AI Reshapes China’s Labor Market: High and Low-Skill Jobs Rise, Middle-Skill Roles Decline
In a landmark study that bridges the gap between macroeconomic theory and micro-level corporate behavior, researchers from Xi’an Jiaotong University have uncovered compelling evidence that artificial intelligence (AI) is not merely replacing human labor—it is actively reconfiguring the structure of employment demand across China’s corporate landscape. The findings, published in the Journal of Xi’an Jiaotong University (Social Sciences), challenge simplistic narratives of AI-driven job loss and instead reveal a nuanced, skill-biased transformation with profound implications for policymakers, business leaders, and workers alike.
At the heart of the research lies a critical question: as AI systems become increasingly capable of performing tasks once reserved for humans, what happens to the demand for labor? Conventional wisdom often leans toward dystopian forecasts—mass unemployment, widespread displacement, and economic instability. Yet the empirical reality, as demonstrated by Hao Meng and Meisha Zhang of the School of Economics and Finance at Xi’an Jiaotong University, is far more complex and, in some respects, counterintuitive.
Analyzing a decade of employment data from Chinese listed companies between 2010 and 2019, the study employs a quasi-natural experimental design using the difference-in-differences (DID) methodology—a rigorous econometric approach that isolates the causal impact of AI adoption by comparing firms that adopted AI technologies with those that did not, while controlling for firm-specific and time-specific factors. This micro-level perspective is particularly significant, as prior research on AI and employment has largely focused on national or sectoral aggregates, especially in developed economies like the United States or Germany. By zooming in on individual enterprises across both manufacturing and service sectors, Meng and Zhang offer one of the first comprehensive views of AI’s labor market effects in a large, developing, and rapidly digitizing economy.
Their central finding is striking: AI adoption leads to an overall reduction in total employment demand at the firm level. This confirms that, at least in the short to medium term, the job-displacing (or “substitution”) effect of AI outweighs its job-creating potential. However, this aggregate decline masks a deeper structural shift—one that aligns with global trends but manifests in uniquely Chinese institutional and economic contexts.
Specifically, the study reveals a pronounced “polarization” or “hollowing out” of the labor market. Firms that adopt AI significantly increase their demand for both high-skill and low-skill workers, while simultaneously reducing their reliance on middle-skill employees. This U-shaped pattern—rising demand at the extremes, falling in the middle—is not new in labor economics; it has been documented in advanced economies under the framework of “skill-biased technological change.” But its emergence in China, a country still undergoing industrialization and urbanization, signals that AI is accelerating structural transformation at an unprecedented pace.
Why does this polarization occur? Meng and Zhang go beyond correlation to identify two distinct mechanisms driving these divergent outcomes.
First, AI enables firms to expand their production scale more efficiently. Lower automation costs boost profitability, which in turn fuels output growth. This expansion creates new demand for labor in non-automatable tasks—particularly those requiring physical presence, manual dexterity, or basic service interactions. These roles are typically filled by low-skill workers. Thus, while AI may replace some routine manual jobs, it simultaneously generates new ones through scale effects, especially in logistics, maintenance, packaging, and customer-facing support roles that resist full automation.
Second, AI drives technological upgrading, which reshapes the nature of work itself. As firms integrate intelligent systems into their operations, they require employees who can develop, manage, interpret, and maintain these technologies. These are high-skill roles—data scientists, AI trainers, systems engineers, compliance monitors—many of which did not exist a decade ago. The study confirms that firms experiencing greater gains in total factor productivity (a proxy for technological advancement) following AI adoption significantly increase their hiring of high-skill workers. Conversely, middle-skill workers—those performing routine, codifiable tasks such as clerical work, machine operation, or mid-level supervision—are most vulnerable. Their roles are precisely the kind that AI excels at automating: predictable, rule-based, and low in emotional or creative content.
This dual-channel mechanism—scale expansion boosting low-skill demand, and tech upgrading boosting high-skill demand—explains the polarized employment response. It also underscores that AI is not simply a labor-saving tool; it is a catalyst for organizational and occupational restructuring.
But the story doesn’t end there. The researchers delve into heterogeneity across ownership, industry, and geography, revealing that AI’s labor market impact is far from uniform.
In state-owned enterprises (SOEs), the classic polarization effect is muted. While high-skill hiring rises, there is no significant decline in middle- or low-skill employment. This likely reflects the social mandate of SOEs in China, which often prioritize employment stability over pure profit maximization. Labor shedding in SOEs faces higher institutional and political costs, leading to “employment inertia” even in the face of automation. In contrast, private firms—driven by market competition and flexible labor practices—exhibit the full U-shaped pattern, closely mirroring findings from Western economies.
Sectoral differences are equally revealing. In manufacturing, AI adoption strongly reduces middle-skill employment while increasing demand at both ends of the skill spectrum. This aligns with the nature of factory work, where routine assembly, quality control, and machine tending are highly automatable. But in the service sector, the pattern diverges: AI boosts demand for high- and middle-skill workers, with little impact on low-skill roles. Why? Because many service jobs involve interpersonal interaction, empathy, or contextual judgment—traits that remain difficult for AI to replicate. Moreover, as manufacturing sheds workers, some may migrate into service roles, artificially inflating demand in that sector.
Geographically, the effects are concentrated in China’s eastern coastal regions—home to tech hubs like Shenzhen, Shanghai, and Hangzhou—where AI adoption is most advanced. Here, firms show the clearest evidence of labor polarization. In central and western regions, the overall employment impact is statistically insignificant, suggesting that AI’s labor market disruption has yet to permeate less developed areas. However, subtle shifts are emerging: eastern firms are reducing low-skill hiring, possibly due to high urban living costs that “squeeze out” low-wage workers, who then relocate to inland cities. This could explain why central regions see rising demand for both high- and low-skill labor, while western firms show unexpected increases in middle-skill hiring—perhaps as they absorb displaced workers or develop localized tech ecosystems.
These findings carry urgent policy implications. The notion that “AI will take all our jobs” is not only inaccurate but dangerously misleading. The real challenge is not mass unemployment, but structural mismatch: a growing surplus of middle-skill workers whose competencies no longer align with market needs, alongside shortages of both high-skill AI specialists and adaptable low-skill service providers.
Meng and Zhang argue for a differentiated policy response. For middle-skill workers—particularly those in routine manufacturing or administrative roles—large-scale reskilling and upskilling programs are essential. Vocational training should emphasize digital literacy, problem-solving, and human-centric skills that complement, rather than compete with, AI. For high-skill talent, China must accelerate the development of AI-focused curricula in universities and foster industry-academia collaboration to close the innovation gap.
At the regional level, one-size-fits-all AI promotion policies are inadequate. Eastern cities should leverage their tech advantage to create high-value jobs and support entrepreneurial ecosystems. Meanwhile, central and western regions must prepare for an influx of displaced workers by investing in infrastructure, basic service industries, and localized AI adoption frameworks that prioritize inclusive growth.
Critically, the study also highlights the limitations of short-term thinking. While AI currently exhibits a net substitution effect, this may reverse as new industries emerge and labor markets adjust. The key is to manage the transition—ensuring that the benefits of AI-driven productivity are broadly shared, and that no segment of the workforce is left behind in the race between man and machine.
As global economies grapple with the same forces, China’s experience offers a cautionary yet hopeful blueprint. Technology does not dictate destiny; institutions, policies, and human agency shape how its impacts unfold. The research by Meng and Zhang provides not just data, but a roadmap for navigating the AI revolution with foresight, equity, and resilience.
Authors: Hao Meng and Meisha Zhang, School of Economics and Finance, Xi’an Jiaotong University
Published in: Journal of Xi’an Jiaotong University (Social Sciences), Vol. 41, No. 5, September 2021
DOI: 10.15896/j.xjtuskxb.202105007