How can industrial intelligence change the employment structure of the floating population? (2024)

Li Siyao, Conceptualization, Data curation, Formal analysis, Funding acquisition, Methodology, Writing – original draft, Writing – review & editing,¹ Wang Ya, Conceptualization, Data curation, Formal analysis, Methodology, Writing – original draft, Writing – review & editing,^2,* and Li Yingfeng, Formal analysis, Methodology, Writing – original draft, Writing – review & editing²

Three effects of AI on the labor market

Digital technologies, closely related to artificial intelligence, have become an important driver of economic growth. It is not only a key direction for future sustainable development, but also has an important impact on the labor market [1,2]. Autor et al. explored the impact of AI on the labor market macroeconomic model and laid the analysis framework foundation for subsequent research [3]. The model for the "task-based model" explains the interaction effect of technology and labor, proving that technological progress on labor demand also has productivity and new employment effects. Acemoglu and Restrepo believed that the substitution effect would have an impact on the labor market [4,5]. Technological advances can bring a lot of industrial robots to replace the workforce in the labor market, thereby reducing costs while increasing productivity and thus potentially reducing the balanced wage level of labor. In terms of the productivity effect, Haltiwanger et al. identified three pathways. First, the large-scale use of AI automation brought about by technological advances in industrial enterprises would greatly save the production costs of enterprises. The larger the scale is, the lower the cost of the final product and services produced. The scale effect can encourage companies to further expand their operations, which in turn increases the demand for labor employment by increasing productivity. Second, the demand for products and services in related industries will also increase. Expanding the scale of related industries will also increase the labor demand for non-AI replacement jobs. Again, the scale effects will drive companies to scale up and increase the number of jobs while increasing labor productivity [6]. The bargaining power of the workforce will also improve, thus increasing the balanced wages in the labor market. In terms of the job creation effect brought by technology, macroeconomic models adopted by Acemoglu and Restrepo demonstrate that under the automation conditions of AI, new industries and services will accelerate in the economic market and produce new jobs [4,7].

Changes in the labor market employment structure brought about by AI

Technological advancements (AI) not only contribute to economic growth and increased productivity but also have an impact on the labor market. Schumpeter’s theory of ’destructive creativity’ highlights the concept of creative destruction as a fundamental process in technological progress and economic development. This process involves the emergence of innovative ideas, processes, products, and organizations, while routine tasks are continuously phased out [8]. Autor et al. put forth theoretical models that highlight the impact of technological progress on different-skilled workforces, revealing a parretn of polarized employment [7,9]. Generally, the work content held by the skilled workforce is most easily replaced by AI because its work involves programming, formatting, and routine activities. In contrast, a low-skilled job, although simple in nature, often requires hands-on machine operation and similar tasks, which leads to increased labor demand rather than being impacted by it. Conversely, high-skilled labor involves complex tasks and requires more mental activities, making it less likely to be replaced by AI. In fact, the progress of AI further promotes the demand for highly skilled personnel.

Quantitative research on the impact of AI on labor structure

Several studies have quantified the likelihood of occupational substitution due to AI, revealing varying impacts on different occupations. For instance, in the United States, approximately 47% of occupations are susceptible to substitution, while the figures are around 6% in Korea, 12% in Austria, 42% in Canada, and roughly 33% overall [10]. Additionally, many studies suggest that AI will also create new employment opportunities. With the rise in popularity of computers and the application of software technology in various industries, new job opportunities have emerged. These include roles in software and application development, database design and analysis, network security and maintenance, among others [11]. In addition to its destructive and creative effects, AI had led to a reshaping of work tasks. This has resulted in a structural restructuring of jobs, which has had a heterogeneous impact on labor with different skills. Scholars have referred to this phenomenon as ’Skilled-Biased Technological Changes’ (SBTC). The analysis of the empirical data of the manufacturing industries in European and American countries confirmed that AI has the strongest substitution effect on the medium-skilled labor force while positively affecting the high- and low-skilled labor forces [12,13]. Lv and Zhang conducted an in-depth analysis of China’s manufacturing industry and showed similar results, with an increased employment structure for high- and low-skilled workforce, an increased proportion of high- and low-technology industries, and a decreased proportion of employment in the technology industry [14]. Hao, using the education degree of labor to analyze the technical level of various industries, found that different industries exhibit an “N”-shaped employment and skills structure. Automation leads to a decline in employment for middle-skilled workers, while high- and low-skilled workforces demonstrate an increasing trend [15]. Qu and Cheng found that the employment structure of migrant workers, namely, the floating population, showed a polarization trend, and the growth rate of high- and low-skilled positions was higher than that of middle-skilled positions [16]. From the perspective of industrial structure transformation and upgrading and the open-trade economy, Li found that the polarization trend of China’s employment structure was verified again [17]. Sun and Hou further integrated industrial structure upgrading, regional differences, industrial intelligence, and employment structure, and a macroeconomic theoretical model was constructed to empirically analyze and demonstrate the polarization and internal mechanisms of labor employment [18].

In summary, foreign studies derived empirical data from European and American countries through macroeconomic theoretical models and reached a basic consensus on the trend of employment polarization. On the basis of the basic ideas and research framework, the existing research in China has used a series of Chinese data to verify the polarization. At present, the unified conclusion is that AI affects the employment structure of the workforce. Compared with the whole labor market, the labor market of the floating population is more sensitive. However, the related literature specifically focusing on the impact of AI on the employment structure of the floating population is limited. Furthermore, a standard theoretical model for analyzing this impact has yet to be established. Building upon this gap, this study proposes an analytical framework to examine the employment structure of the floating population in relation to AI. Drawing inspiration from influential works by foreign scholars Acemoglu and Autor [7] as well as domestic scholars such as Sun and Hou [19], this study developed a macroeconomic theory to derive insights and analyze the mechanisms underlying the impact of AI on the employment structure of the floating population. An empirical analysis is also conducted to explore its internal differences and far-reaching effects.

Equilibrium analysis of the basic mode

The manufacturer used the Z skill level in task i to construct the profit function. See Eq (1) for more details:

Assuming a fully competitive market, all interval tasks needed to be performed. To achieve balance, there is a low-skilled task IL when i <IL, and there is a high-skilled task IH when i >IL. That is, different skilled floating workforces undertake different tasks. The medium- and low-skilled workforce cannot perform the complex work of high-skilled labor, and the high-skilled workforce will not take up low-yield and low-skilled jobs. Hence, m(i) = h(i) = 0 when 0 < i < IL, when IL < i < IH, I(i) = h(i) = 0, and when IH < i < 1, I(i) = m(i) = 0.

Equilibrium analysis of the AI input

The manufacturer invests AI capital to build the production function of task i. See Eq (2) for more details:

where q(i) and Q(i) show the number and efficiency of AI capital in conventional production task i and unconventional production task αQ (i) = 0, respectively. A design was embodied in multiple Acemoglu classics and further categorized to represent the production function. See Eq (3) for more details:

This classified production function is interpreted as the task interval of three different skilled floating population workforces. In (0, IL) range, for unconventional production tasks, but simple tasks, only manual labor can be completed by low-skilled floating population labor. In (IL, IH) range, for routine production tasks, task types are more complex, but a high degree of programming can be completed by the medium-skilled floating population workforce. In (IH, 1) interval, for unconventional production tasks, the tasks are complex, need high mental labor and technical requirements, and should be completed by the high-skilled floating population labor force.

AI capital is reflected in the routine task of the medium-skilled floating population workforce; therefore, the question is exactly how to replace it. The present study constructed a better idea of thought. Within the interval (IL, IH), a subinterval exists (I-ε, Iε). The productivity of AI capital is αQ(i). Corporate decision makers are more inclined to use AI capital to replace the medium-skilled floating population workforce, termed machine replacement. For the subinterval (I-ε, Iε), the productivity of AI capital αQ(i) is not higher than that of skilled migrant workers, and no substitution occurs. When 0 < i < IL`, m (i) = h(i) = q(i) = 0; when I-ε < i < Iε, l(i) = m(i) = h(i) = 0. For part of the interval (IL, IH), except for the subinterval (I-ε, Iε), IL `< i < I-ε and Iε < i < IH`, l (i) = h (i) = q (i) = 0; when IH`< i < 1, l(i) = m(i) = q(i) = 0.

Under the constraints of minimizing production costs and maximizing production efficiency, AI capital will replace the medium-skilled floating population workforce in the corresponding range. The demand function of labor can be represented by the following equation. See Eq (4) for more details:

Given the equal marginal productivity of the tasks performed by the same skilled migrant workforce, see Eq (5) for more details:

Let P be the price index. PL, PM, and PH are the price indexes of unit tasks of low-, medium-, and high-skilled floating populations. See Eq (6) for more details:

The formula is added to obtain three types of skills of the floating population. For the total demand function of the workforce, see Eq (7) for more details:

At this point, a slight change in form can yield the relative needs of the three types of skilled floating labor. See Eq (8) for more details:

AI and floating population labor substitution are related to the subrange of AI ε. The related research data show that the continuous development of AI ensures the goal of minimizing production costs, thus achieving the automation of conventional production tasks. That is, the scope of ε is extended, implying that the demand for a middle-skilled floating population workforce decreases compared with that for a high-skilled and low-skilled floating population workforce. See Eq (9) for more details:

Through a series of derivations and evolutions, we drew the following conclusions.

After theoretical derivation, it can be found that in both Western countries and China, with the development of artificial intelligence technology, automation will gradually replace a greater number of jobs held by a medium-skilled workforce. However, the impact on the replacement of highly skilled and low-skilled floating workforces within the floating population is relatively minimal. It is expected that the positive productivity effect and the employment increase effect may gradually increase.

Based on this, this study proposed two research hypotheses.

• Hypothesis 1: The application of industrial robots replaces the employment of the current floating population workforce.

• Hypothesis 2: The application of industrial robots has the effect of medium-skilled substitution and low- and high-skilled promotion in different skill employment groups. That is, the impact of industrial robots on a heterogeneous skill workforce was bipolar promotion and central substitution.

Model setting

The labor equilibrium model in the AI stage was constructed, which provided theoretical support for the competition equilibrium of different skilled workforces under the automation perspective. This study established the following empirical model of benchmark regression. See Eq (10) for more details:

The aforementioned equation shows the effect of industrial robot permeability on the employment or compensation of the floating population. The dependent variable upper target 1,2…n represents the floating population of different employment identities and skill categories; j and t of each variable represent the province and year, respectively; α and β represent the coefficients; α1 and β1 are the coefficients of the core explanatory variable, indicating the marginal substitution effect of industrial robot permeability on the employment demand of the floating population; X represents a series of control variables; δ and λ represent the provincial fixed effect and year fixed effect, respectively; and ε indicates the error item.

Internal issues

The core independent variable, industrial robot permeability, affected the dependent variables of floating population employment demand and salary. Enterprises’ decisions to introduce industrial robots are often driven by the prevailing "labor shortage" and "talent shortage" challenges in recent years as well as the need to address labor supply and demand contradiction and rising labor costs [19]. Therefore, endogeneity problems might exist in core independent variables that need to be solved using exogenous instrumental variables. This study drew lessons from the findings of Acemoglu and Restrepo and proposed the use of American industrial robot data as a instrumental variable for Chinese industrial robot permeability, thereby mitigating potential endogeneity problems.

Variable description and data sources

The time span of this study was 2012–2018, mainly considering two points. One was only the installation and stock of Chinese industrial robots after 2012. The Global Robot 2019 released by the International Federation of Robotics (IFR) noted that by 2018, the annual sales of China’s industrial robot market ranked first in the world for the sixth consecutive year. In China, the prosperity of the industrial robot market inevitably had an impact on the employment market. Second, after 2012, the labor structure faced the impact of the "superposition effect" of the third phase of population policy with the growth of the working-age population [20]. The secondary industry represented by the manufacturing industry also faced the dilemma of "labor shortages" and "talent shortages”, and the contradiction between supply and demand between the workforce and employment gradually became prominent. The floating population is the main force of the employment market, and the impact was more significant. We should consider how robots affect the employment effect of the floating population, what types of people face "machine replacement", and what the characteristics of "human–computer cooperation" are to promote the organic integration of AI and the employment market. Further research is needed to provide a more accurate and reliable basis for policymaking.

The main data sources of this study are three: (1) Industrial robot data. It came from the installation volume and stock data of industrial robots in the IFR report. To date, the IFR data included industrial robot data from 50 countries during 1993–2019. This study selected four types of industrial robot data as the research objects from the mining industry, manufacturing industry, power, heat, gas, and water production and supply industry, and construction industry during 2012–2018. (2) Floating population data. It came from the China Migrants Dynamic Survey (CMDS) during 2012–2018. These surveys have been carried out annually since 2009; they cover 31 provinces/ autonomous regions/ municipalities (hereinafter referred to as “provinces”). Data was released most recently in 2018, which is used in this paper. The CMDS sample is composed of floating population members aged 15 and above who have lived in their inflow locales for more than one month but do not have household registration (hukou) in those locales. Considering the industry characteristics of the employed floating population in the sample, the floating population in secondary industries such as mining, manufacturing, coal water thermal production and supply industry, and construction was selected as the research object. (3) Statistical Yearbook data. Variable data involving enterprises and regional levels were derived from the National Statistical Yearbook, China Labor Statistical Yearbook, China Industrial Statistical Yearbook, China Science and Technology Statistical Yearbook, and so forth. The specific index selection instructions were as follows. The statistical Yearbook data of China reflects the social and economic characteristics of recent years, and is an effective supplement to the study of this paper.

Explain variable floatingempjt1,2…n: This indicates the employment of the floating population in the secondary industry representing different employment identities and skill categories. In the regression, first, the employment demand and salary effect of the total floating population were measured. Second, employment identity was subdivided, and the employment demand and salary effect of employees, employers, and proprietary workers were measured. Finally, the skill categories were subdivided; education was the most common division. Based on the empirical results of junior high school and below set as low skills, high school/technical secondary school education as medium skills, and a higher degree or above as high skills, the employers and proprietary workers were divided into high-, middle-, and low-skilled categories according to employment status to further measure the effect of independent variables on floating population workforcess with different skill statuses. At the same time, considering the sample size of the sample data, the employment proportion of various floating populations was selected for a comparative analysis to ensure the effectiveness of the results.

Core independent variable PIRexposuretorobotsjt: This indicates the penetration of the industrial robot, that is, the technical impact intensity induced by the robot. The industrial robot stock data of the IFR were mainly used, but these data only reported the robot stock at the national level of subindustries, with no data from subdivided units. Previous scholars have also noted this point. The approach was to measure the robot data at the provincial level [21,22] That is, the industrial employment structure of each province was adopted to form the initial weight and multiplied by the industrial robot penetration (coverage) of employees at the industry level. Hence, the penetration index of the industrial robot at the provincial level was determined. The specific calculation method was as follows:

First, the industrial robot permeability index at the industry level was calculated and recorded as IRit. See Eq (11) for more details:

This metric (IRit) measures the industrial robot penetration of t years in industry i. MRit represents the i industry t annual industrial robot stock, and Labori,2010 indicates the number of employees i in 2010 (base).

Second, the permeability index of the interprovincial industrial robot was estimated and recorded as PIR exposure to robotsjt. See Eq (12) for more details:

This metric (PIR exposure to robotsjt) measured the number of industrial robots used per 10,000 people at the interprovincial level in different years. Empi,j,2011∑iEmpi,j,2011 is the proportion of urban employment in industry i in province j in 2011 to the total urban employment in 2011.

As shown by the aforementioned two types, the penetration of industrial robots at the interprovincial level was the weighted average of the use of robots at the industry level, and the weight was the share of the total employment in the industrial industry in the year before the inspection sample (2011). In theory, if an industry occupies a large share of employment in the region, then the robot growth in the industry has a greater impact on the labor market changes in the region, which directly affects the floating population employment market in the industry. The difference, on the one hand, comes from the difference in robot growth at different industrial industry levels between 2012 and 2018; on the other hand, it comes from the differences in industrial employment structure in different provinces.

Control variables: They mainly control the characteristic variables at the industrial industry and provincial levels. Referring to the study by Cheng et al., the industrial variables include the following: scale characteristics, measured by the average number of employees of industrial enterprises above the designated scale; capital investment, measured by the fixed asset investment of enterprises above the designated scale; and profit level, measured by the total profits of enterprises above the designated scale [23]. Considering the characteristics of the floating population and referring to Han and Zhao and Wei et al., the provincial level included the following: the urbanization rate, measured by the proportion of the urban population in each province; and the housing price level, measured by the average sales price of commercial housing in each province [24,25].

The description analysis of the various variables is shown in Table 1.

Influence of the application of industrial robots on the employment of the floating population in secondary industry

Table 2 reports the impact of industrial robot applications on the employment of the floating population in the secondary industry. The application of industrial robots had a significant negative impact on the employment of the floating population in the secondary industry. Model (1a) and model (1b) used simple benchmark regression, and model (2a) and model (2b) used instrumental variable regression. The two regression results were generally similar in the direction of action. To avoid the bias in the estimated coefficients caused by endogeneity problems such as omitted variables and reverse causality in model construction, this article uses instrumental variable regression to verify the results of the benchmark regression. In the selection of instrumental variables, it is important to ensure that the selected instrumental variables are both related to the core explanatory variable and do not affect the explained variable. The instrumental variable test results, LM statistics (35.912), and F statistics (1184.739) were significant at the 1% level, which justified the selection of instrumental variables. Controlling for various variables and effects, model (2a) showed that for each unit increase in industrial robot permeability, the total floating population employment of the secondary industry decreased by 0.029 units. Model (2b) showed that for each unit increase in industrial robot permeability, the employment proportion of the floating population in the secondary industry declined by 0.086 units. Both absolute and relative quantities indicated that the application of industrial robots had a substitution effect on the employment of the current floating population in the secondary industry. The robustness checks (Adjust the sample range, the samples of Beijing and Shanghai were excluded. And replace the measurement method) also yielded the same results. Thus validating hypothesis 1.

Influence of industrial robot application on the employment of different employment identities

Table 3 reports the impact of industrial robot applications based on instrumental variable regression models on the employment of the floating population with different employment identities. Industrial robot application had a significant negative impact on employees, a significant positive impact on employers, and a positive but not significant impact on self-employed workers. Model (3a) and model (3b) showed that a 1-unit increase in industrial robot permeability resulted in a decrease of 0.320 units in employee employment and a decrease of 0.115 percentage points in the employee employment ratio. Model (3c) and model (3d) indicated that a 1-unit increase led to a 0.002 unit increase in employer employment and a 0.047 percentage point increase in the employer employment ratio. Model (3e) and model (3f) showed that the industrial robot permeability on proprietary worker employment ratio and employment was not significant. Both absolute and relative amounts showed that the application of industrial robots had a substitution effect on the current employment of employees, a promotion effect on employer employment, and no significant impact on the employment of self-employed workers.

Influence of the application of industrial robots on the employment of different skills

Before dividing the floating population with different skills by education as the standard, least squares regression analysis was performed for each education group, and the skills were grouped according to the outcome characteristics. Tables ​Tables44 and ​and55 report the impact of industrial robot applications on the employment of floating populations to different degrees. As a result, the application of industrial robot employment had a significant positive impact on the total employment of unemployed, undergraduate, and graduate students and on the total employment of junior and high school groups. Furthermore, industrial robot application had a significant positive impact on the employment proportion of primary school groups and on the employment proportion of college, undergraduate, and graduate students.

Model (4a) and model (4b) indicated that for each 1 unit increase in industrial robot permeability, the total outstanding employment increased by 0.001 units. The proportion of employment impact was not obvious. Model (4c) and model (4d) indicated that for each 1-unit increase in industrial robot permeability, the impact on the total amount of primary school employment was not obvious. The employment proportion increased by 0.316 percentage points. Model (4e) and model (4f) indicated that for each 1-unit increase in industrial robot permeability, the total employment in junior high schools decreased by 0.042 units. The proportion of employment impact was not obvious. Model (4g) and model (4h) indicated that for each 1-unit increase in industrial robot permeability, the total employment in high school/technical secondary schools decreased by 0.012 units. The proportion of employment impact was not obvious. Model (4i) and model (4j) indicated that for each 1 unit increase in industrial robot permeability, the impact on the total amount of employment in college students was not obvious. The proportion of employment decreased by 0.087 percentage points. Model (4k) and model (4l) indicated that for each 1-unit increase in industrial robot permeability, the total college undergraduate employment increased by 0.004 units. The percentage of employment decreased by 0.050 percentage points. The promotion effect on total employment was higher than the substitution effect of the proportion of employment. Model (4m) and model (4n) indicated that for each 1 unit increase in industrial robot permeability, the total graduate employment increased by 0.0001 units due to decimal retention number issues. The later digits could not be displayed. For each 1-unit increase in industrial robot permeability, the total graduate employment increased by 0.0002332 units. The proportion of employment impact was not obvious. The analysis revealed that the application of industrial robots had a promotion effect on unfinished school and primary school groups, a substitution effect on middle school, high school/technical secondary school, and college specialties, and a promotion effect on college undergraduate and graduate students.

Therefore, the skills of the floating population by education could be divided into three groups: low skills (not school and primary school group), middle skills (high school, high school secondary school. and college specialty group), and high skills (college and graduate group). Industrial robot application had a promoting effect on low- and high-skilled employees, thus validating hypothesis 2.

Cross-impact of the application of industrial robots on the employment of different skills and employment identities

Previous empirical results showed that the application of industrial robots had a significant effect on the floating population with different skills and employment identities. Furthermore, employees and employers were divided into six groups of low, middle, and high skills to explore the cross-impact. Table 6 reports the impact of industrial robot applications on the employment of the floating population with different skills and employment identities. As a result, industrial robot application had a significant negative impact on low-skilled employees and significant positive effects on high-skilled employers.

Whether it is total employment or employment proportion, industrial robot application has a significant negative impact on employees, a significant positive impact on employers, and a positive but not significant impact on self-employed workers. Model (3a) and model (3b) showed that for 1 unit of industrial robot permeability, employee employment decreased by 0.320 units, and the employee employment ratio decreased by 0.115 percentage points. Model (3c) and model (3d) showed that for 1 unit of industrial robot permeability, employee employment increased by 0.002 units, and the employer employment ratio increased by 0.047 percentage points. Model (3e) and model (3f) showed that the industrial robot permeability on proprietary worker employment ratio and employment was not significant. Both the absolute and relative quantities indicated that the application of industrial robots had a substitution effect on current employee employment, promoted the employer employment effect, and had no significant impact on the employment of self-employed workers. Hypothesis 2 was verified.

1. Edziah B K, Sun H, Adom P K, Wang F, Agyemang A O. The role of exogenous technological factors and renewable energy in carbon dioxide emission reduction in Sub-Saharan Africa, Renewable Energy, 2022;196:1418–1428. doi: 10.1016/j.renene.2022.06.130 [CrossRef] [Google Scholar]

2. Sun H, Chen T, Wang C N. Spatial impact of digital finance on carbon productivity, Geoscience Frontiers, 2023;101674. doi: 10.1016/j.gsf.2023.101674 [CrossRef] [Google Scholar]

3. Autor DH, Levy F, Murnane Rj. The Skill Content of Recent Technological Change: An Empirical Exploration. Quarterly Journal of Economics. 2003;4: 1279–1333. doi: 10.3386/W8337 [CrossRef] [Google Scholar]

4. Acemoglu D, Restrepo P. The Race between Machine and Man: Implications of Technology for Growth, Factor Shares and Employment. American Economic Review. 2018;6:1488–1542. doi: 10.1257/aer.20160696 [CrossRef] [Google Scholar]

5. Acemoglu D., Restrepo P.Robots and Jobs: Evidence from US Labor Markets. Journal of Political Economy. 2020;6: 2188–2244. doi: 10.2139/SSRN.2940245 [CrossRef] [Google Scholar]

6. Haltiwanger J., Jarmin R.S., Miranda J.Who Creates Jobs? Small vs. Large vs. Young. Review of Economics and Statistics. 2013;2:347–361. doi: 10.1162/REST_a_00288 [CrossRef] [Google Scholar]

7. Acemoglu D, Autor D H. Skills. Tasks and Technologies: Implications for Employment and Earnings. Handbook of Labor Economics. 2011;4: 1043–1171. doi: 10.1016/S0169-7218(11)02410-5 [CrossRef] [Google Scholar]

8. Bresnahan T F, Computerisation. Wage Dispersion. An Analytical Reinterpretation. The Economic Journal, 1999;109(456): 390–415. doi: 10.1111/1468-0297.00442 [CrossRef] [Google Scholar]

9. Autor D H, Katz L F, Kearney M S. The Polarization of the U.S. Labor Market. American Economic Review. 2006;2: 189–194. doi: 10.3386/w15150 [CrossRef] [Google Scholar]

10. Autor D H, Dorn D. The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market. American Economic Review. 2013;5: 1553–1597. doi: 10.1257/aer.103.5.1553 [CrossRef] [Google Scholar]

11. Frey C B, Osborne M A. The future of employment: How susceptible are jobs to Computerisation? [J]. Technological Forecasting and Social Change, 2017, 114: 254–280. doi: 10.1016/j.techfore.2016.08.019 [CrossRef] [Google Scholar]

12. Lin J.Technological Adaptation, Cities, and New Work [J]. Review of Economics and Statistics, 2011, 93(2): 554–574. doi: 10.1162/REST_a_00079 [CrossRef] [Google Scholar]

13. Goos M, Manning A, Salomons A. Explaining Job Polarization: Routine-Biased Technological Change and Offshoring. American Economic Review. 2014;8: 2509–2526. doi: 10.1257/aer.104.8.2509 [CrossRef] [Google Scholar]

14. Lv S B, Zhang S W. Employment “polarization” in China: an empirical research. China Economic Quarterly. 2015;2: 757–778. doi: 10.1257/aer.104.8.2509 [CrossRef] [Google Scholar]

15. Hao N.An analysis on economic effects of labor force "polarization"——based on two perspectives of economic growth and income inequality. East China Economic Management. 2017;2: 118–125. doi: 10.3969/j.issn.1007-5097.2017.02.016 [CrossRef] [Google Scholar]

16. Qu X B, Cheng J. Changes of employment structure in China: Upgrading or polarization?Studies in Labor Economics. 2015;1: 119–145. CNKI:SUN:LDJJ.0.2015-01-006. [Google Scholar]

17. Li H B, Guo J X, Zhai R R. Does outward foreign direct investment affect job polarization in the China’s labor market?Journal of Finance and Economics. 2017;6: 28–39. doi: 10.16538/j.cnki.jfe.2017.06.003 [CrossRef] [Google Scholar]

18. Sun Z, Hou Y L. How does industrial intelligence reshape the employment structure of Chinese labor force. China Industrial Economics. 2019;5: 61–79. doi: 10.19581/j.cnki.ciejournal.2019.05.004 [CrossRef] [Google Scholar]

19. Li S Q, Wang H C. The impact of artificial intelligence on the manufacturing labor force in the context of agieing: Evidence from China. Science of Science and Management of S.& T. 2021;7: 3–17. [Google Scholar]

20. Dong X S, Xiao X. Does the change of demographic dividend affect China’s industrialization? China Population, Resources and Environment. 2016;9: 20–27. [Google Scholar]

21. Bartik T J.Who Benefits from State and Local Economic Development Policies? Kalamazoo, M.: W.E. Upjohn Institute. doi: 10.17848/9780585223940 [CrossRef] [Google Scholar]

22. Acemoglu D, Akcigit U, Kerr W. Networks and the Macroeconomy: An Empirical Exploration. NBER Macroeconomics Annual. 2016; 1: 273–335. doi: 10.1086/685961 [CrossRef] [Google Scholar]

23. Cheng H, Jia R X, Li D D, Li H B. The Rise of Robots in China. Journal of Economic Perspectives. 2019;3: 71–88. doi: 10.1257/jep.33.2.71 [CrossRef] [Google Scholar]

24. Han M C, Zhao Y F. The employment effect of industrial robots in Chinese manufacturing. Journal of industrial technological economics. 2019;11: 3–12. [Google Scholar]

25. Wei X H, Zhang P K, Du Y H. How robots reshape the urban labor market: From a perspective of migrants’ job tasks. Economic perspectives. 2020;10: 92–109. [Google Scholar]

26. Alexopoulos M, Cohen J. The Medium Is the Measure: Technical Change and Employment, 1909–1949. Review of Economics and Statistics, 2016, 98(4): 792–810. doi: 10.1162/REST_a_00588 [CrossRef] [Google Scholar]

27. Acemoglu D, Restrepo P. Robots and Jobs: Evidence from US Labor Markets. Journal of Political Economy, 2020;128(6): 2188–2244. doi: 10.1086/705716 [CrossRef] [Google Scholar]