In recent years, the new generation of information technology has been widely used in all aspects of social production, and has exerted a complex and far-reaching impact on labor income. Human-machine substitution and human-machine collaboration are two forms of this. Compared with human-machine substitution, human-machine collaboration can expand the capability boundary of workers through the upgrading of labor tools, improve the efficiency of task execution, and more intuitively and logically empower the growth of labor income. However, from the perspective of capital deepening, the application of new-generation information technology will inevitably weaken the role of workers in production and reduce the share of labor income. Therefore, it is unclear whether human-machine collaboration will necessarily increase labor income.

This paper uses more than 19 million online recruitment data to measure human-machine collaboration at the job level, and explores its impact on labor income. The results show that human-machine collaboration generally promotes labor income through two mechanisms: One is to promote labor income through the “labor productivity effect”, and the other is to suppress labor income through the “weakening effect of labor role”. Human-machine collaboration does not necessarily increase labor income, while education-dependent collaboration promotes but non-education-dependent collaboration suppresses labor income. The impact of human-machine collaboration on labor income varies by year, education, and region. Skill specificity strengthens the positive impact of human-machine collaboration on labor income. Human-machine collaboration significantly improves the level of protection for the rights and interests of workers.

This paper makes the following marginal contributions: First, human-machine collaboration is measured at the job level for the first time by using online recruitment data, and the Gaussian mixture clustering method is innovatively introduced to decompose the human-machine collaboration into education-dependent collaboration and non-education-dependent collaboration. Second, it confirms the “labor productivity effect” and “weakening effect of labor role” of human-machine collaboration, and puts forward the view that human-machine collaboration does not necessarily increase labor income. Third, it finds the moderating effect of skill specificity in the impact of human-machine collaboration on labor income, and expands the consequences of human-machine collaboration from the perspective of employment quality.