Unpacking Skill Bias: Automation and New Tasks | 2020
Joint with Daron Acemoglu
The standard approach to modeling inequality, building on Tinbergen’s seminal
work, assumes factor-augmenting technologies and technological change biased in favor of skilled workers. Though this approach has been successful in conceptualizing
and documenting the race between technology and education, it is restrictive in a
number of crucial respects. First, it predicts that technological improvements should
increase the real wages of all workers. Second, it requires sizable productivity growth
to account for realistic changes in relative wages. Third, it is silent on changes in job
and task composition. We extend this framework by modeling the allocation of tasks
to factors and allowing richer forms of technological changes — in particular, automation that displaces workers from tasks they used to perform, and the creation of new
tasks that reinstate workers into the production process. We show that factor prices
depend on the set of tasks that factors perform, and that automation: (i) powerfully
impacts inequality; (ii) can reduce real wages; and (iii) can generate realistic changes
in inequality with small changes in productivity. New tasks, on the other hand, can
increase or reduce inequality depending on whether it is skilled or unskilled workers
that have a comparative advantage in these new activities. Using industry-level estimates of displacement driven by automation and reinstatement due to new tasks, we
show that displacement is associated with significant increases in industry demand for
skills both before 1987 and after 1987, while reinstatement reduced the demand for
skills before 1987, but generated higher demand for skills after 1987. The combined
effects of displacement and reinstatement after 1987 explain a significant part of the
shift towards greater demand for skills in the US economy