In Delhi and Washington and Beijing, it is in the name of growth that policy makers drag their feet when called upon to enact or enforce pollution regulations. Who benefits from this GDP growth remains an afterthought.
Economists deserve their fair share of the blame for stoking this rhetoric. Nothing in either our theory or the data proves the highest GDP per capita is generally desirable. Yet because we fundamentally believe resources can and will be redistributed, we fall into the trap of always trying to make the overall pie as big as possible. This flies directly in the face of what we have learned over the past decades. The evidence is clear—inequality has risen dramatically in recent years, with searing consequences for societies across the world.
CHAPTER 7
PLAYER PIANO
PLAYER PIANO WAS the very first novel published by the great American fabulist Kurt Vonnegut.1 It is a dystopia about a world where most jobs have disappeared. Written in 1952 in the wake of the great postwar expansion of jobs, it was either extremely farsighted or astoundingly misguided, but, either way, it’s a perfect novel for our times.
A player piano is a piano that plays itself. In Vonnegut’s world, machines run themselves and people are no longer needed. They are provided for, and get to do various forms of make-work, but there is nothing meaningful or useful they can do. As Mr. Rosewater, a character in a later (1965) novel by Vonnegut puts it: “The problem is this: How to love people who have no use?”2 Or even have them not hate themselves?
The increasing sophistication of robots and the progress of artificial intelligence has generated considerable anxiety about what would happen to our societies if only a few people had interesting jobs and everyone else had either no work or had a horrible job, and inequality ballooned as a result. Especially if this happened because of forces largely out of their control. Tech moguls are getting desperate to find ideas to solve the problems their technologies might cause. But we don’t need to contemplate the future in order to get a sense of what happens when economic growth leaves behind the majority of a country’s citizens. This has already happened—in the United States since 1980.
ONE FOR THE LUDDITES
An increasing number of economists (and of those who comment on economics) worry that new technologies, such as AI, robots, and automation more generally, will destroy more jobs than they create, making many workers obsolete and causing the share of GDP that goes to pay wages to dwindle. In fact, these days growth optimists and labor pessimists are often the same people; they both imagine future growth will be primarily driven by the replacement of human workers by robots.
In their book The Second Machine Age, our MIT colleagues Erik Brynjolfsson and Andrew McAfee offer a bleak view of the impact of digitization on the future of employment in the United States.3 Digitization, they suspect, will make workers with “ordinary” skills increasingly redundant. As tasks from car painting to spreadsheet manipulation are done by computers or robots, highly educated workers who are adaptable and can program and install the robots will become more and more valuable, but other workers who can be replaced will find themselves without jobs unless they accept extremely low salaries. In this view, artificial intelligence will be the final nail in the coffin of these ordinary workers.
In the first IT revolution, as David Autor has shown, jobs involving routine repetitive tasks were the ones that went.4 Jobs that required quick judgment and initiative stayed put. The number of typists and assembly-line workers diminished, but executive assistants and burger flippers kept their jobs. This time, many say, it is different. Artificial intelligence means machines can learn as they go and are therefore able to carry out increasingly nonroutine tasks, such as playing Go or folding laundry. In June 2018, a restaurant offering robot-made burgers opened in San Francisco. Humans are still taking the orders and cooking the sauces, but the robots cook the gourmet burgers, such as the Tumami Burger (“Smoked oyster aioli, shiitake mushroom sauce, black pepper and salt, pickles, onion, butter lettuce—Designed by Chef Tu, Top Chef Season 15”5), in five minutes and for $6. Esther’s sister Annie Duflo, the CEO of a large NGO, does not have a human assistant; she relies exclusively on an AI-powered assistant named Fin. Fin books her hotels and her plane tickets, manages her calendar, and takes care of her travel reimbursements. Annie is, sadly, much happier with Fin than she was with her human assistants. She pays him (her? it?) much less and gets much more reliable service. To be sure, there are some humans behind Fin, but fewer and fewer, and the business model is clearly to move away from them.
The AI revolution is thus poised to hit people across a wide spectrum of jobs. Accountants, mortgage originators, management consultants, financial planners, paralegals, and sports journalists are already competing with some form of artificial intelligence or, if not, will soon. Cynics might say it is precisely because these more high-end jobs are on the line that we are finally talking about this, and they may be right. But AI will also hurt shelf stackers, office cleaners, restaurant workers, and taxi drivers. Based on the tasks they perform, a McKinsey report6 concludes that 45 percent of US jobs are at risk of being automated, and the OECD estimates that 46 percent of the workers in OECD countries are in occupations at high risk of being either replaced or fundamentally transformed.7
Of course, what this calculation misses is that as some tasks get automatized, and the need for humans gets relieved, people can be put to work elsewhere.
So how bad will it be on net? Economists are of course intrigued by this problem, but in this case they have entirely failed to reach a consensus. The IGM Booth panel of experts were asked their opinion of the following statement: “Holding labor market institutions and job training fixed, rising use of robots and artificial intelligence is likely to increase substantially the number of workers in advanced countries who are unemployed for long periods.” Twenty-eight percent of respondents agreed or strongly agreed with it, 20 percent disagreed or strongly disagreed, and 24 percent were uncertain!8
The difficulty is that doomsday (if it is coming) has not arrived. Robert Gordon, whom as we have seen does not think too highly of today’s innovations, likes to play “spot the robot” when he travels.9 For all the talk, he says, it is still a human clerk who checks him in at the hotel, cleans his room, serves his coffee, and so on.
For the time being, humans have not been made redundant. Unemployment in the United States, as we write this book in the first quarter of 2019, is at a historical low and falling.10 With more and more women joining the labor force, the share of the population in the labor force rose substantially until about 2000 (when it started to plateau or reverse).11 Jobs were found for all those who wanted to work, despite rapid labor-saving technological progress.
Of course, it is true we are probably just at the very beginning of the process of AI-fueled automation. The sense that artificial intelligence is a new class of technology makes it hard to predict what it might do. Futurologists talk about a “singularity,” a dramatic acceleration of the rate of productivity growth fueled by infinitely intelligent machines, although most economists are quite skeptical that we are anywhere close to seeing something like that. But it could well be that if Gordon plays spot the robot in a few years, he will have a more exciting time.
On the other hand, while this particular wave of automation is just starting, there have been others in the past. Like AI today, the spinning jenny, the steam engine, electricity, computer chips, and computer-assisted-learning machinery all automatized and relieved the need for humans in the past.12
What happened then is very much what one might have expected: by replacing workers with machines on some tasks, automation has a powerful displacement effect. It makes the workers redundant. This is what happened to the skilled artisans spinning and weaving at the dawn of the industrial revolution. They were replaced by machines. And as is well known, they did not like it one bit. In the early nineteenth century, the Luddites destroyed machines to protest the mechanization of weaving, which was thre
atening their livelihoods as skilled artisans. The term Luddite is now mostly used pejoratively to describe someone who blindly refuses progress, and their example is often used to dismiss concerns about technology creating unemployment. After all, the Luddites were wrong—jobs did not vanish, and wages and living conditions are much higher today than they were then.
Yet the Luddites were less wrong than we might assume. Their particular jobs did vanish in the industrial revolution, along with the jobs of a whole range of artisans. We are told that in the long run everything was fine, but the long run was very long indeed. Real blue-collar wages in Britain were almost halved between 1755 and 1802. Although 1802 was a particularly low year, they were on a declining trend between 1755 and the turn of the century, and it is only at the turn of the century that they started increasing again. They would recover their 1755 level only in 1820, sixty-five years later.13
This period of intense technological progress in the United Kingdom was also an era of intense deprivation and very difficult living conditions. The economic historian Robert Fogel showed that boys in England during this period were significantly undernourished compared even to slaves in the US South.14 The literature of the time, from Frances Trollope to Charles Dickens, describes what was happening to the economy and society with a certain amount of unmitigated horror. Those were Hard Times indeed.
We know that eventually there was a turnaround in the UK. Even as some workers lost their jobs, the labor-saving innovations raised profitability of other inputs, and hence the demand for workers producing them. Improvements in weaving technology, like John Kay’s flying shuttle, for example, increased demand for yarn, creating jobs for people to produce yarn. And the burgeoning wealth of those profiting from these innovations increased demand for new products and services in a range of sectors (more solicitors, accountants, engineers, bespoke tailors, gardeners, etc.), which created more jobs.
However, nothing tells us the rebound is guaranteed to happen. There may well be no rebound from the fall in demand for labor resulting from this wave of automation and AI. Sectors that become more profitable may invest in new labor-saving technologies instead of hiring more workers. The new wealth may be used to purchase goods made in another country.
We don’t know what will happen this time around, since we haven’t seen the very long run yet, but the impact of the current wave of automation (which started in 1990, giving us a perspective of more than twenty-five years) appears so far to be negative. In a study on the impact of automatization, researchers computed, for each region, a measure of exposure to industrial robots, capturing the spread of robots in the industries in that region.15 They then compared the evolution of employment and wages in the most affected areas to that in the least affected areas. The study found, to the surprise of the authors, who had written a previous paper emphasizing the forces that should lead to a rebound,16 large negative impacts. One more robot in a commuting zone reduces employment by 6.2 workers and also depresses wages. The employment effects are most pronounced in manufacturing and they are particularly strong for workers with lower than a college education, especially those who do routine manual tasks. However, there are no offsetting gains in employment or wages for any other occupation or educational group. These local impacts of robots on employment and wages are reminiscent of the impacts of greater exposure to international trade. They are surprising for the same reasons. As many tasks in a particular industry get automatized, we might have expected displaced workers to find employment in new businesses that would have come to the region to take advantage of the freed-up labor, or to move elsewhere. It is also worrying that the automation of simple tasks did not lead to the hiring of more engineers to supervise the robots. The explanation is probably similar to why competition with China hurt low-skilled workers; in the sticky economy, seamless reallocation is anything but guaranteed.
Even if the total number of jobs does not fall, the current wave of automation tends to displace jobs that require some skills (bookkeepers and accountants) and increase the demand, either for very skilled workers (software programmers for the machines) or for totally unskilled workers (dog walkers, for example), which are both much more difficult to replace with a machine. As software engineers become richer, they have more money to hire dog walkers, who have become relatively cheaper over time, since there is little alternative employment for those with no college education. Even if people remain employed, this leads to an increase in inequality, with higher wages at the top and everyone else pushed to jobs requiring no specific skills; jobs where wages and working conditions can be really bad. This accentuates a trend that has taken place since the 1980s. Workers without a college education have increasingly been pushed out of mid-skill jobs, such as clerical and administrative roles, into low-skill tasks, such as cleaning and security.17
LUDDISM LIGHT?
So should we try to stop the push toward automation? There are in fact good reasons to suspect that some of the recent automation is excessive; corporations seem to decide to automate even when robots are less productive than people. Excessive automation reduces GDP instead of contributing to it.
One reason is the bias in the US tax code, which taxes labor at a higher rate than capital. Employers have to pay payroll taxes (used to finance social security and Medicare) on labor, but not on robots. They get an immediate tax rebate when they invest in the robot, since they can often claim “accelerated depreciation” for a capital expenditure, and if they finance it with a loan they also get to deduct the interest from their earnings. This tax advantage gives employers an incentive to automate, even if it would otherwise cost less to keep the workers.18 Moreover, even without subsidies from the tax code, the many frictions in the labor market may make managers dream of factories without workers. Robots won’t demand maternity leave or protest a wage cut in a recession. It is probably not an accident that automation in the retail sector (such as automatic checkout lines) started first in Europe, where the labor unions are stronger.
The increase in industry concentration and monopolies could also reinforce this tendency. A monopolist does not fear competition. It has no reason to constantly reinvent what it is offering its consumers. Therefore, the monopolist will tend to focus more on cost-cutting innovations, which will increase its profit margins. In contrast, a competitive firm might go for a moonshot to try to take over the market.
Now it is true that even if a business adopts a highly productive new technology that displaces labor, the increase in productivity also creates new resources that could be deployed to find new uses for the freed labor. The technologies most dangerous for the workers are what some researchers have described as “so-so” automation technologies; they are just productive enough to be adopted given the distortions in the tax code, and displace workers, but not productive enough to raise overall productivity.19
Unfortunately, notwithstanding the grandiose talk about singularities, the bulk of R&D resources these days is directed toward machine learning and other big data methods designed to automate existing tasks, rather than the invention of new products that would create new roles for workers, and hence new jobs.20 This may make economic sense for the companies, given the financial gains in replacing workers with robots. But it distracts researchers and engineers from working on the truly pathbreaking innovations. For example, inventing new software or hardware health workers could use to assist patients in doing their rehabilitation therapy at home after a surgery rather than in a hospital could potentially save insurance companies lot of money, improve well-being, and create new jobs. But the bulk of the automation effort today in insurance firms goes toward searching for algorithms that automate the approval of insurance claims. This saves money but destroys jobs. This emphasis on the automation of existing jobs increases the potential for the current wave of innovation to be very damaging for workers.
That unregulated automation could be bad for workers is also the instinct of most Americans on the right and the left. O
ne place, remarkably, where Republican and Democrat poll respondents agree is in their opposition to letting companies decide how much to automate. Eighty-five percent of Americans would support limiting automation to “dangerous and dirty jobs,” with no difference between Democrats and Republicans. Even when the question is posed in a more politically pointed way, asking whether “there should be limits on the number of jobs businesses can replace with machines, even if they are better and cheaper than humans,” 58 percent of Americans, including half of Republicans, say yes.21
This specific force of automation is exacerbating what is always a concern. When a worker is fired, the firm is done with him, but society inherits the liability of his continued well-being. Society does not want him to starve or his family to be homeless; it wants him to find another job he likes. We fear his anger, especially if it leads to a vote for the many lurking extremists in today’s world, whereas the firm does not have to pay for the retraining, the welfare payments, or the social costs of the anger.
This kind of argument has traditionally been used to justify making it difficult to fire workers. Some labor laws, like India’s, make it virtually impossible to fire anyone in larger firms. Others, like the French laws, make it difficult and uncertain. The worker can appeal and possibly be reinstated with back pay. The problem with such firing costs is that they can make life very difficult for a manager faced with a nonperforming worker or an urgent need to downsize in order to survive. As a result, firing costs may discourage hiring in the first place, which would exacerbate unemployment.22
Good Economics for Hard Times Page 28