In the era of artificial intelligence, the proliferation of AI robots is fundamentally reshaping employment and income distribution patterns. Ensuring that this technological advancement benefits a broad spectrum of the population, rather than exacerbating inequalities, presents a critical challenge for sustainable development. This article constructs a task-biased model that incorporates both horizontal penetration and vertical innovation to analyze the economic effects of AI robot penetration on the labor market. Utilizing data from the International Federation of Robotics and the China Household Finance Survey, we empirically examine the impact of AI robot penetration on labor income, wage rates, and labor participation shares. Our findings reveal that AI robot penetration reduces labor income and labor participation shares but increases wage rates. Furthermore, it widens the income gap between high- and low-skilled sectors while narrowing the disparity in labor participation shares between these sectors. The mechanisms through which AI robot penetration influences the labor market include promoting occupational upward mobility, facilitating industrial transformation, and increasing employment stability risks. Overall, AI robot penetration exacerbates income inequality, with more pronounced effects in the service sector. Based on these insights, we propose policy recommendations to synergize AI development with high-quality full employment, promote opportunity equity through skill training and fiscal policies, and proactively empower workers to address new challenges in the labor market.
The rapid development of AI robot technology is transforming production and lifestyle, injecting strong momentum into economic and social progress. In 2015, the Chinese government launched the “Made in China 2025” initiative, aiming to upgrade the manufacturing sector through the integration of informatization and industrialization, leading to the rapid adoption of industrial AI robots. The “14th Five-Year Plan for Robot Industry Development” further emphasized that AI robots, as essential tools for production and daily life, will continue to enhance productivity, improve living standards, and promote sustainable economic and social development. In the digital economy era, China’s AI robot industry, driven by continuous innovation and deep application, demonstrates robust growth. In 2024, China produced 556,400 units of industrial AI robots, and from January to May 2025, production reached 512,100 units. The share of global shipments increased from 14% in 2013 to 51% in 2023. Additionally, service AI robots and special-purpose AI robots have begun to permeate various aspects of production and life, achieving large-scale applications in fields such as warehousing and logistics, education and entertainment, cleaning services, security patrols, and medical rehabilitation. As artificial intelligence advances towards general and generative AI, AI robots exhibit characteristics like deep learning, cross-border integration, human-machine collaboration, group intelligence openness, and autonomous control, inevitably driving innovation and reshaping the current and future labor market landscape. This transformation brings both opportunities and challenges: job content becomes more complex and diverse, while many low-skilled and repetitive tasks are replaced by AI robots, potentially triggering new social inequalities. Therefore, in the context of coexisting opportunities and challenges in AI development, ensuring that AI robot penetration promotes high-quality development in the labor sector and benefits a wider workforce, achieving a balance between AI robot technological advancement and the improvement of traditional labor income levels, has become an urgent issue in the pursuit of common prosperity in the digital age.
The impact of AI robot technology on the labor sector differs from that of traditional technological progress. Traditional skill-biased technological change theory posits that technological development first replaces traditional low-skilled labor, and the gap with frontier technology is the main obstacle to income growth, while higher individual technical literacy is an effective means to bridge this gap and increase income. However, AI robot technology, as a carrier of artificial intelligence, is more likely to exhibit task-biased characteristics. A representative framework is the task-based model proposed by Acemoglu and Restrepo (2020), which suggests that automation and AI robot penetration, on one hand, lead to declines in wage rates and reductions in labor participation shares, and on the other hand, through productivity improvements and the creation of new tasks, generate substitution and creation effects. From a task-biased perspective, as technology costs decrease, machines become more advantageous than humans in an increasing number of production tasks, leading to the polarization of the labor market. This polarization means that the productivity gains from AI robots may not necessarily translate into significant wage growth but instead, due to the “polarization effect,” exacerbate adjustments in the labor market structure, expanding the labor participation shares of high-skilled and low-skilled groups while causing the “disappearance of the middle.” Specifically, AI replaces many simple, repetitive tasks, increasing the risk of unemployment for low- and medium-skilled workers or forcing them to accept lower-paying jobs. Simultaneously, AI creates new occupations that require significantly higher skills and knowledge. Even if AI can enhance the productivity of low- and medium-skilled workers (or workers with limited expertise in complex tasks), this improvement may not reduce inequality; in some cases, increased productivity for low-skilled workers can lead to higher inequality. Thus, the “disappearance of the middle” not only impacts the stability of the labor market but also poses new challenges to income distribution fairness.
In the early stages of AI development, scholars focused more on the productivity-enhancing effects of technological change and automation, generally believing that the development of AI robots would significantly boost productivity, optimize capital structure, and through intelligence and automation, improve enterprise production processes and reduce costs, thereby promoting macroeconomic growth and increasing household income. As research deepened, academia began to pay attention to the “disappearance of the middle” phenomenon, but the actual impact of AI robot development on the labor market and income distribution remains debated. Some studies have explored the heterogeneous effects of AI robots on labor income and market structure across different industries and regions, arguing that AI robot development asymmetrically widens the labor income gap between high- and low-tech sectors and leads to “technological unemployment” that displaces medium- and low-educated workers. While the distinct biased characteristics of AI robots are widely recognized, there is still controversy regarding the actual impact of artificial intelligence on residents’ labor participation shares and income at the micro level. One view holds that AI robot penetration primarily increases total labor demand in society and has substitution effects on different types of labor, reducing labor shares and causing unemployment, but with no significant impact on wages. Another view argues that, in addition to substitution effects, AI robots also have complementary and creative effects, thus not leading to a reduction in total employment but rather bringing about productivity improvements and changes in job structure, and affecting labor wage income through changes in working hours and job types, with this change being uncertain. The controversy between these two views essentially stems from a deeper discussion about whether the substitution effect or the creation effect dominates in AI technology. Some research has noted the uncertainty under the overlap of AI’s substitution and creation effects: automation may either reduce the demand for low-skilled labor and exacerbate inequality or, by creating more complex versions of tasks, give labor a comparative advantage, thereby increasing employment opportunities and reducing skill premiums. This implies that the impact of AI robot penetration on labor income and participation shares is uncertain. Balancing AI development to ensure that its penetration benefits not only a few but covers a broader population and benefits the vast majority is key to addressing the “disappearance of the middle” problem and promoting high-quality development and common prosperity.
Existing research has recognized the task-biased nature of AI technology and its impact on the labor market, but there are still areas requiring further exploration, mainly at three levels. First, although some literature discusses the asymmetric impact of AI technology on the labor market, there is a lack of systematic research on the dynamic evolution of the substitution and creation effects of AI robots, particularly the structural differentiation among occupational groups at the micro level. Second, the dimensions of impact assessment at the micro level are limited; existing studies often adopt a single-dimensional paradigm, focusing either on income distribution effects or on changes in labor participation rates, lacking a comprehensive analysis that integrates income and labor participation into a unified framework, making it difficult to accurately assess the overall economic impact of AI robot penetration. Third, existing studies lack discussion on the synergistic mechanisms of the positive and negative effects of AI robot penetration in the labor market, making it difficult to provide precise decision support for labor market policy formulation in the AI era.
The potential marginal contributions of this article are threefold. First, unlike existing research, this article starts from the dual effects of AI robot penetration to explore its actual impact on labor, combining occupational skill characteristics to reveal its heterogeneous effects on different occupations and skill levels of labor, thereby explaining the “disappearance of the middle” phenomenon and providing a dynamic theoretical framework and micro-level empirical evidence for understanding the internal logic of labor development in the digital era. Second, differing from single-dimensional studies on labor income or employment, this article comprehensively analyzes the impact of AI robot penetration on labor from both income indicators and labor participation share indicators from an input perspective at the micro level, contributing to a deeper understanding of their relationship. Third, this article expands on the mechanisms through which AI robot penetration affects labor, including positive mechanisms such as upward mobility and industrial transformation, and negative mechanisms such as increased employment stability risks, constructing a unified logical framework for synergistic effects, and providing scientific basis and strategic support for addressing intelligent transformation, ensuring robust and fair development of the labor market, and achieving common prosperity.
Theoretical Framework of AI Penetration Effects
In the task-based technological progress framework by Acemoglu (2024), all tasks are initially performed by labor; the automation of existing tasks (i.e., the substitution of labor by AI) is offset by the creation of new tasks performed by labor. Simultaneously, by adding new labor tasks, the share of automated tasks remains unchanged. This model directly links productivity improvements to labor. Jones and Liu (2024) consider the dynamics of exogenous technological progress and simplify the setting of human capital growth. Following this approach, technological innovation embodied in capital occurs in two aspects: vertically, capital inputs become more productive in given tasks; horizontally, capital substitutes for labor in given tasks. Thus, final production can be completed by a series of tasks, and total output is given by:
$$Y_t = \nu \left[ \int_0^1 y_t(i)^\rho \, di \right]^{1/\rho}, \quad \rho < 0$$
where \( y_t(i) \) is the intermediate product, \( \rho \) is the elasticity of substitution between different intermediate products, \( \nu \) is the real value rate of intermediate products, and \( \rho < 0 \) indicates that intermediate products are gross complements. Assume that total social output is composed of countless complementary intermediate outputs. The production of intermediate products \( y_t(i) \) can be entirely achieved by labor, but intermediate products on \( [0, \beta_t] \) can be produced by capital. Therefore, assuming labor \( l_t(i) \) and capital \( x_t(i) \) as production inputs, we have:
$$y_t(i) =
\begin{cases}
A l_t(i), & i \in [0,1] \\
z_t(i)^{\frac{\rho-1}{\rho}} x_t(i), & i \in [0, \beta_t]
\end{cases}$$
where \( z_t(i) \) is the productivity of capital input in task \( i \) at time \( t \), \( A \) is the given technological level of labor, and the capital productivity over the entire period on \( [0, \beta_t] \) is represented by the harmonic mean of capital-embodied technology in intermediate products: \( Z_t = \left[ \frac{1}{\beta_t} \int_0^{\beta_t} \frac{1}{z_t(i)} \, di \right]^{-1} \). Given the prices of final products, labor inputs, and capital inputs, profit maximization under perfect competition leads firms to demand intermediate products at price \( p_t(i) = \nu^\rho \left[ \frac{Y_t}{y_t(i)} \right]^{1-\rho} \). Under the intermediate output conditions in equation (2), the price of non-automated tasks is:
$$p_t(i) = \frac{w_t}{A}$$
For tasks on \( [0, \beta_t] \), firms can use either labor or capital inputs. In the exogenous growth model, competition among intermediate producers drives profits to zero in equilibrium. If firms use capital inputs, the intermediate product price is:
$$p_t(i) = \psi z_t(i)^{\frac{1-\rho}{\rho}}, \quad i \in [0, \beta_t]$$
To simplify the analysis, assume that machines fully depreciate and are consumed within the production cycle (no residual value), so the final capital value rate \( \psi \) equals \( \nu \). Firms, aiming for profit maximization, choose to increase capital input rather than hire new labor because automation-based production has lower costs (intermediate product prices) than labor-based non-automated production, i.e., automation technology is sufficiently productive. Under this condition, according to equations (3) and (4), firms will choose automated production for all \( i \in [0, \beta_t] \) when the following condition is met:
$$w_t \geq \psi A z_t(i)^{\frac{1-\rho}{\rho}}$$
For automated tasks, from the intermediate demand price and equations (2) and (4), the capital allocation per task is:
$$x_t(i) = \nu^{-1} \frac{Y_t}{z_t(i)}, \quad i \in [0, \beta_t]$$
From equations (2) and (6), the intermediate output is:
$$y_t(i) = \nu^{-1} z_t(i)^{\frac{1}{\rho}} Y_t, \quad i \in [0, \beta_t]$$
The share of intermediate sector output in total output is:
$$\frac{p_t(i) y_t(i)}{Y_t} = \frac{1}{z_t(i)}$$
The output share of the intermediate sector is negatively correlated with the horizontal technological progress of capital: improvements in AI robot automation productivity increase output but reduce the sector’s output share. Thus, it can be argued that sectors with higher automation levels participating in production squeeze the living space of traditional labor sectors.
Further consider resource constraints for total investment \( I \) and total labor supply \( L \). From the capital constraint \( I_t = \int_0^1 \psi x_t(i) \, di \) and equation (6), the capital share in total output is:
$$s_{Kt} = \frac{\psi X_t}{Y_t} = \beta_t Z_t^{-1}$$
Under exogenous technology, from the labor constraint \( L_t = \int_0^1 l_t(i) \, di \), the labor participation share in production is:
$$s_{Lt} = \frac{w_t L_t}{Y_t} = 1 – \beta_t Z_t^{-1}$$
According to equation (1), total output is the synthesis of equilibrium intermediate outputs. The total output under equilibrium conditions is:
$$Y_t = \nu A (1 – \beta_t Z_t^{-1})^{-1/\rho} (1 – \beta_t)^{\frac{1-\rho}{\rho}} L_t$$
From equations (10) and (11), the equilibrium wage \( w_t = \nu A (1 – \beta_t Z_t^{-1})^{\frac{\rho-1}{\rho}} (1 – \beta_t)^{\frac{1-\rho}{\rho}} \). Assuming the real value of products and technological level are constant in the short term, taking logarithms and simplifying yields:
$$\ln w_t = \frac{\rho-1}{\rho} [\ln s_{Lt} – \ln(1 – \beta_t)] = \frac{\rho-1}{\rho} [\ln(1 – \beta_t Z_t^{-1}) – \ln(1 – \beta_t)]$$
Changes in wage income are related to changes in labor participation share and AI robot penetration rate, while changes in labor participation share are functions of penetration rate and technological level. When penetration occurs, according to equation (5), the minimum productivity \( z_t^{\min} \) for automated tasks to produce intermediate products at a cost lower than labor should satisfy \( w_t = \psi A z_t^{\min}(i)^{\frac{1-\rho}{\rho}} \). At this point, the productivity of capital input in intermediate tasks should satisfy \( z_t(i) \geq z_t^{\min} = \frac{1-\beta_t}{1-\beta_t Z_t^{-1}} \). Since \( Z_t \) is the harmonic mean of \( z_t(i) \), it follows that \( Z_t \geq z_t^{\min} \geq \frac{1-\beta_t}{1-\beta_t Z_t^{-1}} \), so \( Z_t \geq 1 \) and \( \beta_t Z_t^{-1} \leq 1 \). The logarithm of the equilibrium wage in equation (12) is less than 0, indicating that under short-term substitution effects, AI robot penetration suppresses labor income. Based on this, we propose the following hypothesis:
Hypothesis 1: AI robot penetration will lead to a decrease in labor income.
Household income is determined by both wage rates and working hours. Based on the above analysis, consider the long-term technological progress that crowds out the labor sector. When long-term technological progress leads to a capital-labor substitution elasticity exceeding the sectoral transformation threshold, firms adopt AI robots to substitute production factors, increasing labor costs. This process has two effects: First, the scale effect of automated equipment causes the labor share in total output to show a monotonically decreasing trend; Second, structural transformation driven by Baumol’s cost disease continuously strengthens the human capital premium in retained positions. Therefore, AI robot penetration, by reshaping productivity differences between sectors, pushes up the structural rise in labor costs. In sectors with rapid diffusion of automation technology, capital deepening significantly improves labor productivity, driving up wages for high-skilled positions as marginal output increases. In low-productivity sectors less affected by automation, due to limited human capital mobility and labor-intensive characteristics, wage levels need to align with high-productivity sectors to maintain labor supply, forming a Baumol effect of “stagnant productivity but converging wages.” This cross-sector wage stickiness transmission mechanism causes overall labor costs to exhibit “technology-driven inflation,” forcing firms to further increase AI robot penetration to cope with cost pressures, ultimately forming a reinforcing cycle of “technology penetration—wage rigidity—re-penetration.” This essentially reflects the dynamic reallocation process of labor with different skills during automation. The increase in penetration rate, through multiplier effects, brings about an increase in the wage rate of on-the-job labor, while the accelerating characteristics of wage rates force re-penetration. For residents, if the labor participation share does not change, an increase in wage rate naturally means an improvement in individual utility levels. Further considering changes in labor participation share, from equation (10), it can be seen that the labor participation share is affected by \( \beta_t \) and \( Z_t \), actually reflecting the combined effect of substitution and creation effects of AI robots in the labor market. In the short term, technological stickiness causes \( Z_t \) to remain unchanged, while an increase in AI robot penetration level \( \beta_t \) leads to a reduction in labor participation share. At the micro level, some traditional labor positions are replaced by machines, adversely affecting the employment pattern of labor. In summary, the substitution of labor by AI robots both increases wage rates and, due to the crowding out of labor participation share, reduces labor participation hours. Thus, we propose the following hypotheses:
Hypothesis 2a: AI robot penetration will lead to an increase in labor wage rates.
Hypothesis 2b: AI robot penetration will crowd out the labor participation share of labor.
The above analysis shows that the impact of AI robots on labor income is the result of the combined effect of penetration rate on wage rates and labor participation shares. From equation (8) and dynamic characteristics, according to the diminishing marginal effect of technological progress, when penetration reaches a certain level, the creation effect of technological level \( z_t \) improvement will gradually dominate. This not only creates more new employment opportunities but also promotes the improvement of labor quality and vocational skills, enabling better adaptation to new work environments and demands. Therefore, in the short term, the scale law holds, and an increase in penetration rate crowds out labor participation share. However, in the long term, the scale law does not necessarily hold. As the marginal substitution effect of AI technology diminishes, the positive correlation between the deepening of technological level and labor participation share will gradually emerge, which will become the key to balancing AI development. Thus, although AI robot penetration has a substitution effect to some extent and compresses labor working hours, it is reasonable to believe that in the future when penetration rates are generally high, the improvement in technological level can instead exert a creation effect and increase labor participation share.
Further consider the heterogeneity of different skill sectors engaged in intermediate tasks \( y_t(i) \). When tasks \( i \in [0, \beta_t] \) are replaced by capital due to automation, labor is allocated to tasks \( i \in [\beta_t, 1] \). From the labor constraint, the equilibrium labor allocation is \( l_t(i) = L_t / (1 – \beta_t) \) and output \( y_t(i) = A L_t / (1 – \beta_t) \). Assume that within the same production task \( j \), there are two sectors: high-skilled (h) and low-skilled (l). The difference in labor knowledge and skills only exists in technological productivity, with the high-skilled sector having higher task complexity and technological level \( A_h > A_l \). Capital technological level \( Z_t^j \) is higher and grows relatively slower than in the low-skilled sector, but there is no difference in penetration rate. Therefore, let the wage in production task \( j \) for sector \( j \) be \( w_t^j \). The ratio of wages between high- and low-skilled sectors in production tasks can be expressed as:
$$\frac{w_t^h}{w_t^l} = \frac{A_h}{A_l} \left( \frac{1 – \beta_t Z_t^{h-1}}{1 – \beta_t Z_t^{l-1}} \right)^{\frac{\rho-1}{\rho}}$$
Let the labor participation share in production task \( j \) for sector \( j \) be \( S_t^j \). The ratio of labor participation shares between high- and low-skilled sectors is:
$$\frac{S_t^h}{S_t^l} = \frac{1 – \beta_t Z_t^{h-1}}{1 – \beta_t Z_t^{l-1}}$$
Taking logarithms of equations (13) and (14) respectively yields:
$$\ln \frac{w_t^h}{w_t^l} = \frac{\rho-1}{\rho} \ln(A_h – A_l) [\ln(1 – \beta_t Z_t^{h-1}) – \ln(1 – \beta_t Z_t^{l-1})] > 0$$
$$\ln \frac{S_t^h}{S_t^l} = [\ln(1 – \beta_t Z_t^{h-1}) – \ln(1 – \beta_t Z_t^{l-1})] < 0$$
It is easy to determine that equation (15) is greater than 0 and equation (16) is less than 0, indicating that after considering the differences in technological levels between high- and low-skilled sectors, the labor income gap between high- and low-skilled sectors gradually widens, while the gap in labor participation shares between the two sectors gradually narrows. Additionally, the higher the technological requirements of the sector itself, the greater the impact of technological penetration on wages and labor participation shares. Thus, we propose the following hypothesis:
Hypothesis 3: AI robot penetration will widen the labor income gap between high- and low-skilled sectors but will also narrow the gap in labor participation shares between the two sectors.
Based on the dual effects of creation and substitution, we further discuss the transmission mechanisms through which AI robots affect labor. First, AI robots have the function of task reallocation and working condition improvement, by replacing high-physical, high-risk tasks, promoting the shift of labor to tasks and occupations with lower physical intensity and lower injury risks, while reducing working hours, lowering work intensity, and optimizing the work environment. As tedious and heavy work tasks are taken over by AI robots, non-replaced employees are freed up to focus on more core and advanced tasks, greatly improving work efficiency and providing strong support for upward mobility of labor. Second, AI robot penetration not only promotes close physical connections between machines in the production process but also constructs seamless dual connections of information between human-machine and human-human, promoting a new model of “integration of two industries” where manufacturing and services converge. The convergence of industries not only promotes the “rise of services” but also opens up broad new areas in the producer services market. On one hand, the productivity improvement of traditional manufacturing enterprises leads to substitution of traditional labor inputs; on the other hand, the productivity improvement of traditional manufacturing enterprises gives rise to new service sectors with significant manufacturing characteristics, such as e-commerce and platform-based life services. The renewal of traditional manufacturing and the rise of new intelligent service industries bring production dividends, naturally guiding labor transformation and optimization of labor structure.
Additionally, although AI robots bring many opportunities for economic development, we cannot ignore the structural risks they pose to employment. With accelerating technological iteration and broadening application scenarios, AI robots are triggering a deep reconstruction of the occupational ecology, with many standardized, procedural positions facing potential risks of technological substitution. This transformation has a dual impact on the labor market: on one hand, vulnerable groups such as low-skilled workers and informal workers are the first to be affected, with significantly increased job substitutability; on the other hand, high-skilled groups, empowered by digital technology, continuously optimize human capital structure, effectively resisting automation substitution risks while leveraging technological dividends to achieve career leaps. This differentiated impact of technological empowerment exacerbates structural contradictions in the labor market, leading to increased frequency of job changes and job uncertainty, overall aggravating employment stability risks. Therefore, this risk not only objectively weakens the stability of workers’ wage income but also subjectively affects workers’ employment choices and career development paths, adversely affecting both labor income and participation shares. In summary, we propose the following hypothesis:
Hypothesis 4: AI robot penetration will, through mechanisms such as incentivizing upward mobility of labor, promoting industrial transformation, and increasing employment stability risks, increase labor wage rates while crowding out labor participation shares.
Data Processing and Model Construction
To measure AI robot penetration, we follow the approach of Wang et al. (2020) by matching manufacturing robot data from the International Federation of Robotics (IFR) with manufacturing classifications to construct industrial robot penetration rates. Specifically, we match China’s industry employment data with IFR’s robot stock data, calculate urban industry penetration rates based on the proportion of urban industry employment, and then match with individuals’ industries to obtain urban individual penetration rate indicators. According to IFR robot types and functions, we match robot industry category codes with the results of the “What industry does your work unit belong to?” question in the China Household Finance Survey (CHFS) questionnaire. While maintaining occupational consistency, we ultimately divide into 5 categories: “1. Agriculture, forestry, animal husbandry, and fishery; 2. Mining and manufacturing; 3. Construction; 4. Production and supply of electricity, heat, gas, and water; 5. Tertiary industry including education, research, development, and other fields.” For each type, we obtain the national robot stock for that year from IFR data, and after de-unitizing, calculate the national industry-level robot penetration indicator \( PR_{nt}^{CH} \) matched to the industry in which residents are employed:
$$PR_{nt}^{CH} = \frac{MR_{nt}^{CH}}{L_{n,t=2014}^{CH}}$$
where \( MR_{nt}^{CH} \) is the national industrial robot stock in industry where resident \( n \) is employed in year \( t \), and \( L_{n,t=2014}^{CH} \) is the base period (2014) employment in industry where resident \( n \) is employed. Based on this, for resident \( n \) employed in a certain industry in city \( k \) in year \( t \), the impact of digital penetration in their industry mainly comes from the urban industry penetration rate and their own dependence on work. Therefore, we further construct the AI robot penetration rate at the individual level for the industry in which the resident is employed, \( Ar_{knt} \):
$$Ar_{knt} = \frac{E_{k n t=2015}}{L_{k t=2015}} \times \frac{MR_{nt}^{CH}}{L_{n t=2014}^{CH}}$$
where \( E_{k n t=2015} \) is the total number of people employed in the industry where resident \( n \) is employed in city \( k \) in the base period (2015), and \( L_{k t=2015} \) is the total employment in all industries in city \( k \) in the base period (2015).
The microdata used in this study are from the CHFS questionnaires from 2015 to 2019. In sample screening, we follow the International Labour Organization’s definition of continuous employment status: during the observation period, there is at least one valid employment contract, and the cumulative unemployment duration does not exceed the frictional unemployment threshold (usually set at 4 weeks). Accordingly, we filter samples that answered surveys about the type of work participation, income, hours, etc., in the previous year and had non-zero working hours in the previous year, ensuring that samples are not heterogeneously affected by inconsistent survey time points in continuous employment research. Then, we match individual data with city-level data through city-level codes, finally forming a micro-dataset containing 7,674 samples.
The core explanatory variable is the AI robot penetration rate calculated above, matched to the industry level in which residents are employed.
The core explained variables include total income at the resident level, monthly wage, and annual working months and monthly working days measuring labor participation share. Specifically, total income is measured by the wage income of family members in the previous year in CHFS, and monthly wage is represented by the monthly wage calculated based on the actual number of working months per year of family members, indicating the resident’s wage rate. For the construction of labor participation share, since most residents typically use “month” as the basic labor and remuneration cycle, we use “family members’ working months last year” (i.e., annual working months) in CHFS to assess residents’ labor hours, and supplement with “family members’ average monthly working days last year” (i.e., monthly working days) in the survey to avoid ignoring differences in work intensity by using only “month” as the unit.
Regarding the measurement of labor participation share, the sectoral labor participation share in the theoretical framework essentially reflects the allocation efficiency of production factors, which can be decomposed into two dimensions: the absolute input of labor factors and the marginal efficiency of unit labor output. At the micro level, limited by the macro attributes of the national economic accounting system and the asymmetry of micro-subject data, it is difficult to measure the total labor task volume of individual residents participating in a specific industry. Under the assumption of constant returns to scale production technology, the marginal output elasticity of a single worker approaches 0, meaning that the impact of micro-individual labor input decisions on the aggregate production function \( Y = F(K, L) \) exhibits high-order small characteristics. Based on this, we introduce a rational expectations analysis framework: when economic agents have complete market information, individual labor supply will automatically adapt to the given total production tasks, at which point total output \( Y \) can be considered exogenous, with the core being to achieve logical self-consistency in macro and micro economic analysis through the representative economic agent assumption. Therefore, treating the total production tasks faced by individuals as fixed parameters at the micro level not only conforms to the marginal analysis tradition of neoclassical economics but is also a necessary simplification under data constraints. When total production tasks are deemed exogenous, changes in individual working hours directly reflect the degree to which labor is replaced by robots and are positively correlated with sectoral labor participation share, so changes in individual labor hours can serve as a proxy variable for changes in individual labor share.
Control variables include personal and city dimensions. Personal level: education level, measured by individual years of education; since sample work experience is difficult to observe directly, considering its correlation with age, control for age and age squared; resident political identity (whether Communist Party member or probationary member, yes=1, no=0), health status (physical condition compared to peers, very bad, bad, average, good, very good assigned values -2 to 2 respectively), marital status (unmarried=1, married, cohabiting, separated, divorced, widowed, etc.=0), work overtime incentive (whether overtime pay, yes=1, no=0), unit nature (whether working in government agencies, institutions, state-owned and state-controlled enterprises, yes=1, no=0), household registration type (agricultural=0, urban or other=1), labor dispatch identity (whether work contract signed with “a labor dispatch company”, yes=1, no=0), and work bonus income (logarithm of last year’s work bonus income) as control variables. City level: while controlling for per capita GDP (log), average annual population size (millions), proportion of tertiary industry value added in GDP, urban unemployment rate, and urban annual average wage of employees (log), further consider the role of the skill structure of the employed population in different cities, construct the ratio of the number of high-skilled labor in the city to the total sample in the same city as the urban high-skilled labor share, to control for the impact of differences in urban skill structure; consider the impact of urban digital economy level, using the method of Zhao et al. (2020) to calculate urban digital economy level, to control the impact of urban digitalization process on enterprise digitalization level, employee digital literacy, work methods, and income. Finally, to avoid the impact of extreme values, we use bilateral 1% winsorization for total income and monthly wage to exclude outliers, and to ensure consistency and comparability, all explained variables are added by 1 and then logarithmically processed. Specific descriptive statistical results are shown in Table 1.
| Variable Type | Variable Name | Mean | Std. Dev. | Min | Max |
|---|---|---|---|---|---|
| Core Explained Variables | Total Income | 10.296 | 0.811 | 7.496 | 11.918 |
| Monthly Wage | 7.957 | 0.785 | 5.122 | 9.721 | |
| Annual Working Months | 2.437 | 0.291 | 0.406 | 2.565 | |
| Monthly Working Days | 3.189 | 0.241 | 0.693 | 3.466 | |
| Core Explanatory Variable | Penetration Rate | 0.009 | 0.029 | 0 | 0.702 |
| Personal Control Variables | Age | 44.832 | 9.024 | 17.000 | 79.000 |
| Age Squared | 2091.346 | 816.097 | 289.000 | 6241.000 | |
| Education Level | 10.936 | 3.301 | 0 | 24.000 | |
| Political Identity | 0.208 | 0.406 | 0 | 1.000 | |
| Health Status | 0.664 | 0.845 | -2.000 | 2.000 | |
| Marital Status | 0.975 | 0.158 | 0 | 1.000 | |
| Work Bonus Income | 3.290 | 4.306 | 0 | 10.463 | |
| Work Overtime Incentive | 0.301 | 0.459 | 0 | 1.000 | |
| Unit Nature | 0.383 | 0.486 | 0 | 1.000 | |
| Labor Dispatch Identity | 0.413 | 0.467 | 0 | 1.000 | |
| Household Registration Type | 0.109 | 0.312 | 0 | 1.000 | |
| City Control Variables | Urban High-Skilled Labor Share | 0.092 | 0.076 | 0 | 0.279 |
| Per Capita GDP | 11.039 | 0.575 | 9.256 | 12.281 | |
| Average Annual Population Size | 7.802 | 6.240 | 0.554 | 33.970 | |
| Tertiary Industry Share in GDP | 50.331 | 12.247 | 21.900 | 80.980 | |
| Urban Unemployment Rate | 0.849 | 0.618 | 0.044 | 3.800 | |
| Urban Annual Average Wage of Employees | 11.131 | 0.308 | 10.211 | 11.917 | |
| Urban Digital Economy Level | 0.700 | 1.519 | -1.170 | 10.438 |
The baseline regression part adopts the following two-way fixed effects model controlling for time and individual characteristics:
$$y_{nkt} = \beta_0 + \beta_1 Ar_{nkt} + X_{kt}’ \xi + X_{nkt}’ \sigma + \phi_n + \phi_t + \mu_{nkt}$$
where \( y_{nkt} \) is the core explained variable, representing the annual income, monthly wage, and labor participation share measured by annual working months and monthly working days of resident \( n \) in city \( k \) in year \( t \); \( Ar_{nkt} \) is the core explanatory variable, i.e., the AI robot penetration rate in the industry where resident \( n \) is employed in city \( k \) in year \( t \); \( X_{kt}’ \) and \( X_{nkt}’ \) are city-level and personal-level control variables, respectively; \( \phi_n \) is the individual effect, \( \phi_t \) is the time effect; \( \mu_{nkt} \) is other unobservable random error term; \( \beta_1 \) represents the impact of AI robot penetration. Considering the correlation between different urban economic levels, industrial distribution, and labor, we cluster standard errors at the city level to reduce bias in the estimation of \( \beta_1 \).
Baseline Empirical Analysis
The estimation results of the baseline regression model are shown in Table 2. The coefficient of penetration rate in column (1) is significantly negative, indicating that at this stage, AI robot penetration does suppress total labor income, verifying Hypothesis 1. Total income is jointly determined by wage rate and working time. To determine the source of the decrease in total income under AI robot penetration and verify Hypothesis 2, we substitute monthly wage and labor participation share as explained variables into the baseline model for estimation. The results in column (2) show that an increase in penetration rate反而 increases the monthly wage of on-the-job employees, consistent with theoretical derivation. AI robot penetration makes labor costs increasingly higher, manifested at the resident level as an increase in wage rate, confirming Hypothesis 2a. The results in column (3) show that an increase in penetration rate has a substitution effect, crowding out the labor participation share measured by annual working months. The results in column (4) remain valid when monthly working days are used as the explained variable. Thus, Hypothesis 2b is confirmed: AI robot penetration, while crowding out labor participation share measured by labor hours, also pushes up the wage rate of labor.
| Variable | (1) Total Income | (2) Monthly Wage | (3) Annual Working Months | (4) Monthly Working Days |
|---|---|---|---|---|
| Penetration Rate | -1.201*** (-2.859) | 1.848*** (3.898) | -2.810*** (-12.267) | -0.342*** (-2.635) |
| Adj. R² | 0.464 | 0.407 | 0.398 | 0.309 |
| Number of Observations | 7,674 | 7,674 | 7,674 | 7,674 |
Note: 1. *** indicates significance at the 1% level. 2. t-values are in parentheses. 3. Fixed effects are controlled; control variable estimation results are retained for reference.
Further Analysis
We divide personal occupations into four categories: heads of Party agencies, state agencies, mass organizations, social organizations, enterprises, and institutions, and administrative service personnel (hereinafter referred to as “heads and administrative service personnel”); senior intellectuals and professional technicians, including various social science and natural science researchers, technicians, teaching staff, etc.; social production service and life service personnel; general production manufacturing and related personnel. Among them, senior intellectuals and professional technicians typically have higher skill requirements compared to general production manufacturing personnel, so the two can be used as criteria for dividing high- and low-skilled employees.
First, based on occupational type grouping and estimation according to the baseline model, the results are shown in Table 3. The penetration rate in column (4) is significant. This indicates that as penetration rate increases, the total income of general manufacturing personnel represented by medium and low skills decreases more significantly, further exacerbating the income gap between them and employees at other skill levels.
| Variable | Explained Variable: Total Income | |||
|---|---|---|---|---|
| (1) Heads and Administrative Service Personnel | (2) Senior Intellectuals and Professional Technicians | (3) Production Service Personnel | (4) General Manufacturing Personnel | |
| Penetration Rate | -0.149 (-0.120) | -0.054 (-0.085) | -0.646 (-0.542) | -1.333** (-2.452) |
| Adj. R² | 0.502 | 0.396 | 0.399 | 0.433 |
| Number of Observations | 2,069 | 1,357 | 1,978 | 1,745 |
Note: 1. ** indicates significance at the 5% level. 2. t-values are in parentheses. 3. Fixed effects are controlled; control variable estimation results are retained for reference.
Next, we explore the occupational heterogeneity of the impact of AI robot penetration on monthly wage. The results are shown in Table 4. From the results in columns (2) and (4), as penetration rate increases, the monthly wages of senior intellectuals and professional technicians and general manufacturing personnel increase significantly, indicating that the creation effect of AI robot penetration is heterogeneous, bringing the possibility of higher income to high-skilled groups and general manufacturing labor by increasing monthly wages. At the same time, comparing the coefficient differences between the two, the monthly wage growth of senior intellectuals and professional technicians is more significant than that of general manufacturing personnel, confirming that the widespread application of AI robot technology increases the demand for high-skilled labor and also implying that the growth differences in wage rates represented by monthly wages may lead to further expansion of income gaps rather than narrowing.
| Variable | Explained Variable: Monthly Wage | |||
|---|---|---|---|---|
| (1) Heads and Administrative Service Personnel | (2) Senior Intellectuals and Professional Technicians | (3) Production Service Personnel | (4) General Manufacturing Personnel | |
| Penetration Rate | 2.251 (1.259) | 2.869*** (2.748) | 2.029 (0.985) | 1.704*** (3.030) |
| Adj. R² | 0.475 | 0.332 | 0.303 | 0.338 |
| Number of Observations | 2,069 | 1,357 | 1,978 | 1,745 |
Note: 1. *** indicates significance at the 1% level. 2. t-values are in parentheses. 3. Fixed effects are controlled; control variable estimation results are retained for reference.
Finally, we also conduct grouped estimation for labor participation share. The results are shown in Table 5. The impact of penetration rate on annual working months is significantly negative in all occupational categories, indicating that AI robot penetration does have a substitution effect, squeezing labor participation share. Comparing the coefficient differences in annual working months among different occupational groups, the substitution effect is particularly significant for senior intellectuals and professional technicians and general manufacturing personnel. Although the changes in labor participation share for both are similar, the monthly wage growth for low-skilled general manufacturing personnel is much weaker than for senior intellectuals and professional technicians, validating the actual existence of the “polarization effect.” Additionally, from the perspective of monthly working days, the above conclusion still holds, and it is only significantly negative in the groupings of senior intellectuals and professional technicians and general manufacturing personnel in columns (4) and (8), with the “polarization effect” of labor participation share more obvious. It can be seen that although the difference in labor participation share coefficients between the two is small, the effect of penetration rate on increasing the monthly wage of senior intellectuals and professional technicians is much greater than that of general manufacturing personnel, resulting in no significant impact on the total income of senior intellectuals and professional technicians, while the total income coefficient for general manufacturing personnel is significantly negative. Therefore, the impact of robot penetration is not limited to medium- and low-skilled labor; its heterogeneous effects may be masked by the表象 of unchanged total income.
| Variable | Heads and Administrative Service Personnel | Senior Intellectuals and Professional Technicians | Production Service Personnel | General Manufacturing Personnel | ||||
|---|---|---|---|---|---|---|---|---|
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | |
| Penetration Rate | -1.996** (-2.089) | -0.569 (-0.975) | -3.670*** (-5.683) | -0.526** (-2.196) | -2.822*** (-3.547) | -0.076 (-0.465) | -2.724*** (-9.537) | -0.224* (-1.687) |
| Adj. R² | 0.205 | 0.331 | 0.447 | 0.215 | 0.201 | 0.285 | 0.497 | 0.355 |
| Number of Observations | 2,069 | 2,069 | 1,357 | 1,357 | 1,978 | 1,978 | 1,745 | 1,745 |
Note: 1. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. 2. t-values are in parentheses. 3. Columns (1), (3), (5), (7) have explained variable as annual working months; columns (2), (4), (6), (8) have explained variable as monthly working days. 4. Fixed effects are controlled; control variable estimation results are retained for reference.
In summary, in the context of continuous penetration of task-based technology, the impact of substitution effects on income and labor participation share shows obvious biased characteristics. In contrast to the不断扩大 income gap between high- and low-skilled sectors, robot penetration helps to some extent narrow the gap in labor participation shares among different occupational groups. Thus, Hypothesis 3 is effectively verified.
We introduce the industrial structure characteristics of the industry to which labor belongs to discuss whether there is a moderating effect of different urban industrial structures on the impact of AI robot penetration rate. In urban clusters with high tertiary industry employment shares such as Beijing and Shanghai, industrial development relies more on the service sector labor market, and the human capital-intensive characteristics of services may amplify technological substitution effects;而在以传统制造业为主导、第三产业就业占比较低的传统工业城市中，自动化进程更多受制于资本深化程度下的路径约束，智能机器人渗透的替代效应则相对较弱。为识别产业结构差异的调节作用，按照城市第三产业就业占比划分低、中、高三类服务业就业密度，并分别赋值-1、0、1，与渗透率构建交互项纳入基准模型，估计结果见表6。第（1）列中交互项系数显著为负且与渗透率系数同号，表明第三产业发展程度越高，渗透率对总收入的抑制效应越强，即服务业人力资本密集特征放大了技术替代效应。随着渗透率的提升，第三产业就业占比较高城市中居民的总收入所受的边际影响大于其他城市。类似地，第（2）~（4）列结果显示，渗透率对月工资的影响与基准回归一致，智能机器人渗透导致以月工资指代的工资率上升，而对表示劳动参与份额的年工作月数与月工作天数的影响依然呈现出下降趋势。其中，第（2）列交互项显著为正，呈现出正向调节效应，即第三产业发展较高城市的自动化进程催生技能溢价，带来高于以第一、二产业为主导城市的劳动力报酬。第（3）（4）列交互项系数为负，表明替代效应因产业结构而存在梯度差异，相比以第一、二产业为主导的城市，高第三产业就业占比城市的劳动参与率下降效应更为显著。
| Variable | (1) Total Income | (2) Monthly Wage | (3) Annual Working Months | (4) Monthly Working Days |
|---|---|---|---|---|
| Penetration Rate | -1.393*** (-3.160) | 2.284*** (4.655) | -3.355*** (-17.190) | -0.450*** (-16.562) |
| Is Tertiary Industry | 0.006 (0.172) | -0.090** (-2.291) | 0.082*** (5.200) | 0.012*** (4.736) |
| Penetration Rate × Tertiary Industry | -1.248*** (-2.661) | 3.944*** (6.334) | -4.479*** (-11.856) | -0.564*** (-10.012) |
| Adj. R² | 0.465 | 0.417 | 0.497 | 0.457 |
| Number of Observations | 7,674 | 7,674 | 7,674 | 7,674 |
Note: 1. ** and *** indicate significance at the 5% and 1% levels, respectively. 2. t-values are in parentheses. 3. Fixed effects are controlled; control variable estimation results are retained for reference.
According to theoretical analysis, we further test the mechanisms through which AI robots affect income and labor participation share. First, considering labor unit type and occupational changes, we divide the sample by occupation: when an individual’s occupation changes from general manufacturing personnel, production or administrative service personnel to senior intellectuals and professional technicians, or heads, assign a value of 1, otherwise -1, no change assign 0, thus constructing a dummy variable for upward occupational mobility of labor (Ca) to measure upward professional mobility. Second, examine the industrial structure transformation of labor. Similarly, construct a dummy variable for employment industry transformation (It), if changed from primary or secondary industry to tertiary industry in the previous year assign 1, otherwise -1, no change 0. Finally, explore the connection between technological penetration and personal employment risk. Theoretical analysis suggests that AI robot penetration increases the frequency of job changes, increases job uncertainty, and exacerbates overall employment risk. This means that when faced with job uncertainty, both enterprises and individuals tend to sign short-term contracts to cope with the increased frequency of job rotation. This process is not the traditional “unemployment” risk in the employment process but stems from the “uncertainty” risk caused by increased job rotation frequency in the employment market. Accordingly, we construct a dummy variable for the type of employment contract signed (Ec) to explore the impact of stability risk brought about by the short-term nature of employment contracts: assign values 0 to 3 for signing fixed contracts, long-term contracts, short-term contracts, or no contract with the employer, respectively, to measure employment risk under uncertainty. Based on the data characteristics of categorical variables, we choose the fixed-effects ordered logit (feologit) model for panel data to test the above three mechanisms. The test results are shown in Table 7.
From columns (1) and (2) of Table 7, as penetration rate increases, the trend of upward occupational mobility and transformation to the tertiary industry becomes more obvious. AI robot penetration not only promotes the optimization and upgrading of occupational structure but also facilitates the adjustment and transformation of industrial structure, thereby affecting the total income, monthly wage, and labor participation share of different labor forces. The application of AI robots brings new development opportunities to professional fields such as healthcare, finance, and education, enabling practitioners to achieve upward mobility and improve social status and economic benefits. At the same time, as people’s demand for high-quality life continues to increase, market demand in industries such as tourism, culture, and entertainment is increasingly strong, and the application of artificial intelligence further enhances the service level and competitiveness of these fields, attracting more relevant practitioners. The results in column (3) show that the coefficient of penetration rate is significantly positive, i.e., an increase in penetration rate leads to short-term employment contracts for labor, thereby affecting income and labor participation share. With the popularization of AI robots, some traditional industries and occupations face the risk of being replaced; meanwhile, traditional workers use digital technology to achieve human capital appreciation, causing practitioners to face pressure of unemployment or career change on one hand and motivation for occupational mobility on the other, thereby exacerbating the trend of short-term employment and increasing employment stability risk.
The共同加剧 of occupational mobility and employment stability risk essentially reveals the dual mechanism of AI technology penetration. At the micro level, technological innovation provides upward mobility channels for workers by creating new high-skilled positions and brings development opportunities for human capital appreciation; at the meso level, accelerated technological iteration triggers job reconstruction, significantly shortening employment contract terms, forming systemic risk accumulation. The seemingly contradictory coexistence of “opportunity and risk” is essentially the structural adjustment of the labor market brought about by technological shock. When individuals achieve vertical mobility through skill reshaping, their employment stability depends on continuous technological adaptability; while the market, against the background of compressed technology replacement cycles, is forced to adopt flexible employment mechanisms. Therefore, the individual benefits of upward occupational mobility and the systematic deepening of employment risk are precisely the presentation of the “creative destruction” characteristics of technological progress in different dimensions of the labor market, and the two achieve logical self-consistency through the dynamic game between human capital depreciation and job duration cycles. Thus, Hypothesis 4 is fully verified.
In summary, AI robot penetration promotes upward occupational mobility and transformation to the tertiary industry but also brings employment risks. Its creation and substitution effects jointly affect labor income and participation share. This “opportunity and risk” is essentially the continuation of the structural contradiction between the cycle law of technological revolution and the lag of institutional adjustment in the digital era. Upward occupational mobility reflects the creative power of technological progress, while the deepening of employment risk exposes institutional inertia in the adjustment of production relations. While technological progress enhances total social welfare, it is inevitably accompanied by adaptation pains at the individual level. The core of resolving this contradiction lies in shortening the adaptation time lag between “technology and institution” through institutional innovation.
| Variable | (1) Ca | (2) It | (3) Ec |
|---|---|---|---|
| Penetration Rate | 6.014*** (2.815) | 15.910*** (12.978) | 7.017** (1.987) |
| Pseudo R² | 0.228 | 0.762 | 0.074 |
| Number of Observations | 4,066 | 4,489 | 3,294 |
Note: 1. ** and *** indicate significance at the 5% and 1% levels, respectively. 2. t-values are in parentheses. 3. Fixed effects are controlled; control variable estimation results are retained for reference.
The previous discussion on the differential impact of AI robot penetration between sectors and the “polarization effect” leads us to further examine its impact on income inequality. Referring to the research of Guo and Feng (2024), we calculate the income gap (RD) faced by individuals based on the deprivation function of personal wages as a new explained variable. After controlling for monthly wage and monthly working days to exclude the direct impact of wage rates and labor participation share on income inequality, we estimate according to the baseline model. The results in Table 8 show that AI robot penetration significantly exacerbates income inequality, a trend deeply reflecting changes in socio-economic structure, particularly manifested in the income gap between different occupational groups.
| Variable | Explained Variable: RD | ||||
|---|---|---|---|---|---|
| (1) Full Sample | (2) Heads and Administrative Service Personnel | (3) Senior Intellectuals and Professional Technicians | (4) Production Service Personnel | (5) General Manufacturing Personnel | |
| Penetration Rate | 0.738*** (9.110) | 0.629*** (2.987) | 0.590** (2.526) | 1.089*** (5.320) | 0.646*** (6.052) |
| Monthly Wage | -0.233*** (-67.612) | -0.238*** (-35.254) | -0.213*** (-23.609) | -0.245*** (-42.045) | -0.224*** (-30.832) |
| Working Days | -0.031*** (-3.440) | -0.012 (-0.965) | -0.005 (-0.276) | -0.047*** (-2.886) | -0.051** (-2.569) |
| Adj. R² | 0.871 | 0.903 | 0.808 | 0.890 | 0.828 |
| Number of Observations | 7,674 | 2,069 | 1,357 | 1,978 | 1,745 |
Note: 1. ** and *** indicate significance at the 5% and 1% levels, respectively. 2. t-values are in parentheses. 3. Fixed effects are controlled; control variable estimation results are retained for reference.
The significantly positive coefficient of penetration rate in column (1) of Table 8 indicates that AI robot penetration does exacerbate social income inequality. The analysis by occupational type in columns (2) to (5) shows that although all occupations face pressure of widening income inequality (coefficients are significantly positive), the differences are obvious. The abnormally high coefficient in column (4) for production service personnel confirms the strengthening mechanism of the polarization effect. This grouping has the unique “skill continuum” characteristic of coexisting high, medium, and low-skilled positions, making it an “amplifier” of technological shock. That is, AI robot penetration brings about the bidirectional transformation of middle-level labor, with some upgrading to high-skilled labor and more downgrading to low-skilled operational positions, forming the “disappearance of the middle” in employment structure and the middle collapse of income distribution, ultimately amplifying income inequality. Additionally, the coefficients in columns (2), (3), and (5) are all lower than the full sample, but the causes differ. The coefficient in column (2) for heads and administrative service personnel reflects the buffering effect of institutional rigidity on technological shock, with organizational capital and specific human capital forming a double barrier against technological substitution, delaying the transmission speed of the polarization effect. The significantly positive penetration rate coefficient in column (3) for senior intellectuals and professional technicians mainly stems from high-skilled workers obtaining innovation premiums through technological adaptability, and their widening income gap comes from within-group competition rather than structural substitution. The significantly positive coefficient in column (5) for low-skilled general manufacturing personnel is due to the natural technological barrier formed by human-machine interaction and other non-procedural characteristics, so the adverse impact on this group is also lower than the full sample.
These differences indicate that the social impact of technological penetration depends not only on skill levels but also on the constraints of occupational institutional embeddedness and job structure. With the development of AI technology, the income gap between high-skilled and low-skilled labor will further widen, and the plight of medium-skilled labor will become increasingly severe. This will not only weaken the overall economic efficiency of society but may also trigger social instability and class solidification. How to use institutional advantages to破解 the adverse effects of the polarization effect on “middle” occupational groups will become a key issue to be solved in the future.
Conclusions and Recommendations
From the perspective of task bias, this article constructs a theoretical model considering horizontal and vertical technological progress, focusing on the impact of AI robot penetration on residents’ income and labor participation share, exploring the mechanisms of its creation and substitution effects, verifying its polarization effect on high- and low-skilled labor, and analyzing its impact on income inequality, drawing three conclusions. First, the widespread application of AI robots currently reduces the income of the labor sector, mainly due to its comprehensive impact on wage rates and labor participation shares. The increase in penetration rate on wage rates is insufficient to offset its squeeze on labor participation shares, ultimately leading to a negative impact on residents’ total income. Second, AI robot penetration both widens the income difference between high- and low-skilled sectors and narrows the difference in labor participation shares between the two sectors, and is particularly significant in the tertiary industry. The coexistence of income polarization and participation convergence produces the “disappearance of the middle” phenomenon, exacerbating income inequality and leading to the middle collapse of income distribution. Third, AI robot penetration, while promoting upward occupational mobility of labor and promoting the transformation of employment industrial structure, also brings employment stability risks, thereby affecting labor income and labor participation share. Admittedly, the rapid development of AI robots has promoted significant productivity improvements, but from the perspective of resident utility maximization, its impact presents diversified characteristics. The increase in labor costs is an inevitable trend, and income polarization may be alleviated by the creation effect of AI robot penetration. The key to achieving a balance between AI robots and resident well-being lies in how to effectively increase the “middle” labor return and proportion, i.e., fully leveraging the creation effect of AI robots. Based on the above conclusions, this article proposes the following policy recommendations:
First, promote the efficient synergy and integrated advancement of AI technology development and high-quality full employment. Adhere to the employment priority strategy, incorporate employment goals in the process of promoting the deep integration of AI technological innovation and industrial innovation, guide innovation resources towards directions with great potential for job creation, stabilize and expand employment capacity, and achieve compatibility between technological orientation and labor well-being. Focus on developing human-machine collaboration technologies that require creativity, emotional communication, and other unique human advantages, especially in service sectors such as healthcare, education, elderly care, and high-end consulting, and fundamentally increase labor remuneration and participation share by innovating human-machine collaboration work forms. Use AI and big data technology to prospectively study the potential job creation and unemployment risks of industrial policies and smart manufacturing projects, ensuring the coordination of economic and social benefits of technological innovation. In addition, use new media to break the pessimistic argument that “technological progress equals unemployment,” emphasize lifelong learning, alleviate widespread social anxiety about technological change, and create a positive social employment atmosphere that advocates learning, embraces change, and tolerates failure.
Second, achieve the transition from “income polarization” to “opportunity equity” through skill training and structural fiscal policies. On one hand, strengthen the reshaping and occupational empowerment of middle- and low-level labor positions, and continuously expand the size of the middle-income group to reduce income inequality. Provide micro-courses and micro-certifications through digital platforms to help workers learn in fragmented time, focusing on skill training for small and medium-sized enterprises and medium- and low-skilled workers to enhance the convenience and accessibility of learning. Encourage enterprises to provide professional, targeted skill upgrade training for new human-machine collaboration models, helping workers, especially medium- and low-skilled workers in the service industry, transition from simple task execution to more complex system coordination, data analysis, and human-machine collaborative management, broadening income channels, and promoting opportunity equity. On the other hand, avoid non-optimal employment caused by “excessive automation” and the distortion of AI investment and innovation caused by “direct robot taxation,” consider strengthening capital gains tax and excess profit supplementary tax to ensure that the benefits brought by technological progress are more fairly distributed to society. Further reduce personal income tax rates for low- and middle-income groups and expand the scope of special additional deductions for education, elderly care, child rearing, etc., to increase residents’ disposable income.
Finally, transition from “passive bottoming” to “active empowerment” to smoothly cope with the coexistence of upward labor mobility and employment risks. Jointly with industry associations and leading enterprises, clarify the skill requirements for new positions arising from AI and provide transformation guidance. Establish personal lifelong learning accounts, with joint funding from the government, enterprises, and individuals, allowing workers to independently choose training programs based on their own career development and market changes, achieving “portable” and “sustainable” skill reserves. Build a Chinese occupational information network database to provide personalized career counseling and job transfer support services, alleviate occupational information asymmetry, and ensure that various policies can accurately serve the groups with the greatest transformation potential. Strengthen the “re-employment” function of unemployment insurance, deeply integrate unemployment relief with personalized, intelligent career counseling, skill retraining, entrepreneurship support, and other active employment services, and provide rapid response support from survival security to ability improvement for workers in the transition period. Launch cross-industry retraining programs for groups unemployed or at high risk due to technological change, closely linked to employment positions, provide training allowances, and explore government-subsidized “hire first, train later” new models.