What we get wrong about machines taking our jobs

Conventional wisdom says that if and when the robots arrive, work will disappear. Whether this is good or bad depends on your perspective on fully automated luxury communism, but I digress. My point is that many agree that many jobs will, due to robots, go poof.

But automation is not so simple. Introducing a robot to the factory floor doesn’t automatically send some number of humans to the unemployment line. It can, and it does, but not always. In fact, so-called labor-destroying technologies can actually preserve and create work.

And these counterintuitive scenarios are more common than straight-up job destruction, according to James Bessen, an economist at Boston University.

“One confusion is that automation involves completely replacing workers with machines, and that actually happens very rarely,” Bessen told Emerging Tech Brew.

Turns out, in some situations labor-saving tools can actually lead to more work—sometimes due to human meddling, other times due to the new demands of the technology, and often because of both. Here’s how.

Automation can protect certain jobs

Competitive advantage: a term that anyone who took Econ 101 or played against a really tall person in pickup basketball can understand. It also sums up why automation can sometimes, for select countries and companies, protect or even increasejobs.

Consider the global manufacturing industry. As employment scholar Aaron Benanav pointed out in his recent book Automation and the Future of Work, some of the countries with more robotzed manufacturing industries also employ a relatively high share of their population in the sector.

For example, Benanav writes, in 2016 South Korea had 63 robots per thousand manufacturing workers and 17% of its overall workforce was employed in manufacturing; the US had 19 manufacturing robots per thousand manufacturing workers in the same year, and only 8% of its workforce was in the industry. In other words, South Korea’s manufacturing industry had both a higher robot-human ratio and a higher share of its workforce still employed in manufacturing.

James Bessen said that when labor-saving tech expands work, it’s often “the demand story” at play, a version of which goes something like this: You partially automate the production of a good or service, and it becomes cheaper to produce it at equivalent or higher quality. As a result, that good or service becomes more appealing and accessible to more people, and demand for the product increases.

In some cases, demand increases so much that it results in the expansion of productive capacities—and that can offset any tech-driven job losses, because humans still need to manage certain parts of the process. The famous example here is the bank ATM (it’s so well known that three different sources, including Bessen and Benanav, brought it up to me.)

When ATMs were introduced in 1969, conventional wisdom expected the number of banking locations to shrink, but instead, Benanav said, “It actually made it possible to set up many more of them, it became cost-effective.” There were under 200,000 bank tellers in 1970, but over 400,000 a decade later, according to a paper Bessen wrote. The number of tellers remains high now, at 449,000 as of 2019 per US Bureau of Labor Statistics figures, though it is trending down.

The notion that strategically deployed automation can create jobs is well understood by workers in countries where that’s the case. As Benanav writes, “Unlike workers in the United States, workers in European and East Asian firms believe that automation helps preserve their jobs.”

A secondary force at play here is that countries like Japan, South Korea, Sweden, and Germany have relatively strong worker protections that box in firms’ automation strategies, Benanav told us.

“It's harder to fire workers, and that gives companies an incentive to build out technologies that work with the highly skilled workers that they employ because they can't really get rid of them,” he said. “Those are also the most effective strategies.”

Of course, it doesn’t always work in this direction. Elevator operators are the equally famous example of tech making a job go the way of the dinosaur. Of all the occupations listed in the 1950 US Census, it’s the only one that has disappeared entirely due to automation.

“Automation can increase employment or decrease employment, but it depends on the nature of demand—and demand isn't affected by the technology. Demand is there, it’s only slowly changing over time,” Bessen said.

And nations that underinvest in robotics can instead face a competitive disadvantage. Melonee Wise, CEO of Fetch Robotics—a company that builds autonomous warehouse robots—said that’s what happened to the US manufacturing industry.

“We look at the United States as a desolate manufacturing wasteland, but some of that is because of our resistance to adopt automation,” she said. “And we could have kept many of the jobs that we wanted to keep in the United States if we had been more forward-leaning on automation.”

According to the International Federation of Robotics, the US was ninth in the world in terms of robot density as of 2019, sandwiched between Taipei and Italy, and third in terms of annual robots deployed, with 33,300. China deployed 140,500 the same year, Japan rolled out 49,900, and South Korea—which has a fraction of the US’s GDP and population—deployed nearly as many at 27,900. The US is still second in terms of global manufacturing share, though, accounting for 16.6% of global output as of 2018, down from 29% in the early 1980s.

“At this point, the US is probably underinvested in robotics,” David Autor, a professor of economics at MIT, said. “We are great at innovating them, but in terms of actually building and deploying them at scale, not so much.”

Automation can create brand new jobs

The most intuitive way that automation creates work is by giving way to new jobs that previously didn’t exist.

More than 60% of jobs performed in 2018 had not yet been invented in 1940, according to a 2020 MIT report that Autor co-authored. Tech-focused roles, like computer engineers, are well-represented among jobs created over this period, but as incomes rose new personal services (e.g., fitness coaching, mental health counseling) emerged as well.

So what jobs of the future are being created today? For one, there is plenty of work directly related to building and maintaining the tools for automation.

Peter Brodsky, CEO of automation company Hyperscience, told us that the biggest challenge his company faces is an inability to hire engineers to build its automation software.

“We have a shortage of qualified people, and so it seems like there are more and more jobs being created that we can’t even fill,” he said.

To some extent, Brodsky’s problem is felt sector-wide. A March 2021 O’Reilly survey of enterprise AI adoption found that “the most significant barrier to AI adoption is the lack of skilled people and the difficulty of hiring.” And in 2019 software company Element AI estimated that there were around 144,000 AI-related job openings, but only around 26,000 developers and specialists seeking work.

And on the robotics side, Wise said an entire industry has already cropped up around third-party remote robot support, “where you can log in to the robot remotely and operate it.” Plus One, founded in 2016, is an example of a company that makes software to enable remote robot operations like this. Wise said Fetch uses a third party called Ricoh to remotely handle routine support issues, like navigating a robot to a functioning charging station.

Down the road, she expects to see more and more developers who build apps for robots just like developers currently do for smartphones.

Matthew Hopkins, cofounder of UK-based supply chain consultancy BoxLogic, is himself an example of someone with a new breed of job created by cutting-edge automation. He told us that advising on the adoption of warehouse robotics is a “core service” his three-year-old firm offers to businesses.

You may notice a pattern across all of the jobs mentioned above: they’re all in “knowledge work,” meaning they likely require a college degree or more. That’s because, Autor told us, “New work associated with innovation, at present, is highly skilled. It’s not all just managers, it’s technicians, it’s executives, it’s people who do ideas.”

As of 2018, only about 37% of US adults ages 25–64 had a four-year college degree or higher, according to the US Census Bureau, meaning around 63% did not have a four-year degree. Around 27% had an associates degree or some college courses completed. Just 13% have earned a Master’s or above.

For the 63% without a four-year degree, automation isn’t erasing all job opportunities, but it’s not creating much valuable work either. Autor said there will be plenty of service-oriented jobs available to these workers in the future—for example in healthcare, food service, and the gig economy—but that these roles use generic skill sets, and don’t pay very well as a result. In his book, Benanav calls this phenomenon “mass under-employment.”

“It's not so much that automation will create jobs in food service and cleaning as [much as it is] that these are the residual job categories that are growing as incomes rise and workers are displaced from traditional middle-skilled work such as production, clerical, admin, and sales,” Autor said. “The issue is job bifurcation/polarization. So, we are not running out of jobs, but we are not creating many good jobs for the majority of workers who do not have a BA.”

Hopkins provided an example of this on a micro scale, saying that when a company adopts robotics tech, it generally needs to hire experienced managers who have overseen robotics systems in the past—execs from Amazon, or those who have implemented warehouse robotics at a major retailer like ASOS, for example.

When that happens, companies “pay for that with some of the reduction in people below it.” While the costs vary quite a bit project-to-project, Hopkins said that in his experience, “A decent-sized project would be 10 to 20 million pounds [$13.8 million-$27.6 million],” while a “super project” would exceed 50 million pounds ($68.9 million). There are opportunities for some of the workers laid off in this process to move into jobs in maintenance or in admin, but it’s “not in the same numbers as before and it’s probably not, when you sum it all up, in the same sort of money.”

Automation can make some jobs harder

So, automation can both preserve existing jobs and create brand new ones. But in some scenarios it can also lead to higher output expectations, making life harder for those who work alongside labor-saving tools.

In 1983, technology historian Ruth Schwartz Cowan wrote More Work For Mother, whichtraces a version of this disappointing reality: Technology marketed as reducing housework for women does not always do so.

That’s because, as Cowan puts it, “Modern labor-saving devices eliminated drudgery, not labor.”

Domestic tools make individual tasks easier—it requires less effort to do a load of laundry now than in the 19th century—but they do nothing to eliminate the tyranny of inflating expectations. Sure, that load of laundry is more manageable, but even the least sartorially inclined among us has more clothing than our 19th century counterparts.

Another example from Cowan’s book: When society moved from the open hearth to the stove, it became feasible for homemakers to boil, simmer, and bake all at once. In contrast, the hearth only really accommodated one thing at a time.

That transition transformed expectations of what women would cook for their families. Where a simple one-pot meal used to hit the spot, now a more complex meal was expected. Stoves also required more cleaning and upkeep (e.g., polishing) than a fireplace—tasks that also fell to women at the time.

The net result? The stove “probably increased the amount of time women spent in preparing foodstuffs for cooking,” writes Cowan. Of course, it also led to more varied and likely healthier diets. And it did save labor for men, as “The activity that was radically altered by the presence of a stove was fuel gathering,” a stereotypically male role. But it didn’t eliminate labor for women, as the marketing materials had promised.

It’s Bessen’s demand story, once again: Technology makes it easier to achieve or surpass the same level of quality around the home—and as a result, demand for chores skyrockets. The individual task of cooking or doing laundry becomes easier, but suddenly the palates are more discerning and the piles of dirty clothes are mounting.

The rising expectations that can accompany the adoption of supposedly labor-saving tech are not just ironic and frustrating, but sometimes dangerous too.

A 2020 report from The Center for Investigative Reporting's Reveal publication found that, among the most common type of Amazon warehouse, those with robots had a 50% higher serious injury rate than those without robots.

Amazon didn’t return our request for comment, though a spokesperson previously told the BBC that the high rates of injury were the result of encouraging workers to report even minor incidents.

While that could explain why Amazon has a higher injury rate than the industry average—which it does—it fails to explain why their robotized warehouses see around 9 injuries per 100 workers, while their regular warehouses see around 6 per 100 workers.

Reveal’s reporting offered a clear explanation for this disparity—one that has nothing to do with the technology and everything to do with the managerial choices made alongside it. Much like the husbands looming behind the scenes in Cowan’s book, Amazon bosses paired shiny new tools with sky-high expectations.

Reveal reported that workers at robotic fulfillment centers were expected to grab and scan up to 400 items per hour, up from just 100 items per hour pre-robots.

Kathleen Fagan, a physician who inspected Amazon in her capacity as a medical officer for the federal Occupational Safety and Health Administration (OSHA), put the issue simply to Reveal, “If you’ve got robots that are moving products faster and workers have to then lift or move those products faster, there’ll be increased injuries.”

Hopkins told us that robotics manufacturers typically provide a “throughput” rate in their contracts with clients. This is a guarantee of how much their robots can do in a given amount of time.

But Hopkins said this is usually just a measure of what’spossible for the machine—not necessarily for humans. As a rule of thumb, his firm advises clients to set expectations at 85% of this throughput figure to account for breaks, slowdowns, and “the realities of being human.” He said the 85% rule is common practice among manufacturers as well, “It's setting those expectations that, you know, whilst it says it can do this, that's not sustainable,” he said.

For its part, Fetch avoids throughput guarantees in the first place because the rate “is tied to the performance of the people, which cannot be dictated by Fetch,” Wise told us.

The robots are not in control

We could very well see an AI- and robotics-driven wave of universal job destruction one day, but that day is not today, tomorrow, or even the next. The fields have advanced rapidly, but their applications are still narrow and nowhere approaching the scope of all human capabilities.

While tech advancements matter quite a bit in the near term, human choices like policymaking will play just as large of a role in shaping labor markets. Decisions made in the next decade about how to structure society—e.g., how much (and how well) we invest in retraining, or whether and how we reconfigure employment protections like minimum wage and unemployment insurance—will ultimately set the rules for how automation unfolds and who it affects.

“Tech advances matter greatly. Human policymaking choices also matter greatly. It’s hard to rank them,” Autor said. “The key point is that we collectively have agency, both about what tech advances are made and how they play out across workers and jobs.”