Most near-future fictions are boring.
It’s always dark and always raining, and people are so unhappy.
– Haruki Murakami
Earlier this month I saw a piece on Fox News about McDonald’s testing an automated drive-thru that leverages AI. The story claims the aim is to keep costs down by replacing entry level workers, a means of pushing back against a possible rise in the minimum wage.
It reminded me of a 2013 study by Oxford Martin claiming that 47% of jobs are at high risk of automation.
But is the “employment apocalypse” narrative true? As somebody keenly interested in the nature of exponential technological change, the piece on Fox prompted me to consider a contrarian view.
In a new book, Automation and the Future of Work, Aaron Benanav of the Humboldt University of Berlin makes the case that the story much of society is telling itself about automation (“we’re living in an era of accelerating technological change and jobs are being destroyed at such a pace that the labor market cannot recover”) is misleading.
Benanav claims there is an under discussed plot twist in the story – the macroeconomic context.
The Missing Variable
After each recent recession in the United States, it has taken longer and longer for employment to recover. So called “jobless recoveries” have become the norm. And as unemployment rises, workers’ ability to demand higher wages declines.
Benanav offers the period since 2001 as an example. Output (the volume of manufacturing production) has been going up ~ 1% while productivity (output per worker) has been going up ~ 3% and employment has been declining by ~ 2% (growth in output – growth in productivity = growth in employment). The result of this simple equation might make it seem that productivity growth in the form of machines has been replacing workers.
But if we zoom out and examine the same information with a longer-term historical perspective (back to the 1950s), we’ll notice that over the longer term, the productivity growth rate has been stagnant while the rate of output has been declining steadily. That trend, says Benanav, has been the primary cause of the decline in employment growth. It’s a process called “secular stagnation” – the decades-long tendency for economic growth rates to decline across OECD countries over many business cycles (from about 5% in the early 1960s to about 1.5% in the late 2010s).
In manufacturing specifically, since the late 1960s, deindustrialization has happened in waves across the industrialized world. And the nature of employment has suffered as a result. It’s not the pace of automation that is speeding up, it’s rates of industrial expansion that are slowing down.
For Benanav, slowing growth, not accelerating technological change, is causing the relatively low demand for labor.
So what is the best explanation for this global output-led deindustrialization?
Globalization
Since around 1970 there’s been a rapid increase in international trade, and this theme of globalization might be the most compelling thread in a larger and complicated story. According to many economists, we were supposed to see major gains from trade as each country specialized in different types of production. We were told that everyone would benefit from a shared prosperity (in my memory, all those economics classes in graduate school practically guaranteed it).
Instead, over time, there has been a substantial decrease in the rate of growth of manufacturing across the entire world.
What happened to cause this global slowdown? Generally speaking, it appears that companies have been producing the same sorts of goods, competing in ever more oversupplied markets. As industrial overcapacity has increased, rates of investment have decreased. Global output has been expanding more slowly.
In other words, international trade seems to have generated many more trade redundancies than trade complementarities. Indeed, according to the UN, since the early 2010s, the entire world has been deindustrializing.
Even China and India.
In effect, says Benanav, we’re watching the running down of the industrial growth engine globally. And now we’re in the midst of a shift from an “industrial” to a “service-based” economy.
Onward to the Services Economy
Will the services economy see a rapid automation that replaces human labor that the industrial economy has not? So far it seems the transition to a service-based economy has been associated with a productivity growth slowdown, an economic growth slowdown, and a persistently low demand for labor.
From about 1950 to 1979, the heyday of the industrial economy, productivity growth in core industries was relatively fast, and output growth was even faster, resulting in steady job creation.
But from about 1980 to 2010, the services sector, which was supposed to replace manufacturing as a growth driver, has been growing at a slower rate than the industrial economy did in the earlier period. And productivity growth has been even more of a laggard. The services economy seems to be selecting low productivity jobs.
Another way to think about it is that, so far, we haven’t seen any real productivity growth in services from automation technologies. On the contrary, we’ve seen a significant slowdown in average productivity growth rates across countries like the United States, Japan, and Germany. Overall output growth or GDP has gone down too.
The result is a slow growing economy with a persistently low demand for labor. Of course the nature of this dynamic differs across countries. The United States has low rates of employee protections from dismissal compared to other countries (think France). Workers across the U.S. economy have felt substantially more insecure as a result. And low unionization also means that American workers cannot fight for higher wages as effectively as they can in other places.
In effect, it’s the lack of job creation, not technologically-induced job destruction (the “end of work”), that is at fault for labor market slack. This in turn reduces workers’ bargaining power and frees firms from having to invest in their people as much as they might or should. In many cases, workers end up with little say in how digital work processes are structured.
The Future
Back in October, 2019, economists Jared Bernstein and Dean Baker emphasized the role of the “China shock” (globalization) in America’s lost manufacturing jobs, rather than blaming technology and automation:
From 2000 to 2007 (before the Great Recession), the country lost 3.4 million manufacturing jobs, or 20 percent of the total…. What really happened in those years was that policymakers sat by while millions of U.S. factory workers and their communities were exposed to global competition with no plan for transition or adjustment to the shock, decimating parts of Ohio, Michigan, and Pennsylvania. That was the fault of the policymakers, not the robots.
Baker has also cited evidence that automation has been progressing slowly at best, which is reflected in weak productivity growth. In fact, says Baker, we had much more rapid automation in the period from 1947 to 1973, which actually went along with low rates of unemployment and rapidly rising wages.
As Baker pointed out back in May, 2017, “It’s entirely possible automation will lead to large-scale job loss in the future. It has not in the last decade.”
In an October, 2020, Twitter post, economist Mark Blyth seemed to agree: “There are no self-driving cars, or trucks, Bitcoin is a rich person's gambling asset, there are no blockchain businesses anywhere, the robots who were meant to take your jobs a decade ago are still not here, and yet every year the same crap projections….”
So what now? Will the services economy save us from continuing stagnation?
Benanav has made a compelling case that automation is not responsible for the growing difficulty workers face in finding or holding onto jobs. Rather, the real issue is that economic growth rates have been slowing, generating a persistently low demand for labor. He also points to a wage-productivity gap, and not just in the United States. All across OECD countries there is a growing gap between productivity growth and real average wage growth and real median wage growth, he says.
Simply put, stagnation in wages characterizes the bottom half of the labor market. Today many Americans are both underemployed and underpaid.
And with workers at the mercy of difficult job markets, employers seem less likely to take workers’ perspectives into account when making investment decisions. Insecure work in the digital/platform economy has become more pervasive (think Task Rabbit, Uber, Lyft, etc.). It all helps nudge the digital economy in strange directions.
One question that will need to be asked is whether tomorrow’s economy can provide workers with a sense of meaning and autonomy (ie, self-directed work, mastery of skills, a feeling of purpose).
Or will tomorrow’s jobs end up alienating too many workers?
After all, it seems that a lot of the work currently being done in the digital economy is already alienating in character. The book Uberland (2018) by Alex Rosenblat shows how much Uber drivers’ lives are controlled by algorithms that make their work precarious. Ghost Work (2019) by Mary L. Gray and Siddharth Suri shows how much hidden work there is behind apparently automated processes on platforms such as on Facebook and Instagram, where many people have to look at disturbing images to moderate them because AI still isn’t up to the task.
Benanav goes on to point out that while technology is racing ahead exponentially with faster and faster computers, the way these technologies are actually being implemented depends a lot on what he calls, “social context.” Workers’ autonomy and a feeling of social purpose relates in turn to what kind of solidarity and bargaining power workers will have versus employers. For now, Benanav insists that, “A lot of digital jobs cluster in a place of alienation.”
What to do if we discover the machines aren’t up to the future we’ve planned for them? Then again, when it comes to technological progress, it’s hard to discern exactly what’s going on.
Is it me, or is there a sense of vertigo in the air?
Earlier this month the CEO of Argo AI claimed that self-driving car prototypes are “regularly beating” human drivers, while just this week came news that Japanese conglomerate Softbank Group was pulling the plug on its big plans for Pepper the robot.
Image: from www.neste.us
