The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies

by Erik Brynjolfsson

R, an open source application for statistics.

The journalist A. J. Liebling famously remarked that, “Freedom of the press is limited to those who own one.” It is no exaggeration to say that billions of people will soon have a printing press, reference library, school, and computer all at their fingertips.

the rule of 70 (the time to double a value is roughly equal to 70 divided by its growth rate),

The explanation for this productivity surge is in the lags that we always see when GPTs are installed. The benefits of electrification stretched for nearly a century as more and more complementary innovations were implemented. The digital GPTs of the second machine age are no less profound. Even if Moore’s Law ground to a halt today, we could expect decades of complementary innovations to unfold and continue to boost productivity. However, unlike the steam engine or electricity, second machine age technologies continue to improve at a remarkably rapid exponential pace, replicating their power with digital perfection and creating even more opportunities for combinatorial innovation. The path won’t be smooth—for one thing, we haven’t banished the business cycle—but the fundamentals are in place for bounty that vastly exceeds anything we’ve ever seen before.

In the early days of the 1990s Internet boom, venture capitalists used to joke that there were only two numbers in the new economy: infinity and zero.

Production in the second machine age depends less on physical equipment and structures and more on the four categories of intangible assets: intellectual property, organizational capital, user-generated content, and human capital.

According to Dale Jorgenson and Barbara Fraumeni, the value of human capital in the United States is five to ten times larger than the value of all the physical capital in the United States.24 Human capital has not always been this important to the economy. The great economist Adam Smith understood that one of the great drawbacks of the first machine age was the way it forced workers to do repetitive tasks. In 1776, he noted, “The man whose whole life is spent in performing a few simple operations, of which the effects are perhaps always the same, or very nearly the same, has no occasion to exert his understanding.”

“An imbalance between rich and poor is the oldest and most fatal ailment of all republics.” —Plutarch

The evolution of photography illustrates the bounty of the second machine age, the first great economic consequence of the exponential, digital, combinatorial progress taking place at present. The second one, spread, means there are large and growing differences among people in income, wealth, and other important circumstances of life. We’ve created a cornucopia of images, sharing nearly four hundred billion “Kodak moments” each year with a few clicks of a mouse or taps on a screen. But companies like Instagram and Facebook employ a tiny fraction of the people that were needed at Kodak. Nonetheless, Facebook has a market value several times greater than Kodak ever did and has created at least seven billionaires so far, each of whom has a net worth ten times greater than George Eastman did. The shift from analog to digital has delivered a bounty of digital photos and other goods, but it has also contributed to an income distribution that is far more spread out than before.

In almost every industry, technological progress will bring unprecedented bounty. More wealth will be created with less work. But at least in our current economic system, this progress will also have enormous effects on the distribution of income and wealth. If the work a person produces in one hour can instead be produced by a machine for one dollar, then a profit-maximizing employer won’t offer a wage for that job of more than one dollar. In a free-market system, either that worker must accept a wage of one dollar an hour or find some new way to make a living. Conversely, if a person finds a new way to leverage insights, talents, or skills across one million new customers using digital technologies, then he or she might earn one million times as much as would be possible otherwise.

A slightly more complex model allows for the possibility that technology may not affect all inputs equally, but rather may be ‘biased’ toward some and against others. In particular, in recent years, technologies like payroll processing software, factory automation, computer-controlled machines, automated inventory control, and word processing have been deployed for routine work, substituting for workers in clerical tasks, on the factory floor, and doing rote information processing. By contrast, technologies like big data and analytics, high-speed communications, and rapid prototyping have augmented the contributions made by more abstract and data-driven reasoning, and in turn have increased the value of people with the right engineering, creative, or design skills. The net effect has been to decrease demand for less skilled labor while increasing the demand for skilled labor. Economists including David Autor, Lawrence Katz and Alan Krueger, Frank Levy and Richard Murnane, Daron Acemoglu, and many others have documented this trend in dozens of careful studies.17 They call it skill-biased technical change. By definition, skill-biased technical change favors people with more human capital.

The combination of higher pay despite growing supply can only mean that the relative demand for skilled labor increased even faster than supply. And at the same time, the demand for tasks that could be completed by high school dropouts fell so rapidly that there was a glut of this type of worker, even though their ranks were thinning. The lack of demand for unskilled workers meant ever-lower wages for those who continued to compete for low-skill jobs. And because most of the people with the least education already had the lowest wages, this change increased overall income inequality.

Repetitive work on an assembly line is easier to automate than the work of a janitor. Routine clerical work like processing payments is easier to automate than handling customers’ questions. At present, machines are not very good at walking up stairs, picking up a paperclip from the floor, or reading the emotional cues of a frustrated customer. To capture these distinctions, work by our MIT colleagues Daron Acemoglu and David Autor suggests that work can be divided into a two-by-two matrix: cognitive versus manual and routine versus nonroutine.25 They found that the demand for work has been falling most dramatically for routine tasks, regardless of whether they are cognitive or manual. This leads to job polarization: a collapse in demand for middle-income jobs, while nonroutine cognitive jobs (such as financial analysis) and nonroutine manual jobs (like hairdressing) have held up relatively well.

As we saw in chapter 2, this reflects Moravac’s paradox, the insight that the sensory and motor skills we use in our everyday lives require enormous computation and sophistication.27 Over millions of years, evolution has endowed us with billions of neurons devoted to the subtleties of recognizing a friend’s face, distinguishing different types of sounds, and using fine motor control. In contrast, the abstract reasoning that we associate with ‘higher thought’ like arithmetic or logic is a relatively recent skill, developed over only a few thousand years. It often requires simpler software and less computer power to mimic or even exceed human capabilities on these types of tasks.

WE’VE SEEN THAT SKILL-BIASED technical change has increased the relative demand for highly educated workers while reducing demand for less educated workers whose jobs frequently involve routine cognitive and manual tasks. In addition, capital-biased technological changes that encourage substitution of physical capital for labor have increased the profits earned by capital owners and reduced the share of income going to labor. In each case, historic amounts of wealth have been created. In each case, we also have seen increases in the earnings of the winners relative to the losers. But the biggest changes of all are driven by a third gap between winners and losers: the gap between the superstars in a field and everyone else.

Call it talent-biased technical change.

Technology has supercharged the ability of authors like Rowling to leverage their talents via digitization and globalization. Rowling’s stories can be captured in movies and video games as well as text, but each of those formats, including the original books, can be transmitted globally at trivial cost. She and other superstar storytellers now reach billions of customers through a variety of channels and formats.

At the same time, others working in the content and entertainment industries have not seen a big increase. Only 4 percent of software developers in the burgeoning app economy have made over a million dollars.7 Three-quarters of them made less than thirty thousand dollars. While a handful of writers, actors, or baseball players can become millionaires, many others struggle to make ends meet. A gold-medal winner at the Olympics can earn millions of dollars in endorsements, while the silver medal winner—let alone the person who placed tenth or thirtieth—is quickly forgotten, even if the difference is measured in tenths of a second and could have resulted from a gust of wind or a lucky bounce of the ball.

The ratio of CEO pay to average worker pay increased from seventy in 1990 to three hundred in 2005. Much of this growth is linked to the greater use of information technology,

The bigger the market value of a company, the more compelling the argument for trying to get the very best executive.9 A single decision that increases value by a modest 1 percent is worth $100 million to a ten-billion-dollar company.

In a traditional market, someone who is 90 percent as skilled or works 90 percent as hard creates 90 percent as much value and thus can earn 90 percent as much money. That’s absolute performance. By contrast, a software programmer who writes a slightly better mapping application—one that loads a little faster, has slightly more complete data, or prettier icons—might completely dominate a market. There would likely be little, if any, demand for the tenth-best mapping application, even it got the job done almost as well. This is relative performance. People will not spend time or effort on the tenth-best product when they have access to the best.

In contrast, the economics of personal services (nursing) or physical work (gardening) are very different, since each provider, no matter how skilled or hard-working, can only fulfill a tiny fraction of the overall market demand.

As the great economist Alfred Marshall noted, “a rich client whose reputation or fortune or both are at stake will scarcely count any price too high to secure the services of the best man he can get.”

Although the top 1 percent and 0.01 percent have seen record increases in their earnings, the superstar economy has faced a few headwinds. Perhaps the most important among these is the growth of the long tail—the increased availability of niche products and services. Technology has not just lowered marginal costs; in many cases it has also lowered fixed costs, inventory costs, and the costs of searching. Each of these changes makes it more attractive to offer a greater variety of products and services, filling small niches that previously went unfilled. Instead of going head-to-head with a superstar, some individuals and businesses are instead finding ways to differentiate their products, to find or create an alternative niche where they can be the world’s best.

Instead of being the thousandth-best children’s book author in the world, it may be more profitable to be the number-one author in Science-Based Advice for Ecological Entrepreneurs, or Football Clock Management.

The great economist Joseph Schumpeter wrote of “creative destruction,” where each innovation not only created value for consumers but also wiped out the previous incumbent. The winners scaled up and dominated their markets, but were in turn vulnerable to the next generation of innovators. Schumpeter’s observation describes markets in software, media, and the Internet much better than traditional markets in manufacturing and services. But as more and more industries become increasingly digitized and networked, we can expect the Schumpeterian dynamic to spread.

In a superstar economy, the distribution of income isn’t just more spread out; it has a very different shape. It’s not just that a small group at the top sees big increases. It’s also a change in the fundamental structure of the distribution.

In contrast, superstar (and long tail) markets are often better described by a power law, or Pareto curve, in which a small number of people reap a disproportionate share of sales. This is often characterized as the 80/20 rule, where 20 percent of the participants get 80 percent of the gains, but it can be more extreme than that.

Power law distributions have a ‘fat tail,’ which means the likelihood of extreme events is much greater than one would expect to see in a normal distribution.24 They are also ‘scale invariant,’ which means that the top-selling book accounts for about the same share of the top ten books’ sales as the top ten books do for the top one hundred, or the top one hundred do for the top one thousand. Power laws describe many phenomena, from frequency of earthquakes to the frequency of words in most languages. They also describe the sales distribution of books, DVD, apps, and other information products.

Kim Taipale, founder of the Stilwell Center for Advanced Studies in Science and Technology Policy, has argued that, “The era of bell curve distributions that supported a bulging social middle class is over and we are headed for the power-law distribution of economic opportunities. Education per se is not going to make up the difference.”

In practical terms, this means that when income is distributed according to a power law, most people will be below average—say goodbye, Lake Wobegon! Furthermore, over time, average income can increase without any increase in the median income or, for that matter, without any increase in income for most people. Power-law distributions don’t just increase income inequality; they also mess with our intuitions.

“The test of our progress is not whether we add more to the abundance of those who have much; it is whether we provide enough for those who have little.” —Franklin D. Roosevelt

The fact that technology brings both bounty and spread, and brings more of both over time, leads to an important question: Since there’s so much bounty, should we be concerned about the spread? In other words, we might consider rising inequality less of a problem if people at the bottom are also seeing their lives improve thanks to technology.

This argument acknowledges that highly skilled workers are pulling away from the rest—and that superstars are pulling so far away as to be out of sight—but then essentially asks, “So what? As long as all people’s economic lives are getting better, why should we be concerned if some are getting a lot better?” As Harvard economist Greg Mankiw has argued, the enormous income earned by the “one percent” is not necessarily a problem if it reflects the just deserts of people who are creating value for everyone else.

John Maynard Keynes was less confident that things would always work out so well for workers. His 1930 essay “Economic Possibilities for our Grandchildren,” while mostly optimistic, nicely articulated the position of the second camp—that automation could in fact put people out of work permanently, especially if more and more things kept getting automated. His essay looked past the immediate hard times of the Great Depression and offered a prediction: “We are being afflicted with a new disease of which some readers may not yet have heard the name, but of which they will hear a great deal in the years to come—namely, technological unemployment. This means unemployment due to our discovery of means of economizing the use of labor outrunning the pace at which we can find new uses for labor.”15 The extended joblessness of the Great Depression seemed to confirm Keynes’s ideas, but it eventually eased. Then came World War II and its insatiable demands for labor, both on the battlefield and the home front, and the threat of technological unemployment receded.

There are three economic mechanisms that are candidates for explaining technological unemployment: inelastic demand, rapid change, and severe inequality.

In the long run, the biggest effect of automation is likely to be on workers not in America and other developed nations, but rather in developing nations that currently rely on low-cost labor for their competitive advantage. If you take most of the costs of labor out of the equation by installing robots and other types of automation, then the competitive advantage of low wages largely disappears.

If you look at the types of tasks that have been offshored in the past twenty years, you see that they tend to be relatively routine, well-structured tasks. Interestingly, these are precisely the tasks that are easiest to automate. If you can give precise instructions to someone else on exactly what needs to be done, you can often write a precise computer program to do the same task. In other words, offshoring is often only a way station on the road to automation.

As we look across examples of things we haven’t seen computers do yet, this idea of the “new idea” keeps recurring. We’ve never seen a truly creative machine, or an entrepreneurial one, or an innovative one.

Ideation in its many forms is an area today where humans have a comparative advantage over machines. Scientists come up with new hypotheses. Journalists sniff out a good story. Chefs add a new dish to the menu. Engineers on a factory floor figure out why a machine is no longer working properly. Steve Jobs and his colleagues at Apple figure out what kind of tablet computer we actually want. Many of these activities are supported and accelerated by computers, but none are driven by them.

we believe that employers now and for some time to come will, when looking for talent, follow the advice attributed to the Enlightenment sage Voltaire: “Judge a man by his questions, not his answers.”

As futurist Kevin Kelly put it “You’ll be paid in the future based on how well you work with robots.”

So ideation, large-frame pattern recognition, and the most complex forms of communication are cognitive areas where people still seem to have the advantage, and also seem likely to hold on to it for some time to come.

“self-organizing learning environments” (SOLEs)

Our recommendations about how people can remain valuable knowledge workers in the new machine age are straightforward: work to improve the skills of ideation, large-frame pattern recognition, and complex communication instead of just the three Rs. And whenever possible, take advantage of self-organizing learning environments, which have a track record of developing these skills in people.

“A policy is a temporary creed liable to be changed, but while it holds good it has got to be pursued with apostolic zeal.” —Mahatma Gandhi

“Work saves a man from three great evils: boredom, vice, and need.” —Voltaire

Mechanical Turk, which quickly became popular, was an early instance of what came to be called crowdsourcing, defined by communications scholar Daren Brabham as “an online, distributed problem-solving and production model.”