Artificial intelligence, automation and job content: the implication for wages, job mobility and training

The current wave of automation due to artificial intelligence (AI) is at the heart of public debate because of its impact on the labor market. In contrast to previous waves of automation (software and robots), which essentially replaced middle-skill workers performing specific routine tasks, artificial intelligence has demonstrated its potential to automate both routine and non-routine tasks in a wide range of occupations and sectors. Such a revolution is generating deep modifications in the design of occupations, leading to new patterns of specialization and work organization. Economists have not reached a consensus on the welfare implications of these massive changes. The potential reduction in wages due to the substitution of labor with capital for specific tasks may be counteracted by raising workers’ productivity in other tasks. While theories offer contradicting predictions, evidence of the impact of AI-induced automation on wages
and employment is still limited and inconclusive.
This PhD project will provide new evidence concerning the impact of automation due to AI on workers in France, which is a highly relevant case study. Contrary to the United States, which is the subject of most existing studies, France has strong labor market institutions and high labor costs, generating strong incentives to replace workers with automated processes. In addition, while legal environments are known to influence innovation in firms, French regulations on AI differ from those applied in the United States (e.g., data protection).
This PhD project consists of three empirical studies based on the notion that technology affects the labor market through the task content of jobs. The first two studies will investigate the impact of AI-induced changes in the task content of occupations on wages and job mobility respectively. The key contribution is to identify (i) the extent to which changes in the task content of occupations can be attributed to AI-exposure, and (ii) how such changes have affected the wage and career trajectories for the universe of workers. To do this, I will combine three data sources. First, I will provide a robust measure of exposure of tasks to AI by exploiting indices developed through three different methodologies: The Suitability for Machine Learning index (Brynjolfsson et al., 2018), the AI Occupational Impact (Felten et al., 2018), and the measure developed by Webb (2020). Second, I will exploit the O*NET database to capture variations between 2002 and 2020 in the description of tasks performed in more than 900 occupations. Third, I will merge these two sources of information with workers’ experiences and wages by occupation to French Social Security data. The methodology used to assess the impact of AI on wages and career paths will rely on econometric techniques, which control for the unobservable characteristics of workers, firms, and industries. In addition, machine-learning techniques will be used to analyze the heterogeneity of the impact of AI across subgroups of workers.
While the first two chapters will examine which profiles of workers have been hurt the most by automation, the third study will evaluate which training policies are most beneficial to workers who were displaced because of automation. This chapter will rely on administrative data providing information on training and working experience for the universe of unemployed individuals in France. I will assess the impact of training on unemployment spells using econometric methods, such as regression discontinuity design, difference in differences and matching.
In conclusion, this PhD project is closely aligned with national and international priorities and provides an original scientific contribution to a highly relevant societal issue. It will do so by providing new evidence on the positive and adverse effects of automation and will inform policymakers in the design of new public policies.