David Rotman reports in MIT Technology Review:
The current disruptions are faster and more intensive. It is nothing like what we have seen in the past, and the issue is whether the system can adapt as it did in the past. If AI is going to achieve its full economic potential, we’ll need to pay as much attention to the social and employment challenges as we do to the technical ones.
Last October, Uber had one of its self-driving trucks make a beer run, traveling 200 kilometers down the interstate to deliver a cargo of Budweiser from Fort Collins to Colorado Springs. A person rode in the truck but spent most of the trip in the sleeper berth, monitoring the automated system. (The test came just a few weeks after Uber had announced its driverless car service in Pittsburgh.) The self-driving truck developed by Uber’s recently acquired Otto unit reflects remarkable technological achievements. It also provides yet another indicator of a looming shift in the economy that could have deep political consequences.
It is uncertain how long it will take for driverless trucks and cars to take over the roads. For now, any so-called autonomous vehicle will require a driver, albeit one who is often passive. But the potential loss of millions of jobs is Exhibit A in a report issued by the outgoing U.S. administration in late December. Written by President Obama’s top economic and science advisors, “Artificial Intelligence, Automation, and the Economy” is a clear-eyed look at how fast-developing AI and automation technologies are affecting jobs, and it offers a litany of suggestions for how to deal with the upheaval.
It estimates that automated vehicles could threaten or alter 2.2 million to 3.1 million existing U.S. jobs. That includes the 1.7 million jobs driving tractor-trailers, the heavy rigs that dominate the highways. Long-haul drivers, it says, “currently enjoy a wage premium over others in the labor market with the same level of educational attainment.” In other words, if truck drivers lose their jobs, they’ll be particularly screwed.
It is hard to read the White House report without thinking about the presidential election that happened six weeks before it was published. The election was decided by a few Midwest states in the heart of what has long been called the Rust Belt. And the key issue for many voters there was the economy—or, more precisely, the shortage of relatively well-paying jobs. In the rhetoric of the campaign, much of the blame for lost jobs went to globalization and the movement of manufacturing facilities overseas. “Make America great again” was, in some ways, a lament for the days when steel and other products were made domestically by a thriving middle class.
But many economists argue that automation bears much more blame than globalization for the decline of jobs in the region’s manufacturing sector and the gutting of its middle class. Indeed, in his farewell speech to thousands in a packed convention hall in Chicago, President Obama warned: “The next wave of economic dislocations won’t come from overseas. It will come from the relentless pace of automation that makes a lot of good middle-class jobs obsolete.”
The White House report points in particular to the current wave of AI, which it describes as having begun around 2010. That’s when advances in machine learning and the increasing availability of big data and enhanced computation power began providing computers with unprecedented capabilities such as the ability to accurately recognize images. The report says greater deployment of AI and automation could boost economic growth by creating new types of jobs and improving efficiency in many businesses. But it also points to the negative effects: job destruction and related increases in income inequality. For now at least, “less educated workers are more likely to be replaced by automation than highly educated ones.” The report notes that so far automation has displaced few higher-skill workers, but it adds: “The skills in which humans have maintained a comparative advantage are likely to erode over time as AI and new technologies become more sophisticated.” Labor economists have been pointing out the employment consequences of new digital technologies for several years, and the White House report dutifully lays out many of those findings. As it notes, the imminent problem is not that robots will hasten the day when there is no need for human workers. That end-of-work scenario remains speculative, and the report pays it little heed. Instead, it is far more concerned with the transition in our economy that is already under way: the types of jobs available are rapidly changing. That’s why the report is so timely. It is an attempt to elevate into Washington political circles the discussion of how automation and, increasingly, AI are affecting employment, and why it’s time to finally adopt educational and labor policies to address the plight of workers either displaced by technology or ill suited for the new opportunities.
It is “glaringly obvious,” says Daron Acemoglu, an economist at MIT, that political leaders are “totally unprepared” to deal with how automation is changing employment. Automation has been displacing workers from a variety of occupations, including ones in manufacturing. And now, he says, AI and the quickening deployment of robots in various industries, including auto manufacturing, metal products, pharmaceuticals, food service, and warehouses, could exacerbate the effects. “We haven’t even begun the debate,” he warns. “We’ve just been papering over the issues.” It is often argued that technological progress always leads to massive shifts in employment but that at the end of the day the economy grows as new jobs are created. However, that’s a far too facile way of looking at the impact of AI and automation on jobs today. Joel Mokyr, a leading economic historian at Northwestern University, has spent his career studying how people and societies have experienced the radical transitions spurred by advances in technology, such as the Industrial Revolution that began in the late 18th century. The current disruptions are faster and “more intensive,” Mokyr says. “It is nothing like what we have seen in the past, and the issue is whether the system can adapt as it did in the past.” Mokyr describes himself as “less pessimistic” than others about whether AI will create plenty of jobs and opportunities to make up for the ones that are lost. And even if it does not, the alternative—technological stagnation—is far worse. But that still leaves a troubling quandary: how to help the workers left behind. “There is no question that in the modern capitalist system your occupation is your identity,” he says. And the pain and humiliation felt by those whose jobs have been replaced by automation is “clearly a major issue,” he adds. “I don’t see an easy way of solving it. It’s an inevitable consequence of technological progress.”
The problem is that the United States has been particularly bad over the last few decades at helping people who’ve lost out during periods of technological change. Their social, educational, and financial problems have been largely ignored, at least by the federal government. According to the White House report, the U.S. spends around 0.1 percent of its GDP on programs designed to help people deal with changes in the workplace—far less than other developed economies. And this funding has declined over the last 30 years.
The picture is actually even worse than those numbers alone suggest, says Mark Muro, a senior fellow at the Brookings Institution. Existing federal “readjustment programs,” he says, include a collection of small initiatives—some dating back to the 1960s—addressing everything from military-base closings to the needs of Appalachian coal-mining communities. But none are specifically designed to help people whose jobs have disappeared because of automation. Not only is the overall funding limited, he says, but the help is too piecemeal to take on a broad labor-force disruption like automation. The disparity between the rich and everyone else is larger than ever in the United States and increasing in much of Europe. Why?
Some observers, spearheaded by a clique of Silicon Valley insiders, have begun arguing for a universal basic income as a way to help those unable to find work. Wisely, the White House report rejects such a solution as “giving up on the possibility of workers’ remaining employed.” As an alternative, Muro proposes what he calls a “universal basic adjustment benefit.” Unlike the universal basic income, it would consist of targeted benefits for those seeking new job opportunities. It would provide such support as wage insurance, job counseling, relocation subsidies, and other financial and career help.
Such generous benefits are unlikely to be offered anytime soon, acknowledges Muro, who has worked with manufacturing communities in the Midwest (see “Manufacturing Jobs Aren't Coming Back”). However, the presidential election, he suggests, was a wake-up call for many people. In some ways the result was “secretly about automation,” he says. “There is a great sense of anxiety and frustration out there.”
The question, then, is whether the looming onslaught of AI will make existing tensions even worse.
No one actually knows how AI and advanced automation will affect future job opportunities. Predictions about what types of jobs will be replaced and how fast vary widely. One commonly cited study from 2013 estimated that roughly 47 percent of U.S. jobs could be lost over the next decade or two because they involve work that is easily automated. Other reports—noting that jobs often involve multiple tasks, some of which might be easily automated while others are not—have come up with a smaller percentage of occupations that machines could make obsolete. A recent study by the Organization for Economic Cooperation and Development estimates that around 9 percent of U.S. jobs are at high risk. But the other part of the employment equation—how many jobs will be created—is essentially unknowable. In 1980, who could have predicted this decade’s market for app developers?
In the past, new technologies have greatly expanded overall employment opportunities. But no particular economic rule dictates that this will always be true. And some economists warn that we must not be overly sanguine about the consequences of automation and AI.
“AI is very much in its infancy,” says MIT’s Acemoglu. “We don’t really know what it can do. It’s too soon to know its impact on jobs.” A key part of the answer, he says, will be to what extent the technologies are used to replace humans or, alternatively, to help them carry out their jobs and expand their capabilities. Personal computers, the Internet, and other technologies of the last several decades did replace some bank tellers, cashiers, and others whose jobs involved routine tasks. But mainly these technologies complemented people’s abilities and let them do more at work, says Acemoglu. Will that pattern continue? “With robots, and down the line with artificial intelligence, the replacement part might be far stronger,” he cautions.
Not only might automation and AI prove particularly prone to replacing human workers, but the effects might not be offset by the government policies that have softened the blow of such transitions in the past. Initiatives like improved retraining for workers who have lost their jobs to automation, and increased financial protections for those seeking new careers, are steps recommended by the White House report. But there appears to be no political appetite for such programs. Schemes for giving everyone a guaranteed income are gaining momentum in Silicon Valley and throughout Western Europe. It’s a great idea, until you look closely.
“I’m very worried that the next wave [of AI and automation] will hit and we won’t have the supports in place,” says Lawrence Katz, an economist at Harvard. Katz has published research showing that large investments in secondary education in the early 1900s helped the nation make the shift from an agriculture-based economy to a manufacturing one. And now, he says, we could use our education system much more effectively. For example, some areas of the United States have successfully connected training programs at community colleges to local companies and their needs, he says, but other regions have not, and the federal government has done little in this realm. As a result, he says, “large areas have been left behind.” One problem the growing adoption of AI could make much worse is income inequality (see “Technology and Inequality”) and the sharp divisions between the geographic areas that benefit and those that don’t. We don’t need the expert-written White House report to tell us that the impact of digital technologies and automation in large swaths of the Midwest is very different from the effects in Silicon Valley. A post-election analysis showed that one of the strongest predictors of voting behavior was not a county’s unemployment rate or whether it was wealthy or poor but its share of jobs that are “routine”—economists’ shorthand for ones that are easily automated. Areas with a high percentage of routine jobs overwhelmingly went for Donald Trump and his message of turning back the clock to “make American great again.”We’re in the midst of a jobs crisis, and rapid advances in AI and other technologies may be one culprit. How can we get better at sharing the wealth that technology creates?
The economic anxiety over AI and automation is real and shouldn’t be dismissed. But there is no reversing technological progress. We will need the economic boost from these technologies to improve the lackluster productivity growth that is threatening many people’s financial prospects. Furthermore, the progress AI promises in medicine and other areas could greatly improve how we live. Yet if we fail to use the technology in a way that benefits as many people as possible (see “Who Will Own the Robots?”), we risk fueling public resentment of automation and its creators. The danger is not so much a direct political backlash—though the history of the Luddites suggests it could happen—but, rather, a failure to embrace and invest in the technology’s abundant possibilities.
Despite the excitement around AI, it is still in its early days. Driverless vehicles are fine on sunny days but struggle in the fog or the snow, and they still can’t be trusted in emergency situations. AI systems can spot complex patterns in massive data sets but still lack the common sense of a child or the innate language skills of a two-year-old. There are still very difficult technical challenges ahead. But if AI is going to achieve its full economic potential, we’ll need to pay as much attention to the social and employment challenges as we do to the technical ones.