人工智能渗透与企业技能需求升级 ——来自招聘数据的微观证据

Artificial Intelligence Penetration and the Upgrading of Enterprise Skill Demand ——Micro Evidence from Recruitment Data

摘要:

本文利用2016—2025年A股上市公司在线招聘数据，构建企业层面人工智能渗透度（AIFE）指标，从高技能岗位的招聘规模与结构占比、以及招聘岗位的技能要求三个维度，考察AI渗透与社交型、认知型和创意型技能需求的关系。企业和年份固定效应结果显示，AIFE与三类高技能岗位招聘规模，以及招聘岗位在社交、认知和创意维度上的技能要求均显著正相关；在结构占比方面，AIFE与认知型、创意型高技能岗位占比显著正相关，与社交型岗位占比的关系不显著。不同技能维度的响应存在差异，但不支持“创意—认知—社交”的固定梯度。稳健性检验与Bartik工具变量估计支持主要结论。作用路径分析为生产率提升与常规任务替代两条路径提供了较强证据，研发强化路径在企业固定效应主设定下未获得支持。岗位层面结果表明，AIFE与认知型、创意型岗位工资溢价上升相关，与社交型岗位工资溢价下降相关，价格证据与数量证据相互印证。上述发现为优化AI时代劳动力培养与技能投资方向提供了微观经验依据。

Abstract:

Based on the online recruitment data of A-share listed companies from 2016 to 2025, this paper constructs an index of artificial intelligence penetration (AIFE) at the enterprise level, and investigates the relationship between AI penetration and social, cognitive and creative skills demand from three dimensions: the recruitment scale and structure ratio of high-skilled positions and the skills requirements of recruitment positions. The fixed effect of enterprise and year shows that AIFE is positively correlated with the recruitment scale of three types of high-skilled jobs and the skills requirements of recruitment positions in social, cognitive and creative dimensions. In terms of structural proportion, AIFE has a significant positive correlation with the proportion of cognitive and creative high-skilled jobs, but has no significant relationship with the proportion of social jobs. The responses of different skill dimensions are different, but the fixed gradient of "creativity-cognition-socialization" is not supported. Robustness test and Bartik tool variable estimation support the main conclusions. The action path analysis provides strong evidence for productivity improvement and routine task replacement, and the R&D strengthening path is not supported under the main setting of fixed effect of enterprises. The results at the post level show that AIFE is related to the increase of wage premium in cognitive and creative jobs and the decrease of wage premium in social jobs, and the price evidence and quantitative evidence confirm each other. The above findings provide a micro-empirical basis for optimizing the direction of labor training and skill investment in the AI era.