While some employees have a neutral attitude towards the changes in employment modes brought about by AI (Brougham & Haar, 2018), the increasing use of AI has led to more people being employed in informal, unstable, and non-standard modes, which has weakened the collective power of employees (Boeri et al., 2020). Unlike past automation technologies, the new generation of AI is replacing mental labor and changing the division of labor in unique ways, with differentiating features in the value structure of human capital (Atack et al., 2019). AI has a strong "leader goose" effect with a high spillover driving force, as noted in a speech by Xi Jinping at the Ninth Collective Study of the Political Bureau of the CPC Central Committee in 2018. Due to the weakening of external ability, low-skilled workers have limited autonomy in terms of the time and spatial range of skill improvement, resulting in slower skill improvement than the speed of AI substitution.
In recent years, some industrialized countries have faced the dilemma of declining labor share and slow growth in labor productivity despite the widespread use of AI (Autor, 2019). There is an increasingly deep contradiction between sluggish growth in labor productivity and changing employment demands. Using the wrong type of AI can lead to misallocation of resources by markets, enterprises, and individuals towards labor-intensive tasks, exacerbating labor substitution and causing serious unemployment, weak growth, and income inequality.
The degree to which workers are substituted by AI depends on the match between the types of daily tasks and skills. Workers with high skills who engage in complex work tasks are less likely to be replaced by AI because they possess strong cognitive skills. However, in the process of substitution, high-skilled workers have exacerbated the ripple effect of replacing low-skilled workers. This contradiction leads to significant substitution of AI for medium-skilled labor, affecting long-term employment equilibrium (Acemoglu & Restrepo, 2018b).
According to Acemoglu et al. (2021), workers highly involved in tasks with AI experience negative impacts on their employment and wages. The displacement effect primarily involves the substitution of programmatic work and some non-programmatic work, such as transportation, driving, and image diagnosis, where AI excels (Autor, 2015; Liu Xiangli, 2020). As AI diffuses and penetrates different regions, sectors, industries, and departments of the national economy, the degree of skill premium becomes increasingly prominent. The demand for human capital shifts from medium and low skills to high skills, resulting in employment polarization and exacerbation of income inequality (Acemoglu & Restrepo, 2019).
Low-educated blue-collar workers, medium-skilled blue-collar workers, and white-collar administrative positions are gradually disappearing, with low-educated men facing a sharp decline in employment prospects (Binder & Bound, 2019). Low-skilled workers have lost core skills that cannot compete with technological premium (Agrawal et al., 2019), and technological change has narrowed their scope of work (Acemoglu & Restrepo, 2020c), leading to their replacement by AI in the fast automation process (Autor, 2019).
Under the labor selection mechanism, high-skilled workers are increasingly inclined to work in high-wage industry sectors and are more likely to form partnerships with each other, reinforcing the entry barriers to high-wage industries and isolating medium and low-skilled workers (Song et al., 2019). Income inequality indirectly increases the relative income of capital owners, leading to an increasingly severe class divide (Acemoglu & Restrepo, 2020a).
The root cause of this issue is that AI weakens the ability of low-skilled workers to obtain subsistence materials from labor products, while strengthening the ability of high-skilled workers to obtain subsistence materials from capital products. Through the gap in human capital accumulation, income inequality is indirectly exacerbated.
Labor with scarce skills and low costs will bear the consequences of downward economic and organizational forces, and the chances of returning to formal work organizations are slim (Fleming, 2019). Although the degree of technology application will slow down the spread of the substitution effect (Naude, 2020), labor demand will be stifled by the inability to fully benefit from the creation effect, which cannot generate a strong creation effect to resist the labor demand decline caused by the substitution effect (Acemoglu & Restrepo, 2018a; Clifton et al., 2020).
The job-creation effect takes a long time to materialize, and the creation effect of AI is unlikely to offset the negative pressure on labor demand, employment, and wages caused by the substitution effect in the short term (Lane & Martin, 2021).