The rise and spread of intelligent machines has led to increasing income inequality and anemic job growth. And this dynamic is likely to be permanent. Such is the arresting and depressing thesis proposed by the George Mason University economist Tyler Cowen in his provocative new book, Average Is Over.
The American economy is becoming a "hyper-meritocracy" in which workers will either be big earners or big losers, Cowen believes. He blames this bifurcation on the rise of "genius machines," which are increasingly doing the routine intellectual work that once supported millions of middle-income workers. If your skills enhance the work of ever-more-intelligent machines, you'll likely be a big earner. If your skills do not complement the computer, you're liable to be a big loser. "Ever more people are starting to fall on one side of the divide or the other," writes Cowen. "That's why average is over."
Those middle-class jobs just aren't coming back, Cowen claims. "The financial crash was a very bad one-time event that revealed, rather suddenly, this more fundamental long-term structural problem, namely that a lot of workers had been overemployed relative to their skills," he writes.
Trend data on American household earnings since the 1970s do show that incomes in the top 20 percent rose steeply while those in the bottom 80 percent experienced more modest increases. Writing in the online financial newsletter Advisor Perspectives, economic analyst Doug Short reported in September 2013 that between 1967 and 2012, the incomes of the top 5 percent increased 88 percent in real dollars. Incomes in the top quintile rose 70 percent, the second quintile went up 38 percent, the middle increased 20 percent, the fourth gained 12 percent, and the bottom quintile grew by 20 percent.
While the average real incomes of Americans in all categories have fallen since the Great Recession, those of middle and lower quintiles remain still lower than they were in 2000. As Short reports, the average income of the middle quintile fell 9.1 percent (in constant dollars) between 2000 and 2012; the fourth quintile fell 12.2 percent; and the bottom quintile dropped 15.9 percent. The economics consultancy Sentier Research reported in July 2013 that real median household income is 7.2 percent lower than it was in 2000. Even more tellingly, the real median household income at its height, just before the 2008 recession, was only 0.7 percent higher than it was in 2000.
Cowen interprets all this stagnation and decline as a signal that a "lot of jobs aren't worth as much as before, and they are not being replaced by a comparable number of high-earning slots." And he is particularly alarmed about the trends in men's unemployment and incomes.
Citing data from the Brookings Institution's Hamilton Project, Cowen writes that between 1969 and 2009, "wages for the typical or median male worker have fallen by about 28 percent." This stark claim has been challenged, and Cowen notes that other analysts have argued that men's average wages have fallen by only 9 percent since 1969. That nine percent figure was derived by the Brookings Institution economist Scott Winship. However, Winship has recently re-crunched the wage numbers using a different and more accurate income deflator, and he now finds that there was no decline in men's average wages between 1969 and the peak year of 2007, though since the recession average male wages have dropped 12 percent. While not as terrible as the Hamilton Project numbers, it's still pretty bad if the average wage hasn't increased over four decades.
Cowen further notes that 60 percent of the jobs that disappeared during the recent recession were mid-wage; 73 percent of the new post-recession jobs pay less than $13.52 per hour. In addition, more Americans, especially American men, are exiting the labor force. In the 1960s, only 9 percent of men between the ages of 25 and 64 were not working; today that figure is more than 18 percent. Cowen reports that the civilian labor force participation rate has fallen from over 67 percent in 1999 to below 64 percent now. In addition, in 1990, 63 percent of American national income was paid to labor, but by 2011 that figure had fallen to 58 percent. The implication is that capital, especially including new intelligent machines, is now earning a higher share of the national income.
The recession revealed an interesting paradox: Average productivity per worker soared while unemployment deepened. In the past employment increased in tandem with rising productivity. Cowen explains this contrast by arguing that during the recession companies "laid off a lot of workers who were not producing enough for their level of pay." They have not been hired back.
The middle class still endures in some low-productivity sectors of the economy: government, education, and health care. Of course, all of these areas are shielded from competition, either as monopolies or as highly regulated services. "I wonder how much of the middle class consists of people in government or protected service sector jobs who don't actually produce nearly as much as their pay," writes Cowen. Income data suggest that Cowen's intuition may be correct.
The Bureau of Labor Statistics (BLS) reported in September 2013 that total employer compensation costs for private-sector workers averaged $29.11 per hour (roughly $58,200), whereas total employer compensation costs for state and local government workers averaged $42.09 per hour (roughly $84,200). In other words, the average local and state government employee earns 40 percent more than the average worker.
An August 2013 report by the Cato Institute economist Chris Edwards found that federal civilian worker had an average wage of $81,704. Edwards also reported that the average federal worker "earns 74 percent more in wages and benefits than the average worker in the U.S. private sector. A job-to-job comparison found that federal workers earned higher wages than did private-sector workers in four-fifths of the occupations examined." Between 1975 and 2013, the total number of federal, state, and local government employees grew from 14.8 million to nearly 22 million, remaining steady at about one for every 15 citizens.
According to the BLS, private health care workers accounted for just over 3 percent of the private workforce in 1958; they were nearly 12 percent of the workforce by 2008. Health care constitutes an ever greater proportion of our economy, growing from about 5 percent of GDP in 1960 to 18 percent today, including both public and private expenditures. The health care industry is also bifurcated in terms of pay. The annual average salary of the more than 6 million health care practitioners and technical staff, ranging from physicians to laboratory technicians, is just over $55,000, solidly in the middle quintile of income. Support workers, such as home health aides and dental assistants, earn considerably less, around $23,000.
Cowen also argues that government has boosted the threshold costs of hiring new employees in various ways. For example, the new health care mandate increases the cost of hiring, which ultimately means fewer jobs, especially for entry level and low-skilled workers. Payroll taxes, higher minimum wages, and regulations that increase corporate uncertainty about investments, such as the Sarbanes-Oxley accounting requirements for publicly traded firms, also erect barriers to hiring.
So what is the future of work? Cowen cites Freestyle chess as a model of man-machine integration at work. Freestyle chess is a rapid-fire variation on the game played by teams who consult various computer programs for their assessments of the best moves. The computer outputs are evaluated by the human team members, who are not themselves necessarily highly ranked players. The combined man-machine chess teams regularly outscore the best grandmasters.
Similarly, in the future, the most successful doctors, lawyers, teachers, marketers, and retailers will be members of teams skilled in the use of intelligent machines to inform and guide their decisions. For example, a patient's symptoms and test results could be uploaded into a comprehensive program for an initial diagnosis. This kind of diagnostic procedure need not be done by physicians, but by technicians whose skills enable them to identify when the diagnostic outputs of smart machines need to be supplemented by the insights of a team of doctors.
Cowen predicts that smart machines will constantly monitor and rate every professional and worker on the quality of their outputs, which he believes will dramatically improve services at the high end of the market and yield a hypercompetitive meritocracy. On the other hand, service providers will also have more information about their customers, including past purchases, evaluations by previous service providers, and so forth. "Customers will have to live with the reputations they create for themselves," he observes.
Cowen suggests that we will be living in a "stupider" world in which we reduce the complexity of our work and service environments to take into account the limitations of even smart machines. "Your daily life will become a curious mix of both 'much, much easier' and 'more frustrating,'" he predicts. "It will be like living in a help menu." Our experience of more and more services will increasingly resemble our standardized interactions with bank ATMs, self-service gas stations, and supermarket checkout scanners.
The big earners in our increasingly bifurcated workforce will be self-motivated, conscientious, and highly skilled. Upgraded forms of learning will be key to their success. Cowen notes that the country now spends twice as much per K-12 pupil in real terms than it did four decades ago, yet educational outcomes have barely moved. This indicates a huge drop in schooling productivity. On the other hand, the advent of online courses has democratized learning as never before. Cowen cites the example of the online course taught by the Stanford artificial intelligence researcher Sebastian Thrun in which the best students were not from Stanford but from poor countries such as India. This means that Americans will be competing with the smartest and most motivated students from around the world.
Given the differences in curiosity and self-motivation, Cowen tartly observes that putting the world's very best education online cheaply will likely not matter much at all to most people. Hard-working and disciplined students aiming for high wages will take advantage of the proliferating opportunities to learn. Others lacking motivation will attend schools that enforce discipline and hard work. Individual teachers will function less as experts and instead act more as coaches to motivate students.
At the low end, many who lack motivation or talent will not upgrade their skills and instead opt for what Cowen calls the world of "threshold earners," making just enough income to get by. For high earners, learning will be a constant feature of their lives as they seek newer skills to complement ever-smarter machines.
Cowen also peers into the future of science, predicting that intelligent machines become vital research partners to people. Ultimately, he suggests, smart machines will one day come up with new theories that are so complex no human scientist will understand them. We will take advantage of the resulting predictions to improve our lives, but a lot of science will look like magic to most of us.
As the gap widens between the big earners and the big losers, what kind of country will America become? Cowen imagines a scenario in which 10 to 15 percent of the citizenry is extremely wealthy and has fantastically comfortable and stimulating lives with excellent health care. "Much of the rest of the country will have stagnant or maybe even falling wages in dollar terms, but a lot more opportunities for cheap fun and also cheap education."
The coastal cities, with their existing concentrations of brain-based industry, will become home to the cognitive elites, while the rest of Americans will retreat to the interior with low-cost housing and scant public services. Essentially, Cowen expects members of the lumpenproletariat to peacefully while away their time with bread and circuses cheaply provided by a fabulously productive economy run by an elite class and their smart machines.
"One day soon we will look back and see that we produced two nations, a fantastically successful nation, working in the technologically dynamic sectors, and everyone else," concludes Cowen. "Average is over."
Not everyone agrees. In a September report for the Information Technology and Innovation Foundation, tech analysts Ben Miller and Robert Atkinson reject the idea that robots are killing our jobs, calling the claim "completely wrong and not supported by the data, scholarly evidence or logic." In the past, Miller and Atkinson note, productivity gains generated by technological progress and incomes have generally increased in tandem.
At the center of their counter-argument is the idea that rising productivity-more output per unit of input-produces savings and increases overall wealth in the form of lower prices, higher wages, or higher profits. The increased wealth creates more demand for products and services, which in turn requires more workers. The increased savings also lower interest rates, which encourage more investment in other productive activities, creating more jobs.
Has this virtuous process of producing prosperity for most workers finally broken down? Perhaps not. It could be that the American economy is still suffering from a severe hangover from the financial crisis, exacerbated by the adoption of particularly stupid economic policies.
In their 2012 book Race Against the Machine, the Massachusetts Institute of Technology economists Erik Brynjolfsson and Andrew McAfee also worried that workers are losing in their competition with machine intelligence. They are, however, considerably less fatalistic than Cowen. They recognize that government policies are increasing barriers to both hiring and upgrading the capabilities of low-skilled workers. Consequently, they advocate improving education, aggressively removing regulatory obstacles to business creation, resisting efforts to restrict hiring and firing, decreasing payroll taxes, decoupling benefits from jobs, not rushing to regulate new network businesses, streamlining the patent system, and shortening copyright terms.
Cowen might be right that it's different this time around. Nevertheless, the sensible reforms suggested by Brynjolfsson and McAfee would give workers a better shot at competing with smart machines, and should in any case be adopted whether or not technological unemployment turns out to be a real problem.
In the meantime, Cowen has given you fair warning: If you don't want to welcome robot overlords someday, it can't hurt to take advantage of any opportunity to upgrade your skills.