Does Artificial Intelligence Technology Application have Selective Bias

Jiang Hao,Dong Zhiqing
DOI: https://doi.org/10.19592/j.cnki.scje.410727
2023-01-01
Abstract:Artificial intelligence(AI)technology is not neutrally and equally applied in different sectors of production,and the existing literature has not pay much attention to which industries AI will bias,and what kind of biased impact will it have on the employment structures.This paper constructs an industry-level AI index based on the"TianYanCha"enterprise infor-mation database,combined with AI patent industry data crawled from china Patent Database,to examine the existence and constraints of the labor employment screening effects of AI technology at the listed company level.The research shows that the application of AI exhibits selective bias in terms of jobs and skills preferences,and the gap in the application of AI tech-nology in manufacturing industries is mainly due to the output of intelligent innovation.This,in turn,induces a structural bias towards employment in different jobs and skills,especially a significant increase in the proportion of labor in non-routine cog-nitive jobs,leading to a tendency towards"polarization"in employment structures.The results remain robust after conducting various robustness tests and employing instrumental variables techniques.Simultaneously,there is an asymmetric impact of AI on employment across different technological and ownership-based enterprises.It tends to enhance high-skill labor em-ployment in companies that more"intelligent"and"technological",while reducing low-skill labor employment,promoting skill upgrading in AI-related firms.State-owned enterprise demonstrates a protective stance towards low-level and mid-level skilled labor employment compared to private enterprises.Finally,there is a significant ripple effect of AI technology across industries,further reinforcing the screening effect in the benchmark regression.The COVID-19 pandemic exacerbates the im-pact of AI on mid-level skilled labor.
What problem does this paper attempt to address?