The 1980s and 1990s saw a sharp increase in earnings inequality in the US. As new technologies displaced the routine tasks typically performed by less-educated workers, more-educated workers experienced an upturn in their economic position that has endured ever since. This column argues that artificial intelligence could mitigate or even reverse this disparity between low and highly skilled workers. Because AI is designed to substitute for non-routine tasks, it has the potential to exert downward pressure on the wages of highly skilled workers and the skills premium more broadly.
Earnings differentials among racial, ethnic, gender, age, and school groups are often interpreted as indicators of relative economic hardship and economic fairness. In recent decades, one of the most striking temporal patterns involving earnings differentials in the US has been the general rise in the economic position of more-educated workers in the 1980s and 1990s, with little indication of a subsequent decline (Katz and Murphy 1992, Levy and Murnane 1992, Heathcote et al. 2023).
Much effort has been devoted to testing and measuring how myriad supply, demand, and institutional factors contributed to the growth and sustained high earnings premia of the more educated. The focus has been on assessing variations in immigration and trade flows, the relative supply of more-educated workers, union density, real minimum wages, and technological change. Generally speaking, the strongest evidence highlights the effects of technological change (Krusell et al. 2000). The argument is that new technologies (especially automation) that emerged in the 1980s and 1990s either (1) tended to displace routine tasks performed by less-educated workers (Autor et al. 2008, Acemoglu et al. 2023b), or (2) exhibited a high degree of complementarity with education, resulting in a sizable boost in the relative demand and pay for more-educated (i.e. more-skilled) workers (Bound and Johnson 1992).
More recently, technological change has taken on different forms, especially with the emergence of artificial intelligence or AI (Agrawal et al. 2019, Acemoglu et al. 2023a, Acemoglu 2024). Some scholars suggest the AI revolution will favour more-educated and skilled workers and spur even higher levels of earnings inequality (Korinek and Stiglitz 2019). Others conjecture the opposite (Webb 2019), arguing that AI performs tasks predominantly undertaken by high-skill workers, relaxing the demand for workers with those skills. For example, AI-based models and devices are increasingly used to diagnose disease, develop drugs and vaccines, write and translate various texts and code, or help generate novel ideas in marketing and research and development. Because these tasks are often nonroutine and performed by high-skill workers, AI may exert downward pressure on their wages and thus on the skill premium. Autor (2024) even argues that AI can help to rebuild the middle class by enabling lower-skill individuals to do more complicated tasks.
To analyse the effects of AI on the skill premium at the macroeconomic level, we developed a model that relates aggregate output to the employment of low-skill and high-skill workers and to three different types of productive capital: machines and assembly lines, industrial robots, and AI (Bloom et al. 2024). We emphasise the extent to which these different types of capital substitute for workers with different skills. Humans or industrial robots need to operate machines and assembly lines. Thus, machines and assembly lines are complementary with human labour or industrial robots. By contrast, industrial robots do not need much human input to operate, and are instead designed to substitute for routine, typically low-skill intensive labour, such as workers on an assembly line. Finally, AI is designed to substitute predominantly for nonroutine tasks previously seen as difficult to automate and typically performed by high-skill workers. Overall, this rich structure of the interaction between human capital and physical capital in the production process allows us to explore conditions under which the deployment of industrial robots raises the skill premium, and conditions under which the emergence and increasing use of AI reduces the skill premium.
Relying on standard parameter values used in relevant economic literature and on data for US employment and the stocks of different types of capital deployed in the US economy, we investigate the dependence of the skill premium on the stocks of industrial robots and AI. We define workers with a bachelor’s degree or higher as high-skill workers, and those without a college degree as low-skill workers. We find that the skill premium without the use of AI amounts to about 2.00; in other words, the wages of high-skill workers are on average twice the wages of low-skill workers in the absence of AI. At this value, the skill premium is close to the value observed in the data in the 2000s, which is reassuring insofar as our framework predicts the skill premium correctly.
Next, we investigate the effects of an increase in the stock (and use) of AI capital. Such an increase leads to more intense competition between capital in the form of AI and high-skill workers, implying that the skill premium decreases. This decrease occurs at a diminishing rate; that is, it is strong for increases of the AI capital stock that start at a low level of AI, but levels off as the AI capital stock rises further. Thus, our simulated production framework implies that the deployment and increasing use of AI ceteris paribus puts downward pressure on the gap between the average wages of high-skill and low-skill workers.
Because the ceteris paribus condition is rarely fulfilled in reality, we also allow the stock of industrial robots as a substitute for assembly line workers to accumulate. In this case, we can show that the skill premium is highest when the stock of AI capital is low and the stock of industrial robots is high. A low stock of AI means that high-skill workers face less competition from the production factor capital, which allows their wages to rise, whereas a high stock of industrial robots implies intense competition between capital and low-skill workers, depressing their wages. However, when both types of capital (AI and industrial robots) grow, the rise of the skill premium is strongly moderated. This further implies the inequality-reducing effect of AI.
In addition to effects on the skill premium, we also analyse the effects of industrial robots and AI on the wage levels of low-skill and high-skill workers. In general, increases in the stock of industrial robots reduce the average relative wage of low-skill workers, whereas increases in the stock of AI raise it. The reverse holds true for the wage levels of high-skill workers. Overall, the growing use of both industrial robots and AI can somewhat offset and therefore moderate changes in the skill premium. With respect to income (as opposed to earnings) inequality in general, the returns to the owners of AI capital – which would likely promote increased inequality – are also relevant.
The bottom line is that AI has the potential to mitigate or even reverse the sharp increases in skill premia and earnings inequality that occurred in the 1980s and 1990s and have endured since.
Source : VOXeu
