For nearly one hundred years, science fiction has warned of robotic takeovers and a dystopian societal collapse as pseudo-intelligent, autonomous devices discover that they have little need for humans. Videos from BostonDynamics seem designed to stoke such foreboding, at least for promotional effect. Although the prospect of rogue robots destroying their creators is remote for now, a burgeoning army of industrial robots is displacing workers in a growing number of occupations.
The extent of job losses and their long-term implications on the labor market are topics of hot debate. A recent Brookings Institute study, for example, estimates that several industries are likely to see upwards of 70 percent of their jobs automated away, concluding that
Almost no occupation will be unaffected by the technological change in the AI era.
Well yes. There are effects and there are effects.
While the Brookings report is notable for its occupational specificity, it’s hardly unique in predicting significant disruption to labor markets as the capabilities of industrial robots, autonomous vehicles and other artificially capable machines improve and their use consequently explodes over the coming years.
Fears of mechanically-induced societal chaos are nothing new, and the debate over the future of work has both modern-day Luddites and unrepentant optimists braying into the media wind. However, a review of the economic literature paints a nuanced picture of both benefits and problems. As Brookings writes,
The discourse appears to be arriving at a more complicated understanding, suggesting that automation will bring neither apocalypse nor utopia, but instead both benefits and stress alike. Such is the ambiguous and sometimes disembodied nature of the ‘future of work’ discussion.
Regardless of one’s inclinations, a cursory look at the various policy prescriptions, whether from pundits or academics shows that there are no quick, easy or painless fixes to the robot army that risks creating structural unemployment and an accompanying Marxian reserve army of labor with the potential for societal destabilization.
The Brookings report focuses identifying the specific job occupations most likely to see robot-induced job losses, which it categorizes into those with high (70 percent or greater of jobs susceptible), medium (30-70 percent) or low (less than 30 percent) risk. Overall, it estimates that a quarter of U.S. jobs, or 36 million as of 2016, are greatly exposed to automation-based elimination “in the coming decades” (emphasis added), while 39 percent of positions are not well suited to robotic displacement, although virtually all will see some changes due to the technology.
Brookings goes on to state the obvious by noting that
'Routine’, predictable physical and cognitive tasks will be the most vulnerable to automation in the coming years,
which includes jobs in production, transportation and food preparation along with all types of clerical work.
Jobs with the least exposure to automation-induced job loss are a study in contrasts that includes high-end technical and “creative” positions that require advanced training or human insight and low-paying service work doing personal care, maid and janitorial work or anything requiring what Brookings calls “the need for interpersonal social and emotional intelligence.”
A chart accompanying the report shows an almost perfect inverse correlation between wages and a job’s exposure to automation, i.e. those at the bottom of the wage scale are most at risk. It’s noteworthy, however, that even those at the 90th percentile of income face some threats from automation, however, the report doesn't specify whether the risk is to job loss or merely the AI-enhancement of existing tasks.
Brookings details differences in the risks of automation-fueled job loss by state, metropolitan area and demographics noting that,
Overall, smaller, more rural communities seem significantly more exposed to the automation of current-task content than larger ones. This relationship holds when comparing metropolitan to rural areas as well as among metros of different sizes.
While demographically the automation risk is higher among men, youth and the less educated, the latter of which implies significant racial and ethnic disparities.
Technologies like robotics and various AI methods that are undergoing rapid, nonlinear rates of improvement are inherently unpredictable, which explains the variance in estimates regarding the share of jobs that are vulnerable to displacement. For example, a 2013 paper by Frey and Osborne of Oxford estimated that 47 percent of U.S. employment is at high risk of being replaced by automation. In contrast, a 2018 OECD paper concluded that only 14 percent of jobs in OECD countries are highly automatable (70 percent probability), although it projects that another 32 percent are at medium risk (between a 50 and 70 percent), which the authors say indicates a
While the Brookings, OECD and Oxford papers look at the risk of automation-fueled job losses and the disparity among various demographic groups and geographies, other researchers focus on the implications of automation for business and society. For example, a highly technical IMF working paper works through various economic models that point to much societal disruption ahead, writing (emphasis added),
We analyze a range of variants that reflect widely different views of how automation may transform the labor market. Our main results are surprisingly robust: automation is good for growth and bad for equality; in the benchmark model real wages fall in the short run and eventually rise, but 'eventually' can easily take generations.
The authors conclude that (emphasis added),
Another paper from the Center for Global Development contrasts the proliferation of automation in developed and developing countries by borrowing from both Marxian and neoclassical theories about economic growth and industrialization. Of particular relevance is the concept of a reserve army of labor, namely a surplus of unskilled or low-skilled workers resulting from the substitution of capital (robots and AI software) for labor.
The CGD paper is edifying for its ample use of historical references alone, but of more relevance here it offers insights on the differences between developed countries with a higher concentration of so-called automation-resistant sectors (APS) and the developing world more reliant on agriculture and other so-called automation-prone sectors (APS). Classical economic theory says that as jobs are automated out of existence, workers migrate to new positions in the APS. The authors note a problem that seems likely given such rapid technological improvements in robotics and software.
Given the abundance of cheap labor in developing countries, the dilemma could result in further bifurcation of the labor markets in the developed and developing worlds as the former becomes more technological and the latter more manual. Since labor is already cheap in developing countries relative to high-income countries, the paper posits that there could be two opposing forces at work (emphasis added),
(i) labor is cheaper than in high-income countries, thus more competitive vis-à-vis machines, and there is thus less of an incentive to automate; (ii) conversely, given widespread low-skilled manual routine work, work tasks that are prevalent in developing countries are easier to automate from a technological viewpoint. In other words, the APS will likely be larger in developing countries.
Consequently, in the developing world, automation is simultaneous more technologically feasible, but less economically so, a dichotomy the paper suggests could lead to deindustrialization and agriculturalization in the developing world.
The CGD paper also questions the wisdom of commonly suggested mitigation strategies for robotic-induced labor market disruption in more advanced countries by training displaced workers for jobs in the ARS. It notes (emphasis added),
A widespread policy recommendation is to invest in skills and thus move the labor force away from automatable routine tasks. The problem with this approach is that (i) it is not clear what skills will be automation-resistant for a sufficient time to make the skills investment worthwhile and (ii) whether upskilling is at all realistic given the required time and monetary investment.
We see echoes of this quandary in the current “learn to code” meme on Twitter that has critics suggesting the hundreds of recently fired journalists can solve their employment problem by just learning to program. While done in mockery, the suggestion exposes two flaws with many re-training proposals: (i) the significant amount of time it takes to develop proficiency, i.e. time without a well-paying job, and (ii) the fact that continually moving thousands of people into a field risks an oversupply of labor and hence lower wages.
Labor markets have undergone technology-driven disruption for centuries, starting with the introduction of the Gutenberg press. But the impending changes promises to be different in scale, if not in kind.
The numbers of displaced workers, spanning dozens of occupations and industries, undermines the validity of bromides about society having always successfully adapted to technology-induced labor market disruptions in the past. Not only does retraining 30 to 50 percent of the labor force seem overly ambitious, as the CPG paper points out, with robotic and software technology advancing so rapidly and unpredictably, but the results of many such retraining programs could also be obsolete, or at least much less valuable by the time workers complete them.
The IMF paper aptly sums up the paradoxical dichotomy of robotics: its great for the economy writ large, particularly investors and those with specialized, automation-resistance skills, but terrible for the much larger group of low- and moderately-skilled workers.
Handling this bifurcated labor market will represent a global challenge over the coming decades, but one that will require flexible approaches that can adapt to rapidly changing technologies and business practices. Prescriptive edicts that lock countries and workers into dead-end or failing strategies won't cut it.
And, as a touch of irony, the areas that should see boom times are in economic modeling and labor policy development. Of course.