人工智能暴露度与岗位需求结构重塑 ——来自招聘平台大数据的证据

AI Exposure and the Restructuring of Job Demand:Evidence from Online Recruitment Big Data

摘要:

人工智能的快速发展引发了关于其就业效应是“创造”还是“替代”的广泛讨论。与现有研究多聚焦于就业人数或工资水平不同，本文关注一个更偏向于需求侧的问题：人工智能的持续扩散是否改变了岗位需求结构，而不仅仅影响岗位需求总量？考虑到招聘广告更能反映企业的边际决策，本文基于2018—2024年我国主流招聘平台广告数据，构建了城市—职业层面的人工智能暴露度指标，考察人工智能技术扩散对岗位需求规模、结构及其作用机制的影响。研究发现：人工智能总体上促进了岗位招聘需求的增长，“机器换人”现象并未发生。但需求增长的内部结构并不均衡，即非常规岗位需求增加，常规岗位需求则受到抑制。机制分析表明，人工智能主要通过催生新岗位和提高生产率两条路径推动非常规岗位的扩张；而在劳动力成本较高、工资水平较高、任务可程序化程度较强的情形下，其对常规岗位的替代效应更为突出。异质性分析表明，人工智能对岗位需求的影响随时间推移持续增强，且主要集中于非常规认知和高技能岗位，对常规体力岗位则呈现出较强的替代效应。本文发现，人工智能对劳动力市场的影响更多体现为岗位需求结构的调整，而非仅限于就业规模的增减。

Abstract:

The rapid advancement of artificial intelligence (AI) has intensified the debate over job creation versus displacement. While much of the existing literature focuses on employment levels or wage outcomes, this study shifts attention to a demand-side question: does the diffusion of AI reshape the structure of labor demand rather than merely altering its aggregate scale? Because job postings capture firms’ marginal hiring decisions more directly than traditional employment statistics, we draw on large-scale data from major Chinese online recruitment platforms spanning 2018–2023 to construct a city–occupation-level measure of AI exposure. We then examine how AI diffusion affects both the volume and composition of job demand, along with the underlying mechanisms. Our findings can be summarized as follows. First, AI exposure is associated with a significant increase in overall hiring demand; at the aggregate level, there is no evidence of widespread labor displacement. However, the growth in labor demand is structurally uneven. AI expansion raises demand for non-routine occupations while dampening demand for routine-task-intensive positions. Mechanism analyses suggest that AI fosters the expansion of non-routine jobs through two primary channels: the creation of new occupations and productivity enhancement. In contrast, substitution effects are more pronounced in contexts characterized by higher labor costs, elevated wage levels, and tasks with a high degree of programmability. Heterogeneity analyses further reveal that the impact of AI strengthens over time and is concentrated in non-routine cognitive and high-skill occupations, whereas routine manual jobs experience significant displacement pressures. Overall, our results indicate that AI’s labor market effects manifest first and foremost through the reorganization of task structures on the demand side—observable in firms’ recruitment behavior—rather than through immediate changes in aggregate employment.