By Sheelah Kolhatkar
The encroachment of automation and robotics into the workplace has forced us to rethink the way that certain jobs are done, and it has produced anxiety about whether there will be enough jobs in the future for the human workers who need them. So far, much of the attention has focussed on blue-collar work, as factory assembly lines and warehouses have adopted automated processes more quickly and visibly than other industries. Automation on a factory floor evokes a simple image: robotic arms assembling parts into Tesla cars; mobile robots driving pallets of goods through Amazon distribution centers. In either scenario, the impact on human workers is easy to see. What is harder to visualize is how similar technology might find its way into the aspects of human labor that are invisible and not as easily routinized, such as complex decision-making, strategic planning, and creative thought.
Until recently, the consensus among researchers seemed to be that workers with higher levels of education would be less affected by automation than those lower down on the economic hierarchy. Now, though, new research suggests that higher-educated, white-collar workers may be facing significant disruption, as well. In a study published in November by the Brookings Institution, researchers found that certain higher-wage occupations might be more deeply intertwined with A.I. technology in the future than previously thought. “The big takeaway is that manufacturing professions and occupations will be heavily affected,” one of the study’s co-authors, Mark Muro, told me recently. “But so will white-collar jobs—managerial and office activity.” Muro, a senior fellow at Brookings’ Metropolitan Policy Program, who specializes in economic development and technology and who wears spectacles that lend him a bookish air, noted that whether these jobs are likely to be replaced by A.I. or simply changed by it is still unclear; in some cases, A.I. may end up assisting human workers rather than doing their jobs for them. Either way, the uncertainty is likely to cause alarm in some circles.
One of the challenges of studying the effects of artificial intelligence on white-collar employment is that the integration of algorithms into office work happens slowly, and often imperceptibly. As a starting point, Muro and his co-authors tried to narrow their task by focussing on machine learning, a form of A.I. that uses algorithms to analyze huge amounts of data, find patterns, and then use those patterns to make predictions. “When most people talk about A.I., they’re talking, in many respects, about this,” Muro said. “When they talk about the radiologists reading a scan enhanced by A.I., they’re talking about machine learning.”
After they defined what they were looking for, another challenge presented itself: how to figure out what new kinds of machine-learning applications were likely to be introduced in the coming years. For this, the researchers employed a method developed by Michael Webb, a graduate student in the economics department at Stanford University, who created an algorithm to analyze A.I. patents that had been filed and cross-reference them with tasks performed in various jobs. Webb examined a pool of approximately sixteen thousand patents that contained verb-object pairs such as “diagnose disease” and “predict prognosis,” which correlated with descriptions of occupations used by the Department of Labor. “Patents are a reflection of the things that inventors think are going to be important innovations that will make money in the future,” Webb told me. “The reason you patent something is that you think you could make some money off of this innovation, and you want the right to create that product and not have other people do it instead.”
As a way of testing the effectiveness of this research method, Webb looked back at the previous thirty years or so of patents in software and industrial robotics, to see if the predictions about employment and wage decline one would have found then had panned out. They had: software patents often referred to “recording,” “storing,” and “producing information,” while robot-related patents talked about “cleaning,” “moving,” “welding,” and “assembling.” The words correlated most strongly with the tasks of packers and packagers, hoist and winch operators, machine operators, and those who worked in warehouses—for example, people driving forklifts. “It turns out that the jobs that were highly exposed to those technologies experienced declines in employment and in wages over the next thirty years,” Webb said. That, to him, suggested that software and industrial robots were replacing human labor in those fields (although there were other forces in effect, as well, such as the offshoring of factories).
Webb then analyzed A.I. patent filings and found them using verbs such as “recognize,” “detect,” “control,” “determine,” and “classify,” and nouns like “patterns,” “images,” and “abnormalities.” The jobs that appear to face intrusion by these newer patents are different from the more manual jobs that were affected by industrial robots: intelligent machines may, for example, take on more tasks currently conducted by physicians, such as detecting cancer, making prognoses, and interpreting the results of retinal scans, as well as those of office workers that involve making determinations based on data, such as detecting fraud or investigating insurance claims. People with bachelor’s degrees might be more exposed to the effects of the new technologies than other educational groups, as might those with higher incomes. The findings suggest that nurses, doctors, managers, accountants, financial advisers, computer programmers, and salespeople might see significant shifts in their work. Occupations that require high levels of interpersonal skill seem most insulated. (Notably, jobs at the very top of the earning scale, such as C.E.O., are not shown to be deeply changed.)
When I asked Muro what he thought the implications of his research might be, he told me about a tweet that someone had written in response to the paper’s findings. The author had expressed concern that his work might “muddy the water about the underemployed and vulnerable,” Muro said, by taking attention away from the truck drivers and factory workers who are already being displaced from their jobs, often with less of a financial cushion to fall back on. “There’s no doubt the white-collar part of the story involves some of the most capable, resilient workers in the economy,” Muro said. “For someone in the eightieth percentile of income, they’re probably well trained by, and are being invested in by, their employers.” That kind of investment, Muro added, “doesn’t flow to those at the lower end.” But he noted that the danger of automation coming for white-collar jobs might help make the issue more real for those in charge of policy decisions. “I think maybe this widens the scope of concern,” he said. “This suggests that it isn’t just somebody else’s problem, namely, a black or brown worker in a factory. . . . We’re all going to be contending with tremendous flux and change in the labor market and our work.”
No comments:
Post a Comment