Race Againstthe Machine

How the Digital Revolution isAccelerating Innovation, DrivingProductivity, and IrreversiblyTransforming Employment and theEconomy.

Erik Brynjolfsson and Andrew McAfee

Digital Frontier PressLexington, Massachusetts

© 2011 Erik Brynjolfsson and Andrew McAfee

All rights reserved. No part of the book may bereproduced in any form by any electronic or mechanicalmeans (including photocopying, recording, or informationstorage and retrieval) without permission in writing fromthe publisher.

For information about quantity discounts, emailinfo@raceagainstthemachine.com

www.RaceAgainstTheMachine.com

Library of Congress Cataloging-in-Publication Data

Brynjolfsson, ErikRace against the machine : how the digital revolution isaccelerating innovation, driving productivity, andirreversibly transforming employment and the economy.p. cm.

eISBN 978-0-9847251-0-61. Technological innovations – Economic Aspects.

I. McAfee, Andrew.II. Title.

eBooks created by www.ebookconversion.com

Contents

1. Technology’s Influence on Employmentand the Economy

2. Humanity and Technology on theSecond Half of the Chessboard

3. Creative Destruction: The Economicsof Accelerating Technology andDisappearing Jobs

4. What Is to Be Done? Prescriptions andRecommendations

5. Conclusion: The Digital Frontier

6. Acknowledgments

To my parents, Ari andMarguerite Brynjolfsson, who

always believed in me.

To my father, DavidMcAfee, who showed me that

there’s nothing better than a jobwell done.

Chapter 1.Technology’s Influenceon Employment and theEconomy

If, in like manner, the shuttle wouldweave and the plectrum touch the lyrewithout a hand to guide them, chiefworkmen would not want servants.

—Aristotle

This is a book about how informationtechnologies are affecting jobs, skills,wages, and the economy. To understandwhy this is a vital subject, we need onlylook at the recent statistics about job

growth in the United States.

By the late summer of 2011, the U.S.economy had reached a point where evenbad news seemed good. The governmentreleased a report showing that 117,000jobs had been created in July. Thisrepresented an improvement over Mayand June, when fewer than 100,000 totaljobs had been created, so the report waswell received. A headline in the August 6edition of the New York Times declared,“US Reports Solid Job Growth.”

Behind those rosy headlines, however, laya thorny problem. The 117,000 new jobsweren't even enough to keep up withpopulation growth, let alone reemploy anyof the approximately 12 million

Americans who had lost their jobs in theGreat Recession of 2007-2009. EconomistLaura D’Andrea Tyson calculated thateven if job creation almost doubled, to the208,000 jobs per month experiencedthroughout 2005, it would take until 2023to close the gap opened by the recession.Job creation at the level observed duringJuly of 2011, on the other hand, wouldensure only an ever-smaller percentage ofemployed Americans over time. And inSeptember the government reported thatabsolutely no net new jobs had beencreated in August.

Of all the grim statistics and storiesaccompanying the Great Recession andsubsequent recovery, those related toemployment were the worst. Recessions

always increase joblessness, of course,but between May 2007 and October 2009unemployment jumped by more than 5.7percentage points, the largest increase inthe postwar period.

An Economy That’s Not Putting PeopleBack to Work

An even bigger problem, however, wasthat the unemployed couldn't find workeven after economic growth resumed. InJuly of 2011, 25 months after therecession officially ended, the main U.S.unemployment rate remained at 9.1%, lessthan 1 percentage point better than it wasat its worst point. The mean length of timeunemployed had skyrocketed to 39.9weeks by the middle of 2011, a duration

almost twice as long as that observedduring any previous postwar recovery.And the workforce participation rate, orproportion of working-age adults withjobs, fell below 64%—a level not seensince 1983 when women had not yetentered the labor force in large numbers.

Everyone agreed that this was a direproblem. Nobel Prize-winning economistPaul Krugman described unemployment asa “terrible scourge … a continuingtragedy. … How can we expect to prospertwo decades from now when millions ofyoung graduates are, in effect, beingdenied the chance to get started on theircareers?”

Writing in The Atlantic, Don Peck

described chronic unemployment as “apestilence that slowly eats away atpeople, families, and, if it spreads widelyenough, the fabric of society. Indeed,history suggests that it is perhaps society’smost noxious ill. … This era of highjoblessness … is likely to warp ourpolitics, our culture, and the character ofour society for many years.” His colleagueMegan McArdle asked her readers tovisualize people who had beenunemployed for a long time: “Think aboutwhat is happening to millions of peopleout there … whose savings and socialnetworks are exhausted (or were neververy big to begin with), who are in theirfifties and not young enough to retire, butvery hard to place with an employer whowill pay them as much as they were worth

to their old firm. Think of the people whocan't support their children, or themselves.Think of their despair.”

Many Americans did think of such people.Twenty-four percent of respondents to aJune 2011 Gallup poll identified“unemployment/jobs” as the mostimportant problem facing America (this inaddition to the 36% identifying “economyin general”).

The grim unemployment statistics puzzledmany because other measures of businesshealth rebounded pretty quickly after theGreat Recession officially ended in June2009. GDP growth averaged 2.6% in theseven quarters after the recession’s end, arate 75% as high as the long-term average

over 1948-2007. U.S. corporate profitsreached new records. And by 2010,investment in equipment and softwarereturned to 95% of its historical peak, thefastest recovery of equipment investmentin a generation.

Economic history teaches that whencompanies grow, earn profits, and buyequipment, they also typically hireworkers. But American companies didn’tresume hiring after the Great Recessionended. The volume of layoffs quicklyre turned to pre-recession levels, socompanies stopped shedding workers. Butthe number of new hires remainedseverely depressed. Companies broughtnew machines in, but not new people.

Where Did the Jobs Go?

Why has the scourge of unemploymentbeen so persistent? Analysts offer threealternative explanations: cyclicality,stagnation, and the “end of work.”

The cyclical explanation holds that there’snothing new or mysterious going on;unemployment in America remains so highsimply because the economy is notgrowing quickly enough to put peopleback to work. Paul Krugman is one of theprime advocates of this explanation. As hewrites, “All the facts suggest that highunemployment in America is the result ofinadequate demand—full stop.” FormerOffice of Management and Budget directorPeter Orszag agrees, writing that “the

fundamental impediment to getting joblessAmericans back to work is weak growth.”In the cyclical explanation, an especiallydeep drop in demand like the GreatRecession is bound to be followed by along and frustratingly slow recovery.What America has been experiencingsince 2007, in short, is another case of thebusiness cycle in action, albeit aparticularly painful one.

A second explanation for current hardtimes sees stagnation, not cyclicality, inaction. Stagnation in this context means along-term decline in America’s ability toinnovate and increase productivity.Economist Tyler Cowen articulates thisview in his 2010 book, The GreatStagnation:

We are failing to understand why weare failing. All of these problemshave a single, little noticed rootcause: We have been living off low-hanging fruit for at least threehundred years. … Yet during the lastforty years, that low-hanging fruitstarted disappearing, and we startedpretending it was still there. We havefailed to recognize that we are at atechnological plateau and the treesare more bare than we would like tothink. That’s it. That is what has gonewrong.

To support his view, Cowen cites thedeclining median income of Americanfamilies. Median income is a halfwaypoint; there are as many families making

less than this amount as there are makingmore. The growth of median incomeslowed down significantly at least 30years ago, and actually declined during thefirst decade of this century; the averagefamily in America earned less in 2009than it did in 1999. Cowen attributes thisslowdown to the fact that the economy hasreached a “technological plateau.”

Writing in the Harvard Business Review,Leo Tilman and the Nobel Prize-winningeconomist Edmund Phelps agreed withthis stagnation: “[America’s] dynamism—its ability and proclivity to innovate—hasbrought economic inclusion by creatingnumerous jobs. It has also brought realprosperity—engaging, challenging jobsand careers of self-realization and self-

discovery … [but] dynamism has been indecline over the past decade.”

The stagnation argument doesn’t ignore theGreat Recession, but also doesn’t believethat it’s the principle cause of the currentslow recovery and high joblessness.These woes have a more fundamentalsource: a slowdown in the kinds ofpowerful new ideas that drive economicprogress.

This slowdown pre-dates the GreatRecession. In The Great Stagnation, infact, Cowen maintained that it’s beengoing on since the 1970s, when U.S.productivity growth slowed and themedian income of American familiesstopped rising as quickly as it had in the

past. Cowen, Phelps, and other“stagnationists” hold that only higher ratesof innovation and technical progress willlift the economy out of its currentdoldrums.

A variant on this explanation is not thatAmerica has stagnated, but that othernations such as India and China havebegun to catch up. In a global economy,America businesses and workers can’tearn a premium if they don’t have greaterproductivity than their counterparts inother nations. Technology has eliminatedmany of the barriers of geography andignorance that previously kept capitalistsand consumers from finding the lowestprice inputs and products anywhere in theworld. The result has been a great

equalization in factor prices like wages,raising salaries in developing nations andforcing American labor to compete ondifferent terms. Nobel prize winnerMichael Spence has analyzed thisphenomenon and its implications forconvergence in living standards.

The third explanation for America’scurrent job creation problems flips thestagnation argument on its head, seeing nottoo little recent technological progress,but instead too much. We’ll call this the“end of work” argument, after JeremyRifkin’s 1995 book of the same title. In it,Rifkin laid out a bold and disturbinghypothesis: that “we are entering a newphase in world history—one in whichfewer and fewer workers will be needed

to produce the goods and services for theglobal population.”

Computers caused this important shift. “Inthe years ahead,” Rifkin wrote, “moresophisticated software technologies aregoing to bring civilization ever closer to anear-workerless world. … Today, all …sectors of the economy … areexperiencing technological displacement,forcing millions onto the unemploymentroles.” Coping with this displacement, hewrote, was “likely to be the single mostpressing social issue of the comingcentury.”

The end-of-work argument has been madeby, among many others, economist JohnMaynard Keynes, management theorist

Peter Drucker, and Nobel Prize winnerWassily Leontief, who stated in 1983 that“the role of humans as the most importantfactor of production is bound to diminishin the same way that the role of horses inagricultural production was firstdiminished and then eliminated by theintroduction of tractors.” In his 2009 bookThe Lights in the Tunnel , softwareexecutive Martin Ford agreed, stating that“at some point in the future—it might bemany years or decades from now—machines will be able to do the jobs of alarge percentage of the ‘average’ peoplein our population, and these people willnot be able to find new jobs.” BrianArthur argues that a vast, but largelyinvisible “second economy” alreadyexists in the form of digital automation.

The end-of-work argument is anintuitively appealing one; every time weget cash from an ATM instead of a telleror use an automated kiosk to check in at anairport for a flight, we see evidence thattechnology displaces human labor. Butlow unemployment levels in the UnitedStates throughout the 1980s, ’90s, and firstseven years of the new millennium didmuch to discredit fears of displacement,and it has not been featured in themainstream discussion of today’s joblessrecovery. For example, a 2010 reportpublished by the Federal Reserve Bank ofRichmond, titled “The Rise in Long-TermUnemployment: Potential Causes andImplications,” does not contain the wordscomputer, hardware, software , ortechnology in its text. Working papers

published in 2011 by the InternationalMonetary Fund, titled “New Evidence onCyclical and Structural Sources ofUnemployment” and “Has the GreatRecession Raised U.S. StructuralUnemployment?” are similarly silentabout technology. As technology journalistFarhad Manjoo summarized in the onlinemagazine Slate, “Most economists aren'ttaking these worries very seriously. Theidea that computers might significantlydisrupt human labor markets—and, thus,further weaken the global economy—sofar remains on the fringes.”

Our Goal: Bringing Technology into theDiscussion

We think it’s time to bring this idea into

the mainstream and to pay more attentionto technology’s impact on skills, wages,and employment. We certainly agree that aGreat Recession implies a long recovery,and that current sluggish demand is inlarge part to blame for today’s lack ofjobs. But cyclical weak demand is not thewhole story. The stagnationists are rightthat longer and deeper trends are also atwork. The Great Recession has made themmore visible, but they’ve been going onfor a while.

The stagnationists correctly point out thatmedian income and other importantmeasures of American economic healthstopped growing robustly some time ago,but we disagree with them about why thishas happened. They think it’s because the

pace of technological innovation hasslowed down. We think it’s because thepace has sped up so much that it’s left alot of people behind. Many workers, inshort, are losing the race against themachine.

And it’s not just workers. Technologicalprogress—in particular, improvements incomputer hardware, software, andnetworks—has been so rapid and sosurprising that many present-dayorganizations, institutions, policies, andmindsets are not keeping up. Viewedthrough this lens, the increase inglobalization is not an alternativeexplanation, but rather one of theconsequences of the increased power andubiquity of technology.

So we agree with the end-of-work crowdthat computerization is bringing deepchanges, but we’re not as pessimistic asthey are. We don’t believe in the comingobsolescence of all human workers. Infact, some human skills are more valuablethan ever, even in an age of incrediblypowerful and capable digitaltechnologies. But other skills havebecome worthless, and people who holdthe wrong ones now find that they havelittle to offer employers. They’re losingthe race against the machine, a factreflected in today’s employment statistics.

We wrote this book because we believethat digital technologies are one of themost important driving forces in theeconomy today. They’re transforming the

world of work and are key drivers ofproductivity and growth. Yet their impacton employment is not well understood,and definitely not fully appreciated. Whenpeople talk about jobs in America today,they talk about cyclicality, outsourcing andoffshoring, taxes and regulation, and thewisdom and efficacy of different kinds ofstimulus. We don’t doubt the importanceof all these factors. The economy is acomplex, multifaceted entity.

But there has been relatively little talkabout role of acceleration of technology. Itmay seem paradoxical that faster progresscan hurt wages and jobs for millions ofpeople, but we argue that’s what’s beenhappening. As we’ll show, computers arenow doing many things that used to be the

domain of people only. The pace andscale of this encroachment into humanskills is relatively recent and has profoundeconomic implications. Perhaps the mostimportant of these is that while digitalprogress grows the overall economic pie,it can do so while leaving some people, oreven a lot of them, worse off.

And computers (hardware, software, andnetworks) are only going to get morepowerful and capable in the future, andhave an ever-bigger impact on jobs, skills,and the economy. The root of ourproblems is not that we’re in a GreatRecession, or a Great Stagnation, butrather that we are in the early throes of aGreat Restructuring. Our technologies areracing ahead but many of our skills and

organizations are lagging behind. So it’surgent that we understand thesephenomena, discuss their implications,and come up with strategies that allowhuman workers to race ahead withmachines instead of racing against them.

Here’s how we’ll proceed through the restof this book:

Humanity and Technology on theSecond Half of the Chessboard

Why are computers racing ahead ofworkers now? And what, if anything, canbe done about it? Chapter 2 discussesdigital technology, giving examples of justhow astonishing recent developments havebeen and showing how they have upset

well-established ideas about whatcomputers are and aren’t good at. What’smore, the progress we’ve experiencedaugurs even larger advances in comingyears. We explain the sources of thisprogress, and also its limitations.

Creative Destruction: The Economicsof Accelerating Technology andDisappearing Jobs

Chapter 3 explores the economicimplications of these rapid technologicaladvances and the growing mismatches thatcreate both economic winners and losers.It concentrates on three theories thatexplain how such progress can leave somepeople behind, even as it benefits societyas a whole. There are divergences

between higher-skilled and lower-skilledworkers, between superstars and everyoneelse, and between capital and labor. Wepresent evidence that all three divergencesare taking place.

What Is to Be Done? Prescriptions andRecommendations

Once technical trends and economicprinciples are clear, Chapter 4 considerswhat we can and should do to meet thechallenges of high unemployment andother negative consequences of our currentrace against the machine. We can’t winthat race, especially as computerscontinue to become more powerful andcapable. But we can learn to better racewith machines, using them as allies rather

than adversaries. We discuss ways to putthis principle into practice, concentratingon ways to accelerate organizationalinnovation and enhance human capital.

Conclusion: The Digital Frontier

We conclude in Chapter 5 on an upbeatnote. This might seem odd in a book aboutjobs and the economy written during atime of high unemployment, stagnantwages, and anemic GDP growth. But thisis fundamentally a book about digitaltechnology, and when we look at the fullimpact of computers and networks, nowand in the future, we are very optimisticindeed. These tools are greatly improvingour world and our lives, and will continueto do so. We are strong digital optimists,

and we want to convince you to be one,too.

Chapter 2. Humanityand Technology on theSecond Half of theChessboard

Any sufficiently advanced technology isindistinguishable from magic.

—Arthur C. Clarke, 1962

We used to be pretty confident that weknew the relative strengths andweaknesses of computers vis-à-vishumans. But computers have startedmaking inroads in some unexpected areas.This fact helps us to better understand thepast few turbulent years and the true

impact of digital technologies on jobs.

A good illustration of how much recenttechnology advances have taken us bysurprise comes from comparing acarefully researched book published in2004 with an announcement made in 2010.The book is The New Division of Laborby economists Frank Levy and RichardMurnane. As its title implies, it’s adescription of the comparativecapabilities of computers and humanworkers.

In the book’s second chapter, “WhyPeople Still Matter,” the authors present aspectrum of information-processing tasks.At one end are straightforwardapplications of existing rules. These tasks,

such as performing arithmetic, can beeasily automated. After all, computers aregood at following rules.

At the other end of the complexityspectrum are pattern-recognition taskswhere the rules can’t be inferred. TheNew Division of Labor gives driving intraffic as an example of this type of task,and asserts that it is not automatable:

The … truck driver is processing aconstant stream of [visual, aural, andtactile] information from hisenvironment. … To program thisbehavior we could begin with avideo camera and other sensors tocapture the sensory input. Butexecuting a left turn against oncoming

traffic involves so many factors thatit is hard to imagine discovering theset of rules that can replicate adriver’s behavior. …

Articulating [human] knowledge andembedding it in software for all buthighly structured situations are atpresent enormously difficult tasks. …Computers cannot easily substitutefor humans in [jobs like truckdriving].

The results of the first DARPA GrandChallenge, held in 2004, supported Levyand Murnane’s conclusion. The challengewas to build a driverless vehicle thatcould navigate a 150-mile route throughthe unpopulated Mohave Desert. The

“winning” vehicle couldn’t even make iteight miles into the course and took hoursto go even that far.

In Domain After Domain, ComputersRace Ahead

Just six years later, however, real-worlddriving went from being an example of atask that couldn’t be automated to anexample of one that had. In October of2010, Google announced on its officialblog that it had modified a fleet of ToyotaPriuses to the point that they were fullyautonomous cars, ones that had drivenmore than 1,000 miles on American roadswithout any human involvement at all, andmore than 140,000 miles with only minorinputs from the person behind the wheel.

(To comply with driving laws, Google feltthat it had to have a person sitting behindthe steering wheel at all times).

Levy and Murnane were correct thatautomatic driving on populated roads is anenormously difficult task, and it’s not easyto build a computer that can substitute forhuman perception and pattern matching inthis domain. Not easy, but not impossibleeither—this challenge has largely beenmet.

The Google technologists succeeded notby taking any shortcuts around thechallenges listed by Levy and Murnane,but instead by meeting them head-on. Theyused the staggering amounts of datacollected for Google Maps and Google

Street View to provide as muchinformation as possible about the roadstheir cars were traveling. Their vehiclesalso collected huge volumes of real-timedata using video, radar, and LIDAR (lightdetection and ranging) gear mounted onthe car; these data were fed into softwarethat takes into account the rules of theroad, the presence, trajectory, and likelyidentity of all objects in the vicinity,driving conditions, and so on. Thissoftware controls the car and probablyprovides better awareness, vigilance, andreaction times than any human drivercould. The Google vehicles’ only accidentcame when the driverless car was rear-ended by a car driven by a human driveras it stopped at a traffic light.

None of this is easy. But in a world ofplentiful accurate data, powerful sensors,and massive storage capacity andprocessing power, it is possible. This isthe world we live in now. It’s one wherecomputers improve so quickly that theircapabilities pass from the realm ofscience fiction into the everyday worldnot over the course of a human lifetime, oreven within the span of a professional’scareer, but instead in just a few years.

Levy and Murnane give complexcommunication as another example of ahuman capability that’s very hard formachines to emulate. Complexcommunication entails conversing with ahuman being, especially in situations thatare complicated, emotional, or ambiguous.

Evolution has “programmed” people to dothis effortlessly, but it’s been very hard toprogram computers to do the same.Translating from one human language toanother, for example, has long been a goalof computer science researchers, butprogress has been slow because grammarand vocabulary are so complicated andambiguous.

In January of 2011, however, thetranslation services company Lionbridgeannounced pilot corporate customers forGeoFluent, a technology developed inpartnership with IBM. GeoFluent takeswords written in one language, such as anonline chat message from a customerseeking help with a problem, andtranslates them accurately and

immediately into another language, such asthe one spoken by a customer servicerepresentative in a different country.

GeoFluent is based on statistical machinetranslation software developed at IBM’sThomas J. Watson Research Center. Thissoftware is improved by Lionbridge’sdigital libraries of previous translations.This “translation memory” makesGeoFluent more accurate, particularly forthe kinds of conversations large high-techcompanies are likely to have withcustomers and other parties. One suchcompany tested the quality of GeoFluent’sautomatic translations of online chatmessages. These messages, whichconcerned the company’s products andservices, were sent by Chinese and

Spanish customers to English-speakingemployees. GeoFluent instantly translatedthem, presenting them in the nativelanguage of the receiver. After the chatsession ended, both customers andemployees were asked whether theautomatically translated messages wereuseful—whether they were clear enoughto allow the people to take meaningfulaction. Approximately 90% reported thatthey were. In this case, automatictranslation was good enough for businesspurposes.

The Google driverless car shows how farand how fast digital pattern recognitionabilities have advanced recently.Lionbridge’s GeoFluent shows how muchprogress has been made in computers’

ability to engage in complexcommunication. Another technologydeveloped at IBM’s Watson labs, this oneactually named Watson, shows howpowerful it can be to combine these twoabilities and how far the computers haveadvanced recently into territory thought tobe uniquely human.

Watson is a supercomputer designed toplay the popular game show Jeopardy! inwhich contestants are asked questions on awide variety of topics that are not knownin advance.1 In many cases, thesequestions involve puns and other types ofwordplay. It can be difficult to figure outprecisely what is being asked, or how ananswer should be constructed. PlayingJeopardy! well, in short, requires the

ability to engage in complexcommunication.

The way Watson plays the game alsorequires massive amounts of patternmatching. The supercomputer has beenloaded with hundreds of millions ofunconnected digital documents, includingencyclopedias and other reference works,newspaper stories, and the Bible. When itreceives a question, it immediately goes towork to figure out what is being asked(using algorithms that specialize incomplex communication), then startsquerying all these documents to find andmatch patterns in search of the answer.Watson works with astonishingthoroughness and speed, as IBM researchmanager Eric Brown explained in an

interview:

We start with a single clue, weanalyze the clue, and then we gothrough a candidate generation phase,which actually runs several differentprimary searches, which eachproduce on the order of 50 searchresults. Then, each search result canproduce several candidate answers,and so by the time we’ve generatedall of our candidate answers, wemight have three to five hundredcandidate answers for the clue.

Now, all of these candidate answerscan be processed independently andin parallel, so now they fan out toanswer-scoring analytics [that] score

the answers. Then, we run additionalsearches for the answers to gathermore evidence, and then run deepanalytics on each piece of evidence,so each candidate answer might goand generate 20 pieces of evidenceto support that answer.

Now, all of this evidence can beanalyzed independently and inparallel, so that fans out again. Nowyou have evidence being deeplyanalyzed … and then all of theseanalytics produce scores thatultimately get merged together, usinga machine-learning framework toweight the scores and produce a finalranked order for the candidateanswers, as well as a final

confidence in them. Then, that’s whatcomes out in the end.

What comes out in the end is so fast andaccurate that even the best humanJeopardy! players simply can’t keep up. InFebruary of 2011, Watson played in atelevised tournament against the two mostaccomplished human contestants in theshow’s history. After two rounds of thegame shown over three days, the computerfinished with more than three times asmuch money as its closest flesh-and-bloodcompetitor. One of these competitors, KenJennings, acknowledged that digitaltechnologies had taken over the game ofJeopardy! Underneath his written responseto the tournament’s last question, headded, “I for one welcome our new

computer overlords.”

Moore’s Law and the Second Half ofthe Chessboard

Where did these overlords come from?How has science fiction become businessreality so quickly? Two concepts areessential for understanding thisremarkable progress. The first, and betterknown, is Moore’s Law, which is anexpansion of an observation made byGordon Moore, co-founder ofmicroprocessor maker Intel. In a 1965article in Electronics Magazine, Moorenoted that the number of transistors in aminimum-cost integrated circuit had beendoubling every 12 months, and predictedthat this same rate of improvement would

continue into the future. When this provedto be the case, Moore’s Law was born.

Later modifications changed the timerequired for the doubling to occur; themost widely accepted period at present is18 months. Variations of Moore’s Lawhave been applied to improvement overtime in disk drive capacity, displayresolution, and network bandwidth. Inthese and many other cases of digitalimprovement, doubling happens bothquickly and reliably.

It also seems that software progresses atleast as fast as hardware does, at least insome domains. Computer scientist MartinGrötschel analyzed the speed with whicha standard optimization problem could be

solved by computers over the period1988-2003. He documented a 43millionfold improvement, which he brokedown into two factors: faster processorsand better algorithms embedded insoftware. Processor speeds improved by afactor of 1,000, but these gains weredwarfed by the algorithms, which got43,000 times better over the same period.

The second concept relevant forunderstanding recent computing advancesis closely related to Moore’s Law. Itcomes from an ancient story about mathmade relevant to the present age by theinnovator and futurist Ray Kurzweil. Inone version of the story, the inventor ofthe game of chess shows his creation tohis country’s ruler. The emperor is so

delighted by the game that he allows theinventor to name his own reward. Theclever man asks for a quantity of rice to bedetermined as follows: one grain of rice isplaced on the first square of thechessboard, two grains on the second, fouron the third, and so on, with each squarereceiving twice as many grains as theprevious.

The emperor agrees, thinking that thisreward was too small. He eventually sees,however, that the constant doubling resultsin tremendously large numbers. Theinventor winds up with 264-1 grains ofrice, or a pile bigger than Mount Everest.In some versions of the story the emperoris so displeased at being outsmarted thathe beheads the inventor.

In his 2000 book The Age of SpiritualMachines: When Computers ExceedHuman Intelligence, Kurzweil notes thatthe pile of rice is not that exceptional onthe first half of the chessboard:

After thirty-two squares, the emperorhad given the inventor about 4 billiongrains of rice. That’s a reasonablequantity—about one large field’sworth—and the emperor did start totake notice.

But the emperor could still remain anemperor. And the inventor could stillretain his head. It was as they headedinto the second half of the chessboardthat at least one of them got intotrouble.

Kurzweil’s point is that constant doubling,reflecting exponential growth, isdeceptive because it is initiallyunremarkable. Exponential increasesinitially look a lot like standard linearones, but they’re not. As time goes by—aswe move into the second half of thechessboard—exponential growthconfounds our intuition and expectations.It accelerates far past linear growth,yielding Everest-sized piles of rice andcomputers that can accomplish previouslyimpossible tasks.

So where are we in the history of businessuse of computers? Are we in the secondhalf of the chessboard yet? This is animpossible question to answer precisely,of course, but a reasonable estimate yields

an intriguing conclusion. The U.S. Bureauof Economic Analysis added “InformationTechnology” as a category of businessinvestment in 1958, so let’s use that as ourstarting year. And let’s take the standard18 months as the Moore’s Law doublingperiod. Thirty-two doublings then take usto 2006 and to the second half of thechessboard. Advances like the Googleautonomous car, Watson the Jeopardy!champion supercomputer, and high-qualityinstantaneous machine translation, then,can be seen as the first examples of thekinds of digital innovations we’ll see aswe move further into the second half—into the phase where exponential growthyields jaw-dropping results.

Computing the Economy: The Economic

Power of General PurposeTechnologies

These results will be felt across virtuallyevery task, job, and industry. Suchversatility is a key feature of generalpurpose technologies (GPTs), a termeconomists assign to a small group oftechnological innovations so powerful thatthey interrupt and accelerate the normalmarch of economic progress. Steampower, electricity, and the internalcombustion engine are examples ofprevious GPTs.

It is difficult to overstate their importance.As the economists Timothy Bresnahan andManuel Trajtenberg note:

Whole eras of technical progress and

Whole eras of technical progress andeconomic growth appear to be drivenby … GPTs, [which are]characterized by pervasiveness (theyare used as inputs by manydownstream sectors), inherentpotential for technical improvements,and “innovationalcomplementarities,” meaning that theproductivity of R&D in downstreamsectors increases as a consequenceof innovation in the GPT. Thus, asGPTs improve they spreadthroughout the economy, bringingabout generalized productivity gains.

GPTs, then, not only get better themselvesover time (and as Moore’s Law shows,this is certainly true of computers), theyalso lead to complementary innovations in

the processes, companies, and industriesthat make use of them. They lead, in short,to a cascade of benefits that is both broadand deep.

Computers are the GPT of our era,especially when combined with networksand labeled “information andcommunications technology” (ICT).Economists Susanto Basu and JohnFernald highlight how this GPT allowsdepartures from business as usual.

The availability of cheap ICT capitalallows firms to deploy their otherinputs in radically different andproductivity-enhancing ways. In sodoing, cheap computers andtelecommunications equipment can

foster an ever-expanding sequence ofcomplementary inventions inindustries using ICT.

Note that GPTs don't just benefit their“home” industries. Computers, forexample, increase productivity not only inthe high-tech sector but also in allindustries that purchase and use digitalgear. And these days, that meansessentially all industries; even the leastIT-intensive American sectors likeagriculture and mining are now spendingbillions of dollars each year to digitizethemselves.

Note also the choice of words by Basuand Fernald: computers and networksbring an ever-expanding- set of

opportunities to companies. Digitization,in other words, is not a single projectproviding one-time benefits. Instead, it'san ongoing process of creativedestruction; innovators use both new andestablished technologies to make deepchanges at the level of the task, the job, theprocess, even the organization itself. Andthese changes build and feed on each otherso that the possibilities offered really areconstantly expanding.

This has been the case for as long asbusinesses have been using computers,even when we were still in the front halfof the chessboard. The personal computer,for example, democratized computing inthe early 1980s, putting processing powerin the hands of more and more knowledge

workers. In the mid-1990s two majorinnovations appeared: the World WideWeb and large-scale commercial businesssoftware like enterprise resource planning(ERP) and customer relationshipmanagement (CRM) systems. The formergave companies the ability to tap newmarkets and sales channels, and also madeavailable more of the world’s knowledgethan had ever before been possible; thelatter let firms redesign their processes,monitor and control far-flung operations,and gather and analyze vast amounts ofdata.

These advances don’t expire or fade awayover time. Instead, they get combined withand incorporated into both earlier andlater ones, and benefits keep mounting.

The World Wide Web, for example,became much more useful to people onceGoogle made it easier to search, while anew wave of social, local, and mobileapplications are just emerging. CRMsystems have been extended to smartphones so that salespeople can stayconnected from the road, and tabletcomputers now provide much of thefunctionality of PCs.

The innovations we’re starting to see inthe second half of the chessboard willalso be folded into this ongoing work ofbusiness invention. In fact, they alreadyare. The GeoFluent offering fromLionbridge has brought instantaneousmachine translation to customer serviceinteractions. IBM is working with

Columbia University Medical Center andthe University of Maryland School ofMedicine to adapt Watson to the work ofmedical diagnosis, announcing apartnership in that area with voicerecognition software maker Nuance. Andthe Nevada state legislature directed itsDepartment of Motor Vehicles to come upw i t h regulations covering autonomousvehicles on the state’s roads. Of course,these are only a small sample of themyriad IT-enabled innovations that aretransforming manufacturing, distribution,retailing, media, finance, law, medicine,research, management, marketing, andalmost every other economic sector andbusiness function.

Where People Still Win (at Least for

Now)

Although computers are encroaching intoterritory that used to be occupied bypeople alone, like advanced patternrecognition and complex communication,for now humans still hold the high groundin each of these areas. Experienceddoctors, for example, make diagnoses bycomparing the body of medical knowledgethey’ve accumulated against patients’ labresults and descriptions of symptoms, andalso by employing the advancedsubconscious pattern recognition abilitieswe label “intuition.” (Does this patientseem like they’re holding somethingback? Do they look healthy, or issomething off about their skin tone orenergy level?’) Similarly, the best

therapists, managers, and salespeopleexcel at interacting and communicatingwith others, and their strategies forgathering information and influencingbehavior can be amazingly complex.

But it’s also true, as the examples in thischapter show, that as we move deeper intothe second half of the chessboard,computers are rapidly getting better atboth of these skills. We’re starting to seeevidence that this digital progress isaffecting the business world. A March2011 story by John Markoff in the NewYork Times highlighted how heavilycomputers’ pattern recognition abilitiesare already being exploited by the legalindustry where, according to one estimate,moving from human to digital labor during

the discovery process could let onelawyer do the work of 500.

In January, for example, BlackstoneDiscovery of Palo Alto, Calif.,helped analyze 1.5 milliondocuments for less than $100,000. …

“From a legal staffing viewpoint, itmeans that a lot of people who usedto be allocated to conduct documentreview are no longer able to bebilled out,” said Bill Herr, who as alawyer at a major chemical companyused to muster auditoriums oflawyers to read documents for weekson end. “People get bored, peopleget headaches. Computers don’t.”

The computers seem to be good at

The computers seem to be good attheir new jobs. … Herr … usedediscovery software to reanalyzework his company’s lawyers did inthe 1980s and ’90s. His humancolleagues had been only 60 percentaccurate, he found.“Think about how much money hadbeen spent to be slightly better than acoin toss,” he said.

And an article the same month in the LosAngeles Times by Alena Semuelshighlighted that despite the fact thatclosing a sale often requires complexcommunication, the retail industry hasbeen automating rapidly.

In an industry that employs nearly 1in 10 Americans and has long been a

reliable job generator, companiesincreasingly are looking to peddlemore products with feweremployees. … Virtual assistants aretaking the place of customer servicerepresentatives. Kiosks and self-service machines are reducing theneed for checkout clerks.

Vending machines now sell iPods,bathing suits, gold coins, sunglassesand razors; some will even dispenseprescription drugs and medicalmarijuana to consumers willing tosubmit to a fingerprint scan. Andshoppers are finding information ontouch screen kiosks, rather thantalking to attendants. …

The [machines] cost a fraction of

The [machines] cost a fraction ofbrick-and-mortar stores. They alsoreflect changing consumer buyinghabits. Online shopping has madeAmericans comfortable with the ideaof buying all manner of productswithout the help of a salesman orclerk.

During the Great Recession, nearly 1 in 12people working in sales in America losttheir job, accelerating a trend that hadbegun long before. In 1995, for example,2.08 people were employed in “sales andrelated” occupations for every $1 millionof real GDP generated that year. By 2002(the last year for which consistent data areavailable), that number had fallen to 1.79,a decline of nearly 14 percent.

If, as these examples indicate, both patternrecognition and complex communicationare now so amenable to automation, areany human skills immune? Do peoplehave any sustainable comparativeadvantage as we head ever deeper into thesecond half of the chessboard? In thephysical domain, it seems that we do forthe time being. Humanoid robots are stillquite primitive, with poor fine motorskills and a habit of falling down stairs.So it doesn’t appear that gardeners andrestaurant busboys are in danger of beingreplaced by machines any time soon.

And many physical jobs also requireadvanced mental abilities; plumbers andnurses engage in a great deal of patternrecognition and problem solving

throughout the day, and nurses also do alot of complex communication withcolleagues and patients. The difficulty ofautomating their work reminds us of aquote attributed to a 1965 NASA reportadvocating manned space flight: “Man isthe lowest-cost, 150-pound, nonlinear,all-purpose computer system which can bemass-produced by unskilled labor.”

Even in the domain of pure knowledgework—jobs that don’t have a physicalcomponent—there’s a lot of importantterritory that computers haven’t yet startedto cover. In his 2005 book TheSingularity Is Near: When HumansTranscend Biology, Ray Kurzweilpredicts that future computers will“encompass … the pattern-recognition

powers, problem-solving skills, andemotional and moral intelligence of thehuman brain itself,” but so far only thefirst of these abilities has beendemonstrated. Computers so far haveproved to be great pattern recognizers butlousy general problem solvers; IBM’ssupercomputers, for example, couldn’ttake what they’d learned about chess andapply it to Jeopardy! or any otherchallenge until they were redesigned,reprogrammed, and fed different data bytheir human creators.

And for all their power and speed, today’sdigital machines have shown littlecreative ability. They can’t compose verygood songs, write great novels, orgenerate good ideas for new businesses.

Apparent exceptions here only prove therule. A prankster used an online generatorof abstracts for computer science papersto create a submission that was acceptedfor a technical conference (in fact, theorganizers invited the “author” to chair apanel), but the abstract was simply aseries of somewhat-related technicalterms strung together with a few standardverbal connectors.

Similarly, software that automaticallygenerates summaries of baseball gamesworks well, but this is because muchsports writing is highly formulaic and thusamenable to pattern matching and simplercommunication. Here’s a sample from aprogram called StatsMonkey:

UNIVERSITY PARK — An

UNIVERSITY PARK — Anoutstanding effort by Willie Argocarried the Illini to an 11-5 victoryover the Nittany Lions on Saturday atMedlar Field.

Argo blasted two home runs forIllinois. He went 3-4 in the gamewith five RBIs and two runs scored.

Illini starter Will Strack struggled,allowing five runs in six innings, butthe bullpen allowed only no runs andthe offense banged out 17 hits to pickup the slack and secure the victoryfor the Illini.

The difference between the automaticgeneration of formulaic prose and genuineinsight is still significant, as the history of

a 60-year-old test makes clear. Themathematician and computer sciencepioneer Alan Turing considered thequestion of whether machines could think“too meaningless to deserve discussion,”but in 1950 he proposed a test todetermine how humanlike a machine couldbecome. The “Turing test” involves a testgroup of people having online chats withtwo entities, a human and a computer. Ifthe members of the test group can’t ingeneral tell which entity is the machine,then the machine passes the test.

Turing himself predicted that by 2000computers would be indistinguishablefrom people 70% of the time in his test.However, at the Loebner Prize, an annualTuring test competition held since 1990,

the $25,000 prize for a chat program thatcan persuade half the judges of itshumanity has yet to be awarded. Whateverelse computers may be at present, they arenot yet convincingly human.

But as the examples in this chapter makeclear, computers are now demonstratingskills and abilities that used to belongexclusively to human workers. This trendwill only accelerate as we move deeperinto the second half of the chessboard.What are the economic implications ofthis phenomenon? We’ll turn our attentionto this topic in the next chapter.

1 To be precise, Jeopardy! contestants are

shown answers and must ask questionsthat would yield these answers.

Chapter 3. CreativeDestruction: TheEconomics ofAcceleratingTechnology andDisappearing Jobs

We are being afflicted with a newdisease of which some readers may notyet have heard the name, but of whichthey will hear a great deal in the years tocome—namely, technologicalunemployment. This meansunemployment due to our discovery of

means of economising the use of labouroutrunning the pace at which we can findnew uses for labour.

—John Maynard Keynes, 1930

The individual technologies and thebroader technological accelerationdiscussed in Chapter 2 are creatingenormous value. There is no question thatthey increase productivity, and thus ourcollective wealth. But at the same time,the computer, like all general purposetechnologies, requires parallel innovationin business models, organizationalprocesses structures, institutions, andskills. These intangible assets, comprisingboth organizational and human capital, areoften ignored on companies’ balancesheets and in the official GDP statistics,

but they are no less essential thanhardware and software.

And that’s a problem. Digital technologieschange rapidly, but organizations andskills aren’t keeping pace. As a result,millions of people are being left behind.Their incomes and jobs are beingdestroyed, leaving them worse off inabsolute purchasing power than before thedigital revolution. While the foundation ofour economic system presumes a stronglink between value creation and jobcreation, the Great Recession reveals theweakening or breakage of that link. This isnot merely an artifact of the business cyclebut rather a symptom of a deeperstructural change in the nature ofproduction. As technology accelerates on

the second half of the chessboard, so willthe economic mismatches, underminingour social contract and ultimately hurtingboth rich and poor, not just the first wavesof unemployed.

The economics of technology,productivity, and employment areincreasingly fodder for debate andseemingly filled with paradoxes. How canso much value creation and so mucheconomic misfortune coexist? How cantechnologies accelerate while incomesstagnate? These apparent paradoxes canbe resolved by combining some well-understood economic principles with theobservation that there is a growingmismatch between rapidly advancingdigital technologies and slow-changing

humans.

Growing Productivity

Of the plethora of economic statistics—unemployment, inflation, trade, budgetdeficits, money supply, and so on—one isparamount: productivity growth.Productivity is the amount of output perunit of input. In particular, laborproductivity can be measured as outputper worker or output per hour worked. Inthe long run, productivity growth is almostthe only thing that matters for ensuringrising living standards. Robert Solowearned his Nobel Prize for showing thateconomic growth does not come frompeople working harder but rather fromworking smarter. That means using new

technologies and new techniques ofproduction to create more value withoutincreasing the labor, capital, and otherresources used.

Even a few percentage points of fasterproductivity growth per year can lead tolarge differences in wealth over time. Iflabor productivity grows at 1%, as it didfor much of the 1800s, then it takes about70 years for living standards to double.However, if it grows at 4% per year, as itdid in 2010, then living standards are 16times higher after 70 years. While 4%growth is exceptional, the good news isthat the past decade was a pretty good onefor labor productivity growth—the bestsince the 1960s. The average of over2.5% growth per year is far better than the

1970s and 1980s, and even edges out the1990s (see Figure 3.1). What’s more,there’s now a near consensus amongeconomists about the source of theproductivity surge since the mid-1990s:IT.

Although the official productivitystatistics are encouraging, they are farfrom perfect. They don’t do a very goodjob of accounting for quality, variety,timeliness, customer service, or otherhard-to-measure aspects of output. Whilebushels of wheat and tons of steel are

relatively easy to count, the quality of ateacher’s instruction, the value of morecereal choices in a supermarket, or theability to get money from an ATM 24hours a day is harder to assess.

Compounding this measurement problemis the fact that free digital goods likeFacebook, Wikipedia, and YouTube areessentially invisible to productivitystatistics. As the Internet and mobiletelephony deliver more and more freeservices, and people spend more of theirwaking hours consuming them, this sourceof measurement error becomesincreasingly important. Furthermore, mostgovernment services are simply valued atcost, which implicitly assumes zeroproductivity growth for this entire sector,

regardless of whether true productivity isrising at levels comparable to the rest ofthe economy.

A final source of measurement errorcomes from health care, a particularlylarge and important segment of theeconomy. Health care productivity ispoorly measured and often assumed to bestagnant, yet Americans live on averageabout 10 years longer today than they didin 1960. This is enormously valuable, butit is not counted in our productivity data.According to economist WilliamNordhaus, “to a first approximation, theeconomic value of increases in longevityover the twentieth century is about aslarge as the value of measured growth innon-health goods and services.”

Earlier eras also had significantunmeasured quality components, such asthe welfare gains from telephones, ordisease reductions from antibiotics.Furthermore, there are also areas wherethe productivity statistics overestimategrowth, as when they fail to account forincreases in pollution or when increasedcrime leads people to spend more oncrime-deterring goods and services. Onbalance, the official productivity datalikely underestimate the trueimprovements of our living standards overtime.

Stagnant Median Income

In contrast to labor productivity, medianfamily income has risen only slowly since

the 1970s (Figure 3.2) once the effects ofinflation are taken into account. Asdiscussed in Chapter 1, Tyler Cowen andothers point to this fact as evidence ofeconomy-wide stagnation.

In some ways, Cowen understates hiscase. If you zoom in on the past decadeand focus on working-age households,real median income has actually fallenfrom $60,746 to $55,821. This is the firstdecade to see declining median income

since the figures were first compiled.Median net worth also declined this pastdecade when adjusted for inflation,another first.

Yet at the same time, GDP per person hascontinued to grow fairly steadily (exceptduring recessions). The contrast withmedian income is striking (Figure 3.3).

How can this be? Most of the differencestems from the distinction between themedian and the mean of the distribution.2If 50 construction workers are drinking ata bar and Bill Gates walks in as thepoorest customer walks out, the meanwealth of the customers would soar to $1

billion. However, the wealth of themedian customer, the one exactly in themiddle of the distribution, wouldn’tchange at all.

Something like this has been happening toincomes in the U.S. economy. There havebeen trillions of dollars of wealth createdin recent decades, but most of it went to arelatively small share of the population. Infact, economist Ed Wolff found that over100% of all the wealth increase inAmerica between 1983 and 2009 accruedto the top 20% of households. The otherfour-fifths of the population saw a netdecrease in wealth over nearly 30 years.In turn, the top 5% accounted for over80% of the net increase in wealth and thetop 1% for over 40%. With almost a

fractal-like quality, each successivelyfiner slice at the top of the distributionaccounted for a disproportionately largeshare of the total wealth gains. We havecertainly not increased our GDP in theway that Franklin D. Roosevelt hoped forwhen he said during his second inauguraladdress in 1937, “The test of our progressis not whether we add more to theabundance of those who have much; it iswhether we provide enough for those whohave too little.”

This squares with the evidence fromChapter 2 of the growing performance ofmachines. There has been no stagnation intechnological progress or aggregatewealth creation as is sometimes claimed.Instead, the stagnation of median incomes

primarily reflects a fundamental change inhow the economy apportions income andwealth. The median worker is losing therace against the machine.

Examining other statistics reveals adeeper, more widespread problem. Notonly are income and wages—the price oflabor—suffering, but so is the number ofjobs or the quantity of labor demanded(Figure 3.4). The last decade was the firstdecade since the depths of the GreatDepression that saw no net job creation.

When you consider that the overallpopulation has grown, the lack of jobcreation is even more troubling. Thepopulation of the United States grew by 30million in the past decade, so we wouldneed to create 18 million jobs just to keepthe same share of the population working

as in the year 2000. Instead, we’vecreated virtually none, reducing theemployment to a population ratio fromover 64% to barely 58%.

The lack of jobs is not simply a matter ofmassive layoffs due to the GreatRecession. Instead, it reflects deepstructural issues that have been worseningfor a decade or more. The Bureau ofLabor Statistics Job Openings and LaborTurnover Survey (JOLTS) shows adramatic decrease in hiring since 2000.Lack of hiring, rather than increases inlayoffs, is what accounts for most of thecurrent joblessness. Furthermore, a studyby economists Steven J. Davis, JasonFaberman, and John Haltiwanger foundthat recruiting intensity per job opening

has plummeted during the past decade aswell. Employers just don’t seem to havethe same demand for labor that they oncedid.

This reflects a pattern that was noticeablein the “jobless recovery” of the early1990s, but that has worsened after each ofthe two recessions since then. EconomistsMenzie Chinn and Robert Gordon, inseparate analyses, find that the venerablerelationship between output andemployment known as Okun’s Law hasbeen amended. Historically, increasedoutput meant increased employment, butthe recent recovery created much lessemployment than predicted; GDPrebounded but jobs didn't. The historicallystrong relationship between changes in

GDP and changes in employment appearsto have weakened as digital technologyhas become more pervasive and powerful.As the examples in Chapter 2 make clear,this is not a coincidence.

How Technology Can Destroy Jobs

At least since the followers of Ned Luddsmashed mechanized looms in 1811,workers have worried about automationdestroying jobs. Economists havereassured them that new jobs would becreated even as old ones were eliminated.For over 200 years, the economists wereright. Despite massive automation ofmillions of jobs, more Americans hadjobs at the end of each decade up throughthe end of the 20th century. However, this

empirical fact conceals a dirty secret.There is no economic law that says thateveryone, or even most people,automatically benefit from technologicalprogress.

People with little economics trainingintuitively grasp this point. Theyunderstand that some human workers maylose out in the race against the machine.Ironically, the best-educated economistsare often the most resistant to this idea, asthe standard models of economic growthimplicitly assume that economic growthbenefits all residents of a country.However, just as Nobel Prize-winningeconomist Paul Samuelson showed thatoutsourcing and offshoring do notnecessarily increase the welfare of all

workers, it is also true that technologicalprogress is not a rising tide thatautomatically raises all incomes. Even asoverall wealth increases, there can be,and usually will be, winners and losers.And the losers are not necessarily somesmall segment of the labor force likebuggy whip manufacturers. In principle,they can be a majority or even 90% ormore of the population.

If wages can freely adjust, then the loserskeep their jobs in exchange for acceptingever-lower compensation as technologycontinues to improve. But there’s a limitto this adjustment. Shortly after theLuddites began smashing the machinerythat they thought threatened their jobs, theeconomist David Ricardo, who initially

thought that advances in technology wouldbenefit all, developed an abstract modelthat showed the possibility oftechnological unemployment. The basicidea was that at some point, theequilibrium wages for workers might fallbelow the level needed for subsistence. Arational human would see no point intaking a job at a wage that low, so theworker would go unemployed and thework would be done by a machineinstead.

Of course, this was only an abstractmodel. But in his book A Farewell toAl ms , A Farewell to Alms, economistGregory Clark gives an eerie real-worldexample of this phenomenon in action:

There was a type of employee at thebeginning of the IndustrialRevolution whose job and livelihoodlargely vanished in the earlytwentieth century. This was thehorse. The population of workinghorses actually peaked in Englandlong after the Industrial Revolution,in 1901, when 3.25 million were atwork. Though they had been replacedby rail for long-distance haulage andby steam engines for drivingmachinery, they still plowed fields,hauled wagons and carriages shortdistances, pulled boats on the canals,toiled in the pits, and carried armiesinto battle. But the arrival of theinternal combustion engine in the latenineteenth century rapidly displaced

these workers, so that by 1924 therewere fewer than two million. Therewas always a wage at which allthese horses could have remainedemployed. But that wage was so lowthat it did not pay for their feed.

As technology continues to advance in thesecond half of the chessboard, taking onjobs and tasks that used to belong only tohuman workers, one can imagine a time inthe future when more and more jobs aremore cheaply done by machines thanhumans. And indeed, the wages ofunskilled workers have trendeddownward for over 30 years, at least inthe United States.

We also now understand that

technological unemployment can occureven when wages are still well abovesubsistence if there are downwardrigidities that prevent them from falling asquickly as advances in technology reducethe costs of automation. Minimum wagelaws, unemployment insurance, healthbenefits, prevailing wage laws, and long-term contracts—not to mention custom andpsychology—make it difficult to rapidlyreduce wages.3 Furthermore, employerswill often find wage cuts damaging tomorale. As the efficiency wage literaturenotes, such cuts can be demotivating toemployees and cause companies to losetheir best people.

But complete wage flexibility would beno panacea, either. Ever-falling wages for

significant shares of the workforce is notexactly an appealing solution to the threatof technological employment. Aside fromthe damage it does to the living standardsof the affected workers, lower pay onlypostpones the day of reckoning. Moore’sLaw is not a one-time blip but anaccelerating exponential trend.

The threat of technological unemploymentis real. To understand this threat, we'lldefine three overlapping sets of winnersand losers that technical change creates:(1) high-skilled vs. low-skilled workers,(2) superstars vs. everyone else, and (3)capital vs. labor. Each set has well-documented facts and compelling links todigital technology. What’s more, thesesets are not mutually exclusive. In fact, the

winners in one set are more likely to bewinners in the other two sets as well,which concentrates the consequences.

In each case, economic theory is clear.Even when technological progressincreases productivity and overall wealth,it can also affect the division of rewards,potentially making some people worse offthan they were before the innovation. In agrowing economy, the gains to the winnersmay be larger than the losses of those whoare hurt, but this is a small consolation tothose who come out on the short end of thebargain.

Ultimately, the effects of technology are anempirical question—one that is bestsettled by looking at the data. For all three

sets of winners and losers, the news istroubling. Let’s look at each in turn.

1. High-Skilled vs. Low-Skilled Workers

We’ll start with skill-biased technicalchange, which is perhaps the mostcarefully studied of the three phenomena.This is technical change that increases therelative demand for high-skill labor whilereducing or eliminating the demand forlow-skill labor. A lot of factoryautomation falls into this category, asroutine drudgery is turned over tomachines while more complexprogramming, management, and marketingdecisions remain the purview of humans.

A recent paper by economists Daron

Acemoglu and David Autor highlights thegrowing divergence in earnings betweenthe most-educated and least-educatedworkers. Over the past 40 years, weeklywages for those with a high school degreehave fallen and wages for those with ahigh school degree and some college havestagnated. On the other hand, college-educated workers have seen significantgains, with the biggest gains going to thosewho have completed graduate training(Figure 3.5).

What’s more, this increase in the relativeprice of educated labor—their wages—comes during a period where the supplyof educated workers has also increased.The combination of higher pay in the faceof growing supply points unmistakably to

an increase in the relative demand forskilled labor. Because those with the leasteducation typically already had the lowestwages, this change has increased overallincome inequality.

It’s clear from the chart in Figure 3.5 thatwage divergence accelerated in the digitalera. As documented in careful studies byDavid Autor, Lawrence Katz, and AlanKrueger, as well as Frank Levy andRichard Murnane and many others, the

increase in the relative demand for skilledlabor is closely correlated with advancesin technology, particularly digitaltechnologies. Hence, the moniker “skill-biased technical change,” or SBTC. Thereare two distinct components to recentSBTC. Technologies like robotics,numerically controlled machines,computerized inventory control, andautomatic transcription have beensubstituting for routine tasks, displacingthose workers. Meanwhile othertechnologies like data visualization,analytics, high-speed communications, andrapid prototyping have augmented thecontributions of more abstract and data-driven reasoning, increasing the value ofthose jobs.

Skill-biased technical change has alsobeen important in the past. For most of the19th century, about 25% of all agriculturelabor threshed grain. That job wasautomated in the 1860s. The 20th centurywas marked by an acceleratingmechanization not only of agriculture butalso of factory work. Echoing the firstNobel Prize winner in economics, JanTinbergen, Harvard economists ClaudiaGoldin and Larry Katz described theresulting SBTC as a “race betweeneducation and technology.” Ever-greaterinvestments in education, dramaticallyincreasing the average educational levelof the American workforce, helpedprevent inequality from soaring astechnology automated more and moreunskilled work. While education is

certainly not synonymous with skill, it isone of the most easily measurablecorrelates of skill, so this pattern suggeststhat demand for upskilling has increasedfaster than its supply.

Studies by this book’s co-author ErikBrynjolfsson along with TimothyBresnahan, Lorin Hitt, and Shinku Yangfound that a key aspect of SBTC was notjust the skills of those working withcomputers, but more importantly thebroader changes in work organization thatwere made possible by informationtechnology. The most productive firmsreinvented and reorganized decisionrights, incentives systems, informationflows, hiring systems, and other aspects oforganizational capital to get the most from

the technology. This, in turn, requiredradically different and, generally, higherskill levels in the workforce. It was not somuch that those directly working withcomputers had to be more skilled, butrather that whole production processes,and even industries, were reengineered toexploit powerful new informationtechnologies. What’s more, each dollar ofcomputer hardware was often the catalystfor more than $10 of investment incomplementary organizational capital. Theintangible organizational assets aretypically much harder to change, but theyare also much more important to thesuccess of the organization.

As the 21st century unfolds, automation isaffecting broader swaths of work. Even

the low wages earned by factory workersin China have not insulated them frombeing undercut by new machinery and thecomplementary organizational andinstitutional changes. For instance, TerryGou, the founder and chairman of theelectronics manufacturer Foxconn,announced this year a plan to purchase 1million robots over the next three years toreplace much of his workforce. The robotswill take over routine jobs like sprayingpaint, welding, and basic assembly.Foxconn currently has 10,000 robots, with300,000 expected to be in place by nextyear.

2. Superstars vs. Everyone Else

The second division is between superstars

and everyone else. Many industries arewinner-take-all or winner-take-mostcompetitions, in which a few individualsget the lion’s share of the rewards. Thinkof pop music, professional athletics, andthe market for CEOs. Digital technologiesincrease the size and scope of thesemarkets. These technologies replicate notonly information goods but increasinglybusiness processes as well. As a result,the talents, insights, or decisions of asingle person can now dominate a nationalor even global market. Meanwhile good,but not great, local competitors areincreasingly crowded out of their markets.The superstars in each field can now earnmuch larger rewards than they did inearlier decades.

The effects are evident at the top of theincome distribution. The top 10% of thewage distribution has done much betterthan the rest of the labor force, but evenwithin this group there has been growinginequality. Income has grown faster for thetop 1% than the rest of the top decile. Inturn, the top 0.1% and top 0.01% haveseen their income grow even faster. Thisis not run-of-the-mill skill-biasedtechnical change but rather reflects theunique rewards of superstardom. SherwinRosen, himself a superstar economist, laidout the economics of superstars in aseminal 1981 article. In many markets,consumers are willing to pay a premiumfor the very best. If technology exists for asingle seller to cheaply replicate his orher services, then the top-quality provider

can capture most—or all—of the market.The next-best provider might be almost asgood yet get only a tiny fraction of therevenue.

Technology can convert an ordinarymarket into one that is characterized bysuperstars. Before the era of recordedmusic, the very best singer might havefilled a large concert hall but at mostwould only be able to reach thousands oflisteners over the course of a year. Eachcity might have its own local stars, with afew top performers touring nationally, buteven the best singer in the nation couldreach only a relatively small fraction ofthe potential listening audience. Oncemusic could be recorded and distributedat a very low marginal cost, however, a

small number of top performers couldcapture the majority of revenues in everymarket, from classical music’s Yo-Yo Mato pop’s Lady Gaga.

Economists Robert Frank and Philip Cookdocumented how winner-take-all marketshave proliferated as technologytransformed not only recorded music butalso software, drama, sports, and everyother industry that can be transmitted asdigital bits. This trend has accelerated asmore of the economy is based onsoftware, either implicitly or explicitly.As we discussed in our 2008 HarvardBusiness Review article, digitaltechnologies make it possible to replicatenot only bits but also processes. Forinstance, companies like CVS have

embedded processes like prescriptiondrug ordering into their enterpriseinformation systems. Each time CVSmakes an improvement, it is propagatedacross 4,000 stores nationwide,amplifying its value. As a result, the reachand impact of an executive decision, likehow to organize a process, iscorrespondingly larger.

In fact, the ratio of CEO pay to averageworker pay has increased from 70 in 1990to 300 in 2005, and much of this growth islinked to the greater use of IT, accordingto recent research that Erik did with hisstudent Heekyung Kim. They found thatincreases in the compensation of other topexecutives followed a similar, if lessextreme, pattern. Aided by digital

technologies, entrepreneurs, CEOs,entertainment stars, and financialexecutives have been able to leveragetheir talents across global markets andcapture reward that would have beenunimaginable in earlier times.

To be sure, technology is not the onlyfactor that affects incomes. Politicalfactors, globalization, changes in assetprices, and, in the case of CEOs andfinancial executives, corporategovernance also plays a role. Inparticular, the financial services sectorhas grown dramatically as a share of GDPand even more as a share of profits andcompensation, especially at the top of theincome distribution. While efficientfinance is essential to a modern economy,

it appears that a significant share ofreturns to large human and technologicalinvestments in the past decade, such asthose in sophisticated computerizedprogram trading, were from rentredistribution rather than genuine wealthcreation. Other countries, with differentinstitutions and also slower adoption ofIT, have seen less extreme changes ininequality. But the overall changes in theUnited States have been substantial.According to economist Emmanuel Saez,the top 1% of U.S. households got 65% ofall the growth in the economy since 2002.In fact, Saez reports that the top 0.01% ofhouseholds in the United States—that is,the 14,588 families with income above$11,477,000—saw their share of nationalincome double from 3% to 6% between

1995 and 2007.

3. Capital vs. Labor

The third division is between capital andlabor. Most types of production requireboth machinery and human labor.According to bargaining theory, the wealththey generate is divided according torelative bargaining power, which in turntypically reflects the contribution of eachinput. If the technology decreases therelative importance of human labor in aparticular production process, the ownersof capital equipment will be able tocapture a bigger share of income from thegoods and services produced. To be sure,capital owners are also humans—so it’snot like the wealth disappears from

society—but capital owners are typicallya very different and smaller group than theones doing most of the labor, so thedistribution of income will be affected.

In particular, if technology replaces labor,you might expect that the shares of incomeearned by equipment owners would riserelative to laborers—the classicbargaining battle between capital andlabor.4 This has been happeningincreasingly in recent years. As noted byKathleen Madigan, since the recessionended, real spending on equipment andsoftware has soared by 26% whilepayrolls have remained essentially flat.

Furthermore, there is growing evidencethat capital has captured a growing share

of GDP in recent years. As shown inFigure 3.6, corporate profits have easilysurpassed their pre-recession levels.

According to the recently updated datafrom the U.S. Commerce Department,recent corporate profits accounted for23.8% of total domestic corporateincome, a record high share that is morethan 1 full percentage point above theprevious record. Similarly, corporateprofits as a share of GDP are at 50-year

highs. Meanwhile, compensation to laborin all forms, including wages and benefits,is at a 50-year low. Capital is getting abigger share of the pie, relative to labor.

The recession exacerbated this trend, butit’s part of long-term change in theeconomy. As noted by economists SusanFleck, John Glaser, and Shawn Sprague,the trend line for labor’s share of GDPwas essentially flat between 1974 and1983 but has been falling since then. Whenone thinks about the workers in places likeFoxconn’s factory being replaced bylabor-saving robots, it’s easy to imagine atechnology-driven story for why therelative shares of income might bechanging.

It’s important to note that the “labor”share in the Bureau of Labor Statistics’data includes wages paid to CEOs,finance professionals, professionalathletes, and other “superstars” discussedabove. In this sense, the declining laborshare understates how badly the medianworker has fared. It may also understatethe division of income between capitaland labor, insofar as CEOs and other topexecutives may have bargaining power tocapture some of the “capital’s share” thatwould otherwise accrue to owners ofcommon stock.

Inequality Can Affect the Overall Sizeof the Economy

Technology changes the shares of income

for the skilled vs. unskilled, for superstarsvs. the rest, and for capital vs. labor. Isthis simply a zero-sum game where thelosses of some are exactly offset by gainsto others? Not necessarily. On the positiveside of the ledger, inequality can providebeneficial incentives for skill acquisition,efforts toward superstardom, or capitalaccumulation. However, there are alsoseveral ways it can hurt economic well-being.

First, one of the most basic regularities ofeconomics is the declining marginal utilityof income. A $1,000 windfall is likely toincrease your happiness, or utility, less ifyou already have $10 million than if youonly have $10,000. Second, equality ofopportunity is important to the efficiency

and fairness of a society, even if unequaloutcomes are tolerated or even celebrated.Equality of opportunity, however, can beharder to achieve if children of povertyget inadequate health care, nutrition, oreducation, or people at the bottom areotherwise unable to compete on a level-playing field. Third, inequality inevitablyaffects politics, and this can be damagingand destabilizing. As economist DaronAcemoglu puts it:

Economic power tends to begetpolitical power even in democraticand pluralistic societies. In theUnited States, this tends to workthrough campaign contributions andaccess to politicians that wealth andmoney tend to buy. This political

channel implies another, potentiallymore powerful and distortionary linkbetween inequality and a non-levelplaying field.

Finally, when technology leads torelatively sudden shifts in income betweengroups, it may also dampen overalleconomic growth and potentiallyprecipitate the kind of collapse inaggregate demand reflected in the currentslump.

Consider each of the three sets of winnersand losers discussed earlier. When SBTCincreases the incomes of high-skillworkers and decreases the incomes andemployment of low-skill workers, the neteffect may be a fall in overall demand.

High-skill workers, given extra income,may choose to increase their leisure andsavings rather than work extra hours.Meanwhile, low-skill workers lose theirjobs, go on disability, or otherwise dropout of the labor force. Both groups workless than before, so overall output falls.5

One can tell a similar story for how super-wealthy superstars, given additionalwealth, choose to save most of it whiletheir less-than-stellar competitors have tocut back consumption. Again, overalloutput falls from such a shift. Formersecretary of labor Robert Reich hasargued that such a dynamic was in partresponsible for the Great Depression, andNobel Prize winner Joseph Stiglitz haswritten in detail about how the increasing

concentration of wealth in a relativelysmall group can be corrosive to economicgrowth.

Finally, it’s easy to see how a shift inincome from labor to capital would leadto a similar reduction in overall demand.Capitalists tend to save more of eachmarginal dollar than laborers. In the shortrun, a transfer from laborers to capitalistsreduces total consumption, and thus totalGDP. This phenomenon is summarized ina classic though possibly apocryphalstory: Ford CEO Henry Ford II and UnitedAutomobile Workers president WalterReuther are jointly touring a modern autoplant. Ford jokingly jabs at Reuther:“Walter, how are you going to get theserobots to pay UAW dues?” Not missing a

beat, Reuther responds: “Henry, how areyou going to get them to buy your cars?”

Over time, a well-functioning economyshould be able to adjust to the newconsumption path required by any or all ofthese types of reallocations of income. Forinstance, about 90% of Americans workedin agriculture in 1800; by 1900 it was41%, and by 2000 it was just 2%. Asworkers left farms over the course of twocenturies, there were more than enoughnew jobs created in other sectors, andwhole new industries sprang up to employthem. However, when the changes happenfaster than expectations and/or institutionscan adjust, the transition can becataclysmic. Accelerating technology inthe past decade has disrupted not just one

sector but virtually all of them. A commonway to maintain consumption temporarilyin the face of an adverse shock is toborrow more. While this is feasible in theshort run, and is even rational if the shockis expected to be temporary, it isunsustainable if the trend continues or,worse, grows in magnitude.

Arguably, something like this hashappened in the past decade. Wages formany Americans fell well short ofhistorical growth rates and even fell inreal terms for many groups as technologytransformed their industries. Borrowinghelped mask the problem until the GreatRecession came along. The gradualdemand collapse that might have beenspread over decades was compressed into

a much shorter period, making it harderfor workers to change their skills,entrepreneurs to invent new businessmodels, and managers to make thenecessary adjustments equally quickly.The result has been a dysfunctional seriesof crises. Certainly, much of the recentunemployment is, as past business cycles,simply due to weak demand in the overalleconomy, reflecting an extremely severedownturn. However, this does not negatethe important structural component to thefalling levels of employment, and it isplausible that the Great Recession itselfmay, in part, reflect a delayed response tothese deeper structural issues.

Looking Ahead

As we look ahead, we see these threetrends not only accelerating but alsoevolving. For instance, new research byDavid Autor and David Dorn has put aninteresting twist on the SBTC story. Theyfind that the relationship between skillsand wages has recently become U-shaped.In the most recent decade, demand hasfallen most for those in the middle of theskill distribution. The highest-skilledworkers have done well, but interestinglythose with the lowest skills have sufferedless than those with average skills,reflecting a polarization of labor demand.

This reflects an interesting fact aboutautomation. It can be easier to automatethe work of a bookkeeper, bank teller, orsemi-skilled factory worker than a

gardener, hairdresser, or home healthaide. In particular, over the past 25 years,physical activities that require a degree ofphysical coordination and sensoryperception have proven more resistant toautomation than basic informationprocessing, a phenomenon known asMoravec’s Paradox’. For instance, manytypes of clerical work have beenautomated, and millions of people interactwith robot bank tellers and airport ticketagents each day. More recently, callcenter work—which was widelyoffshored to India, the Philippines, orother low-wage nations in the 1990s—hasincreasingly been replaced by automatedvoice response systems that can recognizean increasingly large domain-specificvocabulary and even complete sentences.

In contrast, vision, fine motor skills, andlocomotion have been much harder toautomate. The human brain can draw onhighly specialized neural circuitry, refinedby millions of years of evolution, torecognize faces, manipulate objects, andwalk through unstructured environments.Although multiplying five-digit numbers isan unnatural and difficult skill for thehuman mind to master, the visual cortexroutinely does far more complexmathematics each time it detects an edgeor uses parallax to locate an object inspace. Machine computation hassurpassed humans in the first task but notyet in the second one.

As digital technologies continue toimprove, we are skeptical that even these

skills will remain bastions of humanexceptionalism in the coming decades.The examples in Chapter 2 of Google’sself-driving car and IBM’s Watson pointto a different path going forward. Thetechnology is rapidly emerging toautomate truck driving in the comingdecade, just as scheduling truck routeswas increasingly automated in the lastdecade. Likewise, the high end of the skillspectrum is also vulnerable, as we see inthe case of ediscovery displacing lawyersand, perhaps, in a Watson-like technology,displacing human medical diagnosticians.

Some Conclusions

Technology has advanced rapidly, and thegood news is that this has radically

increased the economy’s productivecapacity. However, technologicalprogress does not automatically benefiteveryone in a society. In particular,incomes have become more uneven, ashave employment opportunities. Recenttechnological advances have favoredsome skill groups over others, particularly“superstars” in many fields, and probablyalso increased the overall share of GDPaccruing to capital relative to labor.

The stagnation in median income is notbecause of a lack of technologicalprogress. On the contrary, the problem isthat our skills and institutions have notkept up with the rapid changes intechnology. In the 19th and 20th centuries,as each successive wave of automation

eliminated jobs in some sectors andoccupations, entrepreneurs identified newopportunities where labor could beredeployed and workers learned thenecessary skills to succeed. Millions ofpeople left agriculture, but an even largernumber found employment inmanufacturing and services.

In the 21st century, technological changeis both faster and more pervasive. Whilethe steam engine, electric motor, andinternal combustion engine were eachimpressive technologies, they were notsubject to an ongoing level of continuousimprovement anywhere near the pace seenin digital technologies. Already,computers are thousands of times morepowerful than they were 30 years ago, and

all evidence suggests that this pace willcontinue for at least another decade, andprobably more. Furthermore, computersare, in some sense, the “universalmachine” that has applications in almostall industries and tasks. In particular,digital technologies now perform mentaltasks that had been the exclusive domainof humans in the past. General purposecomputers are directly relevant not only tothe 60% of the labor force involved ininformation processing tasks but also tomore and more of the remaining 40%. Asthe technology moves into the second halfof the chessboard, each successivedoubling in power will increase thenumber of applications where it can affectwork and employment. As a result, ourskills and institutions will have to work

harder and harder to keep up lest moreand more of the labor force facestechnological unemployment.

2 The difference also reflects the fact thathouseholds are somewhat smaller nowthan in the past (thus, household incomewill grow less than individual income)and some technical differences betweenthe way GDP and income are calculated.

3 Such wage rigidities have been widelyobserved and lie at the heart of manymacroeconomic models of the businesscycle.

4 The precise economic theory is a bit

more complicated, however. In a well-functioning market, rewards for capital (orlabor) tend to reflect the value of anadditional piece of capital (or additionalw o r ke r ) at the marginthe margin .Depending on how expensive it is toincrease the capital stock, the rewardsearned by capitalists may notautomatically grow with increasedautomation—the predicted effects dependon the exact details of the production,distribution, and governance systems.

5 Economist Arnold Kling describes sucha model on his blog.

Chapter 4. What Is toBe Done? Prescriptionsand Recommendations

The greatest task before civilization atpresent is to make machines what theyought to be, the slaves, instead of themasters of men.

—Havelock Ellis, 1922

In the previous two chapters we showedhow quickly and deeply computers areencroaching into human territory, anddiscussed the economic consequences ofthis phenomenon—how digital progresscan leave some people worse off even as

it improves productivity and grows theoverall pie. Of course, concerns about theinterplay between technology andeconomics are not new. In fact, they’veeven entered American folklore.

The legend of John Henry became popularin the late 19th century as the effects of thesteam-powered Industrial Revolutionaffected every industry and job that reliedheavily on human strength. It’s the story ofa contest between a steam drill and JohnHenry, a powerful railroad worker, to seewhich of the two could bore the longerhole into solid rock.6 Henry wins this raceagainst the machine but loses his life; hisexertions cause his heart to burst. Humansnever directly challenged the steam drillagain.

This legend reflected popular unease atthe time about the potential for technologyto make human labor obsolete. But this isnot at all what happened as the IndustrialRevolution progressed. As steam poweradvanced and spread throughout industry,more human workers were needed, notfewer. They were needed not for their rawphysical strength (as was the case withJohn Henry) but instead for other humanskills: physical ones like locomotion,dexterity, coordination, and perception,and mental ones like communication,pattern matching, and creativity.

The John Henry legend shows us that, inmany contexts, humans will eventuallylose the head-to-head race against themachine. But the broader lesson of the

first Industrial Revolution is more like theIndy 500 than John Henry: economicprogress comes from constant innovationin which people race with machines.Human and machine collaborate togetherin a race to produce more, to capturemarkets, and to beat other teams of humansand machines.

This lesson remains valid and instructivetoday as machines are winning head-to-head mental contests, not just physicalones. Here again, we observe that thingsget really interesting once this contest isover and people start racing withmachines instead of against them.

The game of chess provides a greatexample. In 1997, Gary Kasparov,

humanity’s most brilliant chess master,lost to Deep Blue, a $10 millionspecialized supercomputer programmedby a team from IBM. That was big newswhen it happened, but then developmentsin the world of chess went back to beingreported on and read mainly by chessgeeks. As a result, it’s not well known thatthe best chess player on the planet today isnot a computer. Nor is it a human. Thebest chess player is a team of humansusing computers.

After head-to-head matches betweenhumans and computers becameuninteresting (because the computersalways won), the action moved to“freestyle” competitions, allowing anycombination of people and machines. The

overall winner in a recent freestyletournament had neither the best humanplayers nor the most powerful computers.A s Kasparov writes, it instead consistedof

a pair of amateur American chessplayers using three computers at thesame time. Their skill atmanipulating and “coaching” theircomputers to look very deeply intopositions effectively counteracted thesuperior chess understanding of theirgrandmaster opponents and thegreater computational power of otherparticipants. … Weak human +machine + better process wassuperior to a strong computer aloneand, more remarkably, superior to a

strong human + machine + inferiorprocess.

This pattern is true not only in chess butthroughout the economy. In medicine, law,finance, retailing, manufacturing, and evenscientific discovery, the key to winningthe race is not to compete againstmachines but to compete with machines.As we saw in Chapter 2, while computerswin at routine processing, repetitivearithmetic, and error-free consistency andare quickly getting better at complexcommunication and pattern matching, theylack intuition and creativity and are lostwhen asked to work even a little outside apredefined domain. Fortunately, humansare strongest exactly where computers areweak, creating a potentially beautiful

partnership.

As this partnership advances, we’re nottoo worried about computers holding uptheir end of the bargain. Technologists aredoing an amazing job of making them everfaster, smaller, more energy efficient, andcheaper over time. We are confident thatthese trends will continue even as wemove deeper into the second half of thechessboard.

Digital progress, in fact, is so rapid andrelentless that people and organizationsare having a hard time keeping up. So inthis chapter we want to focus onrecommendations in two areas: improvingthe rate and quality of organizationalinnovation, and increasing human capital

—ensuring that people have the skills theyneed to participate in today’s economy,and tomorrow’s. Making progress in thesetwo areas will be the best way to allowhuman workers and institutions to racewith machines, not against them.

Fostering Organizational Innovation

How can we implement a “race withmachines” strategy? The solution isorganizational innovation: co-inventingnew organizational structures, processes,and business models that leverage ever-advancing technology and human skills.Joseph Schumpeter, the economist,described this as a process of “creativedestruction” and gave entrepreneurs thecentral role in the development and

propagation of the necessary innovations.Entrepreneurs reap rich rewards becausewhat they do, when they do it well, is bothincredibly valuable and far too rare.

To put it another way, the stagnation ofmedian wages and polarization of jobgrowth is an opportunity for creativeentrepreneurs. They can develop newbusiness models that combine the swellingnumbers of mid-skilled workers withever-cheaper technology to create value.There has never been a worse time to becompeting with machines, but there hasnever been a better time to be a talentedentrepreneur.

Entrepreneurial energy in America’s techsector drove the most visible reinvention

of the economy. Google, Facebook,Apple, and Amazon, among others, havecreated hundreds of billions of dollars ofshareholder value by creating whole newproduct categories, ecosystems, and evenindustries. New platforms leveragetechnology to create marketplaces thataddress the employment crisis by bringingtogether machines and human skills in newand unexpected ways:

eBay and Amazon Marketplacespurred over 600,000 people to earntheir livings by dreaming up new,improved, or simply different orcheaper products for a worldwidecustomer base. The Long Tail ofnew products offered enormousconsumer value and is a rapidly

growing segment of the economy.

Apple’s App Store and Google’sAndroid Marketplace make it easyfor people with ideas for mobileapplications to create and distributethem.

Threadless lets people create andsell designs for t-shirts. Amazon’sMechanical Turk makes it easy tofind cheap labor to do a breathtakingarray of simple, well-defined tasks.Kickstarter flips this model on itshead and helps designers andcreative artists find sponsors fortheir projects.

Heartland Robotics providescheap robots-in-a-box that make itpossible for small business peopleto quickly set up their own highlyautomated factory, dramaticallyreducing the costs and increasing theflexibility of manufacturing.

Collectively, these new businessesdirectly create millions of new jobs.7Some of them also create platforms forthousands of other entrepreneurs. None ofthem may ever create billion-dollarbusinesses themselves, but collectivelythey can do more to create jobs andwealth than even the most successful

single venture.

As the great theorist of markets FriedrichHayek noted, some of the most valuableknowledge in an economy is dispersedamong individuals. It is

the knowledge of the particularcircumstances of time and place. …To know of and put to use a machinenot fully employed, or somebody'sskill which could be better utilized,or to be aware of a surplus stockwhich can be drawn upon during aninterruption of supplies, is sociallyquite as useful as the knowledge ofbetter alternative techniques. And theshipper who earns his living fromusing otherwise empty or half-filled

journeys of tramp-steamers, or theestate agent whose whole knowledgeis almost exclusively one oftemporary opportunities, or thearbitrageur who gains from localdifferences of commodity prices, areall performing eminently usefulfunctions based on specialknowledge of circumstances of thefleeting moment not known to others.

Fortunately, digital technologies createenormous opportunities for individuals touse their unique and dispersed knowledgefor the benefit of the whole economy. As aresult, technology enables more and moreopportunities for what Google chiefeconomist Hal Varian calls“micromultinationals”—businesses with

less than a dozen employees that sell tocustomers worldwide and often draw onworldwide supplier and partner networks.While the archetypal 20th-centurymultinational was one of a small numberof megafirms with huge fixed costs andthousands of employees, the comingcentury will give birth to thousands ofsmall multinationals with low fixed costsand a small number of employees each.Both models can conceivably employsimilar numbers of people overall, but thelatter one is likely to be more flexible.

But are there enough opportunities for allthese entrepreneurs? Are we running outof innovations?

When businesses are based on bits instead

of atoms, then each new product adds tothe set of building blocks available to thenext entrepreneur instead of depleting thestock of ideas the way minerals orfarmlands are depleted in the physicalworld. New digital businesses are oftenrecombinations, or mash-ups, of previousones. For example, a student in one of ourclasses at MIT created a simple Facebookapplication for sharing photos. Althoughhe had little formal training inprogramming, he created a robust andprofessional-looking app in a few daysusing standard tools. Within a year he hadover 1 million users. This was possiblebecause his innovation leveraged theFacebook user base, which in turnleveraged the broader World Wide Web,which in turn leveraged the Internet

protocols, which in turn leveraged thecheap computers of Moore's Law andmany other innovations. He could not havecreated value for his million users withoutthe existence of these prior inventions.Because the process of innovation oftenrelies heavily on the combining andrecombining of previous innovations, thebroader and deeper the pool of accessibleideas and individuals, the moreopportunities there are for innovation.

We are in no danger of running out of newcombinations to try. Even if technologyfroze today, we have more possible waysof configuring the different applications,machines, tasks, and distribution channelsto create new processes and products thanwe could ever exhaust.

Here’s a simple proof: suppose the peoplein a small company write down their worktasks— one task per card. If there wereonly 52 tasks in the company, as many asin a standard deck of cards, then therewould be 52! different ways to arrangethese tasks.8 This is far more than thenumber of grains of rice on the second 32squares of a chessboard or even a secondor third full chessboard. Combinatorialexplosion is one of the few mathematicalfunctions that outgrows an exponentialtrend. And that means that combinatorialinnovation is the best way for humaningenuity to stay in the race with Moore’sLaw.

Most of the combinations may be no betterthan what we already have, but some

surely will be, and a few will be “homeruns” that are vast improvements. Thetrick is finding the ones that make apositive difference. Parallelexperimentation by millions ofentrepreneurs is the best and fastest wayto do that. As Thomas Edison once saidwhen trying to find the right combinationof materials for a working lightbulb: “Ihave not failed. I've just found 10,000ways that won't work.” Multiply that by10 million entrepreneurs and you canbegin to see the scale of the economy’sinnovation potential. Most of this potentialremains untapped.

As technology makes it possible for morepeople to start enterprises on a national oreven global scale, more people will be in

the position to earn superstarcompensation. While winner-take-alleconomics can lead to vastlydisproportionate rewards to the topperformer in each market, the key is thatthere is no automatic ceiling to the numberof different markets that can be created. Inprinciple, tens of millions of people couldeach be a leading performer—even the topexpert—in tens of millions of distinct,value-creating fields. Think of them asmicro-experts for macro-markets.Technology scholar Thomas Malone callsthis the age of hyperspecialization. Digitaltechnologies make it possible to scale thatexpertise so that we all benefit from thosetalents and creativity.

Investing in Human Capital

Technology races ahead ever faster as wemove deeper into the second half of thechessboard. To keep up, we need not onlyorganizational innovation, orchestrated byentrepreneurs, but also a second broadstrategy: investments in thecomplementary human capital—theeducation and skills required to get themost out of our racing technology. Smartentrepreneurs can, and will, invent waysto create value by employing even lessskilled workers. However, the messagethe labor market is clearly sending is thatit’s much easier to create value withhighly educated workers.

Unfortunately, our educational progresshas stalled and, as discussed in Chapter 3,this is reflected in stagnating wages and

fewer jobs. The median worker is notkeeping up with cutting-edge technologies.Although the United States once led theworld in the education of its citizens, ithas fallen from first to tenth in the share ofcitizens who are college graduates. Thehigh costs and low performance of theAmerican educational system are classicsymptoms of low productivity in thissector. Despite the importance ofproductivity to overall living standards,and the disproportionate importance ofeducation to productivity, there is far toolittle systematic work done to measure, letalone improve, the productivity ofeducation itself.

It’s not a coincidence that the educationalsector also lags as an adopter of

information technologies. Basicinstructional methods, involving a teacherlecturing to rows of passive students, havechanged little in centuries. As the old jokegoes, it’s a system for transmittinginformation from the notes of the lecturerto the notes of the student without goingthrough the brain of either. In manyclassrooms, the main instructionaltechnology is literally a piece ofyellowish limestone rock scraped across alarger black slate.

The optimistic interpretation is that wehave tremendous upside potential forimprovements in education. As educationbecomes increasingly digitized, educatorscan experiment and track alternativeapproaches, measure and identify what

works, share their findings, and replicatethe best approaches in other subjects andgeographies. This enables a faster pace ofinnovation, leading to furtherimprovements in educational productivity.It also enables unbundling of instruction,evaluation, and certification, whichencourages educational systems to bebased more on delivering genuine,measurable results and less on simplysignaling selection, effort, and prestige.

Furthermore, using IT, both scale andcustomization can be increaseddramatically. A good example is the freeonline course on artificial intelligence atStanford that attracted at least 58,000students. The course uses digital networksto broadcast material and track all

students individually, radically improvingthe productivity of the instructors,lowering costs to students, and, at least inprinciple, delivering a quality product thatwould otherwise be inaccessible to thevast majority of the participants. MIT hasbeen running similar, albeit smaller,classes using a combination of informationand communication technologies for overa decade, most notably in its SystemDesign and Management program.Students at companies around the worlduse a combination of information andcommunication technologies to interactwith professors centrally located at MITand with instructors local to each group ofstudents.

At the K-12 level, Khan Academy offers

over 2,600 short educational videos and144 self-assessment modules for free onthe web. Students can learn at their ownpace, pausing and replaying videos asneeded, earning “badges” to demonstratemastery of various skills and knowledge,and charting their own curricula throughthe ever-growing collection of modules.Students have logged over 70 millionvisits to Khan Academy so far. A growinginfrastructure makes it easy for parents orteachers to track student progress.

An increasingly common approach usesthe Khan Academy’s tools to flip thetraditional classroom model on its head,letting students watch the video lectures athome at their own pace and then havingthem do the “home work” exercises in

class while a teacher circulates amongthem, helping each student individuallywith specific difficulties rather thanproviding a one-size-fits-all lecture to allthe students simultaneously.

Combining videoconferencing, software,and networks with local teachers andtutors has a number of potentialadvantages. The very best “superstar”teachers can be “replicated” viatechnology, giving more students a chanceto learn from them. Furthermore, studentscan learn at their own pace. For instance,software can sense when students arehaving difficulties and need more detail,repetition, and perhaps a slower pace, aswell as when they are quickly grasping thecontent and can be accelerated. Local

human teachers, tutors, and peer tutoringcan easily be incorporated into the systemto provide some of the kinds of value thatthe technology can’t do well, such asemotional support and less-structuredinstruction and assessment.

For instance, creative writing, artsinstruction, and other “soft skills” are notalways as amenable to rule-basedsoftware or distance learning. We concurwith Rhode Island School of Designpresident John Maeda’s vision that amove from STEM (Science, Technology,Engineering, and Mathematics) to STEAM(adding Arts to the mix) is the right visionfor boosting innovation. The technologyand systems for education have to becompatible with that vision.

In particular, softer skills like leadership,team building, and creativity will beincreasingly important. They are the areasleast likely to be automated and most indemand in a dynamic, entrepreneurialeconomy. Conversely, college graduateswho seek the traditional type of job,where someone else tells them what to doeach day, will find themselvesincreasingly in competition withmachines, which excel at followingdetailed instructions.

The Limits to OrganizationalInnovation and Human CapitalInvestment

We’re encouraged by the emergingopportunities to combine digital,

organizational, and human capital tocreate wealth: technology,entrepreneurship, and education are anextraordinarily powerful combination. Butwe want to stress that even thiscombination cannot solve all ourproblems.

First, not everyone can or should be anentrepreneur, and not everyone can orshould spend 16 or more years in school.Second, there are limits to the power ofAmerican entrepreneurship for jobcreation. A 2011 research report for theKauffman Foundation by E. J. Reddy andRobert Litan found that even though thetotal number of new businesses foundedannually in the United States has remainedlargely steady, the total number of people

employed by them at startup has beendeclining in recent years. This could bebecause modern business technology lets acompany start leaner and stay leaner as itgrows.

Third, and most importantly, even whenhumans are racing using machines insteadof against them, there are still winners andlosers as described in Chapter 3. Somepeople, perhaps even a lot, can continue tosee their incomes stagnate or shrink andtheir jobs vanish while overall growthcontinues.

When significant numbers of people seetheir standards of living fall despite anever-growing economic pie, it threatensthe social contract of the economy and

even the social fabric of society. Oneinstinctual response is to simplyredistribute income to those who havebeen hurt. While redistributionameliorates the material costs ofinequality, and that’s not a bad thing, itdoesn’t address the root of the problemsour economy is facing. By itself,redistribution does nothing to makeunemployed workers productive again.Furthermore, the value of gainful work isfar more than the money earned. There isalso the psychological value that almostall people place on doing somethinguseful. Forced idleness is not the same asvoluntary leisure. Franklin D. Rooseveltput this most eloquently:

No country, however rich, can afford

the waste of its human resources.Demoralization caused by vastunemployment is our greatestextravagance. Morally, it is thegreatest menace to our social order.

Thus, we focus our recommendations oncreating ways for everyone to contributeproductively to the economy. Astechnology continues to race ahead, it canwiden the gaps between the swift and theslow on many dimensions. Organizationaland institutional innovations canrecombine human capital with machines tocreate broad-based productivity growth.That’s where we focus ourrecommendations.

Toward an Agenda for Action

By first diagnosing the reason forstagnating median income, we are in aposition to prescribe solutions. Theseinvolve accelerating organizationalinnovation and human capital creation tokeep pace with technology. There are atleast 19 specific steps we can take tothese ends.

Education1 . Invest in education. Start by

simply paying teachers more so thatmore of the best and the brightestsign up for this profession, as theydo in many other nations. Americanteachers make 40% less than theaverage college graduate. Teachersare some of America’s mostimportant wealth creators.

Increasing the quantity and quality ofskilled labor provides a double winby boosting economic growth andreducing income inequality.

2 . Hold teachers accountable forperformance by, for example,eliminating tenure. This should bepart of the bargain for higher pay.

3. Separate student instruction fromtesting and certification. Focusschooling more on verifiableoutcomes and measurableperformance and less on signalingtime, effort or prestige.

4 . Keep K-12 students inclassrooms for more hours. Onereason American students lag behindinternational competitors is theysimply receive about one month lessinstruction per year.

5 . Increase the ratio of skilledworkers in the United States byencouraging skilled immigrants.Offer green cards to foreign studentswhen they complete advanceddegrees, especially in science andengineering subjects at approveduniversities. Expand the H-1B visaprogram. Skilled workers in

America often create more valuewhen working with other skilledworkers. Bringing them together canincrease worldwide innovation andgrowth.

Entrepreneurship

6 . Teach entrepreneurship as askill not just in elite businessschools but throughout highereducation. Foster a broader class ofmid-tech, middle-classentrepreneurs by training them in thefundamentals of business creationand management.

7 . Boost entrepreneurship inAmerican by creating a category offounders’ visas for entrepreneurs,like those in Canada and othercountries.

8 . Create clearinghouses anddatabases to facilitate the creationand dissemination of templates fornew businesses. A set ofstandardized packages for startupscan smooth the path for newentrepreneurs in many industries.These can range from franchiseopportunities to digital “cookbooks”that provide the skeleton structurefor an operation. Job training shouldbe supplemented with

entrepreneurship guidance as thenature of work evolves.

9 . Aggressively lower thegovernmental barriers to businesscreation. In too many industries,elaborate regulatory approvals areneeded from multiple agencies atmultiple levels of government.These too often have the implicitgoal of preserving rents of existingbusiness owners at the expense ofnew businesses and their employees.

Investment

10. Invest to upgrade the country’s

communications and transportationinfrastructure. The AmericanSociety of Civil Engineers gives agrade of D to our overallinfrastructure at present. Improvingit will bring productivity benefits byfacilitating flow and mixing ideas,people, and technologies. It willalso put many people to workdirectly. You don’t have to be anardent Keynesian to believe that thebest time to make these investmentsis when there is plenty of slack inthe labor market.

1 1 . Increase funding for basicresearch and for our preeminentgovernment R&D institutions

including the National ScienceFoundation, the National Institutes ofHealth, and the Defense AdvancedResearch Projects Agency(DARPA) with a renewed focus onintangible assets and businessinnovation. Like other forms ofbasic research, these investmentsare often underfunded by privateinvestors because of the spilloversthey create.

Laws, Regulations, and Taxes

12. Preserve the relative flexibilityof American labor markets byresisting efforts to regulate hiringand firing. Banning layoffs

paradoxically can loweremployment by making it riskier forfirms to hire in the first place,especially if they are experimentingwith new products or businessmodels.

1 3 . Make it comparatively moreattractive to hire a person than tobuy more technology. This can bedone by, among other things,decreasing employer payroll taxesand providing subsidies or taxbreaks for employing people whohave been out of work for a longt i me . Taxes on congestion andpollution can more than make up forthe reduced labor taxes.

14. Decouple benefits from jobs toincrease flexibility and dynamism.Tying health care and othermandated benefits to jobs makes itharder for people to move to newjobs or to quit and start newbusinesses. For instance, many apotential entrepreneur has beenblocked by the need to maintainhealth insurance. Denmark and theNetherlands have led the way here.

1 5 . Don’t rush to regulate newnetwork businesses. Some observersfeel that “crowdsourcing”businesses like Amazon’s

Mechanical Turk exploit theirmembers, who should therefore bebetter protected. However,especially in this early,experimental period, the developersof these innovative platforms shouldbe given maximum freedom toinnovate and experiment, and theirmembers’ freely made decisions toparticipate should be honored, notoverturned.

1 6 . Eliminate or reduce themassive home mortgage subsidy.This costs over $130 billion peryear, which would do much morefor growth if allocated to researchor education. While home

ownership has many laudablebenefits, it likely reduces labormobility and economic flexibility,which conflicts with the economy’sincreased need for flexibility.

17. Reduce the large implicit andexplicit subsidies to financialservices. This sector attracts adisproportionate number of the bestand the brightest minds andtechnologies, in part because thegovernment effectively guarantees“too big to fail” institutions.

18. Reform the patent system. Notonly does it take years to issue good

patents due to the backlog andshortage of qualified examiners, buttoo many low-quality patents areissued, clogging our courts. As aresult, patent trolls are chillinginnovation rather than encouragingit.

1 9 . Shorten, rather than lengthen,copyright periods and increase theflexibility of fair use. Copyrightcovers too much digital content.Rather than encouraging innovation,as specified in the Constitution,excessive restrictions like the SonnyBono Copyright Term Extension Actinhibit mixing and matching ofcontent and using it creatively in

new ways.

These suggestions are only the tip of theiceberg of a broader transformation thatwe need to support, not only to mitigatetechnological unemployment andinequality, but also to fulfill the potentialfor new technologies to grow the economyand create broad-based value. We are notputting forth a complete blueprint for thenext economy—that task is inherentlyimpossible. Instead, we seek to initiate aconversation. That conversation will besuccessful if we accurately diagnose themismatch between acceleratingtechnologies and stagnant organizationsand skills. Successful economies in the21st century will be those that develop the

best ways to foster organizationalinnovation and skill development, and weinvite our readers to contribute to thatagenda.

6 Railroad construction crews of the timeblasted tunnels though mountainsides bydrilling holes into the rock, packing theholes with explosives, and detonatingthem.

7 It’s also worth noting that some valuableenterprises need not create paying jobs togenerate value in the economy. Forinstance, Wikipedia thrives on a modelthat is largely separate from the financialeconomy, but it nonetheless provides

rewards and value. Judging by therevealed preferences of participants,Wikipedia provides sufficient non-monetary rewards to attract millions ofcontributors with diverse talents andexpertise who create tremendous value.When thinking about the evolvingeconomy, we need to remember thatAbraham Maslow’s hierarchy ofneedsaham Maslow’s hierarchy of needsextends beyond material things.

8 52! is shorthand for 52 x 51 x 50 x … x2 x 1, which multiplies to over 8.06 x1067. That is about the number of atoms inour galaxy.

Chapter 5. Conclusion:The Digital Frontier

Technology is a gift of God. After thegift of life it is perhaps the greatest of

God's gifts. It is the mother ofcivilizations, of arts and of sciences.

—Freeman Dyson, 1988

In this book we’ve concentrated on howour increasingly powerful digitaltechnologies affect skills, jobs, and thedemand for human labor. We’ve stressedthat computers are rapidly encroachinginto areas that used to be the domain ofpeople only, like complex communicationand advanced pattern recognition. And

we’ve shown how this encroachment cancause companies to use more computersand fewer people in a growing set oftasks.

We see cause for concern with thisphenomenon because we believe that onesign of a healthy economy is its ability toprovide jobs for all the people who wantto work. As we’ve shown, there’s goodreason to believe that ever-more powerfulcomputers have for some time now beensubstituting for human skills and workersand slowing median incomes and jobgrowth in the United States. As we headdeeper into the second half of thechessboard—into the period wherecontinuing exponential increases incomputing power yield astonishing results

—we expect that economic disruptionswill only grow as well.

We’ve documented our concerns here andsuggested policy changes and otherinterventions to address them. But weclearly are not pessimists abouttechnology and its impacts. In fact, thiswas originally going to be a book aboutall the benefits modern digitaltechnologies have brought to the world.We planned to call it The DigitalFrontier, since the image that keepsoccurring to us is one of a huge amount ofnew territory opening up because oftechnological improvement andinnovation.

This image first occurred to us as we

were conducting research to understandthe impact of digital technology oncompetition in all U.S. industries. Wefound that the more technology an industryhad, the more intense competition within itbecame. In particular, performance gapsgot bigger. The difference in, say, profitmargin between the top and bottomcompanies got a lot larger. This findingimplies that some companies—the topperformers—were racing ahead of the restto explore and exploit new technology-enabled business models. They werehomesteading on a digital frontier,opening up new territory that others wouldeventually settle in.

To build on this research we startedcollecting examples of digital pioneers

and cutting-edge practices, and assembleda group of students and colleagues tobrainstorm and do research with us. Wecalled ourselves the “Digital Frontierteam.”

We changed course with this bookbecause the more we looked, the more webecame convinced of two things. First,that the issue of technology’s impact onemployment was a particularly importantone. The Great Recession and the pace oftechnical progress have combined to makejobs a critical issue at this time, which isa difficult one for many people. When wethink about someone trying to acquirevaluable skills and enter or reenter theworkforce now, we’re reminded of theold Chinese curse, “May you live in

interesting times.”

Second, we saw that very few otherpeople were focusing on the issues raisedhere. When discussing jobs andunemployment, there has been a great dealof attention paid to issues like weakdemand, outsourcing, and labor mobilitybut relatively little attention given totechnology’s role. We felt that this was aserious omission and wanted to correct it.We wanted to show how far and fasttechnology has raced ahead recently, andhighlight that current perspectives andpolicies will need to change to keep upwith it.

But even after writing this book, we stillfirmly believe in the promise of the digital

frontier. Technology has already openedup a huge amount of rich new territory andwill keep doing so. Around the world,economies, societies, and people’s liveshave been improved by digital goods andhigh-tech products; these happy trendswill continue, and likely accelerate.

So we want to conclude this book with aglimpse of the emerging digital frontier—a brief look at some of the benefitsbrought by the still-unfolding computerrevolution. These benefits arise from theconstant improvements summarized byMoore’s Law and discussed above inChapter 2, and also because of thecharacteristics of information itself.

A World of Benefits

Information doesn’t get used up even whenit’s consumed. If Erik eats a meal, Andycan’t eat it, too, but Erik can absolutelyhand a book to Andy once he’s finished,and the book itself (unless Erik has spilledcoffee on it) is not in any way diminishedfor Andy. In fact, it’s probably morevaluable to him after Erik’s done with it,because then they both have its contents intheir heads and can use the information tocollaboratively generate new ideas.

And once a book or other body ofinformation is digitized, even morepossibilities open up. It can be copiedinfinitely and perfectly, and distributedaround the world instantly and at noadditional cost. This is nothing like theeconomics of traditional goods and

services that are the primary focus ofstandard textbooks. It can be a nightmarefor some copyright holders, but it’s greatfor most people. The two of us, forexample, want as many people as possibleto get a copy of this ebook as soon aspossible after we’re done writing it.Thanks to ebook platforms and theInternet, we can accomplish this vision. Inthe previous world of paper-only books,publication and distribution could take ayear, and sales would be limited by thephysical availability of copies of thebook. The digital frontier has eliminatedthis limitation and collapsed timelines.

The economics of digital information, inshort, are the economics not of scarcitybut of abundance. This is a fundamental

shift, and a fundamentally beneficial one.To take just one example, the Internet isnow the largest repository of informationthat has ever existed in the history ofhumankind. It is also a fast, efficient, andcheap worldwide distribution network forall this information. Finally, it is open andaccessible so that more and more peoplecan join it, access all of its ideas, andcontribute their own.

This is incalculably valuable, and groundsfor great optimism, even if some thingslook bleak right now, for computers aremachines that help with ideas, andeconomies run on ideas. As economistPaul Romer writes:

Every generation has perceived the

limits to growth that finite resourcesand undesirable side effects wouldpose if no new … ideas werediscovered. And every generationhas underestimated the potential forfinding new … ideas. Weconsistently fail to grasp how manyideas remain to be discovered. …Possibilities do not merely add up;they multiply.

It might seem as if we’re short on big newideas at present, but this is almostcertainly an illusion. As David Leonhardtnotes, when Bill Clinton assembled thetop minds of the nation to discuss theeconomy in 1992, no one mentioned theInternet.

Romer also makes the point that “perhapsthe most important ideas of all are meta-ideas—ideas about how to support theproduction and transmission of otherideas.” The digital frontier is just such ameta-idea—it generates more ideas andshares them better than anything elsewe’ve ever come up with. So either ahuge amount of basic thinking abouteconomics and growth is wrong, or abumper crop of useful innovations willgrow on this frontier. We’re betting on thelatter possibility.

At a less abstract and more personallevel, the digital frontier is also improvingour lives. If you have Internet access and aconnected device today, it’s both free andeasy to keep in touch with the people who

mean something to you—your kith and kin—even as you and they move around. Youcan use resources like Skype, Facebook,and Twitter to send messages, make voiceand video calls, share still and movingpictures, and let everyone know whatyou’re doing and how you’re doing. Asany lover or grandparent will tell you,these are not trivial capabilities; they’repriceless ones.

Many of us use these resources so oftennow that we take them for granted, butthey’re all less than 10 years old. Thedigital frontier of 2001 was already wide,but it’s gotten immeasurably bigger overthe past decade, and enriched our lives asit has done so.

We see this same phenomenoneverywhere we look. The developingworld, for example, has been transformedby mobile telephones. We in richcountries long ago forgot what it’s like tohave to live in isolation—to have no easyway to communicate farther than ourvoices or bodies could travel. But suchisolation was the sad reality for billionsof people around the world, until mobiletelephony came along.

Once it did, the results were breathtaking.A wonderful study by economist RobertJensen found, for example, that as soon asmobile telephones became available in thefishing regions of Kerala, India, the priceof sardines dropped and stabilized, yetfishermen’s profits actually went up. This

happened because fishermen for the firsttime had access to real-time price anddemand information from the markets onland, which they used to make decisionsthat completely eliminated waste. Resultslike these help explain why there weremore than 3.8 billion mobile phonesubscriptions in the developing world bylate 2010, and why The Economistmagazine wrote that “their spread in poorcountries is not just reshaping the industry—it is changing the world.”

As digital technologies make markets andbusinesses more efficient, they benefit allof us as consumers. As they increasegovernment transparency andaccountability and give us new ways toassemble and make our voices heard, they

benefit us as citizens. And as they put us intouch with ideas, knowledge, friends, andloved ones, they benefit us as humanbeings.

So as we observe the opening up of thedigital frontier, we are hugely optimistic.History has witnessed three industrialrevolutions, each associated with ageneral purpose technology. The first,powered by steam, changed the world somuch that according to historian IanMorris, it “made mockery of all that hadgone before.” It allowed huge andunprecedented increases in population,social development, and standards ofliving. The second, based on electricity,allowed these beneficial trends tocontinue and led to a sharp acceleration of

productivity in the 20th century. In eachcase there were disruptions and crises, butin the end, the mass of humanity wasimmensely better off than before.

The third industrial revolution, which isunfolding now, is fuelled by computersand networks. Like both of the previousones, it will take decades to fully play out.And like each of the first two, it will leadto sharp changes in the path of humandevelopment and history. The twists anddisruptions will not always be easy tonavigate. But we are confident that most ofthese changes will be beneficial ones, andthat we and our world will prosper on thedigital frontier.

Acknowledgments

We’ve been talking about the ideas thatwent into this book for a while now, andwith a lot of people. We found a fantasticgroup of colleagues in the Digital Frontierteam we assembled—a group of studentsand researchers at MIT who volunteered alot of their time over the course of a yearto talk with us, hunt down facts, figures,and examples, and brainstorm about whatwas going on between technology and theeconomy. Team members includedWhitney Braunstein, Claire Calméjane,Greg Gimpel, Tong Li, Liron Wand,George Westerman, and Lynn Wu. We’reextremely grateful to them. In addition,

Mona Masghati and Maya Bustan helpedAndy a great deal with his research, andHeekyung Kim and Jonathan Sidi did thesame for Erik.

We are grateful for conversations ontechnology and employment with our MITcolleagues, including Daron Acemoglu,David Autor, Frank Levy, TodLoofbourrow, Thomas Malone, StuartMadnick, Wanda Orlikowski, MichaelSchrage, Peter Weill, and IrvingWladawsky-Berger. In addition, RobAtkinson, Yannis Bakos, Susanto Basu,Menzie Chinn, Robert Gordon, Lorin Hitt,Rob Huckman, Michael Mandel, DanSnow, Zeynep Ton and Marshall vanAlstyne were very generous with theirinsights. We also benefited greatly from

talking with people in industry who aremaking and using incredible technologies,including Rod Brooks, Paul Hofmann RayKurzweil, Ike Nassi, and Hal Varian.

We presented some of the ideas containedhere to seminar audiences at MIT,Harvard Business School, Northwestern,NYU, UC/Irvine, USC’s AnnenbergSchool, SAP, McKinsey, and theInformation Technology and InnovationFoundation. We also presented relatedwork at conferences including WISE,ICIS, Techonomy, and the Aspen IdeasFestival. We received invaluablefeedback at each of these sessions. Themost important thing we learned was thatthe topic of technology’s influence onemployment immediately engages

people’s interest, so we shaped ourresearch and writing accordingly.Techonomy organizer David Kirkpatrickhas been especially keen to start a high-level conversation about computers,robots, and jobs.

We imposed on a small group of people toread early drafts of the manuscript. MarthaPavlakis, Anna Ivey, George Westerman,David McAfee, Nancy Haller, JohnBrowning, Carol Franco, and Jeff Kehoeall obliged and sharpened up our ideasand our prose. Andrea and Dana Meyer ofWorking Knowledge honed them bothfurther while Jody Berman providedimpeccable copyediting and proofreading.We are grateful to Greg Leutenberg for thecover design.

T h e MIT Center for Digital Business(CDB) has been the ideal home fromwhich to conduct this work, and we’reparticularly grateful to our colleagueDavid Verrill, executive director of theCDB. David makes the place runbeautifully; he’ll be the last person everreplaced by a machine.

We claim sole ownership of virtuallynone of the ideas presented here, butwe’re emphatic that all the mistakes are100% ours.

Authors

Erik Brynjolfsson is a professor at theMIT Sloan School of Management,Director of the MIT Center for DigitalBusiness, Chairman of the SloanManagement Review, a researchassociate at the National Bureau ofEconomic Research, and co-author ofWired for Innovation: How IT IsReshaping the Economy. He graduatedfrom Harvard University and MIT.

Andrew McAfee is a principal researchscientist and associate director at the MITCenter for Digital Business at the SloanSchool of Management. He is the author ofEnterprise 2.0: New Collaborative Tools

for Your Organization’s ToughestChallenges. He graduated from MIT andHarvard University.