How Automation and AI Affect Worker Well-Being: Looking Beyond Displacement and Wages
*Names in bold indicate Presenter
Recent discourse surrounding these emerging technologies has focused largely on labor displacement, arguing that labor substitution is a problem while labor complementarity can be construed as a positive force. In this paper, we look beyond wages and employment rates, arguing that labor complementarity may not be uniformly positive. In particular, increasing levels of automation in work may impact workers’ well-being in the present as well as their expectations about the future. Technology can impact worker stress, difficulty of tasks, levels of monitoring, autonomy, and job security, among other impacts.
To examine such, we rely on two data sets. First, we examine a 2017 Pew Research Center survey on Automation in Everyday Life, which considers how individuals perceive automating technologies have affected their work, as well as their future work expectations. We restrict our attention to employed individuals, and contrast how individuals in manual labor roles are affected compared to those in management labor roles. This analytical approach follows theory emphasizing the difference in technology impacts across ‘low’ and ‘high’ skills (Griliches, 1969).
Second, we use the 2012 and 2013 American Time Use Survey (ATUS) Well-Being Module to ascertain whether workers in highly automated jobs experience different levels of well-being (measured through scales of happiness, stress, tiredness, meaningfulness, etc.,) when performing their jobs. The data set allows us to directly measure well-being of workers while working. To identify the degree of automation in each occupation, we use the Skill-Biased Technological Change literature (Autor, Levy and Murnane, 2003; Jaimovich and Siu 2014) to classify workers by their occupation’s task content as routine or non-routine workers. Alternatively, we adapt the methodology of Frey and Osborne (2017), who classify occupations by their susceptibility to computerization (jobs with significant elements of physical manipulation and social and creative intelligence are classified as less automatable). Both classifications are used to evaluate well-being impacts. These alternative classifications serve as robustness checks on our findings and additionally help to assess the reliability of the automation-based occupation classification approaches themselves.
Our initial results indicate that workers in more automatable occupations perceive meaningful impacts on their well-being, and that these results differ across manual and managerial (cognitive jobs). Manual work appears to be associated with negative impacts on outcomes including how interesting work is, opportunities for advancement, and job security. Standardized effects are reduced but remain significant in most cases when controlled by education, industry, and prior knowledge about automation. We find mixed and weaker associations between well-being (happiness, meaningfulness, stress) and susceptibility to automation. These results may be explained by individuals’ lack of knowledge about the future ‘automatability’ of their work, but further work is needed to disentangle these relations.