is based on integration and innovations, which demand talents with cross-disciplinary skills or diverse backgrounds," said Zhang Chenggang, director of the Research Centre for China's New Employment Patterns at the Capital University of Economics and Business.
"Cross-disciplinary skills or diverse backgrounds are expected to improve the graduates' adaptability to such new jobs," the director noted.
About 69 million jobs are expected to be created in the next five years, and the majority of the fastest-growing roles are technology-related roles, with AI and machine learning specialists topping the list of fast-growing jobs, according to the World Economic Forum's The Future of Jobs Report 2023.