Bessen looks at estimates of employment demand growth.
Taking these considerations into account, I estimate a simple model of occupational demand across industries that allows for changing demand and inter-occupation substitution within industries. As my key independent variable, I measure the extent of computer use by workers in each occupation and industry. These data come from supplements to the Current Population Survey. I assume that occupations that use more computers will have a higher degree of task automation, all else equal. The dependent variable is the relative growth of employment in occupation-industry cells.The last sentence is the takeaway result. Computer automation is not associated with major job losses meaning our modern day Luddites have less to worry about than they first thought.
The estimates contradict popular assumptions about the impact of computer automation. First, computer-using occupations tend to grow faster, not slower. At the sample mean, computer use is associated with a 1.7% increase in occupational employment per year. In other words, the bank teller example may be typical rather than exceptional.
Second, there is a strong substitution effect between occupations. Occupations tend to have declining growth to the extent that other occupations in the same industry use computers. That is, the story is not about machines replacing humans; rather it is one of humans using machines to replace other humans, as graphic designers with computers replaced typesetters.
The substitution effect largely offsets the growth effect. Counting both, at the sample mean, computer use is associated with positive employment growth but the effect is small, 0.45% per year. This association is not necessarily causal—perhaps some other factor caused computer-using occupations to grow. But this finding does show that computer automation is not associated with major job losses.
A downside to increasing computer use is that while the evidence suggests that computers are not causing net job losses, low wage occupations are losing jobs, likely contributing to economic inequality. The policy problem is how to get the workers effected the new skills they need in order to transition to new, well-paying jobs. Developing a workforce with the skills to use new technologies is the real challenge posed by computer automation.