To understand our “Second Machine Age,” as Massachusetts Institute of Technology researchers Erik Brynjolfsson and Andrew McAfee call it, one must look at the First Machine Age, the Industrial Revolution. It started with the steam engine in the 18th century. The increased productivity that this innovation and others generated brought economic growth never seen before in the history of mankind. The Industrial Revolution continued with two new innovations at the end of the 19th century: electricity and the internal combustion engine. The resulting machines replaced muscle power and greatly increased production and incomes, including for the poor.
Brynjolfsson and McAfee’s latest book, The Second Machine Age, endeavors to achieve three tasks: explain the Second Machine Age and the opportunities it creates, analyze the disruptions and problems it will cause, and explore ways to keep the benefits while minimizing the costs.
What can machines do? / In the Second Machine Age, machines will not replace or multiply muscle power, but brainpower. For many, this was heralded in 1997 when Deep Blue, an IBM computer, beat Gary Kasparov, the world chess champion. But that was only the beginning of the visible revolution, which had been smoldering for two or three decades. Just four years later, Watson, another IBM computer, won the knowledge game Jeopardy against two champions.
With hindsight, playing chess should be easy for a computer because the rules are simple. As for winning at Jeopardy, all that was needed was a lot of information that could be processed quickly. As late as 2004, however, some experts thought that driving a car required uniquely human skills because the task seemingly couldn’t be distilled into a computer algorithm. It took less than 10 years for that supposed impossibility to become a reality, with driverless Google cars now cruising Silicon Valley.
“In the next twenty-four months,” Brynjolfsson and McAfee write, “the planet will add more computer power than it did in all previous history.” Three broad factors are driving that development. First, and following Moore’s Law (computing capacity doubles every 18 months), the exponential development of machine intelligence (or artificial intelligence) is now reaching dramatic levels. Second, all information is being digitized: maps, music, books, etc. Third, the new inventions are recombined and multiply into yet more inventions.
There is more to the Second Machine Age than treating information with brute computing force. “If you can give precise instructions to somebody on exactly what needs to be done,” Brynjolfsson and McAfee note, “you can often write a precise computer program to do the same task.” So robots can do many things that, previously, only humans could do. Baxter, a human-shaped robot made by Rethink Robotics in Boston, can work on the factory floor and be taught to make coffee on a house coffeemaker. He does this for $4 an hour, all costs included, which is half the hourly wage at a Dunkin Donuts. This is just one example among many.
Computers are now able to write simple corporate earnings previews. They still cannot write good poetry or analytical books like The Second Machine Age. If and when they do, humans could be in deep trouble. Only at the end of the book do Brynjolfsson and McAfee raise such dystopian possibilities, which had previously been the domain of science-fiction writers.
New Age economics | Despite the dystopian possibilities, the authors of The Second Machine Age are optimistic that smart machines will increase, not decrease, human freedom. Their advice to individuals as to how to prepare for competition with the machines—mainly what to learn—is not useless, although perhaps too dismissive of traditional education.
Brynjolfsson and McAfee predict that the Second Machine Age will bring increased material abundance (though, of course, scarcity won’t be abolished). They envision many other economic consequences of quasi-intelligent machines. But here the book stumbles.
Many of the new digital goods and services, the authors argue, are not accounted for in gross domestic product. They overstate their argument. The value added by social networks, even when free to users, is accounted for in the profits made by the providers (Facebook, Twitter, etc.) and their advertising customers. A service sold to advertisers instead of directly to consumers is not “virtually invisible in GDP.” It is not true, either, that Skype “may add zero to GDP,” since it sells premium services and offers the free services as a promotion. Moreover, Microsoft, which now owns Skype, uses the teleconferencing software to increase its own profits. And when the authors of The Second Machine Age claim in a footnote that an increase in crime can boost GDP, they fall victim to the famous broken-window fallacy. If this were not a fallacy in practice as well as in theory, GDP in Venezuela and Syria would be on a fast-rising curve.
It is true, as Brynjolfsson and McAfee note, that GDP does not measure welfare. But it is not technically meant to be the ultimate measure of human welfare, as all economists know—at least those who have looked at the methodology of GDP and know something of welfare economics. GDP is only a measure of market production. It does not measure consumer surplus (the difference between a good’s price and what a consumer is willing to pay for the good). So the authors make no discovery when they write that “TurboTax has created a great deal of value for its users, not all of which even shows up in the GDP statistics.” It can be easily conceded that the effect of the new economy on consumer welfare is higher than what GDP growth shows, but there is no need to appeal to the magical concept of Gross National Happiness as calculated by the government of Bhutan, which the authors “heartily support” along with social justice and other mantras.
Creative destruction and inequality / Just like the Industrial Revolution and its rebound at the turn of the 19th and 20th centuries, the Second Machine Age will bring much creative destruction in its wake. People unable to compete with the machines will see their wages fall or lose their jobs. Middle-class jobs as well as many lower-class ones will disappear. Only people with cognitive jobs that are not repetitive, and thus can’t be automated, will benefit, as well as the robot owners.
The authors of The Second Machine Age emphasize that the gap between economic winners and losers will deepen, increasing inequality. More and more wealthy people will thrive at the top, and more and more poor will struggle at the bottom, while the middle class will be hollowed out. Oft-repeated statistics suggest that this trend has already started.
There is no doubt that much disruption will occur, and that—especially in the short run—there will be losers as well as winners. But the book overestimates the problem because it neglects or underplays several important mitigating factors.
As the authors readily admit, robots are often complementary to jobs, not always substitutes. Robots need conception, programming, training, maintenance, and support. The robots called “drones” provide an example: according to The Economist, a Reaper drone requires 180 people to keep it flying. There are some jobs that humans will always do better than computers, and they are not all on top of the employment ladder: an advanced robot still needs 24 minutes to fold a towel.
A simple economic fact underplayed by The Second Machine Age is that more robots—that is, more capital—will naturally increase the physical productivity of labor. It is true that, as a consequence, product prices will drop, which may lead to a reduction in the value of labor productivity. But the authors’ fear of a general overproduction of goods and services represents an especially weak part of their analysis.
Say’s Law suggests that, in the long run, supply creates its own demand because people work—that is, supply goods and services—only in order to satisfy their demand. Consequently, workers made redundant by technological progress will, by switching industry, generate a new demand equal to the value of what they now produce. As long as somebody wants to consume more, free markets will absorb what suppliers produce. If Say’s Law did not exist, women, youngsters, or immigrants entering the labor force would steal jobs from others. Brynjolfsson and McAfee barely mention Say’s Law and strangely confuse it with price elasticities in particular markets.
Contra the Luddites, the authors do admit that employment grew with technology nearly nonstop since the Industrial Revolution. Just consider agriculture. From 1810 to 1910, the proportion of the American labor force occupied in agriculture dropped from 84 percent to 31 percent, and then to 8 percent in the following half century; now it is just 2 percent. When the former farmers switched to other industries, they created an equivalent demand for their new production. No technological unemployment developed.
The second machine age may be contributing to the increase in income inequality, but Brynjolfsson and McAfee downplay other causes, even if they admit the existence of many. “Factor equalization” is one of those other causes: globalization is fast increasing wages in underdeveloped countries and dampening increases in the developed world. But there are still other possible causes. A recent study by University of Pennsylvania economist Jeremy Greenwood et al. suggests that the growth of assortative marriage—marrying a person of the same level of education and income—explains all the increase in inequality. The role played by regulation and crony capitalism should also be considered. The recent increase in inequality seems overdetermined many times. Perhaps a perfect storm has developed.
We should not forget that welfare is what counts, not GDP, as Brynjolfsson and McAfee painstakingly argue and then apparently forget. What about the nonquantifiable value that the authors claimed is not recognized in GDP? Is it possible that the median consumer would not want to go back to his 1999 situation, when his money income peaked? In the longer run, moreover, people may not want to work more hours simply because they prefer more leisure instead of consuming more goods and services. Working is not the goal of economic life. Working less and consuming more is a benefit, not a cost.
Future problems will probably not be as serious as The Second Machine Age suggests. Computers cannot have ideas. They cannot be entrepreneurial. They cannot “think out of the box”; they are the box. Here Brynjolfsson and McAfee are brilliant: computers “are still machines for generating answers,” they write, “not posing interesting new questions.” And, citing Voltaire, they implore readers to “judge a man by his questions, not his answers.” So relax.
Market failures? / What should the state do? Brynjolfsson and McAfee vacillate between advocating little and much government intervention. They start by stressing how capitalism and free markets provide the only path to a prosperous future. Their (too) short critique of regulation is well taken. They admit that entrepreneurship is more efficient than government intervention. They argue for reducing barriers to immigration. But what is capitalism? How much can you stretch it before it breaks?
Except for schools, the authors of The Second Machine Age tend to ignore the effect that social, political, and economic institutions have on growth and prosperity. Some institutions are conducive to entrepreneurship and innovation; others are not. The two authors do not mention the role that institutions protecting individual liberty and private property rights had in starting the Industrial Revolution, which, after all, did not happen in Russia or Turkey. They barely mention “freedom” (and never “liberty”), while they seem obsessed with equality. They don’t mention incentives, although they could argue that this concept underlies their analysis.
They want to “grow the economy,” which requires government support and a sort of long-term Keynesian push on aggregate demand. Why not just leave the economy alone? In The Rise and Decline of Nations, University of Maryland economist Mancur Olson argued, more correctly in my opinion, that “an economy with free markets and no government or cartel intervention is like teen-aged growth; it makes a lot of mistakes but nevertheless grows rapidly without special effort or encouragement.”
Brynjolfsson and McAfee tend to see market failures and externalities everywhere, even in unemployment. They argue that we must “support our scientists” (who might not be totally happy to learn they belong to us) with government funding. New or higher taxes are warranted. A minimum wage scheme based on a negative income tax should be introduced. The authors of The Second Machine Age seem to trust that government will do all this efficiently, while reducing regulation and promoting entrepreneurship and economic growth.
The book’s recommendations are based on a naive, pre–Public Choice approach, as if government failures were not at least as prevalent as market failures.
They don’t seem to take seriously the fact that work is a cost, and that if more consumption can be obtained for less work, most people would be happier. It is true that some people find satisfaction and “purpose” in their jobs, although academics and intellectuals sometimes forget that not everybody has as pleasant a job as theirs. Any job has a leisure component, but most people would take the income and drop the job if that were an alternative. Consequently, governments should not support work as such, but only self-reliance and jobs that generate real incomes.
Despite all the pinging and networking, it is taking much time for the conclusions of—or at least the doubts raised by—welfare economics and social choice theory to trickle down to the layman and even to serious analysts occupied in other fields. The Second Machine Age is frustrating with its multitude of collective “we,” as if the political “we” represented anything other than some arbitrary majority or minority. “[T]he future we get,” Brynjolfsson and McAfee write in typical Newspeak, “will depend on the choices we make.” [O]ur values will matter more than ever.” “We must think hard about what it is we really value, what we want more of, and what we want less of.” They even repeat the mantras of “we … as a society” and “society as a whole.” But who is “we”? The authors never tell us.
The book’s policy recommendations are based on a naive, pre–Public Choice approach, as if government failures were not at least as prevalent as market failures. The authors seem to simply assume that the state is geared to reducing externalities for the unanimous good of everybody. They don’t consider the results of the actual political and bureaucratic processes. They don’t envisage that a majority or even a minority can use the state to exploit the rest of the population.
This book is not an economists’ book. Brynjolfsson is a professor of management science and information technology at MIT and director of its Center for Digital Business. One of his graduate degrees is in “managerial economics,” but he appears to be more a management guru than an economist. McAfee has a graduate background in science and management. They are two very educated and no doubt interesting intellectuals, but not necessarily fluent in the fine points of economic theory. Of course, not being a wide-ranging economist is not a fault per se (one might have compensating qualities) and ideas should be judged for what they are, not for their sources. But a good command of economics remains useful in analyzing the economic effects of social or technological events, and in giving economic advice to policymakers.
Once you are aware of its pitfalls, this book is instructive. But it is mainly instructive on the technological front, not for its economic analysis.
