But which workers, exactly, are most likely to suffer lost jobs or reduced income when new technologies arrive?
Bryan Seegmiller, an assistant professor of finance at Kellogg, along with Kellogg finance professor Dimitris Papanikolaou and their colleagues, sought to better understand which types of workers were historically vulnerable to being rendered obsolete by technology, and how career disruptions caused by technology affected their future earnings. They developed a novel way to measure workers’ exposure to emerging technology by identifying similarities between the tasks associated with different occupations and the descriptions in new patents. That allowed them to track how breakthrough technologies impacted the exposure of workers in relevant occupations over time.
As one might expect, they found that manual laborers had the highest exposure to emerging technologies, especially from 1850 to 1970. But other patterns were more surprising. In the 1970s, occupations in which people performed routine “cognitive” tasks, such as clerks, technicians, and programmers, also began to face much larger exposures to technology. And when new inventions showed up, workers who earned the highest salaries within the affected occupations—that is, those with the most advanced skills—saw the biggest slowdowns in their wages.
“The more-skilled workers have the most to lose,” Seegmiller says. They tend to “get hit the hardest in terms of their income.”
Winners and Losers
In general, technology improves productivity and standards of living. But gains and losses aren’t distributed equally. Each advance might help everyone on average, “but there might be a very particular subset of people that just get absolutely hammered by it,” Seegmiller says.
To better understand which workers have been affected by technological advances historically, Seegmiller and Papanikolaou, along with Leonid Kogan and Lawrence Schmidt at the MIT Sloan School of Management, devised a new way to measure how people’s exposure to technology—that is, their risk of being displaced by new inventions—changed over time.
The researchers gathered descriptions of tasks performed in more than 13,000 types of jobs from the Dictionary of Occupation Titles database. Then they developed an algorithm using tools from natural language processing to compare the task descriptions with the text of patents from 1840 to 2010, focusing on breakthrough advances. Based on text similarities, the team could identify patents that were highly related to job tasks associated with specific occupations.
For instance, the algorithm matched a 19th-century patent for a knitting machine to occupations such as textile workers and sewers. A patent for a system to manage financial accounts was matched to financial managers, credit analysts, accountants, bookkeeping clerks, and so on.
A College Degree Won’t Help
The team then examined four broad categories of jobs.
One category was manual occupations, such as electricians and machine operators. Another was interpersonal jobs that required social perceptiveness, or the ability to understand and communicate with other people; these included teachers and psychologists. Routine cognitive jobs involved repeatedly performing tasks that usually followed a set list of instructions—for instance, clerks and technicians. And nonroutine cognitive occupations required skills such as creative thinking, analyzing information, or guiding team members; engineers, surgeons, and managers fell into this category.
As one might expect, manual physical jobs were the most exposed to technological change. But cognitive occupations weren’t immune from risk. Routine cognitive jobs, in particular, started becoming much more exposed starting around the 1970s, as information technology began to take off.
One example was order clerks, whose tasks included taking customers’ orders over the phone, coordinating shipments, and checking order details. In the late 1990s, their exposure to technology rose dramatically. Around this time, many patents were filed for related software, such as a computerized order entry system.
The exposure of workers with a college degree also increased over recent decades. By the early 2000s, it was nearly on par with that of workers without a college degree. “Technologies are creeping into areas they haven’t before,” Seegmiller says. For example, the exposures of various engineering occupations increased in the 1990s due to the introduction of new software and other information technologies that changed required skills and even automated some of the tasks performed by these occupations.“
And this increased exposure presented a tangible risk for all categories of workers. Based on U.S. Census surveys from 1910 to 2010, the team found that an increase in technology exposure was linked to a decline in employment. And wage data starting in the 1980s suggested that more exposure led to lower income. For instance, order clerks’ wages fell by 20 percent relative to other clerk occupations from 1997 to 2010, a time period that saw the rise of e-commerce, which fundamentally changed the occupation.
The team then drilled down deeper to see if there were any differences in the harms experienced by different types of workers within a given level of occupational exposure.
For instance, the researchers compared 45- to 55-year-old workers with 25- to 35-year-old workers. When faced with the same amount of technology exposure, in the same type of job, the older workers’ wages grew 1.8 times more slowly over a five-year period. This may have been partly due to younger workers having invested less time in now-obsolete skills and having more time left in the labor force to pick up new ones.
Again, college-educated workers didn’t fare much better than high-school graduates. For both types of employees, the income slowdown in response to technological advances was similar. “Just having a college degree does not necessarily insulate you,” Seegmiller says.
One of the most striking findings emerged when the team looked at workers who had reached the top income tier within an exposed profession—for example, clerks or machine operators who earned relatively high salaries compared with their peers. These employees saw their wages slow down by more than twice as much as average workers in the same occupation with the same level of technology exposure. “For the people that are really skilled, they have a lot of room to fall,” he says.
This pattern was even stronger among highly paid workers in occupations that required a long track record of specific types of experience, such as skilled trades like tool makers, machinists, and electrical-equipment repairers. For those employees, “you’re really deep into your investment in these particular skills,” he says.
These trends in wages suggested that something more nuanced than automation was going on. In the automation scenario, “technology shows up, and a robot does what you used to do,” Seegmiller says. But a second type of displacement was possible too: rather than directly replacing workers, technology might change the way their jobs were done and require people to pick up new skills.
For instance, a clerk who was highly competent at using a certain record-keeping system might need to learn new software, or an experienced machine operator might be faced with unfamiliar equipment. People who had invested a lot of time and effort into mastering now-obsolete methods could be laid off; or if they stayed at their jobs, their wages could stagnate or decline.
“If something new shows up, and you are really good at the old way of doing things, that can be just as hard for you as a robot coming in to replace workers on the assembly line,” he says.
Learning for Life
The researchers did identify a couple of bright spots. Jobs in the interpersonal category had consistently low exposure to technological change. “One thing that technology can’t do, that it has never been able to replicate, is human-to-human interaction,” Seegmiller says.
And workers who specialized intensively in those interpersonal skills fared better. Even when their technology exposure did go up, their income didn’t slow down as much as it did in other types of occupations.
Technology also wasn’t a uniformly negative force. The team conducted a separate analysis to identify patents in various industries that did not overlap with occupational tasks. Exposure to those advances was actually linked to an increase in workers’ incomes, likely because the inventions had helped them become more productive.
“Not all technology is bad for workers,” Seegmiller says. “But technology hurts particular people.”
So what should workers do to protect themselves from tomorrow’s technologies?
In addition to cultivating interpersonal skills, “being willing to constantly learn and adapt is really important,” he says. Many free or inexpensive online courses can help workers pick up new skills. Policymakers could also develop programs to subsidize training for employees who might soon be displaced.
Additionally, the risk of future technological exposure shouldn’t necessarily discourage people from pursuing an occupation that is valued today. For instance, one emerging concern—which was not addressed in this study—is that AI will take over complex tasks such as data analysis. This might mean that data analysts will see slower wage growth in the future, but they’ll still be paid relatively high salaries compared with many other professions that are more insulated from technology. And if those analysts enjoy their work, the rewards of having a satisfying job could be worth the income risk.
“Thinking that ‘AI is going to take over everything, and therefore I should avoid investing in the technical skills and instead become, say, a baker’—that’s just overly pessimistic,” Seegmiller says.