A Conversation of AI's Impact on Chinese Labor Market
The impact of AI on employment has become a topic of widespread concern for countries around the world. Many institutions in the United States and Europe have already begun researching this issue and reached some preliminary conclusions. However, there seems to be relatively little research on this topic in China, with even less information available to the public. Nevertheless, in June 2024, the National School of Development at Peking University organized a seminar, which led to many interesting discussions.
Panelists:
Lu Ming(陆铭), a member of the 14th National Committee of the Chinese People's Political Consultative Conference, Executive Dean of the China Development Research Institute at Shanghai Jiao Tong University, Distinguished Professor at the Antai College of Economics and Management at Shanghai Jiao Tong University, and Yangtze River Scholar;
Zhang Dandan(张丹丹), a young scholar at Peking University, Vice Dean of the National School of Development, and Yangtze River Young Scholar;
Xiang Kuanhu(向宽虎), Associate Professor at Shanghai University, Distinguished Research Fellow at the China Development Research Institute at Shanghai Jiao Tong University, and Dean of the Lingpeng Institute for Industry and Innovation;
Li Lixing(李力行), a young scholar at Peking University, Professor of Economics at the National School of Development, and Director of the China Public Finance Research Center;
Lu Jun(陆骏), General Manager of Suzhou Dick Vision Co., Ltd.;
Zhu Li(朱丽), Associate Researcher at Peking University's National School of Development;
Hu Jiayin(胡佳胤), Assistant Professor at Peking University's National School of Development;
Yu Hang(于航), Assistant Professor at Peking University's National School of Development and Assistant Professor at the South-South Institute.
Wang Xun(王勋), Associate Research Fellow and Deputy Director of the Think Tank Center at the National School of Development, Peking University moderated the discussion.
Full transcript of the discussion:
Wang Xun: I'd like to ask Professor Lu Ming, given the current economic environment, local governments are generally facing tight financial challenges, with some regions experiencing particularly significant fiscal pressure. In light of this, how should local governments effectively and steadily promote the supply of affordable housing?
Lu Ming: even with limited financial resources, local governments can effectively supply affordable housing by optimizing resource allocation. In the past, to maintain market order, the government may have invested a lot of money in unnecessary areas, such as the high costs of demolishing old houses, regulating illegal basement rentals, and tearing down illegal constructions. If these expenditures can be reduced or reallocated to increase housing supply, it would more directly and efficiently address housing issues. Therefore, I suggest that when finances are tight, the government should focus on core tasks and cut unnecessary spending. Additionally, there should be a differentiated approach for areas with significant population inflow and outflow. Cities with a large influx of people have a high demand for affordable housing. They can raise funds through government debt, while cities experiencing a significant outflow of people have less demand for affordable housing. They may not necessarily need to build affordable housing if finances are tight.
Wang Xun: From my understanding, Professor Zhang Dandan's research seems to suggest that artificial intelligence has a particularly significant impact on mid-skilled workers. However, the general perception is that the substitution effect of AI typically follows a gradual process from low-skilled to mid-skilled and then to high-skilled jobs. Could you elaborate on this, Professor Zhang?
Zhang Dandan: Clearly, this phenomenon challenges some intuitive understandings. According to Professor Lu Jun, his research more clearly finds that the application of AI has increased the demand for low-skilled workers, while mid-skilled workers are more likely to be affected by substitution effects, leading to a growth trend in skill demand at both the low and high ends. This phenomenon is particularly evident in the manufacturing labour market: as machinery and equipment are introduced, human labour is increasingly allocated to tasks that machines find difficult to reach or handle.
Take the polishing of new energy vehicle hubs as an example. Although about 90% of the polishing work has been automated, the edges and complex shapes of the hubs still require manual completion. These tasks require relatively low skills, thus increasing the demand for low-skilled labour to some extent.
When we analyzed the impact of AI and automation technologies on employment using CEGS data, we did not find significant shocks to low-end employment from these technologies. While some positions have been automated, new job opportunities requiring lower skills have also emerged, so overall employment levels have not significantly declined. This finding aligns with observations at the enterprise level. In contrast to the U.S., where mechanization and robotics in manufacturing have reduced considerably jobs, China has not yet experienced a similar large-scale decrease in employment.
Audience Question: Developing innovative drugs is a highly complex and resource-intensive process that requires huge money, capital, and time investments. This affects the survival and growth of companies and significantly impacts the innovation capacity and global competitiveness of China's pharmaceutical industry. In this context, what bottlenecks and potential risks will AI face in pharmaceutical R&D?
Lu Ming: I'll try to analyze your question from an economic perspective. Economists are focusing not just on the specific issues faced by the pharmaceutical industry but are more broadly concerned with overall technological progress in China.
The gap in innovation capability between China in emerging fields like AI and other countries is gradually widening. This is reflected in the slowing growth of unicorn companies and the significant disparity in the number of unicorns compared to the U.S. I submitted a proposal during the Two Sessions addressing this issue, which has garnered widespread attention domestically and internationally.
When exploring the root causes of this problem, we find significant barriers to funding. First, many foreign investments originally directed towards China are choosing to withdraw. While there is no shortage of domestic financing, a considerable proportion of it comes from state-owned enterprises. The management of state-owned funds often comes with strict performance evaluation mechanisms and stringent limits on investment losses, which conflicts with the inherent logic of innovation. Innovation is fundamentally a high-risk, high-reward activity, and allowing for a certain number of failures is necessary for success. However, under the management system of state-owned funds, if most investment projects fail, decision-makers may face accountability risks, leading state-owned funds to shy away from high-risk, potentially failing innovation projects.
Additionally, while this issue may manifest differently in the pharmaceutical sector, in cutting-edge fields like generative AI, there are also notable deficiencies in data foundations and quality. Specifically, the openness of China’s underlying databases is insufficient, and the quality of the data in the open databases also needs improvement. Compared to English databases, Chinese databases show significant scientific rigour, precision, and innovation gaps. This situation may also exist in the pharmaceutical field; for example, the quality of the presentation of the same research findings in Chinese and English literature may differ, thereby limiting innovation development in related areas. I hope this economic perspective partially addresses your question, even though these analyses may not fully target the pharmaceutical industry.
Lu Jun: Although I'm not an expert in the pharmaceutical industry, I’d like to share a few thoughts based on my understanding. The pharmaceutical sector has shown significant advantages in developing small-molecule drugs. This advantage mainly comes from the ability to continuously compute and optimize various structures of drug molecules using advanced technologies during the drug development process, significantly shortening the previously consumed time due to the lack of such technologies.
Currently, there are two significant challenges in this field. First, the application of digital twin technology on humans has not yet reached a mature level, so the essential Phase I, II, and III clinical trials in new drug development still need to be conducted conventionally. If breakthroughs in technology can be achieved in the future, it might reduce the need for animal testing to some extent, and early-stage digital twin technology may play a more significant role in this area.
Second, current drug development is based mainly on existing scientific knowledge, pushing the drug development process forward through modifications of known molecular structure groups based on understanding their functions. The limitations of foundational theories and knowledge somewhat restrict the originality and breakthroughs in drug development.
Audience Question: The first question is for Professor Lu Ming regarding your talk about improving efficiency in near-field services. With the continuous maturation of无人配送 (unmanned delivery) technology, we might even see humanoid robots being applied in this area in the future. Given this trend in AI development, are you concerned that these technologies will gradually replace related labour forces? The second question is for Professor Zhang Dandan. You mentioned significant differences in labour replacement levels between the U.S. and China. What are the reasons behind this difference?
Lu Ming: Your question is very insightful, especially about our report today. First, regarding the impact of AI on employment, I think this statement needs to be nuanced. Employment can refer to individual workers, specific job positions, or entire industries. I lean towards the view that introducing new technologies mainly replaces certain specific positions rather than completely replacing the labour force of an entire industry, let alone causing a widespread negative impact on overall employment demand in that industry.
Based on this, the application of emerging technologies like drones, smart delivery, and unmanned stores will indeed replace certain specific positions. Still, this replacement does not equate to a reduction in the total number of jobs in the industry. On the contrary, introducing new technologies often stimulates new demand, which drives job growth. For example, improved delivery efficiency might lead consumers who previously cooked at home out of efficiency concerns to rely on takeout services, creating more job opportunities in the delivery sector. This phenomenon is particularly evident in the service industry, reflecting the complementarity between different occupations rather than a simple substitution relationship.
Furthermore, when discussing substitution effects, we often overlook that in certain specific industries, especially in the service sector, complementarity may far exceed substitution. Substitution effects may be more pronounced in manufacturing, particularly in standardized production. Still, in the service industry, new technologies often give rise to new service models and employment forms, resulting in a net increase in overall employment. Therefore, while manufacturing jobs may decrease due to technological advancements, the service sector continues to expand due to the complementarity of technology.
Additionally, the characteristics of the near-field service industry, such as the close spatial proximity of supply and demand for services and its reliance on population density, mean that new technologies won't easily replace specific jobs (like domestic workers). This further underscores the profound impact of technological changes on the employment structure in the service sector, which is not merely a substitution but a more complex interplay of complementarity and restructuring.
Finally, I want to emphasize that technological transformation has not diminished the importance of the service sector; instead, it has made it even more critical in certain aspects. While I can’t delve deeply into this field in today’s discussion, my preliminary conclusion is that in the face of technological change, the service sector has not lost its significance; on the contrary, it has highlighted its foundational role in the socio-economic structure.
Zhang Dandan: Regarding the differences in AI research between China and the U.S., in manufacturing, our research seems to show that the application of automated robots has a positive impact on employment in China, while in the U.S., it shows a negative effect. This finding deviates from common perceptions. Furthermore, regarding AI applications, research at the enterprise level in the U.S. indicates that increased AI usage has driven labour demand growth. In contrast, our research (including a joint study with Professor Yu Hang and others) arrived at the opposite conclusion. In the U.S., automation has led to job losses, while in China, automation has promoted job growth; at the same time, AI applications in China have hurt employment, whereas AI applications in the U.S. seem to have a positive effect.
The reasons behind the differences in research between China and the U.S. are worth exploring in depth. One possible reason is that the rapid advancement of AI technology in the U.S. has stimulated new market demand. For example, AI can efficiently handle illustration work, which might lower the market price for illustrators but increase the demand for illustration services overall.
Lu Ming: this phenomenon can be summarized into two main effects. First, from the perspective of an open economy, taking the illustration industry as an example, the U.S. can significantly enhance efficiency in illustration creation through AI. It may attract global demand for illustration services to concentrate in the U.S., thereby generating greater demand for illustrators domestically. This is the demand concentration effect brought about by improved technological efficiency.
The opposite effect appears in China, possibly due to China's relatively conservative approach to the openness and application of AI technology, which has not significantly improved labour productivity. Instead, due to higher technological advancements in countries like the U.S., some jobs that could have been done domestically are outsourced abroad.
The second effect is that with the rise of the service sector in China and the widespread application of AI technology in this field, many new jobs have been created, attracting workers who were originally in manufacturing to shift towards the service industry. In comparison, the impact of AI on manufacturing jobs is more significant.
Zhang Dandan: Professor Li Lixing's research shows that, when comparing China and the U.S., growth in China is more pronounced at urban and industrial levels, which differs from our conclusions. I'm curious about how Professor Li arrived at this conclusion.
Li Lixing: In our report, we explored the multidimensional impacts of technological substitution, covering individual, company, and societal levels. At the company level, we focus on how businesses use new technologies to adapt to market demands, which may involve personnel adjustments or developing and applying new technologies. At the societal level, we examine whether the application of new technologies has spurred new market demands. Based on these different considerations, our research unfolds accordingly at various levels.
Taking publicly listed companies as an example, these companies, as representatives of ongoing operations and firm performance, tend to adopt new technologies to drive growth. If a particular technology fails to deliver the expected growth effects, they may choose not to adopt it. Therefore, in our research on publicly listed companies, we often observe the positive effects of technological substitution. However, to obtain a complete research conclusion, we must consider the dynamic changes within companies, including advancements and exits and specific performances across different levels (such as individual, occupation, company, city, and industry). Our study found a trend of increasing AI investment among publicly listed companies, which aligns with two related studies from the U.S.
Hu Jiayin: Our previous report touched on this topic when discussing the comparison of AI development between China and the U.S. A deeper analysis of the AI development landscape reveals significant differences between the two countries. In short, the U.S. focuses more on AI research and upstream technologies, while China leans towards the practical application and implementation of technology. This difference leads to a more pronounced substitution effect of AI technology in China, as it is often directly applied to transforming and upgrading existing industries, thereby intensifying changes in the job market. In contrast, in the U.S., AI development brings a new starting point for the entire industry, fostering the emergence of new demands, enhancing job creation and relatively weakening direct substitution effects. This phenomenon echoes the principles of global value chain division, indicating that in the current wave of AI technology, China may face more significant pressure to catch up in technology research and upstream industry chain segments.
One key point is particularly worth exploring in depth. Considering the overall proportion of AI researchers, China is undoubtedly relatively low. Moreover, the development in this field involves increasing the number of practitioners and a deep integration of academia, research institutions, and industry, along with comprehensive support from software and hardware technologies, collectively forming a sizeable industrial cluster. Therefore, the current situation and development trends in this area are crucial topics to which we must pay close attention.
Additionally, regarding the service industry, it is well-known that service prices in the U.S. are high, and labour costs are exceptionally high. From a demand-driven perspective, we may need to delve deeper into why, despite service wages in the U.S. being much higher than in China, there hasn't been a drastically different employment or economic situation compared to China.
Audience Question: My question deals with the complex phenomenon of job substitution and complementarity. Right now, our analysis mainly focuses on the jobs themselves, but the choice of this analytical unit is debatable. Artificial intelligence is gradually replacing Certain job functions, and the relationships between the remaining job contents are demonstrating new structural changes. A notable sign of this change is the frequent job adjustments in large companies; finding a large firm that hasn’t made departmental changes in three years is nearly impossible. Therefore, I tend to believe that when we dive deeper into the dynamic relationship between AI and labour, refining our analysis to the level of "job content" might be a more accurate entry point. This approach would help us understand the evolution of job content and profoundly reflect the complex interplay and mutual influence between artificial intelligence and human intelligence in the labour market.
Zhu Li: That’s a great question! Regarding the job content aspect you mentioned, we are further breaking it down to the skills of the individuals involved. Since researching organizational behaviour and human resource management, we’ve focused on skill reconstruction in job content since March 2023, collaborating closely with a significant domestic recruitment platform. Our primary focus is to understand how the emergence of AI affects the demand for employees' hard and soft skills during the hiring process. We’ve observed that with the increasing demands for AI technologies, mastering AI technical skills is no longer enough for employees to thrive; they also need a range of soft skills.
To this end, we’ve calculated an AI exposure index for specific job tasks, aiming to assess the specific impact of AI technologies on these positions through a comprehensive analysis of the various task combinations involved. Furthermore, we’ve researched how this influence reflects on employee compensation and career development for employees who possess strong AI hard skills alongside outstanding soft skills. In our internal practices, we’ve also noted the importance of having both AI hard skills and soft skills. Employees who have technical capabilities in AI and maintain good interactions within the company across departments and with superiors and subordinates tend to receive more development opportunities and broader career spaces.
Yu Hang: Your question has truly energized us, four collaborators. Right from the start of our current research, our core assumption was that AI would profoundly change job content. To validate this assumption, we’ve carefully analyzed the content changes in job advertisements, aiming to reveal the specific impacts of technological changes on job requirements by comparing recruitment ads for the same position five years ago and five years later.
Our analysis shows that in specific industries, especially those with low exposure to AI, like operators and technicians, the AI relevance of job positions is significantly increasing over time. For instance, five years ago, an operator's job might have included using computers or smart devices. In contrast, five years later, job ads increasingly focus on skills like specific parts polishing, reflecting the penetration and integration of AI technology into those roles.
However, we also realize that our current data analysis mainly focuses on positions with strong substitution effects, and our observations on the evolution of white-collar careers are still somewhat lacking. This might be related to the job ads' limitations; as a medium for external display, ads often need to label job titles clearly and are constrained by length and expression, making it difficult to comprehensively showcase the dynamic changes in job content. Therefore, we are eager to obtain more detailed internal data from companies, such as employee rotation records or experiences of the same employee across different job positions, as this data would provide us with a more intuitive and in-depth perspective.
Audience Question: Regarding the issue of network centrality and its relationship with structural holes, the report mentions that mastering individual skills can lead to changes in their network position. I want to explore further whether these position changes are temporary or closely tied to the nature of the job. This question arises from considerations of data characteristics, especially the nature of technology-triggered data. Take ChatGPT as an example—since its launch on November 30, 2022, it has sparked widespread attention and a learning frenzy. Some early adopters may have been able to master relevant skills or abilities quickly. Before these skills become standard requirements, individuals who possess them can occupy advantageous positions and enjoy a corresponding edge. However, as these skills gradually become mainstream and become generic workplace skills, those original advantages might face challenges or disappear. This raises the question: Are the changes we observe merely at the early stages of technological development?
Zhu Li: Your question is very forward-thinking! Currently, we have data covering up to the fourth quarter of 2023, and once we have the full year's data, we'll conduct further matching analysis to see if there's a sustained upward trend. The dynamic analysis based on existing data shows that relevant indicators are indeed on the rise as of the fourth quarter of 2023. Additionally, we've done an in-depth investigation concerning the job adjustment issue you mentioned. Internally, we look at personnel changes and meticulously code the levels employees are in and their job sequences. We can observe how employees transition between different levels and positions through this system. Perhaps this is a change typical of the early stages, and moving forward, we will continue to track the long-term development dynamics in this area.
In this transformative era, how to help people relieve job anxiety and explore better paths for development under the current employment conditions is an important issue I care about in the AI era. I look forward to sharing our findings with everyone when they're ready.
Audience Question: In light of the ongoing automation gradually replacing manual labour, the rise of the gig economy, the general decrease in labour supply in manufacturing, and the return of migrant workers, are the manufacturing sectors in the Yangtze River Delta and the Pearl River Delta still facing labour shortages? Will there still be labour crises?
Xiang Kuanhu: The so-called "labour crisis" in the Yangtze River Delta and Pearl River Delta regions reflects two layers of outcomes: first, the dynamic changes in the population flow, and second, the evolution of labour market demand.
From the perspective of population flow, especially since the end of 2022, there has likely been a short-term adjustment in social attitudes where people's enthusiasm for seeking high-paying jobs away from home has waned. This is mainly due to changes in the economic environment, making it more challenging to make money, prompting migrant workers to adjust and opt for a cheaper, closer lifestyle. However, looking at the long-term trend, as the economy gradually improves and the regional advantages for economic development remain, there’s still a high likelihood for these displaced workers to return to economically developed areas like the Yangtze and Pearl River Deltas. I've seen even friends and family who chose to "lie flat" during the pandemic, eventually opting to seek work outside due to limited local job opportunities. This proves that these regions still hold an attraction in the job market. So, analyzing from the supply side, there may be a short-term labour supply shortage in the manufacturing sectors of these areas, but in the long run, it shouldn’t be an unsolvable issue.
On the demand side, the term "labour crisis" often comes with a lot of unmet demand for workers. However, we are currently in a unique time where labour demand in manufacturing is actually declining in the short term, meaning the main issue in manufacturing labour supply and demand isn't a lack of workers but rather weak demand. Of course, not every industry in the Yangtze and Pearl River Delta regions faces the same difficulties. As Professor Lu Ming pointed out, the service industry might be fostering new points of labour demand growth. From the data we've seen, employment prices in 2024 show a different trend compared to previous years, currently maintaining a high level and even surpassing the historic highs of 2021. This indicates that there could be some degree of imbalance in the labour market overall, closely tied to the growth in demand for services and the relative insufficiency of labour supply. Just looking at manufacturing, the "labour crisis" might not be a widespread problem. In the medium to short term, maintaining employment and worker income through the development of the service industry could be a more pressing issue to focus on.
Zhang Dandan: The term "labour crisis" is frequently seen in the media and can be traced back over twenty years. Since the "Lewis Turning Point" theory was mentioned in the early 2000s, this phrase has appeared often in public discourse. As Professor Xiang mentioned, this phenomenon can be analyzed more deeply from both supply and demand perspectives.
From the supply side, although China faces negative population growth and the labor force has been downward since 2013, our total labor force still ranks first globally, and the extensive labor base remains strong. In fact, wage levels often influence the supply situation in the labour market. When wages exceed workers' psychological expectations, they are willing to offer labour supply. Thus, fluctuations in wage levels become a key mechanism for balancing labour market supply and demand. When the demand for labour from companies exceeds the supply, businesses often raise wage levels to attract more workers, which encourages more people to seek jobs outside.
On the demand side, a notable change is the trend of overseas expansion in China's manufacturing sector, especially pronounced in the Pearl River and Yangtze River Delta regions. Rough estimates suggest that these two regions have lost around one million jobs due to manufacturing transfer, which has decreased labour demand. However, looking from the supply side, the overall labour supply situation remains stable.
Additionally, it's important to note that changes in the international economic environment, such as trade policies and tariff adjustments in the U.S., may have somewhat salvaged manufacturing jobs in countries like the U.S. Still, they also have far-reaching impacts on China's labour market, potentially putting many workers at risk of unemployment.
Audience Question: How will the arrangement of working hours change in the future? What’s the inherent connection between these adjustments in time dimensions and innovation activities?
Zhu Li: From an innovation perspective, we conducted an online survey at the beginning of 2023 to explore the relationship between changes in work content and AI technology. The results showed that the introduction of AI not only facilitates incremental innovation but also drives breakthrough innovation.
Regarding changes in working hours, we have reason to believe that with the ongoing advancements of the Fourth Industrial Revolution and productivity improvements, the overall working hours for humanity are showing a trend toward reduction. From a global viewpoint, shorter working hours are a general trend. We are optimistic about this and believe such changes will create more leisure time, which will help enhance human creativity.
Additionally, we studied how public attitudes have shifted after the increased use of AI technologies. AI is seen as both an opportunity and a challenge, and more people are beginning to view it as a work assistant or complementary partner. This shift in perception significantly impacts creativity development, showcasing AI's multifaceted applications. If we can harness AI correctly and adjust our mindset to face technological challenges positively, our innovation capabilities can indeed be enhanced by AI.
Ultimately, the crux of the matter is that we need to see AI as a tool, assistant, or partner rather than a replacement and proactively explore ways to apply it, taking personal responsibility for innovation.
Audience Question: In the context of the growing prevalence of artificial intelligence, how will China's basic education adjust and respond to better cultivate students' overall qualities and innovative abilities?
Zhang Dandan: The rapid development of AI technology urgently calls for profound reforms in the education sector. The future path of education, especially vocational education, is a pressing issue that needs to be addressed. My research shows that if individuals only engage in low-end jobs in high-end manufacturing, the value of a high education level cannot be fully realized. However, a higher education level provides individuals with a broader range of choices, allowing them to explore more diverse career fields.
In the long run, the labour market demand gradually splits into two ends: highly-educated and low-educated workers. This creates a tricky situation for those with a medium level of education. If they don’t continue to improve themselves, they may fall into the low-end labour market and find it hard to climb back up. Given the large population with medium education levels in China, the future development path of this group deserves particular attention.
At the basic education level, the current education system tends to focus on exam-oriented learning and addressing weaknesses rather than enhancing students' strengths. Students must develop comprehensively; they should achieve a certain level in math, Chinese, and English to enter quality high schools, or else they might shift to vocational education. However, this educational philosophy may need to be adjusted in the AI era. In the face of competition from machines and artificial intelligence, humans should focus more on developing their unique skills to stand out and excel in specific fields. Regarding learning foreign languages like English, while the rise of instant translation technology seems to lessen the urgency of language learning, the unique value of language as a medium for cultural transmission and artistic expression should not be overlooked.
Lu Ming: As an essential part of the education system, the future direction of vocational education must closely align with the changes in human industry and job demands.
First, vocational education shouldn't be limited to training in traditional manufacturing skills but should focus more on the skills needed for service industry positions. For example, with the rise of new industries like live-streaming e-commerce, the shortage of relevant skills is becoming increasingly prominent. It requires vocational education to promptly adjust its direction and cultivate more talent suited to these job demands rather than simply guiding students toward traditional manufacturing roles.
Second, vocational education needs to reassess and strengthen those skill areas that are hard to replace with AI. Traditional vocational education often emphasizes teaching specific operational skills, but AI is gradually taking over these jobs. Instead, social skills, communication abilities, and teamwork skills have become indispensable soft skills in the new era. Therefore, vocational education should increase investment in these areas, such as training students in PPT creation, public speaking, and teamwork, to prepare them for future workplace challenges.
Additionally, I would like to discuss the importance of English education. While English may not be essential for most people, its importance increases for those aiming to become global citizens and engage in international communication. Especially in the AI era, the demand for English has shifted from reading and writing to listening and speaking, which requires us to focus more on oral and listening training in English programs. Therefore, the intensity and specificity of English education need to be enhanced to meet the development needs of different professions and individuals.
Finally, on a broader level, societal differentiation will become more pronounced in the future, with different individuals, industries, and companies facing varying opportunities and challenges due to technological changes. For individuals, those who can keep up with technological trends and master specific skills will experience significant advancements; conversely, those who cannot fall into low-skill, low-income situations. Thus, national policies should increase support for vocational education and skills training, helping those at risk of marginalization to improve themselves and narrow the income gap.
Audience Question: How should vocational education develop upward? At the same time, how can we inspire students' intrinsic motivation and promote substantial improvements in their skill levels?
Lu Ming: I have two suggestions for consideration.
First, vocational education is gradually entering challenging job fields traditionally dominated by university graduates, such as programming positions. With the influence of AI technology, many programming jobs that once required highly specialized knowledge and educational backgrounds can now also be handled by college diploma holders, creating a competitive job market with university graduates to some extent.
Second, we need to reassess and recognize the value of various skills in the service industry. Take live-streaming e-commerce as an example; this emerging industry is not just about hosts selling products live but also involves a whole range of related industries, like studio setup and design. These skills have significant development potential and application value but are often overlooked or underestimated. Therefore, we should enhance our understanding and training of these service industry skills to adapt to and promote the growth of emerging industries.
Audience Question: Some vocational schools have established "live-streaming e-commerce" programs to meet industry demand, however, with the increasing maturity of digital human (virtual hosts, digital characters, etc.) technology. We are concerned about the future direction of the "live-streaming e-commerce" program.
Lu Ming: The core skills needed for live-streaming e-commerce lie in interaction and communication abilities, which are areas that digital human technology currently struggles to fully replace. The success of Dong Yuhui illustrates that a deep cultural background and a profound understanding of the products are key to gaining widespread recognition. Similarly, the high-ticket sales phenomenon on platforms like Xiaohongshu relies on the seller's deep understanding and unique insights into the fashion industry. These qualities are not exclusive to higher education; they can be cultivated through training for vocational students and even a broader audience.
Therefore, in education and training, we should focus on developing individuals' appreciation for beauty, understanding of fashion, and non-technical communication skills—these are human advantages that robots find hard to replicate.
In summary, three points are worth emphasizing:
Vocational education should pay attention to and potentially enter job fields traditionally dominated by university graduates.
The service industry, especially emerging service industries, should become one of the important directions for vocational education.
There should be a strong emphasis on enhancing social skills training, as these are core competencies that digital human technology cannot easily replace.
Audience Question: With the current downturn in the housing market and falling rents, how can the real estate industry find new development space and opportunities in a stagnant environment?
Lu Ming: First, changes in housing demand are heavily influenced by shifts in population distribution. As I discussed in my book "City of Attraction," there's a clear trend of population movement—people are migrating from rural areas to cities, flocking from more minor to larger ones, and within big cities, populations and traffic are increasingly concentrated in the central districts.
For a few mega or super megacities, the suburban housing market might only see growth opportunities if the population increases. If the population remains stable, the suburban housing market will face downward pressure due to the centralization of people and activity.
Secondly, the nature of housing products is undergoing a significant transformation; they're moving beyond just being places to live and increasingly incorporating service attributes. This change is already evident in the market—communities with high-quality property management often see their property values exceeding those with poor management.
So, looking ahead, real estate developers are gradually shifting towards becoming urban life service providers.
Audience Question: Research by Professor Yu Hang shows that the exposure of AI in jobs is declining. This seems contradictory to the reality that AI is widely impacting and enhancing efficiency, which is puzzling. Additionally, the study is based on a sample of a million; was it randomly sampled to reflect the whole? And is measuring AI exposure reasonable and accurately reflecting actual use?
Yu Hang: Regarding the sample's representativeness, our study was conducted in collaboration with the Zhaopin recruitment platform, using random sampling each year, so the data from the recruitment website is reliable. However, whether this sample represents, China's overall labour market needs further exploration.
As for the overall decline in AI exposure, I can't pinpoint the exact direct cause. However, after in-depth analysis, I believe this reflects changes in the demand structure across different occupations rather than simply adjusting job contents within the same occupation. Specifically, there has been a significant increase in positions in the labour market with relatively low AI exposure, such as operators and technicians. In contrast, roles with high AI exposure, like translators and programmers, have not grown simultaneously. This explains why, from an overall perspective, AI exposure appears to be declining.
Regarding the timeline, the emergence of ChatGPT doesn't coincide with the onset of the decline in AI exposure (which began in 2018). That's because the development of large language model technology had been brewing for some time, and companies may have already been making strategic preparations, leading relevant industries to adjust at a macro level.
More importantly, we focus on the unique dynamics displayed in China's labour market on the eve of the large-scale explosion of AI technology. We’re careful with our wording; we don’t directly state that this "eve" caused market changes, but we aim to explore how China's labour market is preparing to welcome the arrival of AI technology.
The research indicates that on the eve of the explosion of AI technology, industries with a high match to AI technology did not see the rapid growth expected; instead, there has been a notable increase in the proportion of jobs with lower AI exposure.
Audience Question: Can we understand this as saying that the structural changes in the labour market are probably more related to fluctuations in macro industry demand rather than being directly determined by the development of AI technology itself? For instance, the recent reduction in hiring by large internet companies is mainly due to the various shocks and challenges the industry has faced over the past few years.
Yu Hang: Macroeconomic factors undoubtedly have a profound impact on various industries.
First, quantifying the specific proportion of AI's influence in the macro context is challenging and can't be generalized. Looking at the U.S. or other countries, we don't see the same clear trends as in China. This reflects different response patterns due to varying national contexts against the backdrop of global technological advancement.
Second, while we didn’t dive deep into specific technical details during today’s discussion, we employed various instrumental variable methods throughout our research for a more in-depth and comprehensive understanding while ensuring the robustness of the results.
Hu Jiayin: Let me add two points.
First, regarding the understanding of "exposure." In this study, we focus on the theoretical possibility of jobs being replaced by AI. This assessment primarily analyzes the tasks and activities required for each position, allowing us to infer the potential level of impact or replacement by AI technology once large language models are introduced. Therefore, exposure is regarded as a theoretical measure.
Second, we analyzed the trends in job postings for positions with relatively high AI exposure in recent years. In doing so, we’ve adequately considered and controlled for macro factors such as time-fixed and industry-fixed effects to ensure the accuracy of our analyses. Unless in extreme cases—like large companies making layoffs strictly based on the level of AI exposure per position—we won’t simply attribute changes in numbers to the influence of AI exposure.
However, one thought-provoking phenomenon is why companies choose positions based on the AI exposure indicators during layoffs. This reflects that driven by AI technology, especially with the application of large language models, companies can optimize their human resource allocation more effectively. For instance, work that initially required two employees to handle might only need one with AI assistance, particularly for entry-level or highly repetitive tasks.
This is precisely the macro trend we aim to depict.