“Artificial Intelligence, Automation and Work,” circulated as NBER Working Paper 24196 in 2018, is the theoretical companion to Daron Acemoglu and Pascual Restrepo’s empirical robot studies. It lays out the task-based model that underpins much of their later work and gives the field a clear vocabulary for thinking about how AI and automation reshape the demand for labor.
The model centers on a tug-of-war between two forces. A displacement effect occurs when machines take over tasks that workers used to do, lowering the demand for labor. Counteracting it are a productivity effect, which raises output and can increase demand for labor in the tasks that are not automated, plus capital deepening and, most importantly, the creation of entirely new labor-intensive tasks that “reinstate” labor in new activities. The authors stress that automation tends to increase output per worker more than it raises wages, which lowers labor’s share of national income, even when the countervailing forces are strong.
The paper also warns that adjustment is not automatic. Skill mismatches can leave displaced workers stranded, and firms may adopt “so-so” automation that is privately profitable but socially wasteful, automating at an excessive rate at the expense of more productivity-enhancing technologies. For business and policy readers, it supplies the conceptual map: the question is never just “will AI automate this,” but whether the economy also generates new tasks fast enough to absorb the workers it displaces.