The effect of automation and artificial intelligence (AI) on future employment is a subject of substantial interest. There are some disagreements about future net employment effects of these technologies, but most researchers broadly agree that nearly all occupations will eventually be affected. Less is known, however, about the interaction between automation and key labor market institutions—policies and organizations that shape the operation of labor markets—and how these institutions might possibly evolve in the face of continued adoption of the newest technology. For example, is the current structure of unemployment insurance programs sufficient for a world with increased job-to-job transitions? What role will private-sector unions play? Or, considering current conversations among presidential candidates, is a universal basic income (UBI) program the best approach to deal with the wage and employment consequences of technological progress? Understanding how such features interact with technological change will have important implications for shaping labor market policies.
At a recent public event hosted by the Future of Middle Class Initiative at Brookings, experts presented new papers on the role these institutions currently play and will continue to play in modern labor markets. The papers spanned a broad set of important topics including the role of unemployment insurance and transfer policies in mitigating harm associated with technologically-induced job disruption, balancing redistributive policy with economic growth goals, how union membership influences the careers progression among people in occupations particularly susceptible to automation, and how minimum wage expansions interact with automation to determine net employment. We summarize these papers below, focusing on their central findings and what they might tell us about current and future directions for labor market policy.
Automation and employment policy
In “Automation: A Guide for Policymakers,” Bessen, Goos, Salomons, and van den Berge (2019) study the impacts of automation on individual workers. First, Bessen et al. argue that beliefs that automation in an industry portends mass unemployment events are largely unsupported by the empirical evidence. The evidence instead points to heterogenous effects across industry sectors—employment in the manufacturing sector typically suffers but employment in service-related industries tends to improve.
They draw on complementary empirical work using unique data on firms from the Netherlands. They present a key stylized fact that guides their empirical work: firm investments in these technologies typically resemble “spikes” wherein the company experiences periods of high automation investment purchase but little expenditure in this area otherwise. Their strategy compares net employment and other worker outcomes for firms that automated earlier in their observation period to outcomes of similar workers in firms that automated at a later point. Figure 2 from their paper illustrates the complementary nature of automation with firm growth. Automating firms were growing and dynamic; by contrast, firms that did not automate were largely stagnant.
Figure 2: Employment at firms making major automation investments and not, Netherlands (Bessen et al. 2019)
Turning to worker outcomes, several findings emerge. First, they find no evidence of large net employment losses at firms after an automation investment spike. They do, however, observe higher rates of job-to-job transitions among workers affected by automation. Workers in automating firms experience income losses about 11 percent of a year’s earnings, on average, over the subsequent 5 years, mostly through increased unemployment spells. These workers are more likely to switch industries, enter self-employment, and retire early after leaving the automating firm. Importantly, these effects are pronounced for older workers.
Although they note that these relatively short-run measured effects may not reflect long-run employment outcomes as technologies continue to improve, they argue that policymakers should not approach these labor market effects through the lens of mass unemployment events. Policy should be aimed at reducing job-to-job transition frictions facing workers. They cite better unemployment insurance, increased access to affordable education and training, limiting firm use of noncompete agreements, and providing relocation assistance as key examples of where policy can help mitigate the potential harm associated with automation-induced job disruptions.
The macroeconomics of automation
Nir Jaimovich, Itay Saporta-Eksten, Henry Siu, and Yaniv Yedid-Levi contributed “The macroeconomics of automation: Data, theory, and policy analysis,” which develops a macroeconomic model that serves as a theoretical laboratory to examine the net effects of a menu of potential policy responses to the labor market effects of automation.
Their analysis proceeds in two stages. First, they apply machine -learning (ML) techniques to micro-level Current Population Survey (CPS) data in the 1980s to identify the characteristics of prime-age workers typically employed in occupations characterized by routine tasks. They then apply their “trained” model to a contemporary sample to better understand how workers with characteristics associated with routine work are faring in the modern labor market, focusing on job type, labor force participation rates, and unemployment rates. The paper finds that both male and female workers that would have been likely employed in highly routinized occupations in the past were now more likely to either be in manual jobs or more likely to be out of the labor force in 2017. These “stylized facts” are used to build a heterogenous agent macro model that incorporates occupational choice.
The authors use this model to study the equilibrium outcomes for several distinct policies financed by increased tax collection. These policies include a “retraining program” targeted at the low-skilled, transfer policies aimed at improving wellbeing and reducing inequality, and a more progressive tax system. Their results, conditional on the assumptions of their model, suggest substantial different equilibrium outcomes emerging across policies.
In their model:
- Implementing a retraining program for lower-skilled workers paid for by increased taxes on the higher skilled increases overall economic growth, productivity, and labor force participation.
- Increasing unemployment insurance (UI) redistributes more income downward to lower-skilled workers, therefore lowering inequality and results in higher wages for employed lower-skilled workers. There is, however, no real change in output overall due to negative employment effects on higher-skilled workers.
- Implementing a Universal Basic Income (UBI) scheme again redistributes more income downward to lower-skilled workers, lowering inequality. Employed lower-skilled workers enjoy higher wages. However, distortions from substantial tax increases needed to fund the program ultimately lead to reductions in labor force participation and lower output in the economy.
- Reducing labor tax rates on lower-skill workers increases their labor force participation and reduces transfers. Reduced transfers, however, do not outweigh the tax revenue losses. Tax increases on higher-skill workers reduce their labor supply. These counteracting effects cancel each other resulting to roughly no change in output.
While these policy outcomes take place within a highly-stylized context, the authors’ results highlight potential tradeoffs associated with these policy proposals, finding that the best policies that tend to help low-wage workers and promote economic growth are those that help displaced workers move to new jobs such as retraining and unemployment insurance.
Union membership and automation
In “Automation, organized labor, and the employment trajectories of workers in routine jobs,” Zachary Parolin argues that organized labor affects the pace and consequences of technological change. Using individual-level Panel Study of Income Dynamics (PSID) data, he studies how union membership affects the likelihood that a routine worker (1) remains employed in a routine job for a longer duration of time, (2) avoids unemployment, and (3) achieves higher earnings over time relative to non-unionized routine workers.
Employing difference-in-differences, propensity-score matching, and event study empirical strategies, Parolin’s key findings are summarized in Figure 4 of the paper:
Figure 4: Conditional effect of union membership on likelihood of employment and low-income status after first year of work in routine occupation
Note: Coefficients on the effect of union membership presented. Models control for age, education, sex, race, state, and year of observation. Low earnings excludes unemployed.
There is a positive effect of unionization on the employment stability of workers in routine jobs. 72 percent of unionized workers remained in routine jobs at 10 years following the start of their routine employment. In contrast, only 46 percent of non-unionized routine workers remained in routine jobs after 10 years. Non-unionized routine workers more likely to be unemployed (20 percent) relative to the unionized group (9 percent). Union membership reduces the likelihood that an adult in a routine job will transition to a non-routine cognitive job. Among the non-unionized group, adults were more likely to end up in non-routine cognitive jobs. In addition, unionized workers are less likely to have low earnings relative to non-unionized routine workers.
Parolin notes that while technological innovation may be inevitable, its effects on the financial wellbeing of the workers in routine occupations is in large part a product of policy, politics, and power relations. While focusing exclusively on the role of worker power in shaping the pace and consequences of technological change, it is likely other labor market and welfare state institutions—such as minimum wage policies, access to education or job training, and social insurance/assistance programs for jobless adults—could similarly help cushion the blow of labor market change for adults working in routine jobs.
The minimum wage and technological substitution
In their paper “The Evolution of Technological Substitution in Low-Wage Labor Markets,” Daniel Aaronson and Brian Phelan study how minimum wage expansions affect employment in lower-wage occupations that are more susceptible to automation. Focusing on the period just prior to and after the financial crisis of 2008, they use variation in minimum wages expansions to examine how employment and wages changed in occupations classified as routine.
Their principal findings can be summarized in Figures 3 and 4 of their paper. Figure 3 suggests that the negative net employment impact on more heavily routine occupations has been largely concentrated in States that experienced minimum wage increases in the post-financial crisis era. This negative employment effect is largely imperceptible in States that did not. By contrast, Figure 4 illustrates that the jobs that have a more heavily interpersonal component have increased monotonically in both States that had minimum wage increases and those that did not.
Their results also suggest that increases in interpersonal jobs are smaller than losses of routine jobs in the post-financial crisis period. Hence, there has been a net decrease in employment in low-wage markets over their observation period. This effect varied by geography as well: the impact of automation in rural areas is generally larger than the corresponding effects in larger metropolitan areas.
They also show that job disruption has also been more pronounced among workers with low education levels. Among workers with a high school diploma or less, the estimated loss of routine jobs and gain of interpersonal jobs is twice as large compared to the full sample (meaning that lower-educated workers may be more likely to experience job change as a result from automation). Among this less-educated group, the results appear to be driven by the young (less than 30), minorities, and men. The authors find especially large job loss estimates for less-educated minority workers.
Their work complements recent research focusing on the impact of automation on middle-skill routine employment. The results on net-employment losses in the low-wage labor market attributable to automation suggest that offsetting employment growth in other sectors may not be enough and may be an area that requires attention from policymakers.
Lessons to consider
In the 21st century labor market, the degree to which workers, their families, and their communities adapt to the new technologies of automation and AI will depend on how public policy, private institutions, and businesses evolve to support them. Notably, a future in which all workers are displaced technology remains implausible (at least for now). New evidence, however, points to significant disruption and transitions for some workers.
The findings in these papers and conversations surrounding them during the conference point us in several directions relevant to the overall health and maintenance of the middle class in a labor market characterized by technological change. First, automation and other technological progress often reflect dynamism and economic growth. We should be careful when choosing policies that may potentially interfere with this dynamism, which will negatively impact future job creation. Adopting policies to help workers adapt faster is a better priority. Second, income transfer policies aimed at helping workers disrupted by automation should be carefully considered and targeted in a manner consistent with the empirical evidence given the scale, expense, and potential tradeoffs of such programs. Finally, in the US context, there needs to be more discussion about the role of worker power that served as a counterbalance to the power held by firms. The Future of Middle Class Initiative at Brookings will continue to explore these issues over the following months so continue to visit this space and sign up for our newsletter if you have not done so.
Research Support: Sarah Nzau