Sunteți pe pagina 1din 263
ERIKBRYNJOLFSSONANDREWMCAFEE

ERIKBRYNJOLFSSONANDREWMCAFEE

ERIKBRYNJOLFSSONANDREWMCAFEE

ToMarthaPavlakis,theloveofmylife.

Tomyparents,DavidMcAfeeandNancyHaller,whopreparedmeforthe

secondmachineagebygivingmeeveryadvantageapersoncouldhave.

Chapter1THEBIGSTORIES Chapter2THESKILLSOFTHENEWMACHINES:TECHNOLOGY RACESAHEAD Chapter3MOORE’SLAWANDTHESECONDHALFOFTHE

Chapter1THEBIGSTORIES

Chapter2THESKILLSOFTHENEWMACHINES:TECHNOLOGY

RACESAHEAD

Chapter3MOORE’SLAWANDTHESECONDHALFOFTHE

CHESSBOARD

Chapter4THEDIGITIZATIONOFJUSTABOUTEVERYTHING

Chapter5INNOVATION:DECLININGORRECOMBINING?

Chapter6ARTIFICIALANDHUMANINTELLIGENCEINTHESECOND

MACHINEAGE

Chapter7COMPUTINGBOUNTY

Chapter8BEYONDGDP

Chapter9THESPREAD

Chapter10THEBIGGESTWINNERS:STARSANDSUPERSTARS

Chapter11IMPLICATIONSOFTHEBOUNTYANDTHESPREAD

Chapter12LEARNINGTORACEWITHMACHINES:

RECOMMENDATIONSFORINDIVIDUALS

Chapter13POLICYRECOMMENDATIONS

Chapter14LONG-TERMRECOMMENDATIONS

Chapter15TECHNOLOGYANDTHEFUTURE

(WhichIsVeryDifferentfrom“TechnologyIstheFuture”)

Acknowledgments

Notes

IllustrationSources

Index

“TechnologyisagiftofGod.AfterthegiftoflifeitisperhapsthegreatestofGod’sgifts.It

“TechnologyisagiftofGod.AfterthegiftoflifeitisperhapsthegreatestofGod’sgifts.It

isthemotherofcivilizations,ofartsandofsciences.”

—FreemanDyson

WHATHAVEBEENTHEmostimportantdevelopmentsinhumanhistory? Asanyoneinvestigatingthisquestionsoonlearns,it’sdifficulttoanswer.For one thing, when does ‘human history’ even begin? Anatomically and behaviorallymodernHomosapiens,equippedwithlanguage,fannedoutfrom theirAfricanhomelandsomesixtythousandyearsago. 1 By25,000BCE 2 they hadwipedouttheNeanderthalsandotherhominids,andthereafterfacedno competitionfromotherbig-brained,upright-walkingspecies.

Wemightconsider25,000BCEareasonabletimetostarttrackingthebig

storiesofhumankind,wereitnotforthedevelopment-retardingiceageearth was experiencing at the time. 3 In his book Why the West Rules—For Now, anthropologist Ian Morris starts tracking human societal progress in 14,000 BCE,whentheworldclearlystartedgettingwarmer. Anotherreasonit’sahardquestiontoansweristhatit’snotclearwhatcriteria weshoulduse:whatconstitutesatrulyimportantdevelopment?Mostofusshare asensethatitwouldbeaneventoradvancethatsignificantlychangesthecourse ofthings—onethat‘bendsthecurve’ofhumanhistory.Manyhavearguedthat thedomesticationofanimalsdidjustthis,andisoneofourearliestimportant achievements. The dog might well have been domesticated before 14,000 BCE, but the horsewasnot;eightthousandmoreyearswouldpassbeforewestartedbreeding themandkeepingthemincorrals.Theox,too,hadbeentamedbythattime(ca.

6,000BCE)andhitchedtoaplow.Domesticationofworkanimalshastenedthe

transitionfromforagingtofarming,animportantdevelopmentalreadyunderway by8,000BCE. 4 Agricultureensuresplentifulandreliablefoodsources,whichinturnenable largerhumansettlementsand,eventually,cities.Citiesinturnmaketempting targetsforplunderandconquest.Alistofimportanthumandevelopmentsshould thereforeincludegreatwarsandtheempirestheyyielded.TheMongol,Roman, Arab, and Ottoman empires—to name just four—were transformative; they affectedkingdoms,commerce,andcustomsoverimmenseareas. Ofcourse,someimportantdevelopmentshavenothingtodowithanimals, plants,orfightingmen;somearesimplyideas.PhilosopherKarlJaspersnotes

thatBuddha(563–483BCE),Confucius(551–479BCE),andSocrates(469–399

BCE)alllived quitecloseto oneanotherin time(butnotinplace).In his

analysisthesemenarethecentralthinkersofan‘AxialAge’spanning800–200

BCE. Jaspers calls this age “a deep breath bringing the most lucid consciousness”andholdsthatitsphilosophersbroughttransformativeschoolsof thoughttothreemajorcivilizations:Indian,Chinese,andEuropean. 5 TheBuddhaalsofoundedoneoftheworld’smajorreligions,andcommon sense demands that any list of major human developments include the establishmentofothermajorfaithslikeHinduism,Judaism,Christianity,and Islam. Each has influenced the lives and ideals of hundreds of millions of people. 6 Manyofthesereligions’ideasandrevelationswerespreadbythewritten word, itself a fundamental innovation in human history. Debate rages about preciselywhen,where,andhowwritingwasinvented,butasafeestimateputsit

inMesopotamiaaround3,200BCE.Writtensymbolstofacilitatecountingalso

existedthen,buttheydidnotincludetheconceptofzero,asbasicasthatseems tousnow.Themodernnumberingsystem,whichwecallArabic,arrivedaround 830CE. 7 Thelistofimportantdevelopmentsgoesonandon.TheAtheniansbeganto practice democracy around 500 BCE. The Black Death reduced Europe’s

populationbyatleast30percentduringthelatterhalfofthe1300s.Columbus

sailedtheoceanbluein1492,beginninginteractionsbetweentheNewWorld

andtheOldthatwouldtransformboth.

TheHistoryofHumanityinOneGraph

Howcanweevergetclarityaboutwhichofthesedevelopmentsisthemost important?Allofthecandidateslistedabovehavepassionateadvocates—people whoargueforcefullyandpersuasivelyforonedevelopment’ssovereigntyover alltheothers.AndinWhytheWestRules—ForNowMorrisconfrontsamore fundamentaldebate:whetheranyattempttorankorcomparehumaneventsand developmentsismeaningfulorlegitimate.Manyanthropologistsandothersocial scientistssayitisnot.Morrisdisagrees,andhisbookboldlyattemptstoquantify human development. As he writes, “reducing the ocean of facts to simple numericalscoreshasdrawbacksbutitalsohastheonegreatmeritofforcing everyone to confront the same evidence—with surprising results.” 8 In other words,ifwewanttoknowwhichdevelopmentsbentthecurveofhumanhistory, itmakessensetotrytodrawthatcurve. Morrishasdonethoughtfulandcarefulworktoquantifywhathetermssocial

development (“a group’s ability to master its physical and intellectual environmenttogetthingsdone”)overtime.*AsMorrissuggests,theresultsare surprising.Infact,they’reastonishing.Theyshowthatnoneofthedevelopments discussedsofarhasmatteredverymuch,atleastincomparisontosomething else—somethingthatbentthecurveofhumanhistorylikenothingbeforeor since.Here’sthegraph,withtotalworldwidehumanpopulationgraphedover timealongwithsocialdevelopment;asyoucansee,thetwolinesarenearly identical:

FIGURE1.1NumericallySpeaking,MostofHumanHistoryIsBoring.

FIGURE1.1 NumericallySpeaking,MostofHumanHistoryIsBoring.

Formanythousandsofyears,humanitywasaverygradualupwardtrajectory. Progress was achingly slow, almost invisible. Animals and farms, wars and empires,philosophiesandreligionsallfailedtoexertmuchinfluence.Butjust overtwohundredyearsago,somethingsuddenandprofoundarrivedandbent the curve of human history—of population and social development—almost ninetydegrees.

EnginesofProgress

Bynowyou’veprobablyguessedwhatitwas.Thisisabookabouttheimpactof

technology,afterall,soit’sasafebetthatwe’reopeningitthiswayinorderto demonstratehowimportanttechnologyhasbeen.Andthesuddenchangeinthe graphinthelateeighteenthcenturycorrespondstoadevelopmentwe’vehearda lot about: the Industrial Revolution, which was the sum of several nearly simultaneousdevelopmentsinmechanicalengineering,chemistry,metallurgy, andotherdisciplines.Soyou’vemostlikelyfiguredoutthatthesetechnological developmentsunderliethesudden,sharp,andsustainedjumpinhumanprogress. Ifso,yourguessisexactlyright.Andwecanbeevenmorepreciseabout whichtechnologywasmostimportant.Itwasthesteamengineor,tobemore precise,onedevelopedandimprovedbyJamesWattandhiscolleaguesinthe secondhalfoftheeighteenthcentury. PriortoWatt,steamengineswerehighlyinefficient,harnessingonlyabout onepercentoftheenergyreleasedbyburningcoal.Watt’sbrillianttinkering between1765and1776increasedthismorethanthreefold. 9 AsMorriswrites, thismadeallthedifference:“Eventhough[thesteam]revolutiontookseveral

itwasnonethelessthebiggestandfastesttransformationin

decadestounfold

theentirehistoryoftheworld.” 10 TheIndustrialRevolution,ofcourse,isnotonlythestoryofsteampower,but steamstarteditall.Morethananythingelse,itallowedustoovercomethe limitationsofmusclepower,humanandanimal,andgeneratemassiveamounts ofusefulenergyatwill.Thisledtofactoriesandmassproduction,torailways andmasstransportation.Itled,inotherwords,tomodernlife.TheIndustrial Revolutionusheredinhumanity’sfirstmachineage—thefirsttimeourprogress wasdrivenprimarilybytechnologicalinnovation—anditwasthemostprofound timeoftransformationourworldhaseverseen.*Theabilitytogeneratemassive amountsofmechanicalpowerwassoimportantthat,inMorris’swords,it“made mockeryofallthedramaoftheworld’searlierhistory.” 11

FIGURE1.2WhatBenttheCurveofHumanHistory?TheIndustrialRevolution.

Nowcomesthesecondmachineage.Computersandotherdigitaladvances

Nowcomesthesecondmachineage.Computersandotherdigitaladvances aredoingformentalpower—theabilitytouseourbrainstounderstandand shapeourenvironments—whatthesteamengineanditsdescendantsdidfor musclepower.They’reallowingustoblowpastpreviouslimitationsandtaking usintonewterritory.Howexactlythistransitionwillplayoutremainsunknown, butwhetherornotthenewmachineagebendsthecurveasdramaticallyas Watt’ssteamengine,itisaverybigdealindeed.Thisbookexplainshowand why. For now, a very short and simple answer: mental power is at least as important for progress and development—for mastering our physical and intellectualenvironmenttogetthingsdone—asphysicalpower.Soavastand unprecedentedboosttomentalpowershouldbeagreatboosttohumanity,justas theealierboosttophysicalpowersoclearlywas.

PlayingCatch-Up

Wewrotethisbookbecausewegotconfused.Foryearswehavestudiedthe impactofdigitaltechnologieslikecomputers,software,andcommunications networks,andwethoughtwehadadecentunderstandingoftheircapabilities and limitations. But over the past few years, they started surprising us.

Computersstarteddiagnosingdiseases,listeningandspeakingtous,andwriting high-qualityprose,whilerobotsstartedscurryingaroundwarehousesanddriving carswithminimalornoguidance.Digitaltechnologieshadbeenlaughablybad atalotofthesethingsforalongtime—thentheysuddenlygotverygood.How didthishappen?Andwhatweretheimplicationsofthisprogress,whichwas astonishingandyetcametobeconsideredamatterofcourse? Wedecidedtoteamupandseeifwecouldanswerthesequestions.Wedid thenormalthingsbusinessacademicsdo:readlotsofpapersandbooks,looked atmanydifferentkindsofdata,andbattedaroundideasandhypotheseswith eachother.Thiswasnecessaryandvaluable,butthereallearning,andthereal fun, started when we went out into the world. We spoke with inventors, investors, entrepreneurs, engineers, scientists, and many others who make technologyandputittowork. Thankstotheiropennessandgenerosity,wehadsomefuturisticexperiences in today’s incredible environment of digital innovation. We’ve ridden in a driverlesscar,watchedacomputerbeatteamsofHarvardandMITstudentsina gameofJeopardy!,trainedanindustrialrobotbygrabbingitswristandguiding itthroughaseriesofsteps,handledabeautifulmetalbowlthatwasmadeina

3Dprinter,andhadcountlessothermind-meltingencounterswithtechnology.

WhereWeAre

Thisworkledustothreebroadconclusions. Thefirstisthatwe’relivinginatimeofastonishingprogresswithdigital technologies—thosethathavecomputerhardware,software,andnetworksat theircore.Thesetechnologiesarenotbrand-new;businesseshavebeenbuying computersformorethanhalfacentury,andTimemagazinedeclaredthepersonal

computerits“MachineoftheYear”in1982.Butjustasittookgenerationsto

improve the steam engine to the point that it could power the Industrial Revolution,it’salsotakentimetorefineourdigitalengines. We’llshowwhyandhowthefullforceofthesetechnologieshasrecently beenachievedandgiveexamplesofitspower.“Full,”though,doesn’tmean “mature.” Computers are going to continue to improve and to do new and unprecedentedthings.By“fullforce,”wemeansimplythatthekeybuilding blocks are already in place for digital technologies to be as important and transformationaltosocietyandtheeconomyasthesteamengine.Inshort,we’re ataninflectionpoint—apointwherethecurvestartstobendalot—becauseof

computers.Weareenteringasecondmachineage. Oursecondconclusionisthatthetransformationsbroughtaboutbydigital technologywillbeprofoundlybeneficialones.We’reheadingintoanerathat won’tjustbedifferent;itwillbebetter,becausewe’llbeabletoincreaseboth thevarietyandthevolumeofourconsumption.Whenwephraseitthatway—in thedryvocabularyofeconomics—italmostsoundsunappealing.Whowantsto consumemoreandmoreallthetime?Butwedon’tjustconsumecaloriesand gasoline.Wealsoconsumeinformationfrombooksandfriends,entertainment fromsuperstarsandamateurs,expertisefromteachersanddoctors,andcountless otherthingsthatarenotmadeofatoms.Technologycanbringusmorechoice andevenfreedom. Whenthesethingsaredigitized—whenthey’reconvertedintobitsthatcanbe storedonacomputerandsentoveranetwork—theyacquiresomeweirdand wonderfulproperties.They’resubjecttodifferenteconomics,whereabundance is the norm rather than scarcity. As we’ll show, digital goods are not like physicalones,andthesedifferencesmatter. Ofcourse,physicalgoodsarestillessential,andmostofuswouldlikethem tohavegreatervolume,variety,andquality.Whetherornotwewanttoeatmore, we’dliketoeatbetterordifferentmeals.Whetherornotwewanttoburnmore fossil fuels, we’d like to visit more places with less hassle. Computers are helpingaccomplishthesegoals,andmanyothers.Digitizationisimprovingthe physical world, and these improvements are only going to become more important.Amongeconomichistoriansthere’swideagreementthat,asMartin Weitzmanputsit,“thelong-termgrowthofanadvancedeconomyisdominated bythebehavioroftechnicalprogress.” 12 Aswe’llshow,technicalprogressis improvingexponentially. Ourthirdconclusionislessoptimistic:digitizationisgoingtobringwithit somethornychallenges.Thisinitselfshouldnotbetoosurprisingoralarming; eventhemostbeneficialdevelopmentshaveunpleasantconsequencesthatmust bemanaged.TheIndustrialRevolutionwasaccompaniedbysoot-filledLondon skies and horrific exploitation of child labor. What will be their modern equivalents? Rapid and accelerating digitization is likely to bring economic ratherthanenvironmentaldisruption,stemmingfromthefactthatascomputers get more powerful, companies have less need for some kinds of workers. Technologicalprogressisgoingtoleavebehindsomepeople,perhapsevenalot ofpeople,asitracesahead.Aswe’lldemonstrate,there’sneverbeenabetter timetobeaworkerwithspecialskillsortherighteducation,becausethese

peoplecanusetechnologytocreateandcapturevalue.However,there’snever beenaworsetimetobeaworkerwithonly‘ordinary’skillsandabilitiestooffer, becausecomputers,robots,andotherdigitaltechnologiesareacquiringthese skillsandabilitiesatanextraordinaryrate. Overtime,thepeopleofEnglandandothercountriesconcludedthatsome aspectsoftheIndustrialRevolutionwereunacceptableandtookstepstoend them(democraticgovernmentandtechnologicalprogressbothhelpedwiththis). ChildlabornolongerexistsintheUK,andLondonaircontainslesssmokeand sulfur dioxide now than at any time since at least the late 1500s. 13 The challengesofthedigitalrevolutioncanalsobemet,butfirstwehavetobeclear onwhattheyare.It’simportanttodiscussthelikelynegativeconsequencesof thesecondmachineageandstartadialogueabouthowtomitigatethem—weare confidentthatthey’renotinsurmountable.Buttheywon’tfixthemselves,either. We’llofferourthoughtsonthisimportanttopicinthechapterstocome. Sothisisabookaboutthesecondmachineageunfoldingrightnow—an inflection point in the history of our economies and societies because of digitization. It’s an inflection point in the right direction—bounty instead of scarcity,freedominsteadofconstraint—butonethatwillbringwithitsome difficultchallengesandchoices.

Thisbookisdividedintothreesections.Thefirst,composedofchapters1

through6,describesthefundamentalcharacteristicsofthesecondmachineage.

Thesechaptersgivemanyexamplesofrecenttechnologicalprogressthatseem likethestuffofsciencefiction,explainwhythey’rehappeningnow(afterall, we’vehadcomputersfordecades),andrevealwhyweshouldbeconfidentthat thescaleandpaceofinnovationincomputers,robots,andotherdigitalgearis onlygoingtoaccelerateinthefuture.

Thesecondpart,consistingofchapters7through11,exploresbountyand

spread,thetwoeconomicconsequencesofthisprogress.Bountyistheincrease involume,variety,andqualityandthedecreaseincostofthemanyofferings broughtonbymoderntechnologicalprogress.It’sthebesteconomicnewsinthe worldtoday.Spread,however,isnotsogreat;it’sever-biggerdifferencesamong peopleineconomicsuccess—inwealth,income,mobility,andotherimportant measures. Spread has been increasing in recent years. This is a troubling development for many reasons, and one that will accelerate in the second machineageunlessweintervene.

Thefinalsection—chapters12through15—discusseswhatinterventionswill

be appropriate and effective for this age. Our economic goals should be to

maximizethebountywhilemitigatingthenegativeeffectsofthespread.We’ll offerourideasabouthowtobestaccomplishtheseaims,bothintheshortterm andinthemoredistantfuture,whenprogressreallyhasbroughtusintoaworld sotechnologicallyadvancedthatitseemstobethestuffofsciencefiction.Aswe stress in our concluding chapter, the choices we make from now on will determinewhatkindofworldthatis.

we make from now on will determinewhatkindofworldthatis. *

*Morrisdefineshumansocialdevelopmentasconsistingoffourattributes:energycapture(per-

person calories obtained from the environment for food, home and commerce, industry and agriculture,andtransportation),organization(thesizeofthelargestcity),war-makingcapacity (numberoftroops,powerandspeedofweapons,logisticalcapabilities,andothersimilarfactors), and information technology (the sophistication of available tools for sharing and processing information,andtheextentoftheiruse).Eachoftheseisconvertedintoanumberthatvariesover time from zero to 250. Overall social development is simply the sum of these four numbers. BecausehewasinterestedincomparisonsbetweentheWest(Europe,Mesopotamia,andNorth America at various times, depending on which was most advanced) and the East (China and

Japan),hecalculatedsocialdevelopmentseparatelyforeachareafrom14,000BCEto2000CE.

In2000,theEastwashigheronlyinorganization(sinceTokyowastheworld’slargestcity)and

hadasocialdevelopmentscoreof564.83.TheWest’sscorein2000was906.37.Weaveragethe

twoscores.

*WerefertotheIndustrialRevolutionasthefirstmachineage.However,“themachineage”is

alsoalabelusedbysomeeconomichistorianstorefertoaperiodofrapidtechnologicalprogress

spanningthelatenineteenthandearlytwentiethcenturies.Thissameperiodiscalledbyothers

theSecondIndustrialRevolution,whichishowwe’llrefertoitinlaterchapters.

“Anysufficientlyadvancedtechnologyisindistinguishablefrommagic.” —ArthurC.Clarke

“Anysufficientlyadvancedtechnologyisindistinguishablefrommagic.”

—ArthurC.Clarke

INTHESUMMEROF2012,wewentforadriveinacarthathadnodriver.

DuringaresearchvisittoGoogle’sSiliconValleyheadquarters,wegottoride inoneofthecompany’sautonomousvehicles,developedaspartofitsChauffeur project.Initiallywehadvisionsofcruisinginthebackseatofacarthathadno one in the front seat, but Google is understandably skittish about putting obviouslyautonomousautosontheroad.Doingsomightfreakoutpedestrians andotherdrivers,orattracttheattentionofthepolice.Sowesatintheback whiletwomembersoftheChauffeurteamrodeupfront. WhenoneoftheGooglershitthebuttonthatswitchedthecarintofully automatic driving mode while we were headed down Highway 101, our

curiosities—andself-preservationinstincts—engaged.The101isnotalwaysa

predictableorcalmenvironment.It’sniceandstraight,butit’salsocrowded mostofthetime,anditstrafficflowshavelittleobviousrhymeorreason.At highwayspeedstheconsequencesofdrivingmistakescanbeseriousones.Since wewerenowpartoftheongoingChauffeurexperiment,theseconsequences weresuddenlyofmorethanjustintellectualinteresttous. Thecarperformedflawlessly.Infact,itactuallyprovidedaboringride.It didn’tspeedorslalomamongtheothercars;itdroveexactlythewaywe’reall taught to in driver’s ed. A laptop in the car provided a real-time visual representationofwhattheGooglecar‘saw’asitproceededalongthehighway— allthenearbyobjectsofwhichitssensorswereaware.Thecarrecognizedallthe surroundingvehicles,notjustthenearestones,anditremainedawareofthemno matterwheretheymoved.Itwasacarwithoutblindspots.Butthesoftware doingthedrivingwasawarethatcarsandtrucksdrivenbyhumansdohaveblind spots.Thelaptopscreendisplayedthesoftware’sbestguessaboutwhereall theseblindspotswereandworkedtostayoutofthem. Wewerestaringatthescreen,payingnoattentiontotheactualroad,when traffic ahead of us came to a complete stop. The autonomous car braked

smoothlyinresponse,comingtoastopasafedistancebehindthecarinfront, and started moving again once the rest of the traffic did. All the while the Googlers in the front seat never stopped their conversation or showed any nervousness,orindeedmuchinterestatallincurrenthighwayconditions.Their

hundredsofhoursinthecarhadconvincedthemthatitcouldhandlealittlestop-

and-gotraffic.Bythetimewepulledbackintotheparkinglot,wesharedtheir

confidence.

TheNewNewDivisionofLabor

Ourridethatdayonthe101wasespeciallyweirdforusbecause,onlyafew

years earlier, we were sure that computers would not be able to drive cars. Excellentresearchandanalysis,conductedbycolleagueswhowerespectagreat deal,concludedthatdrivingwouldremainahumantaskfortheforeseeable future.Howtheyreachedthisconclusion,andhowtechnologieslikeChauffeur startedtooverturnitinjustafewyears,offersimportantlessonsaboutdigital progress.

In2004FrankLevyandRichardMurnanepublishedtheirbookTheNew

DivisionofLabor. 1 Thedivisiontheyfocusedonwasbetweenhumananddigital labor—inotherwords,betweenpeopleandcomputers.Inanysensibleeconomic system, people should focus on the tasks and jobs where they have a comparativeadvantageovercomputers,leavingcomputerstheworkforwhich theyarebettersuited.IntheirbookLevyandMurnaneofferedawaytothink aboutwhichtasksfellintoeachcategory. One hundred years ago the previous paragraph wouldn’t have made any sense.Backthen,computerswerehumans.Thewordwasoriginallyajobtitle, notalabelforatypeofmachine.Computersintheearlytwentiethcenturywere people,usuallywomen,whospentalldaydoingarithmeticandtabulatingthe results.Overthecourseofdecades,innovatorsdesignedmachinesthatcould

takeovermoreandmoreofthiswork;theywerefirstmechanical,thenelectro-

mechanical, and eventually digital. Today, few people if any are employed

simplytodoarithmeticandrecordtheresults.Eveninthelowest-wagecountries

therearenohumancomputers,becausethenonhumanonesarefarcheaper,

faster,andmoreaccurate.

Ifyouexaminetheirinnerworkings,yourealizethatcomputersaren’tjust

numbercrunchers,they’resymbolsprocessors.Theircircuitrycanbeinterpreted

inthelanguageofonesandzeroes,butequallyvalidlyastrueorfalse,yesorno,

oranyothersymbolicsystem.Inprinciple,theycandoallmannerofsymbolic

work,frommathtologictolanguage.Butdigitalnovelistsarenotyetavailable,

sopeoplestillwriteallthebooksthatappearonfictionbestsellerlists.Wealso

haven’tyetcomputerizedtheworkofentrepreneurs,CEOs,scientists,nurses,

restaurantbusboys,ormanyothertypesofworkers.Whynot?Whatisitabout

theirworkthatmakesithardertodigitizethanwhathumancomputersusedto

do?

ComputersAreGoodatFollowingRules

These are the questions Levy and Murnane tackled in TheNewDivision of Labor,andtheanswerstheycameupwithmadeagreatdealofsense.The authorsputinformationprocessingtasks—thefoundationofallknowledgework —on a spectrum. At one end are tasks like arithmetic that require only the application of well-understood rules. Since computers are really good at followingrules,itfollowsthattheyshoulddoarithmeticandsimilartasks. LevyandMurnanegoontohighlightothertypesofknowledgeworkthatcan alsobeexpressedasrules.Forexample,aperson’screditscoreisagoodgeneral predictor of whether they’ll pay back their mortgage as promised, as is the amountofthemortgagerelativetotheperson’swealth,income,andotherdebts. So the decision about whether or not to give someone a mortgage can be effectivelyboileddowntoarule. Expressedinwords,amortgagerulemightsay,“Ifapersonisrequestinga mortgageofamountMandtheyhaveacreditscoreofVorhigher, annual incomegreaterthanIortotalwealthgreaterthanW,andtotaldebtnogreater thanD,thenapprovetherequest.”Whenexpressedincomputercode,wecalla mortgagerulelikethisanalgorithm.Algorithmsaresimplifications;theycan’t anddon’ttakeeverythingintoaccount(likeabillionaireunclewhohasincluded theapplicantinhiswillandlikestorock-climbwithoutropes).Algorithmsdo, however,includethemostcommonandimportantthings,andtheygenerally workquitewellattaskslikepredictingpaybackrates.Computers,therefore,can andshouldbeusedformortgageapproval.*

ButLousyatPatternRecognition

AttheotherendofLevyandMurnane’sspectrum,however,lieinformation processingtasksthatcannotbeboileddowntorulesoralgorithms.Accordingto the authors, these are tasks that draw on the human capacity for pattern recognition.Ourbrainsareextraordinarilygoodattakingininformationviaour senses and examining it for patterns, but we’re quite bad at describing or figuringouthowwe’redoingit,especiallywhenalargevolumeoffast-changing information arrives at a rapid pace. As the philosopher Michael Polanyi famouslyobserved,“Weknowmorethanwecantell.” 2 Whenthisisthecase,

accordingtoLevyandMurnane,taskscan’tbecomputerizedandwillremainin

thedomainofhumanworkers.Theauthorscitedrivingavehicleintrafficasan

exampleofsuchastask.Astheywrite,

Asthedrivermakeshisleftturnagainsttraffic,heconfrontsawallofimagesandsounds

generated by oncoming cars, traffic lights, storefronts, billboards, trees, and a traffic policeman.Usinghisknowledge,hemustestimatethesizeandpositionofeachofthese

Thetruckdriver[has]theschemato

recognize what [he is] confronting. But articulating this knowledge and embedding it in softwareforallbuthighlystructuredsituationsareatpresentenormouslydifficulttasks Computerscannoteasilysubstituteforhumansin[jobslikedriving].

objectsandthelikelihoodthattheyposeahazard

SoMuchforThatDistinction

WewereconvincedbyLevyandMurnane’sargumentswhenwereadTheNew

DivisionofLaborin2004.Wewerefurtherconvincedthatyearbytheinitial

resultsoftheDARPAGrandChallengefordriverlesscars. DARPA,theDefenseAdvancedResearchProjectsAgency,wasfoundedin 1958(inresponsetotheSovietUnion’slaunchoftheSputnik satellite) and tasked with spurring technological progress that might have military

applications.In2002theagencyannounceditsfirstGrandChallenge,whichwas

tobuildacompletelyautonomousvehiclethatcouldcompletea150-milecourse

throughCalifornia’sMojaveDesert.Fifteenentrantsperformedwellenoughina

qualifyingruntocompeteinthemainevent,whichwasheldonMarch13,2004.

Theresultswerelessthanencouraging.Twovehiclesdidn’tmakeittothe startingarea,oneflippedoverinthestartingarea,andthreehoursintotherace only four cars were still operational. The “winning” Sandstorm car from

CarnegieMellonUniversitycovered7.4miles(lessthan5percentofthetotal)

before veering off the course during a hairpin turn and getting stuck on an embankment. The contest’s $1 million prize went unclaimed, and Popular Sciencecalledtheevent“DARPA’sDebacleintheDesert.” 3 Withinafewyears,however,thedebacleinthedesertbecamethe‘funonthe

101’thatweexperienced.GoogleannouncedinanOctober2010blogpostthat

itscompletelyautonomouscarshadforsometimebeendrivingsuccessfully,in traffic,onAmericanroadsandhighways.Bythetimewetookourrideinthe

summerof2012theChauffeurprojecthadgrownintoasmallfleetofvehicles

that had collectively logged hundreds of thousands of miles with no human involvementandwithonlytwoaccidents.Oneoccurredwhenapersonwas drivingtheChauffeurcar;theotherhappenedwhenaGooglecarwasrear-ended

(byahumandriver)whilestoppedataredlight. 4 Tobesure,therearestillmany situationsthatGoogle’scarscan’thandle,particularlycomplicatedcitytrafficor off-road driving or, for that matter, any location that has not already been meticulouslymappedinadvancebyGoogle.Butourexperienceonthehighway convinced us that it’s a viable approach for the large and growing set of everydaydrivingsituations. Self-drivingcarswentfrombeingthestuffofsciencefictiontoon-the-road realityinafewshortyears.Cutting-edgeresearchexplainingwhytheywerenot cominganytimesoonwasoutpacedbycutting-edgescienceandengineeringthat broughtthemintoexistence,againinthespaceofafewshortyears.Thisscience andengineeringacceleratedrapidly,goingfromadebacletoatriumphinalittle morethanhalfadecade. Improvement in autonomous vehicles reminds us of Hemingway’s quote abouthowamangoesbroke:“Graduallyandthensuddenly.” 5 Andself-driving carsarenotananomaly;they’repartofabroad,fascinatingpattern.Progresson someoftheoldestandtoughestchallengesassociatedwithcomputers,robots, andotherdigitalgearwasgradualforalongtime.Theninthepastfewyearsit becamesudden;digitalgearstartedracingahead,accomplishingtasksithad always been lousy at and displaying skills it was not supposed to acquire anytime soon. Let’s look at a few more examples of surprising recent technologicalprogress.

GoodListenersandSmoothTalkers

In addition to pattern recognition, Levy and Murnane highlight complex communication as a domain that would stay on the human side in the new divisionoflabor.Theywritethat,“Conversationscriticaltoeffectiveteaching, managing, selling, and many other occupations require the transfer and interpretationofabroadrangeofinformation.Inthesecases,thepossibilityof exchanginginformationwithacomputer,ratherthananotherhuman,isalong wayoff.” 6 In the fall of 2011, Apple introduced the iPhone 4S featuring “Siri,” an intelligentpersonalassistantthatworkedviaanatural-languageuserinterface.In otherwords,peopletalkedtoitjustastheywouldtalktoanotherhumanbeing. The software underlying Siri, which originated at the California research

instituteSRIInternationalandwaspurchasedbyApplein2010,listenedtowhat

iPhoneusersweresayingtoit,triedtoidentifywhattheywanted,thentook

actionandreportedbacktotheminasyntheticvoice. AfterSirihadbeenoutforabouteightmonths,KyleWagneroftechnology blogGizmodolistedsomeofitsmostusefulcapabilities:“Youcanaskaboutthe scores of live games—‘What’s the score of the Giants game?’—or about individualplayerstats.YoucanalsomakeOpenTablereservations,getYelp scores,askaboutwhatmoviesareplayingatalocaltheaterandthenseeatrailer. Ifyou’rebusyandcan’ttakeacall,youcanaskSiritoremindyoutocallthe personbacklater.Thisisthekindofeverydaytaskforwhichvoicecommands canactuallybeincrediblyuseful.” 7 TheGizmodopostendedwithcaution:“Thatactuallysoundsprettycool.Just withtheobviousSiricriterion:Ifitactuallyworks.8 Uponitsrelease,alotof people found that Apple’s intelligent personal assistant didn’t work well. It didn’tunderstandwhattheyweresaying,askedforrepeatedclarifications,gave strangeorinaccurateanswers,andputthemoffwithresponseslike“I’mreally sorryaboutthis,butIcan’ttakeanyrequestsrightnow.Pleasetryagainina littlewhile.”AnalystGeneMunstercataloguedquestionswithwhichSirihad trouble:

WhereisElvisburied?Responded,“Ican’tanswerthatforyou.” Itthoughttheperson’snamewasElvisBuried. •WhendidthemovieCinderellacomeout?Respondedwitha movietheatersearchonYelp. •When is the next Halley’s Comet? Responded, “You have no meetingsmatchingHalley’s.” •IwanttogotoLakeSuperior.Respondedwithdirectionstothe companyLakeSuperiorX-Ray. 9

Siri’ssometimesbizarreandfrustratingresponsesbecamewellknown,butthe powerofthetechnologyisundeniable.Itcancometoyouraidexactlywhenyou needit.Onthesametripthataffordedussometimeinanautonomouscar,we sawthisfirsthand.AfterameetinginSanFrancisco,wehoppedinourrentalcar todrivedowntoGoogle’sheadquartersinMountainView.Wehadaportable GPSdevicewithus,butdidn’tplugitinandturnitonbecausewethoughtwe knewhowtogettoournextdestination. We didn’t, of course. Confronted with an Escherian maze of elevated

highways,off-ramps,andsurfacestreets,wedrovearoundlookingforanon-

ramp while tensions mounted. Just when our meeting at Google, this book

project,andourprofessionalrelationshipseemedinseriousjeopardy,Erikpulled

outhisphoneandaskedSirifor“directionstoU.S.101South.”Thephone

respondedinstantlyandflawlessly:thescreenturnedintoamapshowingwhere wewereandhowtofindtheelusiveon-ramp. Wecouldhavepulledover,foundtheportableGPSandturnediton,typedin ourdestination,andwaitedforourrouting,butwedidn’twanttoexchange informationthatway.Wewantedtospeakaquestionandhearandsee(becausea mapwasinvolved)areply.Siriprovidedexactlythenaturallanguageinteraction

wewerelookingfor.A2004reviewoftheprevioushalf-century’sresearchin

automatic speech recognition (a critical part of natural language processing) openedwiththeadmissionthat“Human-levelspeechrecognitionhasprovedto beanelusivegoal,”butlessthanadecadelatermajorelementsofthatgoalhave beenreached.Appleandothercompanieshavemaderobustnaturallanguage processing technology available to hundreds of millions of people via their mobilephones. 10 AsnotedbyTomMitchell,whoheadsthemachine-learning departmentatCarnegieMellonUniversity:“We’reatthebeginningofaten-year periodwherewe’regoingtotransitionfromcomputersthatcan’tunderstand language to a point where computers can understand quite a bit about language.” 11

DigitalFluency:TheBabelFishGoestoWork

Naturallanguageprocessingsoftwareisstillfarfromperfect,andcomputersare notyetasgoodaspeopleatcomplexcommunication,butthey’regettingbetter all the time. And in tasks like translation from one language to another, surprising developments are underway: while computers’ communication abilitiesarenotasdeepasthoseoftheaveragehumanbeing,they’remuch broader. Apersonwhospeaksmorethanonelanguagecanusuallytranslatebetween them with reasonable accuracy. Automatic translation services, on the other hand,areimpressivebutrarelyerror-free.EvenifyourFrenchisrusty,youcan probablydobetterthanGoogleTranslatewiththesentence“MontyPython’s ‘Dirty Hungarian Phrasebook’sketch is one of their funniest ones.” Google offered,“SketchdesMontyPython‘Phrasebooksalehongrois’estl’undesplus drôles les leurs.” This conveys the main gist, but has serious grammatical problems. Thereislesschanceyoucouldhavemadeprogresstranslatingthissentence

(oranyother)intoHungarian,Arabic,Chinese,Russian,Norwegian,Malay, Yiddish,Swahili,Esperanto,oranyoftheothersixty-threelanguagesbesides FrenchthatarepartoftheGoogleTranslateservice.ButGooglewillattempta translationoftextfromanyoftheselanguagesintoanyother,instantaneously andatnocostforanyonewithWebaccess. 12 TheTranslateservice’ssmartphone appletsusersspeakmorethanfifteenoftheselanguagesintothephoneand,in response,willproducesynthesized,translatedspeechinmorethanhalfofthe fifteen.It’sasafebetthateventheworld’smostmultilingualpersoncan’tmatch thisbreadth. Foryearsinstantaneoustranslationutilitieshavebeenthestuffofscience fiction (most notably The Hitchhiker’s Guide to the Galaxy’s Babel Fish, a strange creature that once inserted in the ear allows a person to understand speechinanylanguage). 13 GoogleTranslateandsimilarservicesaremakingita reality today. In fact, at least one such service is being used right now to facilitate international customer service interactions. The translation services company Lionbridge has partnered with IBM to offer GeoFluent, an online applicationthatinstantlytranslateschatsbetweencustomersandtroubleshooters

whodonotsharealanguage.Inaninitialtrial,approximately90percentof

GeoFluentusersreportedthatitwasgoodenoughforbusinesspurposes. 14

HumanSuperiorityinJeopardy!

Computersarenowcombiningpatternmatchingwithcomplexcommunication

toquiteliterallybeatpeopleattheirowngames.In2011,theFebruary14and15

episodesoftheTVgameshowJeopardy!includedacontestantthatwasnota human being. It was a supercomputer called Watson, developed by IBM specifically to play the game (and named in honor of legendary IBM CEO

ThomasWatson,Sr.).Jeopardy!debutedin1964andin2012wasthefifthmost

popularsyndicatedTVprograminAmerica. 15 Onatypicaldayalmost7million people watch host Alex Trebek ask trivia questions on various topics as contestantsvietobethefirsttoanswerthemcorrectly.* Theshow’slongevityandpopularitystemfromitsbeingeasytounderstand yetextremelyhardtoplaywell.Almosteveryoneknowstheanswerstosomeof thequestionsinagivenepisode,butveryfewpeopleknowtheanswersto almostallofthem.Questionscoverawiderangeoftopics,andcontestantsare not told in advance what those topics will be. Players also have to be simultaneouslyfast,bold,andaccurate—fastbecausetheycompeteagainstone

anotherforthechancetoanswereachquestion;boldbecausetheyhavetotryto answeralotofquestions,especiallyharderones,inordertoaccumulateenough money to win; and accurate because money is subtracted for each incorrect answer. Jeopardy!’sproducersfurtherchallengecontestantswithpuns,rhymes,and other kinds of wordplay. A clue might ask, for example, for “A rhyming reminderofthepastinthecityoftheNBA’sKings.” 16 Toanswercorrectly,a playerwouldhavetoknowwhattheacronymNBAstoodfor(inthiscase,it’s the National Basketball Association, not the National Bank Act or chemical compoundn-Butylamine),whichcitytheNBA’sKingsplayin(Sacramento), andthattheclue’sdemandforarhymingreminderofthepastmeantthatthe rightansweris“WhatisaSacramentomemento?”insteadofa“Sacramento souvenir”oranyotherfactuallycorrectresponse.Respondingcorrectlytoclues liketheserequiresmasteryofpatternmatchingandcomplexcommunication. AndwinningatJeopardy!requiresdoingboththingsrepeatedly,accurately,and almostinstantaneously.

Duringthe2011shows,WatsoncompetedagainstKenJenningsandBrad

Rutter,twoofthebestknowledgeworkersinthisesotericindustry.Jenningswon

Jeopardy!arecordseventy-fourtimesinarowin2004,takinghomemorethan

$3,170,000inprizemoneyandbecomingsomethingofafolkheroalongthe

way. 17 Infact,JenningsissometimesgivencreditfortheexistenceofWatson. 18 According to one story circulating within IBM, Charles Lickel, a research manager at the company interested in pushing the frontiers of artificial intelligence,washavingdinnerinasteakhouseinFishkill,NewYork,onenight

inthefallof2004.At7p.m.,henoticedthatmanyofhisfellowdinersgotup

andwentintotheadjacentbar.Whenhefollowedthemtofindoutwhatwas goingon,hesawthattheywereclusteredinfrontofthebar’sTVwatching Jennings extend his winning streak beyond fifty matches. Lickel saw that a match between Jennings and a Jeopardy!-playing supercomputer would be extremely popular, in addition to being a stern test of a computer’s pattern matchingandcomplexcommunicationabilities. SinceJeopardy!isathree-waycontest,theidealthirdcontestantwouldbe

BradRutter,whobeatJenningsintheshow’s2005UltimateTournamentof

Championsandwonmorethan$3,400,000. 19 Bothmenhadpackedtheirbrains withinformationofallkinds,weredeeplyfamiliarwiththegameandallofits idiosyncrasies,andknewhowtohandlepressure. Thesetwohumanswouldbetoughforanymachinetobeat,andthefirst

versions of Watson weren’t even close. Watson could be ‘tuned’ by its programmerstobeeithermoreaggressiveinansweringquestions(andhence

morelikelytobewrong)ormoreconservativeandaccurate.InDecember2006,

shortlyaftertheprojectstarted,whenWatsonwastunedtotrytoanswer70

percentofthetime(arelativelyaggressiveapproach)itwasonlyabletocome

upwiththerightresponseapproximately15percentofthetime.Jennings,in

sharpcontrast,answeredabout90percentofquestionscorrectlyingameswhen

hebuzzedinfirst(inotherwords,wontherighttorespond)70percentofthe

time. 20

But Watson turned out to be a very quick learner. The supercomputer’s performanceontheaggressionvs.accuracytradeoffimprovedquickly,andby

November2010,whenitwasaggressiveenoughtowintherighttoanswer70

percentofasimulatedmatch’stotalquestions,itansweredabout85percentof

them correctly. This was impressive improvement, but it still didn’t put the computerinthesameleagueasthebesthumanplayers.TheWatsonteamkept working until mid-January of 2011, when the matches were recorded for broadcastinFebruary,butnooneknewhowwelltheircreationwoulddoagainst JenningsandRutter. Watsontrouncedthemboth.Itcorrectlyansweredquestionsontopicsranging from “Olympic Oddities” (responding “pentathlon” to “A1976 entry in the ‘modern’thiswaskickedoutforwiringhisepeetoscorepointswithouttouching hisfoe”)to“ChurchandState”(realizingthattheanswersallcontainedoneor theotherofthesewords,thecomputeranswered“gestate”whentold“Itcan meantodevelopgraduallyinthemindortocarryduringpregnancy”).Whilethe supercomputer was not perfect (for example, it answered “chic” instead of “class”whenaskedabout“stylishelegance,orstudentswhoallgraduatedinthe sameyear”aspartofthecategory“AlternateMeanings”),itwasverygood. Watsonwasalsoextremelyfast,repeatedlybuzzinginbeforeJenningsand Ruttertowintherighttoanswerquestions.Inthefirstofthetwogamesplayed,

forexample,Watsonbuzzedinfirst43times,thenansweredcorrectly38times.

JenningsandRuttercombinedtobuzzinonly33timesoverthecourseofthe

samegame. 21

Attheendofthetwo-daytournament,Watsonhadamassed$77,147,more

thanthreetimesasmuchaseitherofitshumanopponents.Jennings,whocame in second, added a personal note on his answer to the tournament’s final question:“Iforonewelcomeournewcomputeroverlords.”Helaterelaborated,

“Justasfactoryjobswereeliminatedinthetwentiethcenturybynewassembly-

linerobots,BradandIwerethefirstknowledge-industryworkersputoutof workbythenewgenerationof‘thinking’machines.‘Quizshowcontestant’may bethefirstjobmaderedundantbyWatson,butI’msureitwon’tbethelast.” 22

TheParadoxofRobotic‘Progress’

A final important area where we see a rapid recent acceleration in digital improvement is robotics—building machines that can navigate through and interact with the physical world of factories, warehouses, battlefields, and offices.Hereagainweseeprogressthatwasverygradual,thensudden.

ThewordrobotenteredtheEnglishlanguageviathe1921Czechplay,R.U.R.

(Rossum’s“Universal”Robots)byKarelCapek,andautomatonshavebeenan object of human fascination ever since. 23 During the Great Depression, magazineandnewspaperstoriesspeculatedthatrobotswouldwagewar,commit crimes,displaceworkers,andevenbeatboxerJackDempsey. 24 IsaacAsimov coined the term robotics in 1941 and provided ground rules for the young disciplinethefollowingyearwithhisfamousThreeLawsofRobotics:

1.Arobotmaynotinjureahumanbeingor,throughinaction,allowa

humanbeingtocometoharm.

2.Arobotmustobeytheordersgiventoitbyhumanbeings,except

wheresuchorderswouldconflictwiththeFirstLaw.

3.Arobotmustprotectitsownexistenceaslongassuchprotection

doesnotconflictwiththeFirstorSecondLaws. 25

Asimov’senormousinfluenceonbothsciencefictionandreal-worldrobot-

makinghaspersistedforseventyyears.Butoneofthosetwocommunitieshas

racedfaraheadoftheother.SciencefictionhasgivenusthechattyandloyalR2-

D2andC-3PO,BattlestarGalactica’sominousCylons,theterribleTerminator,

andendlessvarietiesofandroids,cyborgs,andreplicants.Decadesofrobotics research,incontrast,gaveusHonda’sASIMO,ahumanoidrobotbestknownfor aspectacularlyfaileddemothatshowcaseditsinabilitytofollowAsimov’sthird

law.Ata2006presentationtoaliveaudienceinTokyo,ASIMOattemptedto

walkupashallowflightofstairsthathadbeenplacedonthestage.Onthethird step,therobot’skneesbuckledanditfelloverbackward,smashingitsfaceplate onthefloor. 26

ASIMOhassincerecoveredanddemonstratedskillslikewalkingupand

downstairs,kickingasoccerball,anddancing,butitsshortcomingshighlighta broadtruth:alotofthethingshumansfindeasyandnaturaltodointhephysical worldhavebeenremarkablydifficultforrobotstomaster.AstheroboticistHans Moravechasobserved,“Itiscomparativelyeasytomakecomputersexhibit adult-levelperformanceonintelligencetestsorplayingcheckers,anddifficultor impossibletogivethemtheskillsofaone-year-oldwhenitcomestoperception andmobility.” 27 This situation has come to be known as Moravec’s paradox, nicely summarized by Wikipedia as “the discovery by artificial intelligence and

robotics researchers that, contrary to traditional assumptions, high-level reasoning requires very little computation, but low-level sensorimotor skills require enormous computational resources.” 28 * Moravec’s insight is broadly accurate,andimportant.AsthecognitivescientistStevenPinkerputsit,“The mainlessonofthirty-fiveyearsofAIresearchisthatthehardproblemsareeasy

andtheeasyproblemsarehard

appears,itwillbethestockanalystsandpetrochemicalengineersandparole board members who are in danger of being replaced by machines. The gardeners, receptionists, and cooks are secure in their jobs for decades to come.” 29 Pinker’spointisthatroboticsexpertshavefounditfiendishlydifficultto buildmachinesthatmatchtheskillsofeventheleast-trainedmanualworker. iRobot’sRoomba,forexample,can’tdoeverythingamaiddoes;itjustvacuums thefloor.MorethantenmillionRoombashavebeensold,butnoneofthemis goingtostraightenthemagazinesonacoffeetable. Whenitcomestoworkinthephysicalworld,humansalsohaveahuge flexibilityadvantageovermachines.Automatingasingleactivity,likesoldering awireontoacircuitboardorfasteningtwopartstogetherwithscrews,ispretty easy,butthattaskmustremainconstantovertimeandtakeplaceina‘regular’ environment.Forexample,thecircuitboardmustshowupinexactlythesame orientationeverytime.Companiesbuyspecializedmachinesfortaskslikethese, havetheirengineersprogramandtestthem,thenaddthemtotheirassembly lines.Eachtimethetaskchanges—eachtimethelocationofthescrewholes move,forexample—productionmuststopuntilthemachineryisreprogrammed. Today’s factories, especially large ones in high-wage countries, are highly automated, but they’re not full of general-purpose robots. They’re full of dedicated, specialized machinery that’s expensive to buy, configure, and reconfigure.

Asthenewgenerationofintelligentdevices

RethinkingFactoryAutomation

RodneyBrooks,whoco-foundediRobot,noticedsomethingelseaboutmodern, highlyautomatedfactoryfloors:peoplearescarce,butthey’renotabsent.Anda lotoftheworktheydoisrepetitiveandmindless.Onalinethatfillsupjelly jars,forexample,machinessquirtapreciseamountofjellyintoeachjar,screw onthetop,andstickonthelabel,butapersonplacestheemptyjarsonthe conveyorbelttostarttheprocess.Whyhasn’tthisstepbeenautomated?Because inthiscasethejarsaredeliveredtothelinetwelveatatimeincardboardboxes thatdon’tholdthemfirmlyinplace.Thisimprecisionpresentsnoproblemtoa person(whosimplyseesthejarsinthebox,grabsthem,andputsthemonthe conveyorbelt),buttraditionalindustrialautomation hasgreatdifficulty with jellyjarsthatdon’tshowupinexactlythesameplaceeverytime.

In2008Brooksfoundedanewcompany,RethinkRobotics,topursueand

builduntraditionalindustrialautomation:robotsthatcanpickandplacejellyjars and handle the countless other imprecise tasks currently done by people in today’s factories. His ambition is to make some progress against Moravec’s paradox.What’smore,Brooksenvisionscreatingrobotsthatwon’tneedtobe programmedbyhigh-paidengineers;instead,themachinescanbetaughttodoa task(orretaughttodoanewone)byshopfloorworkers,eachofwhomneed less than an hour of training to learn how to instruct their new mechanical

colleagues.Brooks’smachinesarecheap,too.Atabout$20,000,they’reasmall

fractionofthecostofcurrentindustrialrobots.Wegotasneakpeekatthese potentialparadox-bustersshortlybeforeRethink’spublicunveilingoftheirfirst lineofrobots,namedBaxter.Brooksinvitedustothecompany’sheadquartersin Bostontoseetheseautomatons,andtoseewhattheycoulddo. Baxterisinstantlyrecognizableasahumanoidrobot.Ithastwoburly,jointed armswithclaw-likegripsforhands;atorso;andaheadwithanLCDfacethat swivelsto‘lookat’thenearestperson.Itdoesn’thavelegs,though;Rethink sidesteppedtheenormouschallengesofautomaticlocomotionbyputtingBaxter on wheels and having it rely on people to get from place to place. The company’sanalysessuggestthatitcanstilldolotsofusefulworkwithoutthe abilitytomoveunderhisownpower. To train Baxter, you grab it by the wrist and guide the arm through the motionsyouwantittocarryout.Asyoudothis,thearmseemsweightless;its motorsareworkingsoyoudon’thaveto.Therobotalsomaintainssafety;the twoarmscan’tcollide(themotorsresistyouifyoutrytomakethishappen)and

theyautomaticallyslowdownifBaxtersensesapersonwithintheirrange.These andmanyotherdesignfeaturesmakeworkingwiththisautomatonanatural, intuitive,andnonthreateningexperience.Whenwefirstapproachedit,wewere nervousaboutcatchingarobotarmtotheface,butthisapprehensionfaded quickly,replacedbycuriosity. BrooksshowedusseveralBaxtersatworkinthecompany’sdemoarea.They were blowing past Moravec’s paradox—sensing and manipulating lots of differentobjectswith‘hands’rangingfromgripstosuctioncups.Therobots aren’tasfastorfluidasawell-trainedhumanworkeratfullspeed,butthey mightnotneedtobe.Mostconveyorbeltsandassemblylinesdonotoperateat fullhumanspeed;theywouldtirepeopleoutiftheydid. Baxterhasafewobviousadvantagesoverhumanworkers.Itcanworkallday everydaywithoutneedingsleep,lunch,orcoffeebreaks.Italsowon’tdemand healthcarefromitsemployeroraddtothepayrolltaxburden.Anditcandotwo completely unrelated things at once; its two arms are capable of operating independently.

ComingSoontoAssemblyLines,Warehouses,and

HallwaysNearYou

AftervisitingRethinkandseeingBaxterinaction,weunderstoodwhyTexas

InstrumentsVicePresidentRemiEl-Ouazzanesaidinearly2012,“Wehavea

firmbeliefthattheroboticsmarketisonthecuspofexploding.”There’salotof evidence to support his view. The volume and variety of robots in use at companiesisexpandingrapidly,andinnovatorsandentrepreneurshaverecently madedeepinroadsagainstMoravec’sparadox. 30 Kiva,anotheryoungBoston-areacompany,hastaughtitsautomatonstomove aroundwarehousessafely,quickly,andeffectively.Kivarobotslooklikemetal

ottomansorsquashedR2-D2s.Theyscuttlearoundbuildingsataboutknee-

height,stayingoutofthewayofhumansandoneanother.They’relowtothe groundsotheycanscootunderneathshelvingunits,liftthemup,andbringthem tohumanworkers.Aftertheseworkersgrabtheproductstheyneed,therobot whiskstheshelfawayandanothershelf-bearingrobottakesitsplace.Software trackswherealltheproducts,shelves,robots,andpeopleareinthewarehouse, and orchestrates the continuous dance of the Kiva automatons. In March of

2012,KivawasacquiredbyAmazon—aleaderinadvancedwarehouselogistics

BostonDynamics,yetanotherNewEnglandstartup,hastackledMoravec’s paradox head-on. The company builds robots aimed at supporting American troopsinthefieldby,amongotherthings,carryingheavyloadsoverrough terrain.ItsBigDog,whichlookslikeagiantmetalmastiffwithlongskinnylegs, cangoupsteephills,recoverfromslipsonice,anddootherverydog-like things.Balancingaheavyloadonfourpointswhilemovingoveranuneven landscapeisatrulynastyengineeringproblem,butBostonDynamicshasbeen makinggoodprogress. Asafinalexampleofrecentroboticprogress,considertheDouble,whichis aboutasdifferentfromtheBigDogaspossible.Insteadoftrottingthroughrough enemy terrain, the Double rolls over cubicle carpets and hospital hallways carrying an iPad. It’s essentially an upside-down pendulum with motorized wheelsatthebottomandatabletatthetopofafour-tofive-footstick.The Double provides telepresence—it lets the operator ‘walk around’ a distant buildingandseeandhearwhat’sgoingon.Thecamera,microphone,andscreen oftheiPadserveastheeyes,ears,andfaceoftheoperator,whoseesandhears whattheiPadseesandhears.TheDoubleitselfactsasthelegs,transportingthe wholeassemblyaroundinresponsetocommandsfromtheoperator.Double Roboticscallsit“thesimplest,mostelegantwaytobesomewhereelseinthe

worldwithoutflyingthere.”ThefirstbatchofDoubles,pricedat$2,499,sold

outsoonafterthetechnologywasannouncedinthefallof2012. 32 ThenextroundofroboticinnovationmightputthebiggestdentinMoravec’s

paradoxever.In2012DARPAannouncedanotherGrandChallenge;insteadof

autonomous cars, this one was about automatons. The DARPA Robotics Challenge(DRC)combinedtooluse,mobility,sensing,telepresence,andmany other long-standing challenges in the field. According to the website of the agency’sTacticalTechnologyOffice,

TheprimarytechnicalgoaloftheDRCistodevelopgroundrobotscapableofexecuting complex tasks in dangerous, degraded, human-engineered environments. Competitors in the DRC are expected to focus on robots that can use standard tools and equipment commonlyavailableinhumanenvironments,rangingfromhandtoolstovehicles,withan

emphasisonadaptabilitytotoolswithdiversespecifications.33

With the DRC, DARPA is asking the robotics community to build and

demonstratehigh-functioninghumanoidrobotsbytheendof2014.Accordingto

aninitialspecificationsuppliedbytheagency,theywillhavetobeabletodrive autilityvehicle,removedebrisblockinganentryway,climbaladder,closea valve, and replace a pump. 34 These seem like impossible requirements, but

we’vebeenassuredbyhighlyknowledgeablecolleagues—onescompetinginthe DRC,infact—thatthey’llbemet. Manysawthe 2004GrandChallenge as instrumental in accelerating progress with autonomous vehicles. There’s an excellentchancethattheDRCwillbesimilarlyimportantatgettinguspast Moravec’sparadox.

MoreEvidenceThatWe’reatanInflectionPoint

Self-drivingcars,Jeopardy!championsupercomputers,andavarietyofuseful

robotshaveallappearedjustinthepastfewyears.Andtheseinnovationsarenot

justlabdemos;they’reshowingofftheirskillsandabilitiesinthemessyreal

world.Theycontributetotheimpressionthatwe’reataninflectionpoint—a

bendinthecurvewheremanytechnologiesthatusedtobefoundonlyinscience

fictionarebecomingeverydayreality.Asmanyotherexamplesshow,thisisan

accurateimpression.

OntheStarTrektelevisionseries,devicescalledtricorderswereusedtoscan andrecordthreekindsofdata:geological,meteorological,andmedical.Today’s consumersmartphonesserveallthesepurposes;theycanbeputtoworkas seismographs, real-time weather radar maps, and heart- and breathing-rate monitors. 35 And,ofcourse,they’renotlimitedtothesedomains.Theyalsowork asmediaplayers,gameplatforms,referenceworks,cameras,andGPSdevices. On Star Trek, tricorders and person-to-person communicators were separate devices,butintherealworldthetwohavemergedinthesmartphone.They enable their users to simultaneously access and generate huge amounts of informationastheymovearound.Thisopensuptheopportunityforinnovations thatventurecapitalistJohnDoerrcalls“SoLoMo”—social,local,andmobile. 36 Computershistoricallyhavebeenverybadatwritingrealprose.Inrecent timestheyhavebeenabletogenerategrammaticallycorrectbutmeaningless sentences,astateofaffairsthat’sbeenmercilesslyexploitedbypranksters.In 2008, for example, the International Conference on Computer Science and Software Engineering accepted the paper “Towards the Simulation of E- commerce”andinviteditsauthortochairasession.Thispaperwas‘written’by SCIgen,aprogramfromtheMITComputerScienceandArtificialIntelligence Lab that “generates random Computer Science research papers.” SCIgen’s authors wrote that, “Our aim here is to maximize amusement, rather than

coherence,”andafterreadingtheabstractof“TowardstheSimulationofE-

Recentadvancesincooperativetechnologyandclassicalcommunicationarebasedentirely on the assumption that the Internet and active networks are not in conflict with object- orientedlanguages.Infact,fewinformationtheoristswoulddisagreewiththevisualizationof

DHTsthatmaderefiningandpossiblysimulating8bitarchitecturesareality,whichembodies

thecompellingprinciplesofelectricalengineering.38

Recentdevelopmentsmakeclear,though,thatnotallcomputer-generatedprose isnonsensical.Forbes.comhascontractedwiththecompanyNarrativeScience to write the corporate earnings previews that appear on the website. These storiesareallgeneratedbyalgorithmswithouthumaninvolvement.Andthey’re indistinguishablefromwhatahumanwouldwrite:

ForbesEarningPreview:H.J.Heinz AqualityfirstquarterearningsannouncementcouldpushsharesofH.J.Heinz(HNZ)toa

new52-weekhighasthepriceisjust49centsoffthemilestoneheadingintothecompany’s

earningsreleaseonWednesday,August29,2012.

TheWallStreetconsensusis80centspershare,up2.6percentfromayearagowhen

H.Jreportedearningsof78centspershare.

Theconsensusestimateremainsunchangedoverthepastmonth,butithasdecreased

fromthreemonthsagowhenitwas82cents.Analystsareexpectingearningsof$3.52per

shareforthefiscalyear.Analystsprojectrevenuetofall0.3percentyear-over-yearto$2.84

billionforthequarter,afterbeing$2.85billionayearago.Fortheyear,revenueisprojected

torollinat$11.82billion.39

Evencomputerperipheralslikeprintersaregettinginontheact,demonstrating usefulcapabilitiesthatseemstraightoutofsciencefiction.Insteadofjustputting inkonpaper,theyaremakingcomplicatedthree-dimensionalpartsoutofplastic, metal, and other materials. 3D printing, also sometimes called “additive manufacturing,” takes advantage of the way computer printers work: they depositaverythinlayerofmaterial(ink,traditionally)onabase(paper)ina patterndeterminedbythecomputer. Innovatorsreasonedthatthereisnothingstoppingprintersfromdepositing layersoneontopoftheother.Andinsteadofink,printerscanalsodeposit materialslikeliquidplasticthatgetscuredintoasolidbyultravioletlight.Each layerisverythin—somewherearoundone-tenthofamillimeter—butovertime athree-dimensionalobjecttakesshape.Andbecauseofthewayitisbuiltup, thisshapecanbequitecomplicated—itcanhavevoidsandtunnelsinit,and even parts that move independently of one another. At the San Francisco headquartersofAutodesk,aleadingdesignsoftwarecompany,wehandleda working adjustable wrench that was printed as a single part, no assembly required. 40

Thiswrenchwasademonstrationproductmadeoutofplastic,but3Dprinting

hasexpandedintometalsaswell.AutodeskCEOCarlBassispartofthelarge and growing community of additive manufacturing hobbyists and tinkerers. Duringourtourofhiscompany’sgallery,ashowcaseofalltheproductsand projectsenabledbyAutodesksoftware,heshowedusabeautifulmetalbowlhe designedonacomputerandhadprintedout.Thebowlhadanelaboratelattice pattern on its sides. Bass said that he’d asked friends of his who were experiencedinworkingwithmetal—sculptors,ironworkers,welders,andsoon —howthebowlwasmade.Noneofthemcouldfigureouthowthelatticewas produced. The answer was that a laser had built up each layer by fusing powderedmetal.

3DprintingtodayisnotjustforartprojectslikeBass’sbowl.It’susedby

countlesscompanieseverydaytomakeprototypesandmodelparts.It’salso

beingusedforfinalpartsrangingfromplasticventsandhousingsonNASA’s

next-generationMoonrovertoametalprostheticjawboneforaneighty-three-

year-oldwoman.Inthenearfuture,itmightbeusedtoprintoutreplacement partsforfaultyenginesonthespotinsteadofmaintainingstockpilesofthemin inventory.Demonstrationprojectshaveevenshownthatthetechniquecouldbe usedtobuildconcretehouses. 41 Mostoftheinnovationsdescribedinthischapterhaveoccurredinjustthe past few years. They’ve taken place in areas where improvement had been frustratinglyslowforalongtime,andwherethebestthinkingoftenledtothe conclusionthatitwouldn’tspeedup.Butthendigitalprogressbecamesudden afterbeinggradualforsolong.Thishappenedinmultipleareas,fromartificial intelligencetoself-drivingcarstorobotics.

Howdidthishappen?Wasitafluke—aconfluenceofanumberofluckyone-

time advances? No, it was not. The digital progress we’ve seen recently is certainlyimpressive,butit’sjustasmallindicationofwhat’stocome.It’sthe dawnofthesecondmachineage.Tounderstandwhyit’sunfoldingnow,weneed tounderstandthenatureoftechnologicalprogressintheeraofdigitalhardware, software, and networks. In particular, we need to understand its three key characteristics:thatitisexponential,digital,andcombinatorial.Thenextthree chapterswilldiscusseachoftheseinturn.

.Thenextthree chapterswilldiscusseachoftheseinturn. * In the years leading up to the Great Recession that began

* In the years leading up to the Great Recession that began in 2007, companies were giving

mortgages to people with lower and lower credit scores, income, and wealth, and higher and higherdebtlevels.Inotherwords,theyeitherrewroteorignoredtheirpreviousmortgageapproval algorithms.Itwasn’tthattheoldmortgagealgorithmsstoppedworking;itwasthattheystopped beingused.

*Tobeprecise,Trebekreadsanswersandthecontestantshavetostatethequestionthatwould

giverisetothisanswer.

*Sensorimotorskillsarethosethatinvolvesensingthephysicalworldandcontrollingthebodyto

movethroughit.

“Thegreatestshortcomingofthehumanraceisourinabilitytounderstandthe exponentialfunction.” —AlbertA.Bartlett

“Thegreatestshortcomingofthehumanraceisourinabilitytounderstandthe

exponentialfunction.”

—AlbertA.Bartlett

ALTHOUGHHESCOFOUNDEROFIntel,amajorphilanthropist,andrecipientofthe PresidentialMedalofFreedom,GordonMooreisbestknownforapredictionhe

made,almostasanaside,ina1965article.Moore,thenworkingatFairchild

Semiconductor,wroteanarticleforElectronicsmagazinewiththeadmirably directtitle“CrammingMoreComponentsontoIntegratedCircuits.”Atthetime, circuits of this type—which combined many different kinds of electrical components onto a single chip made primarily of silicon—were less than a decadeold,butMooresawtheirpotential.Hewrotethat,“Integratedcircuits willleadtosuchwondersashomecomputers—oratleastterminalsconnectedto acentralcomputer—automaticcontrolsforautomobiles,andpersonalportable communicationsequipment.” 1 The article’s most famous forecast, however, and the one that has made Mooreahouseholdname,concernedthecomponentcrammingofthetitle:

Thecomplexityforminimumcomponentcostshasincreasedatarateofroughlyafactorof

twoperyear

increase.Overthelongerterm,therateofincreaseisabitmoreuncertain,althoughthereis

noreasontobelieveitwillnotremainnearlyconstantforatleasttenyears.2

Certainlyovertheshorttermthisratecanbeexpectedtocontinue,ifnotto

ThisistheoriginalstatementofMoore’sLaw,andit’sworthdwellingfora momentonitsimplications.“Complexityforminimumcomponentcosts”here essentiallymeanstheamountofintegratedcircuitcomputingpoweryoucould buyforonedollar.Mooreobservedthatovertherelativelybriefhistoryofhis industrythisamounthaddoubledeachyear:youcouldbuytwiceasmuchpower

perdollarin1963asyoucouldin1962,twiceasmuchagainin1964,andtwice

asmuchagainin1965.

Moore predicted this state of affairs would continue, perhaps with some changetotiming,foratleastanothertenyears.Thisboldstatementforecast

circuitsthatwouldbemorethanfivehundredtimesaspowerfulin1975asthey

werein1965.*

As it turned out, however, Moore’s biggest mistake was in being too conservative.His“law”hasheldupastonishinglywellforoverfourdecades,not justone,andhasbeentruefordigitalprogressinareasotherthanintegrated circuits.It’sworthnotingthatthetimerequiredfordigitaldoublingremainsa

matterofdispute.In1975Moorerevisedhisestimateupwardfromoneyearto

two,andtodayit’scommontouseeighteenmonthsasthedoublingperiodfor

generalcomputingpower.Still,there’snodisputethatMoore’sLawhasproved remarkablyprescientforalmosthalfacentury. 3

It’sNotaLaw:It’saBunchofGoodIdeas

Moore’s Law is very different from the laws of physics that govern thermodynamicsorNewtonianclassicalmechanics.Thoselawsdescribehow theuniverseworks;they’retruenomatterwhatwedo.Moore’sLaw,incontrast, isastatementabouttheworkofthecomputerindustry’sengineersandscientists; it’sanobservationabouthowconstantandsuccessfultheireffortshavebeen.We simplydon’tseethiskindofsustainedsuccessinotherdomains. Therewasnoperiodoftimewhencarsgottwiceasfastortwiceasfuel efficienteveryyearortwoforfiftyyears.Airplanesdon’tconsistentlyhavethe abilitytoflytwiceasfar,ortrainstheabilitytohaultwiceasmuch.Olympic runnersandswimmersdon’tcuttheirtimesinhalfoverageneration,letalonea coupleofyears. So how has the computer industry kept up this amazing pace of improvement? Therearetwomainreasons.First,whiletransistorsandtheotherelementsof computingareconstrainedbythelawsofphysicsjustlikecars,airplanes,and swimmers,theconstraintsinthedigitalworldaremuchlooser.Theyhavetodo withhowmanyelectronspersecondcanbeputthroughachanneletchedinan integratedcircuit,orhowfastbeamsoflightcantravelthroughfiber-opticcable. AtsomepointdigitalprogressbumpsupagainstitsconstraintsandMoore’sLaw mustslowdown,butittakesawhile.HenrySamueli,chieftechnologyofficerof chipmaker Broadcom Corporation, predicted in 2013 that “Moore’s Law is comingtoanend—inthenextdecadeitwillprettymuchcometoanendsowe have15yearsorso.” 4 ButsmartpeoplehavebeenpredictingtheendofMoore’sLawforawhile now,andthey’vebeenprovedwrongoverandoveragain. 5 Thisisnotbecause theymisunderstoodthephysicsinvolved,butbecausetheyunderestimatedthe peopleworkinginthecomputerindustry.ThesecondreasonthatMoore’sLaw hasheldupsowellforsolongiswhatwemightcall‘brillianttinkering’— findingengineeringdetoursaroundtheroadblocksthrownupbyphysics.When itbecamedifficulttocramintegratedcircuitsmoretightlytogether,forexample, chipmakersinsteadlayeredthemontopofoneanother,openingupagreatdeal of new real estate. When communications traffic threatened to outstrip the

capacity even of fiber-optic cable, engineers developed wavelength division multiplexing (WDM), a technique for transmitting many beams of light of differentwavelengthsdownthesamesingleglassfiberatthesametime.Over andoveragainbrillianttinkeringhasfoundwaystoskirtthelimitationsimposed byphysics.AsIntelexecutiveMikeMarberryputsit,“Ifyou’reonlyusingthe sametechnology,theninprincipleyourunintolimits.Thetruthiswe’vebeen

modifyingthetechnologyeveryfiveorsevenyearsfor40years,andthere’sno

endinsightforbeingabletodothat.” 6 Thisconstantmodificationhasmade Moore’sLawthecentralphenomenonofthecomputerage.Thinkofitasa steadydrumbeatinthebackgroundoftheeconomy.

ChartingthePowerofConstantDoubling

Once this doubling has been going on for some time, the later numbers overwhelmtheearlierones,makingthemappearirrelevant.Toseethis,let’s lookatahypotheticalexample.ImaginethatErikgivesAndyatribble,thefuzzy creaturewithahighreproductiveratemadefamousinanepisodeofStarTrek. Every day each tribble gives birth to another tribble, so Andy’s menagerie doublesinsizeeachday.Ageekwouldsayinthiscasethatthetribblefamilyis experiencingexponentialgrowth.That’sbecausethemathematicalexpression forhowmanytribblesthereareondayxis2x –1 ,wherethex1isreferredtoas anexponent.Exponentialgrowthlikethisisfastgrowth;aftertwoweeksAndy hasmorethansixteenthousandofthecreatures.Here’sagraphofhowhis tribblefamilygrowsovertime:

FIGURE3.1TribblesoverTime:ThePowerofConstantDoubling

Thisgraphisaccurate,butmisleadinginanimportantsense.Itseemstoshow

Thisgraphisaccurate,butmisleadinginanimportantsense.Itseemstoshow thatalltheactionoccursinthelastcoupleofdays,withnothingmuchhappening inthefirstweek.Butthesamephenomenon—thedailydoublingoftribbles— hasbeengoingonthewholetimewithnoaccelerationsordisruptions.This steadyexponentialgrowthiswhat’sreallyinterestingaboutErik’s‘gift’toAndy. Tomakeitmoreobvious,wehavetochangethespacingofthenumbersonthe graph. Thegraphwe’vealreadydrawnhasstandardlinearspacing;eachsegmentof theverticalaxisindicatestwothousandmoretribbles.Thisisfineformany purposesbut,aswe’veseen,it’snotgreatforshowingexponentialgrowth.To highlightitbetter,we’llchangetologarithmicspacing,whereeachsegmentof

theverticalaxisrepresentsatenfoldincreaseintribbles:anincreasefirstfrom1

to10,thenfrom10to100,thenfrom100to1,000,andsoon.Inotherwords,we

scaletheaxisbypowersof10orordersofmagnitude.

Logarithmicgraphshaveawonderfulproperty:theyshowexponentialgrowth

asaperfectlystraightline.Here’swhatthegrowthofAndy’stribblefamily

lookslikeonalogarithmicscale:

FIGURE3.2TribblesoverTime:ThePowerofConstantDoubling

Thisviewemphasizesthesteadinessofthedoublingovertimeratherthanthe

Thisviewemphasizesthesteadinessofthedoublingovertimeratherthanthe largenumbersattheend.Becauseofthis,weoftenuselogarithmicscalesfor graphing doublings and other exponential growth series. They show up as straightlinesandtheirspeediseasiertoevaluate;thebiggertheexponent,the fastertheygrow,andthesteepertheline.

ImpoverishedEmperors,HeadlessInventors,andthe

SecondHalfoftheChessboard

Ourbrainsarenotwellequippedtounderstandsustainedexponentialgrowth.In particular,weseverelyunderestimatehowbigthenumberscanget.Inventorand futuristRayKurzweilretellsanoldstorytodrivethispointhome.Thegameof chessoriginatedinpresent-dayIndiaduringthesixthcenturyCE,thetimeofthe GuptaEmpire. 7 Asthestorygoes,itwasinventedbyaveryclevermanwho traveled to Pataliputra, the capital city, and presented his brainchild to the emperor.Therulerwassoimpressedbythedifficult,beautifulgamethathe invitedtheinventortonamehisreward. Theinventorpraisedtheemperor’sgenerosityandsaid,“AllIdesireissome rice to feed my family.” Since the emperor’s largess was spurred by the inventionofchess,theinventorsuggestedtheyusethechessboardtodetermine theamountofricehewouldbegiven.“Placeonesinglegrainofriceonthefirst square of the board, two on the second, four on the third, and so on,” the inventorproposed,“sothateachsquarereceivestwiceasmanygrainsasthe previous.”

“Make it so,” the emperor replied, impressed by the inventor’s apparent modesty. Moore’sLawandthetribbleexerciseallowustoseewhattheemperordid not:sixty-threeinstancesofdoublingyieldsafantasticallybignumber,even whenstartingwithasingleunit.Ifhisrequestwerefullyhonored,theinventor wouldwindupwith2 641 ,ormorethaneighteenquintilliongrainsofrice.A pileofricethisbigwoulddwarfMountEverest;it’smorericethanhasbeen producedinthehistoryoftheworld.Ofcourse,theemperorcouldnothonor sucharequest.Insomeversionsofthestory,onceherealizesthathe’sbeen tricked,hehastheinventorbeheaded.

Kurzweiltellsthestoryoftheinventorandtheemperorinhis2000bookThe

AgeofSpiritualMachines:WhenComputersExceedHumanIntelligence.He aimsnotonlytoillustratethepowerofsustainedexponentialgrowthbutalsoto highlight the point at which the numbers start to become so big they are inconceivable:

Afterthirty-twosquares,theemperorhadgiventheinventorabout4billiongrainsofrice.

That’sareasonablequantity—aboutonelargefield’sworth—andtheemperordidstartto takenotice. Buttheemperorcouldstillremainanemperor.Andtheinventorcouldstillretainhishead. Itwasastheyheadedintothesecondhalfofthechessboardthatatleastoneofthemgot

intotrouble.8

Kurzweil’sgreatinsightisthatwhilenumbersdogetlargeinthefirsthalfof thechessboard,westillcomeacrossthemintherealworld.Fourbilliondoesnot necessarily outstrip our intuition. We experience it when harvesting grain, assessingthefortunesoftheworld’srichestpeopletoday,ortallyingupnational debtlevels.Inthesecondhalfofthechessboard,however—asnumbersmount intotrillions,quadrillions,andquintillions—weloseallsenseofthem.Wealso lose sense of how quickly numbers like these appear as exponential growth continues. Kurzweil’sdistinctionbetweenthefirstandsecondhalvesofthechessboard inspired a quick calculation. Among many other things, the U.S. Bureau of EconomicAnalysis(BEA)tracksAmericancompanies’expenditures.TheBEA firstnoted“informationtechnology”asadistinctcorporateinvestmentcategory

in1958.WetookthatyearasthestartingpointforwhenMoore’sLawentered

the business world, and used eighteen months as the doubling period. After thirty-twoofthesedoublings,U.S.businessesenteredthesecondhalfofthe

chessboardwhenitcomestotheuseofdigitalgear.Thatwasin2006.

Ofcourse,thiscalculationisjustafunlittleexercise,notanythinglikea seriousattempttoidentifytheonepointatwhicheverythingchangedinthe worldofcorporatecomputing.Youcouldeasilyarguewiththestartingpointof

1958andadoublingperiodofeighteenmonths.Changestoeitherassumption

wouldyieldadifferentbreakpointbetweenthefirstandsecondhalvesofthe chessboard.Andbusinesstechnologistswerenotonlyinnovatinginthesecond half;aswe’lldiscusslater,thebreakthroughsoftodayandtomorrowrelyon,and wouldbeimpossiblewithout,thoseofthepast. Wepresentthiscalculationherebecauseitunderscoresanimportantidea:that exponentialgrowtheventuallyleadstostaggeringlybignumbers,onesthatleave ourintuitionandexperiencebehind.Inotherwords,thingsgetweirdinthe secondhalfofthechessboard.Andliketheemperor,mostofushavetrouble keepingup. Oneofthethingsthatsetsthesecondmachineageapartishowquicklythat second half of the chessboard can arrive. We’re not claiming that no other technologyhaseverimprovedexponentially.Infact,aftertheone-timeburstof improvement in the steam engine Watt’s innovations created, additional tinkeringledtoexponentialimprovementovertheensuingtwohundredyears. Buttheexponentswererelativelysmall,soitonlywentthroughaboutthreeor fourdoublingsinefficiencyduringthatperiod. 9 Itwouldtakeamillenniumto reachthesecondhalfofthechessboardatthatrate.Inthesecondmachineage, thedoublingshappenmuchfasterandexponentialgrowthismuchmoresalient.

Second-HalfTechnologies

Our quick doubling calculation also helps us understand why progress with digitaltechnologiesfeelssomuchfasterthesedaysandwhywe’veseensomany recentexamplesofsciencefictionbecomingbusinessreality.It’sbecausethe steadyandrapidexponentialgrowthofMoore’sLawhasaddeduptothepoint thatwe’renowinadifferentregimeofcomputing:we’renowinthesecondhalf ofthechessboard.Theinnovationswedescribedinthepreviouschapter—cars that drive themselves in traffic; Jeopardy!-champion supercomputers; auto- generatednewsstories;cheap,flexiblefactoryrobots;andinexpensiveconsumer devices that are simultaneously communicators, tricorders, and computers—

haveallappearedsince2006,ashavecountlessothermarvelsthatseemquite

differentfromwhatcamebefore.

Oneofthereasonsthey’reallappearingnowisthatthedigitalgearattheir

heartsisfinallybothfastandcheapenoughtoenablethem.Thiswasn’tthecase

justadecadeago.Whatdoesdigitalprogresslooklikeonalogarithmicscale?

Let’stakealook.

FIGURE3.3TheManyDimensionsofMoore’sLaw

Let’stakealook. FIGURE3.3 TheManyDimensionsofMoore’sLaw

ThisgraphshowsthatMoore’sLawisbothconsistentandbroad;it’sbeenin forceforalongtime(decades,insomecases)andappliestomanytypesof digitalprogress.Asyoulookatit,keepinmindthatifitusedstandardlinear scalingontheverticalaxis,allofthosestraight-ishlineswouldlooklikethefirst graph above of Andy’s tribble family—horizontal most of the way, then suddenlyclosetoverticalattheend.Andtherewouldreallybenowaytograph them all together—the numbers involved are just too different. Logarithmic scalingtakescareoftheseissuesandallowsustogetaclearoverallpictureof improvementindigitalgear. It’sclearthatmanyofthecriticalbuildingblocksofcomputing—microchip density,processingspeed,storagecapacity,energyefficiency,downloadspeed, and so on—have been improving at exponential rates for a long time. To understandthereal-worldimpactsofMoore’sLaw,let’scomparethecapabilities ofcomputersseparatedbyonlyafewdoublingperiods.TheASCIRed,thefirst productoftheU.S.government’sAcceleratedStrategicComputingInitiative,

wastheworld’sfastestsupercomputerwhenitwasintroducedin1996.Itcost

square feet of floor space (80 percent of a tennis court) at Sandia National Laboratories in New Mexico. 10 Designed for calculation-intensive tasks like simulatingnucleartests,ASCIRedwasthefirstcomputertoscoreaboveone teraflop—onetrillionfloatingpointoperations*persecond—onthestandard benchmarktestforcomputerspeed.Toreachthisspeeditusedeighthundred

kilowattsperhour,aboutasmuchaseighthundredhomeswould.By1997,it

hadreached1.8teraflops.

Nineyearslateranothercomputerhit1.8teraflops.Butinsteadofsimulating

nuclearexplosions,itwasdevotedtodrawingthemandothercomplexgraphics in all their realistic, real-time, three-dimensional glory. It did this not for physicists,butforvideogameplayers.ThiscomputerwastheSonyPlayStation

3,whichmatchedtheASCIRedinperformance,yetcostaboutfivehundred

dollars,tookuplessthanatenthofasquaremeter,anddrewabouttwohundred watts. 11 In less than ten years exponential digital progress brought teraflop calculatingpowerfromasinglegovernmentlabtolivingroomsandcollege

dormsallaroundtheworld.ThePlayStation3soldapproximately64million

units.TheASCIRedwastakenoutofservicein2006.

Exponentialprogresshasmadepossiblemanyoftheadvancesdiscussedin thepreviouschapter.IBM’sWatsondrawsonaplethoraofcleveralgorithms, butitwouldbeuncompetitivewithoutcomputerhardwarethatisaboutone hundredtimesmorepowerfulthanDeepBlue,itschess-playingpredecessorthat beat the human world champion, Garry Kasparov, in a 1997 match. Speech recognitionapplicationslikeSirirequirelotsofcomputingpower,whichbecame

availableonmobilephoneslikeApple’siPhone4S(thefirstphonethatcame

withSiriinstalled).TheiPhone4Swasaboutaspowerful,infact,asApple’s

top-of-the-linePowerbookG4laptophadbeenadecadeearlier.Asallofthese

innovationsshow,exponentialprogressallowstechnologytokeepracingahead

andmakessciencefictionrealityinthesecondhalfofthechessboard.

NotJustforComputersAnymore:TheSpreadofMoore’s

Law

Anothercomparisonacrosscomputergenerationshighlightsnotonlythepower ofMoore’sLawbutalsoitswidereach.AsisthecasewiththeASCIRedand

thePlayStation3,theCray-2supercomputer(introducedin1985)andiPad2

tablet(introducedin2011)hadalmostidenticalpeakcalculationspeeds.Butthe

iPadalsohadaspeaker,microphone,andheadphonejack.Ithadtwocameras;

theoneonthefrontofthedevicewasVideoGraphicsArray(VGA)quality, whiletheoneonthebackcouldcapturehigh-definitionvideo.Bothcouldalso

takestillphotographs,andthebackcamerahada5xdigitalzoom.Thetablethad

receivers that allowed it to participate in both wireless telephone and Wi-Fi networks.ItalsohadaGPSreceiver,digitalcompass,accelerometer,gyroscope, andlightsensor.Ithadnobuilt-inkeyboard,relyinginsteadonahigh-definition touchscreenthatcouldtrackuptoelevenpointsofcontactsimultaneously. 12 It

fitallofthiscapabilityintoadevicethatcostmuchlessthan$1,000andwas

smaller,thinner,andlighterthanmanymagazines.TheCray-2,whichcostmore

than$35million(in2011dollars),wasbycomparisondeaf,dumb,blind,and

immobile. 13

ApplewasabletocramallofthisfunctionalityintheiPad2becauseabroad

shifthastakenplaceinrecentdecades:sensorslikemicrophones,cameras,and accelerometers have moved from the analog world to the digital one. They became,inessence,computerchips.Astheydidso,theybecamesubjecttothe exponentialimprovementtrajectoriesofMoore’sLaw.

Digitalgearforrecordingsoundswasinusebythe1960s,andanEastman

Kodakengineerbuiltthefirstmoderndigitalcamerain1975. 14 Earlydevices wereexpensiveandclunky,butqualityquicklyimprovedandpricesdropped.

Kodak’sfirstdigitalsingle-lensreflexcamera,theDCS100,costabout$13,000

whenitwasintroducedin1991;ithadamaximumresolutionof1.3megapixels

andstoreditsimagesinaseparate,ten-poundharddrivethatusersslungover theirshoulders.However,thepixelsperdollaravailablefromdigitalcameras doubledabouteveryyear(aphenomenonknownas“Hendy’sLaw”afterKodak AustraliaemployeeBarryHendy,whodocumentedit),andallrelatedgeargot exponentially smaller, lighter, cheaper, and better over time. 15 Accumulated improvement in digital sensors meant that twenty years after the DCS 100, Apple could include two tiny cameras, capable of both still and video

photography,ontheiPad2.AndwhenitintroducedanewiPadthefollowing

year,therearcamera’sresolutionhadimprovedbyafactorofmorethanseven.

MachineEyes

AsMoore’sLawworksovertimeonprocessors,memory,sensors,andmany otherelementsofcomputerhardware(anotableexceptionisbatteries,which haven’t improved their performance at an exponential rate because they’re essentially chemical devices, not digital ones), it does more than just make

computingdevicesfaster,cheaper,smaller,andlighter.Italsoallowsthemtodo thingsthatpreviouslyseemedoutofreach. Researchersinartificialintelligencehavelongbeenfascinated(somewould sayobsessed)withtheproblemofsimultaneouslocalizationandmapping,which they refer to as SLAM. SLAM is the process of building up a map of an unfamiliarbuildingasyou’renavigatingthroughit—wherearethedoors?where arestairs?whatareallthethingsImighttripover?—andalsokeepingtrackof whereyouarewithinit(soyoucanfindyourwaybackdownstairsandoutthe frontdoor).Forthegreatmajorityofhumans,SLAMhappenswithminimal consciousthought.Butteachingmachinestodoithasbeenahugechallenge. Researchers thought a great deal about which sensors to give a robot (cameras?lasers?sonar?)andhowtointerpretthereamsofdatatheyprovide,

butprogresswasslow.Asa2008reviewofthetopicsummarized,SLAM“is

oneofthefundamentalchallengesofrobotics

thecurrentapproachescannotperformconsistentmapsforlargeareas,mainly duetotheincreaseofthecomputationalcostandduetotheuncertaintiesthat becomeprohibitivewhenthescenariobecomeslarger.” 16 Inshort,sensinga sizable area and immediately crunching all the resulting data were thorny

problemspreventingrealprogresswithSLAM.Until,thatis,a$150video-game

accessorycamealongjusttwoyearsafterthesentencesabovewerepublished.

InNovember2010MicrosoftfirstofferedtheKinectsensingdeviceasan

additiontoitsXboxgamingplatform.TheKinectcouldkeeptrackoftwoactive players,monitoringasmanyastwentyjointsoneach.Ifoneplayermovedin frontoftheother,thedevicemadeabestguessabouttheobscuredperson’s movements,thenseamlesslypickedupalljointsonceheorshecamebackinto view.Kinectcouldalsorecognizeplayers’faces,voices,andgesturesanddoso acrossawiderangeoflightingandnoiseconditions.Itaccomplishedthiswith digitalsensorsincludingamicrophonearray(whichpinpointedthesourceof soundbetterthanasinglemicrophonecould),astandardvideocamera,anda depthperceptionsystemthatbothprojectedanddetectedinfraredlight.Several onboardprocessorsandagreatdealofproprietarysoftwareconvertedtheoutput ofthesesensorsintoinformationthatgamedesignerscoulduse. 17 Atlaunch,all ofthiscapabilitywaspackedintoafour-inch-talldevicelessthanafootwide

thatretailedfor$149.99.

The Kinect sold more than eight million units in the sixty days after its release(morethaneithertheiPhoneoriPad)andcurrentlyholdstheGuinness WorldRecordforthefastest-sellingconsumerelectronicsdeviceofalltime. 18

[butit]seemsthatalmostall

TheinitialfamilyofKinect-specificgamesletplayersplaydarts,exercise,brawl inthestreets,andcastspellsàlaHarryPotter. 19 These,however,didnotcome close to exhausting the system’s possibilities. In August of 2011 at the SIGGRAPH (short for the Association of Computing Machinery’s Special Interest Group on Graphics and Interactive Techniques) conference in Vancouver,BritishColumbia,ateamofMicrosoftemployeesandacademics usedKinectto“SLAM”thedoorshutonalong-standingchallengeinrobotics. SIGGRAPHisthelargestandmostprestigiousgatheringdevotedtoresearch and practice on digital graphics, attended by researchers, game designers, journalists,entrepreneurs,andmostothersinterestedinthefield.Thismadeitan appropriate place for Microsoft to unveil what the Creators Project website called “The Self-Hack That Could Change Everything.”* 20 This was the KinectFusion,aprojectthatusedtheKinecttotackletheSLAMproblem.

InavideoshownatSIGGRAPH2011,apersonpicksupaKinectandpoints

it around a typical office containing chairs, a potted plant, and a desktop computerandmonitor. 21 Ashedoes,thevideosplitsintomultiplescreensthat showwhattheKinectisabletosense.Itimmediatelybecomesclearthatifthe KinectisnotcompletelysolvingtheSLAMproblemfortheroom,it’scoming close.Inrealtime,Kinectdrawsathree-dimensionalmapoftheroomandallthe objectsinit,includingacoworker.ItpicksupthewordDELLpressedintothe plasticonthebackofthecomputermonitor,eventhoughthelettersarenot coloredandonlyonemillimeterdeeperthattherestofthemonitor’ssurface. Thedeviceknowswhereitisintheroomatalltimes,andevenknowshow virtual ping-pong balls would bounce around if they were dropped into the scene.AsthetechnologyblogEngadgetputitinapost-SIGGRAPHentry,“The

Kinecttook3Dsensingtothemainstream,andmoreover,allowedresearchersto

pickupacommodityproductandgoabsolutelynuts.” 22

InJuneof2011,shortlybeforeSIGGRAPH,Microsofthadmadeavailablea

Kinectsoftwaredevelopmentkit(SDK)givingprogrammerseverythingthey needed to start writing PC software that made use of the device. After the conferencetherewasagreatdealofinterestinusingtheKinectforSLAM,and manyteamsinroboticsandAIresearchdownloadedtheSDKandwenttowork. Inlessthanayear,ateamofIrishandAmericanresearchersledbyour colleagueJohnLeonardofMIT’sComputerScienceandArtificialIntelligence Lab announced Kintinuous, a “spatially extended” version of KinectFusion. WithKintinuous,userscoulduseaKinecttoscanlargeindoorvolumeslike apartmentbuildingsandevenoutdoorenvironments(whichtheteamscannedby

holdingaKinectoutsideacarwindowduringanighttimedrive).Attheendof thepaperdescribingtheirwork,theKintinuousresearcherswrote,“Inthefuture wewillextendthesystemtoimplementafullSLAMapproach.” 23 Wedon’t think it will be long until they announce success. When given to capable technologists,theexponentialpowerofMoore’sLaweventuallymakeseventhe toughestproblemstractable. Cheapandpowerfuldigitalsensorsareessentialcomponentsofsomeofthe science-fictiontechnologiesdiscussedinthepreviouschapter.TheBaxterrobot hasmultipledigitalcamerasandanarrayofforceandpositiondetectors.Allof thesewouldhavebeenunworkablyexpensive,clunky,andimprecisejustashort timeago.AGoogleautonomouscarincorporatesseveralsensingtechnologies, butitsmostimportant‘eye’isaCyclopeanLIDAR(acombinationof“LIght” and “raDAR”) assembly mounted on the roof. This rig, manufactured by

Velodyne, contains sixty-four separate laser beams and an equal number of detectors,allmountedinahousingthatrotatestentimesasecond.Itgenerates

about1.3milliondatapointspersecond,whichcanbeassembledbyonboard

computers into a real-time 3D picture extending one hundred meters in all directions.SomeearlycommercialLIDARsystemsavailablearoundtheyear

2000costupto$35million,butinmid-2013Velodyne’sassemblyforself-

navigatingvehicleswaspricedatapproximately$80,000,afigurethatwillfall

much further in the future. David Hall, the company’s founder and CEO, estimatesthatmassproductionwouldallowhisproduct’spriceto“droptothe levelofacamera,afewhundreddollars.” 24 Alltheseexamplesillustratethefirstelementofourthree-partexplanationof whywe’renowinthesecondmachineage:steadyexponentialimprovementhas broughtusintothesecondhalfofthechessboard—intoatimewhenwhat’s comebeforeisnolongeraparticularlyreliableguidetowhatwillhappennext. The accumulated doubling of Moore’s Law, and the ample doubling still to come,givesusaworldwheresupercomputerpowerbecomesavailabletotoysin justafewyears,whereever-cheapersensorsenableinexpensivesolutionsto previously intractable problems, and where science fiction keeps becoming reality. Sometimesadifferenceindegree(inotherwords,moreofthesame)becomes adifferenceinkind(inotherwords,differentthananythingelse).Thestoryof thesecondhalfofthechessboardalertsusthatweshouldbeawarethatenough exponentialprogresscantakeustoastonishingplaces.Multiplerecentexamples convinceusthatwe’realreadythere.

* Since29=512 * Multiplying62.34by24358.9274isanexampleofafloatingpointoperation.Thedecimalpointin

*Since29=512

*Multiplying62.34by24358.9274isanexampleofafloatingpointoperation.Thedecimalpointin

suchoperationsisallowedto‘float’insteadofbeingfixedinthesameplaceforbothnumbers.

*Inthiscontext,a“hack”isanefforttogetinsidethegutsofapieceofdigitalgearanduseitfor

anunorthodoxpurpose.Aself-hackisonecarriedoutbythecompanythatmadethegearinthe

firstplace.

“Whenyoucanmeasurewhatyouarespeakingabout,andexpressitinnumbers,you

“Whenyoucanmeasurewhatyouarespeakingabout,andexpressitinnumbers,you

knowsomethingaboutit;butwhenyoucannotexpressitinnumbers,yourknowledgeis

ofameagreandunsatisfactorykind.”

—LordKelvin

“HEY,HAVEYOUHEARDabout

“You’vegottocheckout

?”

Questions and recommendations like these are the stuff of everyday life. They’rehowwelearnaboutnewthingsfromourfriends,family,andcolleagues, and how we spread the word about exciting things we’ve come across. Traditionally,suchcoolhuntingendedwiththenameofaband,restaurant,place tovisit,TVshow,book,ormovie. Inthedigitalage,sentenceslikethesefrequentlyendwiththenameofa website or a gadget. And right now, they’re often about a smartphone application.Bothofthemajortechnologyplatformsinthismarket—Apple’siOS and Google’s Android—have more than five hundred thousand applications available. 1 Thereareplentyof“Top10”and“Bestof”listsavailabletohelp usersfindthecreamofthesmartphoneappcrop,buttraditionalwordofmouth hasretaineditspower.

NotlongagoMattBeane,adoctoralstudentattheMITSloanSchoolof

ManagementandamemberofourDigitalFrontierteam,gaveusatip.“You’ve

gottocheckoutWaze;it’samazing.”ButwhenwefoundoutitwasaGPS-

basedappthatprovideddrivingdirections,weweren’timmediatelyimpressed.

OurcarshavenavigationsystemsandouriPhonescangivedrivingdirections

throughtheMapsapplication.Wecouldnotseeaneedforyetanotherhow-do-I-

get-theretechnology.

AsMattpatientlyexplained,usingWazeislikebringingaDucatitoadrag

raceagainstanoxcart.UnliketraditionalGPSnavigation,Wazedoesn’ttellyou

whatroutetoyourdestinationisbestingeneral;ittellsyouwhatrouteisbest

rightnow.Asthecompanywebsiteexplains:

wasgivenaPDAwithan

externalGPSdevicepre-installedwithnavigationsoftware.Ehud’sinitialexcitementquickly gave way to disappointment—the product didn’t reflect the dynamic changes that characterizerealconditionsontheroad

TheideaforWazeoriginatedyearsago,whenEhudShabtai

Ehudtookmattersintohisownhands

Hisgoal?Toaccuratelyreflecttheroadsystem,

stateoftrafficandalltheinformationrelevanttodriversatanygivenmoment.2

Anyone who has used a traditional GPS system will recognize Shabtai’s

frustration.Yes,theyknowyourpreciselocationthankstoanetworkoftwenty-

four geosynchronous GPS satellites built and maintained by the U.S. government.Theyalsoknowaboutroads—whichonesarehighways,one-way streets,andsoon—becausetheyhaveaccesstoadatabasewiththisinformation. Butthat’saboutit.Thethingsadriverreallywantstoknowabout—trafficjams, accidents, road closures, and other factors that affect travel time—escape a traditionalsystem.Whenasked,forexample,tocalculatethebestroutefrom Andy’shousetoErik’s,itsimplytakesthestartingpoint(Andy’scar’scurrent location)andtheendingpoint(Erik’shouse)andconsultsitsroaddatabaseto calculate the theoretically “quickest” route between the two. This route will includemajorroadsandhighways,sincetheyhavethehighestspeedlimits. Ifit’srushhour,however,thistheoreticallyquickestroutewillnotactuallybe thequickestone;withthousandsofcarssqueezingontothemajorroadsand

highways, traffic speed will not approach, let alone eclipse, the speed limit. Andy should instead seek out all the sneaky little back roads that longtime

commutersknowabout.Andy’sGPSknowsthattheseroadsexist(ifit’sup-to-

date,itknowsaboutallroads),butdoesn’tknowthatthey’rethebestoptionat eightforty-fiveonaTuesdaymorning.Evenifhestartsoutonbackroads,his well-meaningGPSwillkeepreroutinghimontothehighway. ShabtairecognizedthatatrulyusefulGPSsystemneededtoknowmorethan wherethecarwasontheroad.Italsoneededtoknowwhereothercarswereand howfasttheyweremoving.Whenthefirstsmartphonesappearedhesawan

opportunity,foundingWazein2008alongwithUriLevineandAmirShinar.The

software’s genius is to turn all the smartphones running it into sensors that uploadconstantlytothecompany’sserverstheirlocationandspeedinformation. Asmoreandmoresmartphonesruntheapplication,therefore,Wazegetsamore andmorecompletesenseofhowtrafficisflowingthroughoutagivenarea. Insteadofjustastaticmapofroads,italsohasalwayscurrentupdatesontraffic conditions.Itsserversusethemap,theseupdates,andasetofsophisticated

algorithmstogeneratedrivingdirections.IfAndywantstodrivetoErik’sat8:45

a.m.onaTuesday,Wazeisnotgoingtoputhimonthehighway.It’sgoingto

keephimonsurfacestreetswheretrafficiscomparativelylightatthathour.

ThatWazegetsmoreusefultoallofitsmembersasitgetsmoremembersisa classicexampleofwhateconomistscallanetworkeffect—asituationwherethe valueofaresourceforeachofitsusersincreaseswitheachadditionaluser.And

thenumberofWazers,asthey’recalled,isincreasingquickly.InJulyof2012

thecompanyreportedthatithaddoubleditsuserbasetotwentymillionpeople intheprevioussixmonths. 3 Thiscommunityhadcollectivelydrivenmorethan

3.2billionmilesandhadtypedinmanythousandsofupdatesaboutaccidents,

suddentrafficjams,policespeedtraps,roadclosings,newfreewayexitsand entrances,cheapgas,andotheritemsofinteresttotheirfellowdrivers. WazemakesGPSwhatitshouldbefordrivers:asystemforgettingwhere youwanttogoasquicklyandeasilyaspossible,regardlessofhowmuchyou know about local roads and conditions. It instantly turns you into the most knowledgeabledriverintown.

TheEconomicsofBits

Waze is possible in no small part because of Moore’s Law and exponential technologicalprogress,thesubjectsofthepreviouschapter.Theservicerelieson vastnumbersofpowerfulbutcheapdevices(thesmartphonesofitsusers),each ofthemequippedwithanarrayofprocessors,sensors,andtransmitters.Such technologysimplydidn’texistadecadeago,andsoneitherdidWaze.Itonly becamefeasibleinthepastfewyearsbecauseofaccumulateddigitalpower

increasesandcostdeclines.Aswesawinchapter3,exponentialimprovementin

computer gear is one of the three fundamental forces enabling the second machineage. Wazealsodependscriticallyonthesecondofthesethreeforces:digitization.

Intheirlandmark1998bookInformationRules,economistsCarlShapiroand

HalVariandefinethisphenomenonas“encod[inginformation]asastreamof bits.” 4 Digitization, in other words, is the work of turning all kinds of informationandmedia—text,sounds,photos,video,datafrominstrumentsand sensors,andsoon—intotheonesandzeroesthatarethenativelanguageof computersandtheirkin.Waze,forexample,usesseveralstreamsofinformation:

digitizedstreetmaps,locationcoordinatesforcarsbroadcastbytheapp,and alertsabouttrafficjams,amongothers.It’sWaze’sabilitytobringthesestreams togetherandmakethemusefulforitsusersthatcausestheservicetobeso popular. We thought we understood digitization pretty well based on the work of Shapiro,Varian,andothers,andbasedonouralmostconstantexposuretoonline content, but in the past few years the phenomenon has evolved in some unexpecteddirections.Ithasalsoexplodedinvolume,velocity,andvariety.This surgeindigitizationhashadtwoprofoundconsequences:newwaysofacquiring knowledge(inotherwords,ofdoingscience)andhigherratesofinnovation. Thischapterwillexplorethefascinatingrecenthistoryofdigitization.

Likesomanyothermodernonlineservices,Wazeexploitstwoofthewell-

understood and unique economic properties of digital information: such informationisnon-rival,andithasclosetozeromarginalcostofreproduction. Ineverydaylanguage,wemightsaythatdigitalinformationisnot“usedup” whenitgetsused,anditisextremelycheaptomakeanothercopyofadigitized resource.Let’slookateachofthesepropertiesinabitmoredetail. Rivalgoods,whichweencountereveryday,canonlybeconsumedbyone personorthingatatime.IfthetwoofusflyfromBostontoCalifornia,theplane thattakesoffafteruscannotuseourfuel.Andycan’talsohavetheseatthatErik issittingin(airlinerulesprohibitsuchsharing,evenifwewereupforit)and can’tusehiscolleague’sheadphonesifErikhasalreadyputthemontolistento music on his smartphone. The digitized music itself, however, is non-rival. Erik’slisteningtoitdoesn’tkeepanyoneelsefromdoingso,atthesametimeor later. IfAndybuysandreadsanoldhardcovercopyofthecollectedworksof science-fictionwriterJulesVerne,hedoesn’t“useitup”;hecanpassitonto Erikoncehe’sdone.ButifthetwoofuswanttodipintoTwentyThousand LeaguesUndertheSeaatthesametime,weeitherhavetofindanothercopyor Andyhastomakeacopyofthebookheowns.Hemightbelegallyentitledtodo thisbecauseit’snotundercopyright,buthe’dstillhavetospendalotoftimeat thephotocopierorpaysomeoneelsetodoso.Ineithercase,makingthatcopy wouldnotbecheap. 5 Inaddition,aphotocopyofaphotocopyofaphotocopy startstogethardtoread. ButifAndyhasacquiredadigitalcopyofthebook,withacouplekeystrokes ormouseclickshecancreateaduplicate,saveittoaphysicaldisk,andgivethe copy to Erik. Unlike photocopies, bits cloned from bits are usually exactly identicaltotheoriginal.Copyingbitsisalsoextremelycheap,fast,andeasyto do.Whiletheveryfirstcopyofabookormoviemightcostalottocreate, makingadditionalcopiescostalmostnothing.Thisiswhatismeantby“zero marginalcostofreproduction.” Thesedays,ofcourse,insteadofhandingErikadisk,Andyismorelikelyto attachthefiletoane-mailmessageorshareitthroughacloudservicelike Dropbox.Onewayoranother,though,he’sgoingtousetheInternet.He’lltake thisapproachbecauseit’sfaster,moreconvenient,and,inanimportantsense, essentiallyfree.Likemostpeople,wepayaflatfeeforInternetaccessathome andonourmobiledevices(MITpaysforouraccessatwork).Ifweexceeda certaindatalimit,ourInternetServiceProvidermightstartchargingusextra,but

untilthatpointwedon’tpaybythebit;wepaythesamenomatterhowmany bitsweuploadordownload.Assuch,there’snoadditionalcostforsendingor receivingonemorechunkofdataovertheNet.Unlikegoodsmadeofatoms, goodsmadeofbitscanbereplicatedperfectlyandsentacrosstheroomoracross the planet almost instantaneously and almost costlessly. Making things free, perfect, and instant might seem like unreasonable expectations for most products,butasmoreinformationisdigitized,moreproductswillfallintothese categories.

BusinessModelsWhentheFirstCopyisStillExpensive

ShapiroandVarianelegantlysummarizetheseattributesbystatingthatinanage of computers and networks, “Information is costly to produce but cheap to reproduce.” 6 Instantaneousonlinetranslationservices,oneofthescience-fiction-

into-realitytechnologiesdiscussedinchapter2,takeadvantageofthisfact.They

makeuseofpairedsetsofdocumentsthatweretranslated,oftenatconsiderable expense,byahumanfromonelanguageintoanother.Forexample,theEuropean

Unionanditspredecessorbodieshavesince1957issuedallofficialdocuments

inallthemainlanguagesofitsmembercountries,andtheUnitedNationshas beensimilarlyprolificinwritingtextsinallsixofitsofficiallanguages. This huge body of information was not cheap to generate, but once it’s digitizedit’sverycheaptoreplicate,chopup,andsharewidelyandrepeatedly. This is exactly what a service like Google Translate does. When it gets an EnglishsentenceandarequestforitsGermanequivalent,itessentiallyscansall thedocumentsitknowsaboutinbothEnglishandGerman,lookingforaclose match (or a few fragments that add up to a close match), then returns the corresponding German text. Today’s most advanced automatic translation services, then, are not the result of any recent insight about how to teach computersalltherulesofhumanlanguagesandhowtoapplythem.Instead, they’reapplicationsthatdostatisticalpatternmatchingoverhugepoolsofdigital contentthatwascostlytoproduce,butcheaptoreproduce.

WhatHappensWhentheContentComesFreely?

Butwhatwouldhappentothedigitalworldifinformationwerenolongercostly

toproduce?Whatwouldhappenifitwerefreerightfromthestart?We’vebeen

learningtheanswerstothesequestionsintheyearssinceInformationRules

cameout,andthey’rehighlyencouraging.

Theoldbusinesssayingisthat“timeismoney,”butwhat’samazingaboutthe modern Internet is how many people are willing to devote their time to producing online content without seeking any money in return. Wikipedia’s content,forexample,isgeneratedforfreebyvolunteersallaroundtheworld. It’sbyfartheworld’slargestandmostconsultedreferencework,butnoonegets paidtowriteoredititsarticles.Thesameistrueforcountlesswebsites,blogs, discussion boards, forums, and other sources of online information. Their creators expect no direct monetary reward and offer the information free of charge.

WhenShapiroandVarianpublishedInformationRulesin1998,theriseof

suchuser-generatedcontent,muchofwhichiscreatedwithoutmoneychanging hands,hadyettooccur.Blogger,oneofthefirstweblogservices,debutedin August 1999, Wikipedia in January 2001, and Friendster, an early social

networkingsite,in2002.FriendsterwassooneclipsedbyFacebook,whichwas

foundedin2004andhassincegrownintothemostpopularInternetsiteinthe

world. 7 Infact,sixofthetenmostpopularcontentsitesthroughouttheworldare primarilyuser-generated,asaresixofthetoptenintheUnitedStates. 8 Allthisuser-generatedcontentisn’tjustmakingusfeelgoodbylettingus expressourselvesandcommunicatewithoneanother;it’salsocontributingto someoftherecentscience-fiction-into-realitytechnologieswe’veseen.Siri,for example,improvesitselfovertimebyanalyzingtheever-largercollectionof soundfilesitsusersgeneratewheninteractingwiththevoicerecognitionsystem. AndWatson’sdatabase,whichconsistedofapproximatelytwohundredmillion pagesofdocumentstakingupfourterabytesofdiskspace,includedanentire copyofWikipedia. 9 Forawhileitalsoincludedthesaltylanguage–filledUrban Dictionary,butthisarchiveofuser-generatedcontentwasremovedafter,tothe dismayofitscreators,Watsonstartedtoincludecursewordsinitsresponses. 10

Perhapsweshouldn’tbetoosurprisedbythegrowthandpopularityofuser-

generatedcontentontheInternet.Afterall,wehumansliketoshareandinteract.

What’sabitmoresurprisingishowmuchourmachinesalsoapparentlylike

talkingtoeachother.

Machine-to-machine(M2M)communicationisacatch-alltermfordevices

sharingdatawithoneanotherovernetworksliketheInternet.Wazemakesuse

ofM2M;whentheappisactiveonasmartphone,itconstantlysendsinformation

toWaze’sserverswithoutanyhumaninvolvement.Similarly,whenyousearch

thepopulartravelsiteKayakforcheapairfares,Kayak’sserversimmediately

sendrequeststotheircounterpartsatvariousairlines,whichwritebackinreal

timewithoutanyhumaninvolvement.ATMsasktheirbankshowmuchmoney

wehaveinouraccountsbeforelettinguswithdrawcash;digitalthermometersin

refrigerated trucks constantly reassure supermarkets that the produce isn’t gettingtoohotintransit;sensorsinsemiconductorfactoriesletheadquarters

knoweverytimeadefectoccurs;andcountlessotherM2Mcommunications

takeplaceinrealtime,allthetime.AccordingtoaJuly2012storyintheNew

YorkTimes, “The combined level of robotic chatter on the world’s wireless

networks

voiceconversationstakingplaceonwirelessgrids.” 11

islikelysoontoexceedthatgeneratedbythesumofallhuman

RunningOutofMetricSystem:TheDataExplosion

The digitization of just about everything—documents, news, music, photos, video,maps,personalupdates,socialnetworks,requestsforinformationand responsestothoserequests,datafromallkindsofsensors,andsoon—isoneof themostimportantphenomenaofrecentyears.Aswemovedeeperintothe secondmachineage,digitizationcontinuestospreadandaccelerate,yielding somejaw-droppingstatistics.AccordingtoCiscoSystems,worldwideInternet

trafficincreasedbyafactoroftwelveinjustthefiveyearsbetween2006and

2011,reaching23.9exabytespermonth. 12 Anexabyteisaridiculouslybignumber,theequivalentofmorethantwo hundredthousandofWatson’sentiredatabase.However,eventhisisnotenough tocapturethemagnitudeofcurrentandfuturedigitization.Technologyresearch

firmIDCestimatesthattherewere2.7zettabytes,or2.7sextillionbytes,of

digitaldataintheworldin2012,almosthalfasmuchagainasexistedin2011.

Andthisdatawon’tjustsitondiskdrives;it’llalsomovearound.Ciscopredicts thatglobalInternetProtocoltrafficwillreach1.3zettabytesby2016. 13 That’s over250billionDVDsofinformation. 14 Asthesefiguresmakeclear,digitizationyieldstrulybigdata.Infact,ifthis kindofgrowthkeepsupformuchlongerwe’regoingtorunoutofmetric

system.Whenitssetofprefixeswasexpandedin1991atthenineteenthGeneral

ConferenceonWeightsandMeasures,thelargestonewasyotta,signifyingone septillion,or10 24 . 15 We’reonlyoneprefixawayfromthatinthe‘zettabyteera.’

BinaryScience

Therecentexplosionofdigitizationisclearlyimpressive,butisitimportant? Arealloftheseexa-andzettabytesofdigitaldataactuallyuseful? They’reincrediblyuseful.Oneofthemainreasonswecitedigitizationasa main force shaping the second machine age is that digitization increases understanding.Itdoesthisbymakinghugeamountsofdatareadilyaccessible, anddataarethelifebloodofscience.By“science”here,wemeantheworkof formulatingtheoriesandhypotheses,thenevaluatingthem.Or,lessformally, guessinghowsomethingworks,thencheckingtoseeiftheguessisright. AwhilebackErikguessedthatdataaboutInternetsearchesmightsignal futurechangesinhousingsalesandpricesaroundthecountry.Hereasonedthat ifacoupleisgoingtomovetoanothercityandbuyahouse,theyarenotgoing tocompletetheprocessinjustafewdays.They’regoingtostartinvestigating the move and purchase months in advance. These days those initial investigationswilltakeplaceovertheInternetandconsistoftypingintoasearch enginephraseslike“Phoenixrealestateagent,”“Phoenixneighborhoods,”and “Phoenixtwo-bedroomhouseprices.” Totestthishypothesis,ErikaskedGoogleifhecouldaccessdataaboutits searchterms.Hewastoldthathedidn’thavetoask;thecompanymadethese datafreelyavailable overtheWeb. Erikandhis doctoralstudentLynn Wu, neither of whom was versed in the economics of housing, built a simple statisticalmodeltolookatthedatautilizingtheuser-generatedcontentofsearch termsmadeavailablebyGoogle.Theirmodellinkedchangesinsearch-term volumetolaterhousingsalesandpricechanges,predictingthatifsearchterms liketheonesabovewereontheincreasetoday,thenhousingsalesandpricesin Phoenixwouldrisethreemonthsfromnow.Theyfoundtheirsimplemodel

worked.Infact,itpredictedsales23.6percentmoreaccuratelythanpredictions

publishedbytheexpertsattheNationalAssociationofRealtors. Researchershavehadsimilarsuccessusingnewlyavailabledigitaldatain otherdomains.AteamledbyRumiChunaraofHarvardMedicalSchoolfound thattweetswerejustasaccurateasofficialreportswhenitcametotrackingthe

spreadofcholeraafterthe2010earthquakeinHaiti;theywerealsoatleasttwo

weeksfaster. 16 SitaramAsurandBernardoHubermanofHP’sSocialComputing Labfoundthattweetscouldalsobeusedtopredictmoviebox-officerevenue. Theyconcludedthat“thisworkshowshowsocialmediaexpressesacollective wisdom which, when properly tapped, can yield an extremely powerful and accurateindicatoroffutureoutcomes.” 17

Digitizationcanalsohelpusbetterunderstandthepast.AsofMarch2012

Googlehadscannedmorethantwentymillionbookspublishedoverseveral centuries. 18 Thishugepoolofdigitalwordsandphrasesformsabaseforwhat’s beingcalledculturomics,or“theapplicationofhigh-throughputdatacollection andanalysistothestudyofhumanculture.” 19 Amultidisciplinaryteamledby Jean-Baptiste Michel and Erez Lieberman Aiden analyzed over five million

bookspublishedinEnglishsince1800.Amongotherthings,theyfoundthatthe

numberofwordsinEnglishincreasedbymorethan70percentbetween1950

and2000,thatfamenowcomestopeoplemorequicklythaninthepastbutalso

fadesfaster,andthatinthetwentiethcenturyinterestinevolutionwasdeclining untilWatsonandCrickdiscoveredthestructureofDNA. 20 Alloftheseareexamplesofbetterunderstandingandprediction—inother words,ofbetterscience—viadigitization.HalVarian,who’snowGoogle’schief economist,hasforyearsenjoyedafront-rowseatforthisphenomenon.Healso hasawaywithwords.Oneofourfavoritequotesofhisis,“Ikeepsayingthat thesexyjobinthenexttenyearswillbestatisticians.AndI’mnotkidding.” 21 Whenwelookattheamountofdigitaldatabeingcreatedandthinkabouthow muchmoreinsightthereistobegained,we’reprettysurehe’snotwrong,either.

NewLayersYieldNewRecipes

Digitalinformationisn’tjustthelifebloodfornewkindsofscience;it’sthe secondfundamentalforce(afterexponentialimprovement)shapingthesecond machineagebecauseofitsroleinfosteringinnovation.Wazeisagreatexample here.Theserviceisbuiltonmultiplelayersandgenerationsofdigitization,none ofwhichhavedecayedorbeenusedupsincedigitalgoodsarenon-rival. Thefirstandoldestlayerisdigitalmaps,whichareatleastasoldaspersonal computers. 22 ThesecondisGPSlocationinformation,whichbecamemuchmore useful for driving when the U.S. government increased its GPS accuracy in 2000. 23 The third is social data; Waze users help each other by providing informationoneverythingfromaccidentstopolicespeedtrapstocheapgas; theycanevenusetheapptochatwithoneanother.Andfinally,Wazemakes extensiveuseofsensordata;infact,itessentiallyconvertseverycarusingitinto atraffic-speedsensorandusesthesedatatocalculatethequickestroutes. In-carnavigationsystemsthatuseonlythefirsttwogenerationsofdigital data—mapsandGPSlocationinformation—havebeenaroundforawhile.They canbeextremelyuseful,especiallyinunfamiliarcities,butaswe’veseen,they haveseriousshortcomings.ThefoundersofWazerealizedthatasdigitization

advancedandspreadtheycouldovercometheshortcomingsoftraditionalGPS

navigation.Theseinnovatorsmadeprogressbyaddingsocialandsensordatato

anexistingsystem,greatlyincreasingitspowerandusefulness.Aswe’llseein

thenextchapter,thisstyleofinnovationisoneofthehallmarksofourcurrent

time.It’ssoimportant,infact,thatit’sthethirdandlastoftheforcesshapingthe

secondmachineage.Thenextchapterexplainswhythisis.

“Ifyouwanttohavegoodideasyoumusthavemanyideas.” —LinusPauling

“Ifyouwanttohavegoodideasyoumusthavemanyideas.”

—LinusPauling

EVERYONE AGREES THAT IT would be troubling news if America’s rate of innovationweretodecrease.Butwecan’tseemtoagreeatallaboutwhetherthis isactuallyhappening. Wecareaboutinnovationsomuchnotsimplybecausewelikenewstuff, although we certainly do. As the novelist William Makepeace Thackeray observed,“Noveltyhascharmsthatourmindcanhardlywithstand.” 1 Someof uscanhardlywithstandtheallureofnewgadgets;othersarecharmedbythe latest fashion styles or places to see and be seen. From an economist’s perspective,satisfyingthesedesiresisgreat—takingcareofconsumerdemandis usuallyseenasagoodthing.Butinnovationisalsothemostimportantforcethat makesoursocietywealthier.

WhyInnovationis(Almost)Everything

Paul Krugman speaks for many, if not most, economists when he says, “Productivityisn’teverything,butinthelongrunitisalmosteverything.”Why? Because,heexplains,“Acountry’sabilitytoimproveitsstandardoflivingover timedependsalmostentirelyonitsabilitytoraiseitsoutputperworker”—in otherwords,thenumberofhoursoflaborittakestoproduceeverything,from automobilestozippers,thatweproduce. 2 Mostcountriesdon’thaveextensive mineralwealthoroilreserves,andthuscan’tgetrichbyexportingthem.*Sothe onlyviablewayforsocietiestobecomewealthier—toimprovethestandardof livingavailabletoitspeople—isfortheircompaniesandworkerstokeepgetting moreoutputfromthesamenumberofinputs,inotherwordsmoregoodsand servicesfromthesamenumberofpeople. Innovationishowthisproductivitygrowthhappens.Economistslovetoargue withoneanother,butthere’sgreatconsensusamongthemaboutthefundamental importance of innovation for growth and prosperity. Most in the profession wouldagreewithJosephSchumpeter,thetopic’sgreatscholar,whowrotethat, “Innovationistheoutstandingfactintheeconomichistoryofcapitalistsociety. andalsoitislargelyresponsibleformostofwhatwewouldatfirstsight attributetootherfactors.” 3 Itisherethattheconsensusends.Howmuchofthis “outstandingfact”istakingplacerightnow,andwhetherit’sonanupwardor downwardtrend,isamatterofgreatdispute.

WhyWeShouldBeWorried:InnovationsGetUsedUp

Economist Bob Gordon, one of the most thoughtful, thorough, and widely respectedresearchersofproductivityandeconomicgrowth,recentlycompleteda majorstudyofhowtheAmericanstandardoflivinghaschangedoverthepast

150years.Hisworklefthimconvincedthatinnovationisslowingdown.

Gordon emphasizes—as do we—the role of new technologies in driving economic growth. And like us, he’s impressed by the productive power unleashed by the steam engine and the other technologies of the Industrial Revolution.AccordingtoGordon,itwasthefirsttrulysignificanteventinthe economichistoryoftheworld.Ashewrites,“therewasalmostnoeconomic growthforfourcenturiesandprobablyforthepreviousmillennium”priorto 1750,orroughlywhentheIndustrialRevolutionstarted. 4 Aswesawinthefirst chapter,humanpopulationgrowthandsocialdevelopmentwereverynearlyflat untilthesteamenginecamealong.Unsurprisingly,itturnsoutthateconomic growthwas,too. AsGordonshows,however,oncethisgrowthgotstarteditstayedonasharp upwardtrajectoryfortwohundredyears.Thiswasduenotonlytotheoriginal IndustrialRevolution,butalsotoasecondone,ittooreliantontechnological innovation.Threenoveltieswerecentralhere:electricity,theinternalcombustion engine,andindoorplumbingwithrunningwater,allofwhichcameontothe

scenebetween1870and1900.

The ‘great inventions’ of this second industrial revolution, in Gordon’s

estimation,“weresoimportantandfar-reachingthattheytookafull100yearsto

havetheirmaineffect.”Butoncethateffecthadbeenrealized,anewproblem

emerged.Growthstalledout,andevenbegantodecline.Attheriskofbeing

flippant,whenthesteamengineranoutofsteam,theinternalcombustionengine

wastheretoreplaceit.Butoncetheinternalcombustionengineranoutoffuel,

weweren’tleftwithmuch.TouseGordon’swords,

Thegrowthofproductivity(outputperhour)slowedmarkedlyafter1970.Whilepuzzlingat

thetime,itseemsincreasinglyclearthattheone-time-onlybenefitsoftheGreatInventions

Allthatremainedafter

1970 were second-round improvements, such as developing short-haul regional jets, extendingtheoriginalinterstatehighwaynetworkwithsuburbanringroads,andconverting

andtheirspin-offshadoccurredandcouldnothappenagain

residentialAmericafromwindowunitairconditionerstocentralairconditioning.5

Gordonisfarfromaloneinthisview.Inhis2011bookTheGreatStagnation,

economistTylerCowenisdefinitiveaboutthesourceofAmerica’seconomic

woes:

Wearefailingtounderstandwhywearefailing.Alloftheseproblemshaveasingle,little

noticed root cause: We have been living off low-hanging fruit for at least three hundred

years

started pretending it was still there. We have failed to recognize that we are at a

Yetduringthelastfortyyears,thatlow-hangingfruitstarteddisappearing,andwe

technologicalplateauandthetreesaremorebarethanwewouldliketothink.6

GeneralPurposeTechnologies:TheOnesThatReally

Matter

Clearly, Gordon and Cowen see the invention of powerful technologies as centraltoeconomicprogress.Indeed,there’sbroadagreementamongeconomic historiansthatsometechnologiesaresignificantenoughtoacceleratethenormal marchofeconomicprogress.Todothis,theyhavetospreadthroughoutmany,if notmost,industries;theycan’tremaininjustone.Thecottongin,forexample, was unquestionably important within the textile sector at the start of the nineteenthcentury,butprettyinsignificantoutsideofit.* Thesteamengineandelectricalpower,bycontrast,quicklyspreadjustabout everywhere. The steam engine didn’t just massively increase the amount of poweravailabletofactoriesandfreethemfromtheneedtobelocatedneara streamorrivertopowerthewaterwheel;italsorevolutionizedlandtravelby enablingrailroadsandseatravelviathesteamship.Electricitygaveafurther boosttomanufacturingbyenablingindividuallypoweredmachines.Italsolit factories,officebuildings,andwarehousesandledtofurtherinnovationslikeair conditioning,whichmadepreviouslyswelteringworkplacespleasant. Withtheirtypicalverbalflair,economistscallinnovationslikesteampower andelectricitygeneralpurposetechnologies(GPTs).EconomichistorianGavin Wrightoffersaconcisedefinition:“deepnewideasortechniquesthathavethe potentialforimportantimpactsonmanysectorsoftheeconomy.” 7 “Impacts” heremeansignificantbooststooutputduetolargeproductivitygains.GPTsare important because they are economically significant—they interrupt and acceleratethenormalmarchofeconomicprogress. Inadditiontoagreeingontheirimportance,scholarshavealsocometoa consensusonhowtorecognizeGPTs:theyshouldbepervasive,improvingover time,andabletospawnnewinnovations. 8 Theprecedingchaptershavebuilta casethatdigitaltechnologiesmeetallthreeoftheserequirements.Theyimprove alongaMoore’sLawtrajectory,areusedineveryindustryintheworld,andlead

toinnovationslikeautonomouscarsandnonhumanJeopardy!champions.Are we alone in thinking that information and communication technology (ICT) belongsinthesamecategoryassteamandelectricity?Arewetheonlyoneswho think,inshort,thatICTisaGPT? Absolutelynot.Mosteconomichistoriansconcurwiththeassessmentthat ICTmeetsallofthecriteriagivenabove,andsoshouldjointheclubofgeneral purposetechnologies.Infact,inalistofallthecandidatesforthisclassification compiledbytheeconomistAlexanderField,onlysteampowergotmorevotes thanICT,whichwastiedwithelectricityasthesecondmostcommonlyaccepted GPT. 9 If we are all in agreement, then why the debate over whether ICTs are usheringinanewgoldenageofinnovationandgrowth?Because,theargument goes,theireconomicbenefitshavealreadybeencapturedandnowmostnew ‘innovation’involvesentertainingourselvesinexpensivelyonline.Accordingto RobertGordon:

ThefirstindustrialrobotwasintroducedbyGeneralMotorsin1961.Telephoneoperators

wentawayinthe1960s

Airlinereservationssystemscameinthe1970s,andby1980

bar-code scanners and cash machines were spreading through the retail and banking

The first personal computers arrived in the early 1980s with their word

processing,wordwrap,andspreadsheets

Morerecentandthusmorefamiliarwasthe

rapiddevelopmentofthewebande-commerceafter1995,aprocesslargelycompletedby

2005.10

Atpresent,saysCowen,“ThegainsoftheInternetareveryrealandIamhereto

praisethem,notdamnthem

morefun,inpartbecauseoftheInternet.Wearealsohavingmorecheapfun. [But]wearecomingupshortontherevenueside,soitishardertopayour debts,whetherindividuals,businesses,orgovernments.” 11 Twenty-firstcentury

ICT,inshort,isfailingtheprimetestofbeingeconomicallysignificant.

Still,theoverallpictureisthis:Wearehaving

WhyWeShouldn’tBeWorried:InnovationsDon’tGet

UsedUp

Foranygoodscientist,ofcourse,dataaretheultimatedeciderofhypotheses.So whatdothedatasayhere?Dotheproductivitynumbersbackupthispessimistic

viewofthepowerofdigitization?We’llgettothedatainchapter7.First,

though,wewanttopresentaverydifferentviewofhowinnovationworks—an

alternativetothenotionthatinnovationsget‘usedup.’

Gordonwritesthat“itisusefultothinkoftheinnovativeprocessasaseries ofdiscreteinventionsfollowedbyincrementalimprovementswhichultimately tapthefullpotentialoftheinitialinvention.” 12 Thisseemssensibleenough.An inventionlikethesteamengineorcomputercomesalongandwereapeconomic benefitsfromit.Thosebenefitsstartsmallwhilethetechnologyisimmatureand notwidelyused,growtobequitebigastheGPTimprovesandpropagates,then taperoffastheimprovement—andespeciallythepropagation—diedown.When multipleGPTsappearatthesametime,orinasteadysequence,wesustainhigh rates of growth over a long period. But if there’s a big gap between major innovations, economic growth will eventually peter out. We’ll call this the ‘innovation-as-fruit’viewofthings,inhonorofTylerCowen’simageryofallthe low-hanging fruit being picked. In this perspective, coming up with an innovationislikegrowingfruit,andexploitinganinnovationislikeeatingthe fruitovertime. Anotherschoolofthought,though,holdsthatthetrueworkofinnovationis notcomingupwithsomethingbigandnew,butinsteadrecombiningthingsthat alreadyexist.Andthemorecloselywelookathowmajorstepsforwardinour knowledgeandabilitytoaccomplishthingshaveactuallyoccurred,themorethis recombinantviewmakessense.Forexample,it’sexactlyhowatleastoneNobel Prize–winninginnovationcameabout.

KaryMulliswonthe1993NobelPrizeinChemistryforthedevelopmentof

thepolymerasechainreaction(PCR),anowubiquitoustechniqueforreplicating DNAsequences. When the idea first came to him on a nighttime drive in California,though,healmostdismisseditoutofhand.Asherecountedinhis

NobelAwardspeech,“Somehow,Ithought,ithadtobeanillusion

Therewasnotasingleunknowninthescheme.Everystepinvolved

tooeasy

Itwas

had been done already.” 13 “All” Mullis did was recombine well-understood techniquesinbiochemistrytogenerateanewone.Andyetit’sobviousMullis’s recombinationisanenormouslyvaluableone. Afterexaminingmanyexamplesofinvention,innovation,andtechnological progress,complexityscholarBrianArthurbecameconvincedthatstorieslikethe inventionofPCRaretherule,nottheexception.Ashesummarizesinhisbook TheNatureofTechnology,“Toinventsomethingistofinditinwhatpreviously exists.” 14 EconomistPaulRomerhasarguedforcefullyinfavorofthisview,the so-called‘newgrowththeory’withineconomics,inordertodistinguishitfrom perspectives like Gordon’s. Romer’s inherently optimistic theory stresses the importanceofrecombinantinnovation.Ashewrites:

Economicgrowthoccurswheneverpeopletakeresourcesandrearrangetheminwaysthat

makethemmorevaluable

ideaswerediscovered.And

ideas.Weconsistently

Possibilitiesdonotmerelyadd

up;theymultiply.15

failtograsphowmanyideasremaintobediscovered

everygenerationhasunderestimatedthepotentialforfindingnew

resourcesandundesirablesideeffectswouldposeifnonew

Everygenerationhasperceivedthelimitstogrowththatfinite

Romeralsomakesavitalpointaboutaparticularlyimportantcategoryofidea,

whichhecalls“meta-ideas”:

Perhapsthemostimportantideasofallaremeta-ideas—ideasabouthowtosupportthe

twosafepredictions.First,the

countrythattakestheleadinthetwenty-firstcenturywillbetheonethatimplementsan innovationthatmoreeffectivelysupportstheproductionofnewideasintheprivatesector.

Second,newmeta-ideasofthiskindwillbefound.16

productionandtransmissionofotherideas

Thereare

DigitalTechnologies:TheMostGeneralPurposeofAll

GordonandCowenareworld-classeconomists,butthey’renotgivingdigital technologiestheirdue.Thenextgreatmeta-idea,invokedbyRomer,hasalready beenfound:itcanbeseeninthenewcommunitiesofmindsandmachinesmade possiblebynetworkeddigitaldevicesrunninganastonishingvarietyofsoftware. The GPT of ICT has given birth to radically new ways to combine and recombineideas.Likelanguage,printing,thelibrary,oruniversaleducation,the globaldigitalnetworkfostersrecombinantinnovation.Wecanmixandremix ideas,botholdandrecent,inwayswenevercouldbefore.Let’slookatafew examples. Google’sChauffeurprojectgivesnewlifetoanearlierGPT:theinternal combustionengine.Whenaneverydaycarisequippedwithafastcomputerand abunchofsensors(allofwhichgetcheaperaccordingtoMoore’sLaw)anda hugeamountofmapandstreetinformation(availablethankstothedigitization ofeverything)itbecomesanautopilotedvehiclestraightoutofsciencefiction. Whilewehumansarestilltheonesdoingthedriving,innovationslikeWazewill helpusgetaroundmorequicklyandeasetrafficjams.Wazeisarecombination ofalocationsensor,datatransmissiondevice(thatis,aphone),GPSsystem,and socialnetwork.TheteamatWazeinventednoneofthesetechnologies;theyjust putthemtogetherinanewway.Moore’sLawmadeallinvolveddevicescheap enough,anddigitizationmadeallnecessarydataavailabletofacilitatetheWaze system. TheWebitselfisaprettystraightforwardcombinationoftheInternet’smuch

olderTCP/IPdatatransmissionnetwork;amarkuplanguagecalledHTMLthat specifiedhowtext,pictures,andsoonshouldbelaidout;andasimplePC applicationcalleda‘browser’todisplaytheresults.Noneoftheseelementswas particularlynovel.Theircombinationwasrevolutionary. FacebookhasbuiltontheWebinfrastructurebyallowingpeopletodigitize their social network and put media online without having to learn HTML. Whetherornotthiswasanintellectuallyprofoundcombinationoftechnological

capabilities,itwasapopularandeconomicallysignificantlyone—byJuly2013,

thecompanywasvaluedatover$60billion. 17 Whenphotosharingbecameone ofthemostpopularactivitiesonFacebook,KevinSystromandMikeKrieger decided to build a smartphone application that mimicked this capability, combiningitwiththeoptiontomodifyaphoto’sappearancewithdigitalfilters. This seems like a minor innovation, especially since Facebook already had

enabledmobilephotosharingin2010whenSystromandKriegerstartedtheir

project.However,theapplicationtheybuilt,calledInstagram,attractedmore

than30millionusersbythespringof2012,userswhohadcollectivelyuploaded

more than 100 million photographs. Facebook acquired Instagram for

approximately$1billioninAprilof2012.

Thisprogressiondriveshomethepointthatdigitalinnovationisrecombinant innovationinitspurestform.Eachdevelopmentbecomesabuildingblockfor future innovations. Progress doesn’t run out; it accumulates. And the digital worlddoesn’trespectanyboundaries.Itextendsintothephysicalone,leadingto cars and planes that drive themselves, printers that make parts, and so on. Moore’sLawmakescomputingdevicesandsensorsexponentiallycheaperover time,enablingthemtobebuilteconomicallyintomoreandmoregear,from doorknobstogreetingcards.Digitizationmakesavailablemassivebodiesofdata relevant to almost any situation, and this information can be infinitely reproducedandreusedbecauseitisnon-rival.Asaresultofthesetwoforces,the numberofpotentiallyvaluablebuildingblocksisexplodingaroundtheworld, and the possibilities are multiplying as never before. We’ll call this the ‘innovation-as-building-block’viewoftheworld;it’stheoneheldbyArthur, Romer,andthetwoofus.Fromthisperspective,unlikeintheinnovation-as-fruit view,buildingblocksdon’tevergeteatenorotherwiseusedup.Infact,they increasetheopportunitiesforfuturerecombinations.

LimitstoRecombinantGrowth

Ifthisrecombinantviewofinnovationiscorrect,thenaproblemlooms:asthe number of building blocks explodes, the main difficulty is knowing which combinationsofthemwillbevaluable.Inhispaper“RecombinantGrowth,”the economistMartinWeitzmandevelopedamathematicalmodelofnewgrowth theory in which the ‘fixed factors’ in an economy—machine tools, trucks, laboratories,andsoon—areaugmentedovertimebypiecesofknowledgethat hecalls‘seedideas,’andknowledgeitselfincreasesovertimeaspreviousseed ideasarerecombinedintonewones. 18 Thisisaninnovation-as-building-block viewoftheworld,whereboththeknowledgepiecesandtheseedideascanbe combinedandrecombinedovertime. This model has a fascinating result: because combinatorial possibilities explode so quickly there is soon a virtually infinite number of potentially valuablerecombinationsoftheexistingknowledgepieces.*Theconstrainton theeconomy’sgrowththenbecomesitsabilitytogothroughallthesepotential recombinationstofindthetrulyvaluableones. AsWeitzmanwrites,

Insuchaworldthecoreofeconomiclifecouldappearincreasinglytobecenteredonthe

more and more intensive processing of ever-greater numbers of new seed ideas into

In the early stages of development, growth is constrained by

numberofpotentialnewideas,butlateronitisconstrainedonlybytheabilitytoprocess

them.19

workable

Gordonaskstheprovocativequestion,“Isgrowthover?”We’llrespondon

behalfofWeitzman,Romer,andtheothernewgrowththeoristswith“Nota

chance.It’sjustbeingheldbackbyourinabilitytoprocessallthenewideasfast

enough.”

WhatThisProblemNeedsAreMoreEyeballsandBigger

Computers

Ifthisresponseisatleastsomewhataccurate—ifitcapturessomethingabout howinnovationandeconomicgrowthworkintherealworld—thenthebestway toaccelerateprogressistoincreaseourcapacitytotestoutnewcombinationsof ideas.Oneexcellentwaytodothisistoinvolvemorepeopleinthistesting process,anddigitaltechnologiesaremakingitpossibleforevermorepeopleto participate.We’reinterlinkedbyglobalICT,andwehaveaffordableaccessto massesofdataandvastcomputingpower.Today’sdigitalenvironment,inshort, is a playground for large-scale recombination. The open source software

advocateEricRaymondhasanoptimisticobservation:“Givenenougheyeballs, allbugsareshallow.” 20 Theinnovationequivalenttothismightbe,“Withmore eyeballs,morepowerfulcombinationswillbefound.” NASA experienced this effect as it was trying to improve its ability to forecastsolarflares,oreruptionsonthesun’ssurface.Accuracyandplentyof advancewarningarebothimportanthere,sincesolarparticleevents(orSPEs,as flaresareproperlyknown)canbringharmfullevelsofradiationtounshielded gearandpeopleinspace.Despitethirty-fiveyearsofresearchanddataonSPEs, however,NASAacknowledgedthatithad“nomethodavailabletopredictthe onset,intensityordurationofasolarparticleevent.” 21 Theagencyeventuallyposteditsdataandadescriptionofthechallengeof predictingSPEsonInnocentive,anonlineclearinghouseforscientificproblems. Innocentiveis‘non-credentialist’;peopledon’thavetobePhDsorworkinlabs inordertobrowsetheproblems,downloaddata,oruploadasolution.Anyone can work on problems from any discipline; physicists, for example, are not excludedfromdigginginonbiologyproblems. Asitturnedout,thepersonwiththeinsightandexpertiseneededtoimprove SPEpredictionwasnotpartofanyrecognizableastrophysicscommunity.He wasBruceCragin,aretiredradiofrequencyengineerlivinginasmalltownin NewHampshire.Craginsaidthat,“ThoughIhadn’tworkedintheareaofsolar physics as such, I had thought a lot about the theory of magnetic reconnection.” 22 This was evidently the right theory for the job, because

Cragin’sapproachenabledpredictionofSPEseighthoursinadvancewith85

percentaccuracy,andtwenty-fourhoursinadvancewith75percentaccuracy.

Hisrecombinationoftheoryanddataearnedhimathirty-thousand-dollarreward fromthespaceagency. Inrecentyears,manyorganizationshaveadoptedNASA’sstrategyofusing technologytoopenuptheirinnovationchallengesandopportunitiestomore eyeballs.Thisphenomenongoesbyseveralnames,including‘openinnovation’ and‘crowdsourcing,’anditcanberemarkablyeffective.Theinnovationscholars

LarsBoJeppesenandKarimLakhanistudied166scientificproblemspostedto

Innocentive,allofwhichhadstumpedtheirhomeorganizations.Theyfoundthat thecrowdassembledaroundInnocentivewasabletosolveforty-nineofthem, for a success rate of nearly 30 percent. They also found that people whose expertisewasfarawayfromtheapparentdomainoftheproblemweremore likelytosubmitwinningsolutions.Inotherwords,itseemedtoactuallyhelpa solvertobe‘marginal’—tohaveeducation,training,andexperiencethatwere

notobviouslyrelevantfortheproblem.JeppesenandLakhaniprovidevivid

examplesofthis:

[Therewere]differentwinningsolutionstothesamescientificchallengeofidentifyingafood-

gradepolymerdeliverysystembyanaerospacephysicist,asmallagribusinessowner,a

All four submissions

transdermal drug delivery specialist, and an industrial

successfullyachievedtherequiredchallengeobjectiveswithdifferingscientificmechanisms.

[Anothercaseinvolved]anR&Dlabthat,evenafterconsultingwithinternalandexternal

specialists,didnotunderstandthetoxicologicalsignificanceofaparticularpathologythat

Itwaseventuallysolved,using

methods common in her field, by a scientist with a Ph.D. in protein crystallography who wouldnotnormallybeexposedtotoxicologyproblemsorsolvesuchproblemsonaroutine

basis.23

hadbeenobservedinanongoingresearchprogram

Like Innocentive, the online startup Kaggle also assembles a diverse, non- credentialistgroupofpeoplefromaroundtheworldtoworkontoughproblems submittedbyorganizations.Insteadofscientificchallenges,Kagglespecializes indata-intensiveoneswherethegoalistoarriveatabetterpredictionthanthe submittingorganization’sstartingbaselineprediction.Hereagain,theresultsare strikinginacoupleofways.Foronething,improvementsoverthebaselineare usually substantial. In one case, Allstate submitted a dataset of vehicle characteristicsandaskedtheKagglecommunitytopredictwhichofthemwould have later personal liability claims filed against them. 24 The contest lasted approximatelythreemonthsanddrewinmorethanonehundredcontestants.The winning prediction was more than 270 percent better than the insurance company’sbaseline. AnotherinterestingfactisthatthemajorityofKagglecontestsarewonby peoplewhoaremarginaltothedomainofthechallenge—who,forexample, made the best prediction about hospital readmission rates despite having no experienceinhealthcare—andsowouldnothavebeenconsultedaspartofany traditionalsearchforsolutions.Inmanycases,thesedemonstrablycapableand successfuldatascientistsacquiredtheirexpertiseinnewanddecidedlydigital ways.

BetweenFebruaryandSeptemberof2012Kagglehostedtwocompetitions

aboutcomputergradingofstudentessays,whichweresponsoredbytheHewlett

Foundation.*KaggleandHewlettworkedwithmultipleeducationexpertstoset

upthecompetitions,andastheywerepreparingtolaunchmanyofthesepeople

wereworried.Thefirstcontestwastoconsistoftworounds.Elevenestablished

educationaltestingcompanieswouldcompeteagainstoneanotherinthefirst

round,withmembersofKaggle’scommunityofdatascientistsinvitedtojoinin,

individuallyorinteams,inthesecond.TheexpertswereworriedthattheKaggle

crowdwouldsimplynotbecompetitiveinthesecondround.Afterall,eachof

thetestingcompanieshadbeenworkingonautomaticgradingforsometimeand

haddevotedsubstantialresourcestotheproblem.Theirhundredsofperson-

yearsofaccumulatedexperienceandexpertiseseemedlikeaninsurmountable advantageoverabunchofnovices. Theyneedn’thaveworried.Manyofthe‘novices’drawntothechallenge outperformedallofthetestingcompaniesintheessaycompetition.Thesurprises continued when Kaggle investigated who the top performers were. In both competitions, none of the top three finishers had any previous significant experiencewitheitheressaygradingornaturallanguageprocessing.Andinthe secondcompetition,noneofthetopthreefinishershadanyformaltrainingin artificialintelligencebeyondafreeonlinecourseofferedbyStanfordAIfaculty andopentoanyoneintheworldwhowantedtotakeit.Peopleallovertheworld did,andevidentlytheylearnedalot.Thetopthreeindividualfinisherswere from,respectively,theUnitedStates,Slovenia,andSingapore. Quirky, another Web-based startup, enlists people to participate in both phasesofWeitzman’srecombinantinnovation—firstgeneratingnewideas,then filteringthem.Itdoesthisbyharnessingthepowerofmanyeyeballsnotonlyto comeupwithinnovationsbutalsotofilterthemandgetthemreadyformarket. Quirkyseeksideasfornewconsumerproductsfromitscrowd,andalsorelieson the crowd to vote on submissions, conduct research, suggest improvements, figureouthowtonameandbrandtheproducts,anddrivesales.Quirkyitself makes the final decisions about which products to launch and handles

engineering,manufacturing,anddistribution.Itkeeps70percentofallrevenue

madethroughitswebsiteanddistributestheremaining30percenttoallcrowd

members involved in the development effort; of this 30 percent, the person

submittingtheoriginalideagets42percent,thosewhohelpwithpricingshare

10percent,thosewhocontributetonamingshare5percent,andsoon.Bythe

fallof2012,Quirkyhadraisedover$90millioninventurecapitalfinancingand

hadagreementstosellitsproductsatseveralmajorretailers,includingTarget and Bed Bath & Beyond. One of its most successful products, a flexible

electricalpowerstripcalledPivotPower,soldmorethan373thousandunitsin

lessthantwoyearsandearnedthecrowdresponsibleforitsdevelopmentover

$400,000.

Affinnova,yetanotheryoungcompanysupportingrecombinantinnovation,

helpsitscustomerswiththesecondofWeitzman’stwophases:sortingthrough

thepossiblecombinationsofbuildingblockstofindthemostvaluableones.It does this by combining crowdsourcing with Nobel Prize–worthy algorithms. WhenCarlsbergbrewerieswantedtoupdatethebottleandlabelforBelgium’s Grimbergen,theworld’soldestcontinuallyproducedabbeybeer,itknewithad to proceed carefully. The company wanted to update the brand without sacrificingitsstrongreputationordownplayingitsninehundredyearsofhistory. Itknewthattheredesignwouldmeangeneratingmanycandidatesforeachof severalattributes—bottleshape,embossments,labelcolor,labelplacement,cap design,andsoon—thensettlingontherightcombinationofallofthese.The ‘right’combinationfromamongthethousandsofpossibilities,however,wasnot obviousattheoutset. The standard approach to this kind of problem is for the design team to generateafewcombinationsthattheythinkaregood,thenusefocusgroupsor other small-scale methods to finalize which is best. Affinnova offers a very different approach. It makes use of the mathematics of choice modeling, an advance significant enough to have earned a Nobel Prize for its intellectual godfather, economist Daniel McFadden. Choice modeling quickly identifies people’s preferences—do they prefer a brown embossed bottle with a small label, or a green non-embossed one with a large label?—by repeatedly presentingthemwithasmallsetofoptionsandaskingthemtoselectwhichthey like best. Affinnova presents these options via the Web and can find the mathematically optimal set of options (or at least come close to it) after involvingonlyafewhundredpeopleintheevaluationprocess.ForGrimbergen, thedesignthatresultedfromthisexplicitlyrecombinantprocesshadanapproval rating3.5timesgreaterthanthatofthepreviousbottle. 25 Whenweadopttheperspectiveofthenewgrowththeoristsandmatchit againstwhatweseewithWaze,Innocentive,Kaggle,Quirky,Affinnova,and manyothers,webecomeoptimisticaboutthecurrentandfutureofinnovation. Andthesedigitaldevelopmentsarenotconfinedtothehigh-techsector—they’re notjustmakingcomputersandnetworksbetterandfaster.They’rehelpingus driveourcarsbetter(andmaysoonmakeitunnecessaryforustodriveatall), allowingustoarriveatbetterpredictionsofsolarflares,solvingproblemsin foodscienceandtoxicology,andgivingusbetterpowerstripsandbeerbottles. Theseandcountlessotherinnovationswilladdupovertime,andthey’llkeep comingandkeepaddingup.Unlikesomeofourcolleagues,weareconfident thatinnovationandproductivitywillcontinuetogrowathealthyratesinthe future.Plentyofbuildingblocksareinplace,andthey’rebeingrecombinedin

betterandbetterwaysallthetime.

betterandbetterwaysallthetime. * Inreality,manyofthecountriesthat do havelargeamountsofmineralandcommoditywealth

*Inreality,manyofthecountriesthatdohavelargeamountsofmineralandcommoditywealth

areoftencrippledbythetwinterrorsofthe“resourcecurse”:lowgrowthratesandlotsofpoverty.

* Some have tied the invention of the cotton gin to increased demand for slave labor in the AmericanSouthandthereforetotheCivilWar,butitsdirecteconomiceffectoutsidethetextile industrywasminimal.

*Keepinmindthatifthereareonlyfifty-twoseedideasinsuchaneconomy,theyhavemany

morepotentialcombinationsthanthereareatomsinoursolarsystem.

*Improvementsinthisareaareimportantbecauseessaysarebetteratcapturingstudentleaning

thanmultiple-choicequestions,butmuchmoreexpensivetogradewhenhumanratersareused.

Automaticgradingofessayswouldbothimprovethequalityofeducationandloweritscost.

“AndhereIamthinkingofthoseastonishingelectronicmachines bywhichour

“AndhereIamthinkingofthoseastonishingelectronicmachines

bywhichour

mentalcapacitytocalculateandcombineisreinforcedandmultipliedbytheprocessand

toadegreethatherald

astonishingadvances.”

—PierreTeilharddeChardin

THE PREVIOUS FIVE CHAPTERS laidouttheoutstandingfeaturesofthesecond machineage:sustainedexponentialimprovementinmostaspectsofcomputing, extraordinarily large amounts of digitized information, and recombinant innovation.Thesethreeforcesareyieldingbreakthroughsthatconvertscience fiction into everyday reality, outstripping even our recent expectations and theories.What’smore,there’snoendinsight. Theadvanceswe’veseeninthepastfewyears,andintheearlysectionsof this book—cars that drive themselves, useful humanoid robots, speech

recognitionandsynthesissystems,3Dprinters,Jeopardy!-championcomputers

—arenotthecrowningachievementsofthecomputerera.They’rethewarm-up acts.Aswemovedeeperintothesecondmachineagewe’llseemoreandmore suchwonders,andthey’llbecomemoreandmoreimpressive. Howcanwebesosure?Becausetheexponential,digital,andrecombinant powersofthesecondmachineagehavemadeitpossibleforhumanitytocreate twoofthemostimportantone-timeeventsinourhistory:theemergenceofreal, usefulartificialintelligence(AI)andtheconnectionofmostofthepeopleonthe planetviaacommondigitalnetwork. Either of these advances alone would fundamentally change our growth prospects. When combined, they’re more important than anything since the IndustrialRevolution,whichforevertransformedhowphysicalworkwasdone.

ThinkingMachines,AvailableNow

Machines that can complete cognitive tasks are even more important than machinesthatcanaccomplishphysicalones.AndthankstomodernAIwenow havethem.Ourdigitalmachineshaveescapedtheirnarrowconfinesandstarted todemonstratebroadabilitiesinpatternrecognition,complexcommunication, andotherdomainsthatusedtobeexclusivelyhuman. We’ve also recently seen great progress in natural language processing, machinelearning(theabilityofacomputertoautomaticallyrefineitsmethods andimproveitsresultsasitgetsmoredata),computervision,simultaneous localizationandmapping,andmanyoftheotherfundamentalchallengesofthe discipline. We’re going to see artificial intelligence do more and more, and as this

happenscostswillgodown,outcomeswillimprove,andourliveswillgetbetter. Soon countless pieces of AI will be working on our behalf, often in the background.They’llhelpusinareasrangingfromtrivialtosubstantivetolife changing.TrivialusesofAIincluderecognizingourfriends’facesinphotosand recommendingproducts.Moresubstantiveonesincludeautomaticallydriving carsontheroad,guidingrobotsinwarehouses,andbettermatchingjobsandjob seekers.Buttheseremarkableadvancespaleagainstthelife-changingpotential ofartificialintelligence. Totakejustonerecentexample,innovatorsattheIsraelicompanyOrCam havecombinedasmallbutpowerfulcomputer,digitalsensors,andexcellent algorithmstogivekeyaspectsofsighttothevisuallyimpaired(apopulation numberingmorethantwentymillionintheUnitedStatesalone).Auserofthe OrCam system, which was introduced in 2013, clips onto her glasses a combination of a tiny digital camera and speaker that works by conducting soundwavesthroughthebonesofthehead. 1 Ifshepointsherfingeratasource oftextsuchasabillboard,packageoffood,ornewspaperarticle,thecomputer immediatelyanalyzestheimagesthecamerasendstoit,thenreadsthetextto herviathespeaker. Readingtext‘inthewild’—inavarietyoffonts,sizes,surfaces,andlighting conditions—hashistoricallybeenyetanotherareawherehumansoutpacedeven themostadvancedhardwareandsoftware.OrCamandsimilarinnovationsshow thatthisisnolongerthecase,andthathereagaintechnologyisracingahead.As itdoes,itwillhelpmillionsofpeopleleadfullerlives.TheOrCamcostsabout

$2,500—thepriceofagoodhearingaid—andiscertaintobecomecheaperover

time. Digital technologies are also restoring hearing to the deaf via cochlear implantsandwillprobablybringsightbacktothefullyblind;theFDArecently approved a first-generation retinal implant. 2 AI’s benefits extend even to quadriplegics,sincewheelchairscannowbecontrolledbythoughts. 3 Considered objectively,theseadvancesaresomethingclosetomiracles—andthey’restillin theirinfancy. Artificialintelligencewillnotjustimprovelives;itwillalsosavethem.After winningJeopardy!,forexample,Watsonenrolledinmedicalschool.Tobeabit moreprecise,IBMisapplyingthesameinnovationsthatallowedWatsonto answertoughquestionscorrectlytothetaskofhelpingdoctorsbetterdiagnose what’swrongwiththeirpatients.Insteadofvolumesandvolumesofgeneral knowledge,thesupercomputerisbeingtrainedtositontopofalloftheworld’s

high-qualitypublishedmedicalinformation;matchitagainstpatients’symptoms, medical histories, and test results; and formulate both a diagnosis and a treatmentplan.Thehugeamountsofinformationinvolvedinmodernmedicine makethistypeofadvancecriticallyimportant.IBMestimatesthatitwouldtake

ahumandoctor160hoursofreadingeachandeveryweekjusttokeepupwith

relevantnewliterature. 4 IBMandpartnersincludingMemorialSloan-KetteringCancerCenterandthe ClevelandClinicareworkingtobuildDr.Watson.Theorganizationsinvolvedin this program are careful to stress that the AI technologies will be used to augmentphysicians’clinicalexpertiseandjudgment,notreplacethem.Still,itis notimplausiblethatDr.Watsonmightonedaybetheworld’sbestdiagnostician. We’realreadyseeingAI-aideddiagnosesinsomemedicalspecialties.Ateam led by pathologist Andrew Beck developed the C-Path (computational pathologist)systemtoautomaticallydiagnosebreastcancerandpredictsurvival ratesbyexaminingimagesoftissue,justashumanpathologistsdo. 5 Sincethe

1920s,thesehumanshavebeentrainedtolookatthesamesmallsetofcancer

cellfeatures. 6 TheC-Pathteam,incontrast,haditssoftwarelookatimageswith afresheye—withoutanypre-programmednotionsaboutwhichfeatureswere associatedwithcancerseverityorpatientprognosis.Notonlywasthissoftware atleastasaccurateashumans,italsoidentifiedthreefeaturesofbreastcancer tissue that turned out to be good predictors of survival rates. Pathologists, however,hadnotbeentrainedtolookforthem. Asitracesahead,artificialintelligencemightbringwithitsometroubles, whichwe’lldiscussinourconclusion.Butfundamentally,thedevelopmentof thinkingmachinesisanincrediblypositiveone.

BillionsofInnovators,ComingSoon

In addition to powerful and useful AI, the other recent development that promises to further accelerate the second machine age is the digital interconnectionoftheplanet’speople.Thereisnobetterresourceforimproving

theworldandbetteringthestateofhumanitythantheworld’shumans—all7.1

billionofus.Ourgoodideasandinnovationswilladdressthechallengesthat arise,improvethequalityofourlives,allowustolivemorelightlyontheplanet, andhelpustakebettercareofoneanother.Itisaremarkableandunmistakable fact that, with the exception of climate change, virtually all environmental, social,andindividualindicatorsofhealthhaveimprovedovertime,evenas

humanpopulationhasincreased. Thisimprovementisnotaluckycoincidence;itiscauseandeffect.Things havegottenbetterbecausetherearemorepeople,whointotalhavemoregood ideasthatimproveouroveralllot.TheeconomistJulianSimonwasoneofthe first to make this optimistic argument, and he advanced it repeatedly and forcefully throughout his career. He wrote, “It is your mind that matters economically,asmuchormorethanyourmouthorhands.Inthelongrun,the mostimportanteconomiceffectofpopulationsizeandgrowthisthecontribution ofadditionalpeopletoourstockofusefulknowledge.Andthiscontributionis largeenoughinthelongruntoovercomeallthecostsofpopulationgrowth.” 7 BoththeoryanddatabearoutSimon’sinsight.Thetheoryofrecombinant innovation stresses how important it is to have more eyeballs looking at challengesandmorebrainsthinkingabouthowexistingbuildingblockscanbe rearrangedtomeetthem.Thistheoryfurtherholdsthatpeoplealsoplaythevital role of filtering and improving the innovations of others. And the data on everything from air quality to commodity prices to levels of violence show improvement over time. These data, in other words, show humanity’s remarkableabilitytomeetitschallenges. WedohaveonequibblewithSimon,however.Hewrotethat,“Themainfuel tospeedtheworld’sprogressisourstockofknowledge,andthebrakeisour lackofimagination.” 8 Weagreeaboutthefuelbutdisagreeaboutthebrake.The mainimpedimenttoprogresshasbeenthat,untilquiterecently,asizableportion of the world’s people had no effective way to access the world’s stock of knowledgeortoaddtoit. IntheindustrializedWestwehavelongbeenaccustomedtohavinglibraries, telephones,andcomputersatourdisposal,butthesehavebeenunimaginable luxuries to the people of the developing world. That situation is rapidly

changing.In2000,forexample,therewereapproximatelysevenhundredmillion

mobilephonesubscriptionsintheworld,fewerthan30percentofwhichwerein

developingcountries. 9 By2012thereweremorethansixbillionsubscriptions, over 75 percent of which were in the developing world. The World Bank estimatesthatthree-quartersofthepeopleontheplanetnowhaveaccesstoa mobilephone,andthatinsomecountriesmobiletelephonyismorewidespread thanelectricityorcleanwater. The first mobile phones bought and sold in the developing world were capableoflittlemorethanvoicecallsandtextmessages,yeteventhesesimple devices could make a significant difference. Between 1997 and 2001 the

economistRobertJensenstudiedasetofcoastalvillagesinKerala,India,where fishing was the main industry. 10 Jensen gathered data both before and after mobile phone service was introduced, and the changes he documented are remarkable.Fishpricesstabilizedimmediatelyafterphoneswereintroduced, andeventhoughthesepricesdroppedonaverage,fishermen’sprofitsactually increasedbecausetheywereabletoeliminatethewastethatoccurredwhenthey tooktheirfishtomarketsthatalreadyhadenoughsupplyfortheday.Theoverall economicwell-beingofbothbuyersandsellersimproved,andJensenwasable totiethesegainsdirectlytothephonesthemselves. Now,ofcourse,eventhemostbasicphonessoldinthedevelopingworldare morepowerfulthantheonesusedbyKerala’sfishermanoveradecadeago.

Approximately70percentofallphonessoldworldwidein2012were‘feature

phones’—lesscapablethantheAppleiPhoneandSamsungGalaxysmartphones oftherichworld,butstillabletotakepictures(andoftenvideos),browsethe Web, and run at least some applications. 11 And cheap mobile devices keep improving.TechnologyanalysisfirmIDCforecaststhatsmartphoneswilloutsell featurephonesinthenearfuture,andwillmakeupabouttwo-thirdsofallsales by2017. 12 Thisshiftisduetocontinuedsimultaneousperformanceimprovementsand costdeclinesinbothmobilephonedevicesandnetworks,andithasanimportant consequence:itwillbringbillionsofpeopleintothecommunityofpotential knowledgecreators,problemsolvers,andinnovators. Today,peoplewithconnectedsmartphonesortabletsanywhereintheworld haveaccesstomany(ifnotmost)ofthesamecommunicationresourcesand informationthatwedowhilesittinginourofficesatMIT.Theycansearchthe Web and browse Wikipedia. They can follow online courses, some of them taughtbythebestintheacademicworld.Theycansharetheirinsightsonblogs, Facebook,Twitter,andmanyotherservices,mostofwhicharefree.Theycan evenconductsophisticateddataanalysesusingcloudresourcessuchasAmazon WebServicesandR,anopensourceapplicationforstatistics. 13 Inshort,they can be full contributors in the work of innovation and knowledge creation, takingadvantageofwhatAutodeskCEOCarlBasscalls“infinitecomputing.” 14 Until quite recently rapid communication, information acquisition, and knowledgesharing,especiallyoverlongdistances,wereessentiallylimitedto the planet’s elite. Now they’re much more democratic and egalitarian, and gettingmoresoallthetime.ThejournalistA.J.Lieblingfamouslyremarked that, “Freedom of the press is limited to those who own one.” It is no

exaggeration to say that billions of people will soon have a printing press, referencelibrary,school,andcomputerallattheirfingertips. 15 Thoseofuswhobelieveinthepowerofrecombinantinnovationbelievethat thisdevelopmentwillboosthumanprogress.Wecan’tpredictexactlywhatnew insights,products,andsolutionswillarriveinthecomingyears,butwearefully confident that they’ll be impressive. The second machine age will be characterized by countless instances of machine intelligence and billions of interconnectedbrainsworkingtogethertobetterunderstandandimproveour world.Itwillmakemockeryoutofallthatcamebefore.

“Mosteconomicfallaciesderivefromthetendencytoassumethatthereisafixedpie,

“Mosteconomicfallaciesderivefromthetendencytoassumethatthereisafixedpie,

thatonepartycangainonlyattheexpenseofanother.”

—MiltonFriedman

EACHDAYGOVERNMENTAGENCIES,thinktanks,NGOs,andacademicresearchers generate more statistics than any person could read, let alone absorb. On television,inthepagesofthebusinesspress,andintheblogosphere,achorusof analystsdebateandpredicttrendsininterestrates,unemployment,stockprices, deficitsandmyriadotherindicators.Butwhenyouzoomoutandconsidertrends overthepastcentury,oneoverwhelmingfactloomsaboveallothers:overall livingstandardshaveincreasedenormouslyintheUnitedStatesandworldwide.

IntheUnitedStates,therateofGDPgrowthperpersonhasaveraged1.9percent

peryeargoingbacktotheearly1800s. 1 Applyingtheruleof70(thetimeto

doubleavalueisroughlyequalto70dividedbyitsgrowthrate),weseethatthis

wasenoughtodoublelivingstandardseverythirty-sixyears,quadruplingthem overthecourseofatypicallifetime.* Thisincreaseisimportantbecauseeconomicgrowthcanhelpsolveahostof

otherchallenges.IfGDPoftheUnitedStatesgrowsjust1percentfastereach

yearthancurrentlyprojected,Americanswouldbefivetrilliondollarsricherby 2033. 2 IfGDPgrowsjust0.5percentfaster,theU.S.budgetproblemwouldbe solvedwithoutanychangestopolicy. 3 Ofcourse,slowergrowthwouldmakeit significantlyhardertoclosethedeficit,letaloneincreasespendingonanynew initiativesorcuttaxes.

ProductivityGrowth

ButwhatdrivesincreasesinGDPperperson?Partofitcomesfromusingmore resources.Butmostofitcomesfromincreasesinourabilitytogetmoreoutput fromthegivenlevelofinputs—inotherwords,increasesinproductivity.(Most commonly, this term is used as shorthand for ‘labor productivity,’which is outputperhourworked[oroutputperworker].)*Inturn,productivitygrowth comesfrominnovationsintechnologyandtechniquesofproduction. Simply working more hours does not increase productivity. Indeed, Americansonceroutinelyworkedfifty,sixty,orevenseventyhoursperweek. Whilesomestilldo,theaverageworkweekisshorternow(thirty-fivehoursper week),andyetlivingstandardsarehigher.RobertSolowgothisNobelPrizein Economicsforshowingthatincreasesinlaborinputandcapitalinputcouldnot explainmostoftheincreaseinthetotaloutputoftheeconomy.Infact,itwould

taketheaverageAmericanonlyelevenhoursoflaborperweektoproduceas

muchasheorsheproducedinfortyhoursin1950.Thatrateofimprovementis

comparable for workers in Europe and Japan, and even higher in some developingnations.*

FIGURE7.1LaborProductivity

in some developingnations. * FIGURE7.1 LaborProductivity

Productivityimprovementwasparticularlyrapidinthemiddlepartofthe

twentiethcentury,especiallythe1940s,50s,and60s,asthetechnologiesofthe

firstmachineage,fromelectricitytotheinternalcombustionengine,started

firingonallcylinders.However,in1973productivitygrowthsloweddown(see

figure7.1).

In1987,BobSolowhimselfnotedthattheslowdownseemedtocoincidewith

theearlydaysofthecomputerrevolution,famouslyremarking,“Weseethe computerageeverywhere,exceptintheproductivitystatistics.” 4 In1993,Erik published an article evaluating the “Productivity Paradox” that noted the computers were still a small share of the economy and that complementary innovationsweretypicallyneededbeforegeneralpurposetechnologieslikeIT hadtheirrealimpact. 5 Laterworktakingintoaccountmoredetaileddataon productivityandITuseamongindividualfirmsrevealedastrongandsignificant correlation:theheaviestITusersweredramaticallymoreproductivethantheir competitors. 6 By the mid-1990s, these benefits were big enough to become visibleintheoverallU.S.economy,whichexperiencedageneralproductivity surge.Whilethisrisehadanumberofcauses,economistsnowattributethe lion’sshareofthosegainstothepowerofIT. 7

Theproductivityslowdowninthe1970s,andthesubsequentspeed-uptwenty

yearslater,hadaninterestingprecedent.Inthelate1890s,electricitywasbeing

introducedtoAmericanfactories.Butthe“productivityparadox”ofthaterawas thatlaborproductivitygrowthdidnottakeoffforovertwentyyears.Whilethe technologiesinvolvedwereverydifferent,manyoftheunderlyingdynamics werequitesimilar. University of Chicago economist Chad Syverson looked closely at the underlyingproductivitydataandshowedhoweerilyclosethisanalogyis. 8 As

showninfigure7.2,theslowstartandsubsequentaccelerationofproductivity

growthintheelectricityeramatcheswellwiththespeed-upthatbeganinthe

1990s.Thekeytounderstandingthispatternistherealizationthat,asdiscussed

inchapter5,GPTsalwaysneedcomplements.Comingupwiththosecantake

years, or even decades, and this creates lags between the introduction of a technology and the productivity benefits. We’ve clearly seen this with both electrificationandcomputerization.

FIGURE7.2LaborProductivityinTwoEras

FIGURE7.2 LaborProductivityinTwoEras Perhaps the most important complementary innovations are the

Perhaps the most important complementary innovations are the business processchangesandorganizationalcoinventionsthatnewtechnologiesmake possible. Paul David, an economic historian at Stanford University and the UniversityofOxford,examinedtherecordsofAmericanfactorieswhenthey first electrified and found that they often retained a similar layout and organizationtothosethatwerepoweredbysteamengines. 9 Inasteamengine– drivenplant,powerwastransmittedviaalargecentralaxle,whichinturndrove aseriesofpulleys,gears,andsmallercrankshafts.Iftheaxlewastoolongthe torsioninvolvedwouldbreakit,somachinesneededtobeclusterednearthe main power source, with those requiring the most power positioned closest.

Exploitingallthreedimensions,industrialengineersputequipmentonfloors aboveandbelowthecentralsteamenginestominimizethedistancesinvolved. Yearslater,whenthathallowedGPTelectricityreplacedthesteamengine, engineerssimplyboughtthelargestelectricmotorstheycouldfindandstuck themwherethesteamenginesusedtobe.Evenwhenbrand-newfactorieswere built,theyfollowedthesamedesign.Perhapsunsurprisingly,recordsshowthat theelectricmotorsdidnotleadtomuchofanimprovementinperformance. Theremighthavebeenlesssmokeandalittlelessnoise,butthenewtechnology wasnotalwaysreliable.Overall,productivitybarelybudged. Onlyafterthirtyyears—longenoughfortheoriginalmanagerstoretireand bereplacedbyanewgeneration—didfactorylayoutschange.Thenewfactories lookedmuchlikethoseweseetoday:asinglestoryspreadoutoveranacreor more.Insteadofasinglemassiveengine,eachpieceofequipmenthaditsown smallelectricmotor.Insteadofputtingthemachinesneedingthemostpower closesttothepowersource,thelayoutwasbasedonasimpleandpowerfulnew principle:thenaturalworkflowofmaterials. Productivitydidn’tmerelyinchupwardontheresultingassemblylines;it doubled or even tripled. What’s more, for most of the subsequent century, additional complementary innovations, from lean manufacturing and steel minimillstoTotalQualityManagementandSixSigmaprinciples,continuedto boostmanufacturingproductivity. As with earlier GPTs, significant organizational innovation is required to capturethefullbenefitofsecondmachineagetechnologies.TimBerners-Lee’s

inventionoftheWorldWideWebin1989,totakeanobviousexample,initially

benefitedonlyasmallgroupofparticlephysicists.Butdueinparttothepower of digitization and networks to speed the diffusion of ideas, complementary innovationsarehappeningfasterthantheydidinthefirstmachineage.Lessthan tenyearsafteritsintroduction,entrepreneurswerefindingwaystousetheWeb toreinventpublishingandretailing. Whilelessvisible,thelargeenterprise-wideITsystemsthatcompaniesrolled outinthe1990shavehadanevenbiggerimpactonproductivity. 10 Theydidthis mainlybymakingpossibleawaveofbusinessprocessredesign.Forexample, Walmartdroveremarkableefficienciesinretailingbyintroducingsystemsthat sharedpoint-of-saledatawiththeirsuppliers.Therealkeywastheintroduction

ofcomplementaryprocessinnovationslikevendormanagedinventory,cross-

docking, and efficient consumer response that have become staple business-

schoolcasestudies.Theynotonlymadeitpossibletoincreasesalesfrom$1

billionaweekin1993to$1billioneverythirty-sixhoursin2001,butalso

helpeddrivedramaticincreasesintheentireretailinganddistributionindustries, accountingformuchoftheadditionalproductivitygrowthnationwideduring thisperiod. 11

ITinvestmentsoaredinthe1990s,peakingwithasurgeofinvestmentinthe

latterhalfofthedecadeasmanycompaniesupgradedtheirsystemstotake advantageoftheInternet,implementlargeenterprisesystems,andavoidthe

much-hypedY2Kbug.Atthesametime,innovationinsemiconductorstook

gigantic leaps, so the surging spending on IT delivered even more rapidly increasinglevelsofcomputerpower.Adecadeafterthecomputerproductivity paradoxwaspopularized,Harvard’sDaleJorgenson,workingwithKevinStiroh attheNewYorkFederalReserveBankdidacarefulgrowthaccountingand concluded,“Aconsensushasemergedthatalargeportionoftheacceleration through 2000 can be traced to the sectors of the economy that produce informationtechnologyoruseITequipmentandsoftwaremostintensively.” 12 Butit’snotjustthecomputer-producingsectorsthataredoingwell.KevinStiroh oftheNewYorkFederalReserveBankfoundthatindustriesthatwereheavier

usersofITtendedtobemoreproductivethroughoutthe1990s.Thispatternwas

evenmoreevidentinrecentyears,accordingtoacarefulstudybyHarvard’s DaleJorgensonandtwocoauthors.Theyfoundthattotalfactorproductivity growth increased more between the 1990s and 2000s in IT-using industries, while it fell slightly in those sectors of the economy that did not use IT extensively. 13 It’simportanttonotethatthecorrelationbetweencomputersandproductivity isnotjustevidentattheindustrylevel;itoccursatthelevelofindividualfirms aswell.InworkErikdidwithLorinHittoftheUniversityofPennsylvania WhartonSchool,hefoundthatfirmsthatusemoreITtendtohavehigherlevels ofproductivityandfasterproductivitygrowththantheirindustrycompetitors. 14 The first five years of the twenty-first century saw a renewed wave of innovationandinvestment,thistimelessfocusedoncomputerhardwareand morefocusedonadiversifiedsetofapplicationsandprocessinnovations.For instance,asAndydescribedinacasestudyhedidforHarvardBusinessSchool, CVS found that their prescription drug ordering process was a source of customerfrustration,sotheyredesignedandsimplifiedit. 15 Byembeddingthe stepsinanenterprise-widesoftwaresystem,theywereabletoreplicatethedrug orderingprocessinoverfourthousandlocations,dramaticallyboostingcustomer satisfactionandultimatelyprofits.CVSwasnotatypical.Inastatisticalanalysis

ofoversixhundredfirmsthatErikdidwithLorinHitt,hefoundittakesan averagefivetosevenyearsbeforefullproductivitybenefitsofcomputersare visibleintheproductivityofthefirmsmakingtheinvestments.Thisreflectsthe timeandeffortrequiredtomaketheothercomplementaryinvestmentsthatbring a computerization effort success. In fact, for every dollar of investment in computer hardware, companies need to invest up to another nine dollars in software,training,andbusinessprocessredesign. 16 Theeffectsoforganizationalchangeslikethesebecameincreasinglyvisible intheindustry-levelproductivitystatistics. 17 Theproductivitysurgeinthe1990s wasmostvisibleincomputer-producingindustries,butoverallproductivitygrew evenfasterintheearlyyearsofthetwenty-firstcentury,whenamuchbroader setofindustriessawsignificantproductivitygains.LikeearlierGPTs,thepower of computers was their ability to affect productivity far from their ‘home’ industry.

Overall,Americanproductivitygrowthinthedecadefollowingtheyear2000

exceededeventhehighgrowthratesoftheroaring1990s,whichinturnwas

higherthan1970sor1980sgrowthrateshadbeen. 18 TodayAmericanworkersaremoreproductivethanthey’veeverbeen,buta closerlookatrecentnumberstellsamorenuancedstory.Thegoodperformance

sincetheyear2000wasclusteredintheearlyyearsofthedecade.Since2005,

productivitygrowthhasnotbeenasstrong.Asnotedinchapter5,thishasledto

anewwaveofworriesaboutthe“endofgrowth”byeconomists,journalists,and bloggers.Wearenotconvincedbythepessimists.Theproductivitylullafterthe introductionofelectricitydidnotmeantheendofgrowth,nordidthelullinthe

1970s.

Part of the recent slowdown simply reflects the Great Recession and its aftermath.Recessionsarealwaystimesofpessimism,whichisunderstandable, andthepessimisminvariablyspillsoverintopredictionsabouttechnologyand thefuture.Thefinancialcrisisandburstofthehousingbubbleledtoacollapse ofconsumerconfidenceandwealth,whichtranslatedintodramaticallylower

demandandGDP.WhiletherecessiontechnicallyendedinJune2009,aswe

writethisin2013theU.S.economyisstilloperatingwellbelowitspotential,

withunemploymentat7.6percentandcapacityutilizationat78percent.During

suchaslump,anymetricthatincludesoutputinthenumerator,suchaslabor productivity,willoftenbeatleasttemporarilydepressed.Infact,whenyoulook

athistory,youseethatintheearlyyearsoftheGreatDepression,inthe1930s,

productivitydidn’tjustslowbutactuallyfellfortwoyearsinarow—something

itneverdidintherecentslump.Growthpessimistshadevenmorecompanyin

the1930sthantheydotoday,butthefollowingthreedecadesprovedtobethe

bestonesofthetwentiethcentury.Gobacktofigure7.2andlookmostcloselyat

thedashedlinechartingtheyearsfollowingthedipinproductivityintheearly

1930s.You’llseethebiggestwaveofgrowthandbountythatthefirstmachine

ageeverdelivered. Theexplanationforthisproductivitysurgeisinthelagsthatwealwayssee whenGPTsareinstalled.Thebenefitsofelectrificationstretchedfornearlya centuryasmoreandmorecomplementaryinnovationswereimplemented.The digitalGPTsofthesecondmachineagearenolessprofound.EvenifMoore’s Law ground to a halt today, we could expect decades of complementary innovationstounfoldandcontinuetoboostproductivity.However,unlikethe steam engine or electricity, second machine age technologies continue to improveataremarkablyrapidexponentialpace,replicatingtheirpowerwith digital perfection and creating even more opportunities for combinatorial innovation.Thepathwon’tbesmooth—foronething,wehaven’tbanishedthe businesscycle—butthefundamentalsareinplaceforbountythatvastlyexceeds anythingwe’veeverseenbefore.

anythingwe’veeverseenbefore. *

*TheRuleof70(or,moreprecisely,theruleof69.3percent)isbasedonthefollowingequation:

(1+x)y = 2 where x is the rate of growth and y is the number of years. Taking the natural

logarithmofbothsidesgivesyln(1+x)=ln2.Theln(2)is0.693andforsmallx,ln(1+x)is

roughlyequaltox,sotheequationsimplifiestoxy=70percent.

*Onecanalsomeasurecapitalproductivity,whichisoutputperunitofcapitalinput;ormultifactor productivity, which is output divided by a weighted average of both capital and labor inputs. Economistssometimesuseanothertermformultifactorproductivity,the“SolowResidual,”which betterreflectsthefactthatwedon’tnecessarilyknowitsorigins.RobertSolowhimselfnotedthat itwaslessaconcretemeasureoftechnologicalprogressthana“measureofourignorance.”

That’s a good thing, because there are natural limits to how much we can increase inputs, especially labor. They’re subject to diminishing returns—no one is going to work more than twenty-four hours a day, or employ more than 100 percent of the labor force. In contrast, productivitygrowthreflectsabilitytoinnovate—it’slimitedonlybyourimaginations.

*Outputdividedbylaborandphysicalcapitalinputsisoftenmoreambitiouslycalled‘totalfactor productivity.’ However, that term can be a bit misleading, because there are other inputs to production. For instance, companies can make major investments in intangible organizational capital.Themorekindsofinputsweareabletomeasure,thebetterwecanaccountforoverall outputgrowth.Asaresult,theresidualthatwelabel“productivity”(notexplainedbygrowthof inputs)willgetsmaller.

“TheGrossNationalProductdoesnotincludethebeautyofourpoetryortheintelligence

“TheGrossNationalProductdoesnotincludethebeautyofourpoetryortheintelligence

ofourpublicdebate.Itmeasuresneitherourwitnorourcourage,neitherourwisdomnor

ourlearning,neitherourcompassionnorourdevotion.Itmeasureseverything,inshort,

exceptthatwhichmakeslifeworthwhile.”

—RobertF.Kennedy

WHENPRESIDENTHOOVERWAStryingtounderstandwhatwashappeningduring theGreatDepressionanddesignaprogramtofightit,acomprehensivesystem ofnationalaccountsdidnotexist.Hehadtorelyonscattereddatalikefreight car loadings, commodity prices, and stock price indexes that gave only an incomplete and often unreliable view of economic activity. The first set of

nationalaccountswaspresentedtoCongressin1937basedonthepioneering

workofNobelPrizewinnerSimonKuznets,whoworkedwithresearchersatthe NationalBureauofEconomicResearchandateamattheU.S.Departmentof Commerce. The resulting set of metrics have served as beacons that helped illuminate many of the dramatic changes that transformed the economy throughoutthetwentiethcentury. Butastheeconomyhaschangedso,too,mustourmetrics.Moreandmore whatwecareaboutinthesecondmachineageareideas,notthings—mind,not matter;bits,notatoms;andinteractions,nottransactions.Thegreatironyofthis informationageisthat,inmanyways,weactuallyknowlessaboutthesources ofvalueintheeconomythanwedidfiftyyearsago.Infact,muchofthechange hasbeeninvisibleforalongtimesimplybecausewedidnotknowwhattolook for.There’sahugelayeroftheeconomyunseenintheofficialdataand,forthat matter,unaccountedforontheincomestatementsandbalancesheetsofmost companies.Freedigitalgoods,thesharingeconomy,intangiblesandchangesin ourrelationshipshavealreadyhadbigeffectsonourwell-being.Theyalsocall forneworganizationalstructures,newskills,newinstitutions,andperhapseven areassessmentofsomeofourvalues.

MusictoYourEars

Thestoryofmusic’smovefromphysicalmediatocomputerfileshasbeentold often and well, but one of that transition’s most interesting aspects is less discussed.Musicishidingitselffromourtraditionaleconomicstatistics.Salesof

musiconphysicalmediadeclinedfrom800millionunitsin2004tolessthan

400millionunitsin2008.Yetoverthesametimeperiodtotalunitsofmusic

purchasedstillgrew,reflectinganevenfasterincreaseinthepurchasesofdigital downloads.DigitalstreamssuchasiTunes,Spotify,orPandoraalsocameto prominence, and, of course, the purchase data don’t reflect the even larger

numberofsongsthatwereshared,streamed,ordownloadedforfree,oftenvia

piracy.BeforetheriseoftheMP3,eventhemostfanaticalmusicfan,witha

basementstackedhighwithLPs,tapes,andCDs,wouldn’thavehadafractionof thetwentymillionsongsavailableonachild’ssmartphoneviaserviceslike SpotifyorRhapsody.What’smore,cleverresearchbyJoelWaldfogelatthe UniversityofMinnesotafindsquantitativeevidencethattheoverallqualityof musichasnotdeclinedoverthepastdecadeandis,ifanything,higherthan ever. 1 Ifyou’relikemostpeople,youarelisteningtomoreandbettermusicthan everbefore. Sohowdidmusicdisappear?Thevalueofmusichasnotchanged,onlythe

price.From2004to2008,thecombinedrevenuefromsalesofmusicdropped

from$12.3billionto$7.4billion—that’sadeclineof40percent.Evenwhenwe

include all digital sales, throwing in ringtones on mobile phones for good

measure,thetotalrevenuestotherecordcompaniesarestilldown30percent.

SimilareconomicsapplywhenyoureadtheNew York Times,Bloomberg Businessweek,orMITSloanManagementReviewonlineatareducedpriceorfor free instead of buying a physical copy at the newsstand, or when you use Craigslistinsteadoftheclassifiedads,orwhenyousharephotosviaFacebook instead of mailing prints around to friends and relatives. Analog dollars are becomingdigitalpennies. By now, the number of pages of digital text and images on the Web is estimatedtoexceedonetrillion. 2 Asdiscussedinchapter4,bitsarecreatedat virtuallyzerocostandtransmittedalmostinstantaneouslyworldwide.What’s more,acopyofadigitalgoodisexactlyidenticaltotheoriginal.Thisleadsto someverydifferenteconomicsandsomespecialmeasurementproblems.When abusinesstravelercallshometotalktoherchildrenviaSkype,thatmayadd zerotoGDP,butit’shardlyworthless.Eventhewealthiestrobberbaronwould havebeenunabletobuythisservice.Howdowemeasurethebenefitsoffree goodsorservicesthatwereunavailableatanypriceinpreviouseras?

WhatGDPLeavesOut

Despite all the attention it gets from economists, pundits, journalist, and politicians, GDP, even if it were perfectly measured, does not quantify our

welfare.ThetrendsinGDPgrowthandproductivitygrowthcoveredinchapter7

areimportant,buttheyarenotsufficientmeasuresofouroverallwell-being,or

evenoureconomicwell-being.RobertKennedyputthispoeticallyinhisquote

atthebeginningofthischapter. Whileitwouldbeunrealistictoputadollarvalueonstirringoratorylike RFK’s,wecandoabetterjobofunderstandingourbasiceconomicprogressby consideringsomeofthechangesinthegoodsandservicesthatweareableto consume.Itsoonbecomesclearthatthetrendsintheofficialstatisticsnotonly underestimateourbounty,butinthesecondmachineagetheyhavealsobecome increasinglymisleading. Inadditiontotheirvastlibraryofmusic,childrenwithsmartphonestoday have access to more information in real time via the mobile web than the presidentoftheUnitedStateshadtwentyyearsago.Wikipediaaloneclaimsto haveoverfiftytimesasmuchinformationasEncyclopaediaBritannica, the premier compilation of knowledge for most of the twentieth century. 3 Like WikipediabutunlikeBritannica,muchoftheinformationandentertainment availabletodayisfree,asareoveronemillionappsonsmartphones. 4 Becausetheyhavezeroprice,theseservicesarevirtuallyinvisibleinthe officialstatistics.Theyaddvaluetotheeconomy,butnotdollarstoGDP.And becauseourproductivitydataare,inturn,basedonGDPmetrics,theburgeoning availability of free goods does not move the productivity dial. There’s little doubt,however,thattheyhaverealvalue.WhenagirlclicksonaYouTube videoinsteadofgoingtothemovies,she’ssayingthatshegetsmorenetvalue from YouTube than traditional cinema. When her brother downloads a free gamingapponhisiPadinsteadofbuyinganewvideogame,he’smakinga similarstatement.

Free:GoodforWell-Being,BadforGDP

Insomeways,theproliferationoffreeproductsevenpushesGDPdownward.If thecostofcreatinganddeliveringanencyclopediatoyourdesktopisafew penniesinsteadofthousandsofdollars,thenyou’recertainlybetteroff.Butthis decreaseincostslowersGDPevenasourpersonalwell-beingincreases,leaving GDPtotravelintheoppositedirectionofourtruewell-being.Asimpleswitchto usingafreetextingservicelikeApple’siChatinsteadofSMS,freeclassifieds likeCraigslistinsteadofnewspaperads,orfreecallslikeSkypeinsteadofa traditional telephone service can make billions of dollars disappear from companies’revenuesandtheGDPstatistics. 5 Astheseexamplesshow,oureconomicwelfareisonlylooselyrelatedto GDP. Unfortunately many economists, journalists, and much of the general

publicstilluse“GDPgrowth”asasynonymfor“economicgrowth.”Formuch ofthetwentiethcentury,thiswasafaircomparison.Ifoneassumesthateach additionalunitofproductioncreatedasimilarincrementinwell-being,then countinguphowmanyunitswereproduced,asGDPdoes,wouldbeafine approximationofwelfare.Anationthatsellsmorecars,morebushelsofwheat, andmoretonsofsteelprobablycorrespondstoanationwhosepeoplearebetter off. Withagreatervolumeofdigitalgoodsintroducedeachyearthatdonothave adollarprice,thistraditionalGDPheuristicisbecominglessuseful.Aswe discussed in chapter 4, the second machine age is often described as an “informationeconomy,”andwithgoodreason.Morepeoplethaneverareusing Wikipedia,Facebook,Craigslist,Pandora,Hulu,andGoogle,withthousandsof newdigitalgoodsintroducedeachyear. The U.S. Bureau of Economic Analysis defines the information sector’s contributiontotheeconomyasthesumofthesalesofsoftware,publishing, motion pictures, sound recording, broadcasting, telecommunications, and informationanddataprocessingservices.Accordingtotheofficialmeasures,

theseaccountforjust4percentofourGDPtoday,almostpreciselythesame

share of GDPas in the late 1980s, before the World Wide Web was even invented.Butclearlythisisn’tright.Theofficialstatisticsaremissingagrowing shareoftherealvaluecreatedinoureconomy.

MeasuringGrowthwithaTimeMachine:WouldYou

Rather

?

CanweimproveonGDPasameasureofwell-being?Economistssometimesuse analternateapproachthatresemblesthechildren’sgame“Wouldyourather

?”The1912Searsshoppingcataloghadthousandsofitemsforsale,froma

“SearsMotorCar”for$335(page1,213)todozensofpairsofwomen’sshoes,

someavailableforaslittleas$1.50(pages371–79).SupposeIgaveyouan

expandedversionofthiscatalogthatlistedallthegoodsandservicesavailable

in1912,notjustfromSears,butfromanysellerintheeconomyof1912,andall

thesamepricesas1912. 6 Wouldyourathershopexclusivelyinthatoldcatalog, withnootherchoices,orwouldyouratherpaytoday’spricesforafullselection oftoday’sgoodsandservices? Ortomakethecomparisonlessdifficult,picktwomorerecentcatalogs,like

1993versus2013.Ifyouhadfiftythousanddollarstospend,wouldyourather

beabletobuyany1993-modelcar(itwouldbebrand-new)andpay1993prices,

ora2013carandpay2013prices?Wouldyouratherbeabletobuythebananas,

contactlenses,chickenwings,shirts,chairs,bankingservices,airlinetickets, movies,telephoneservice,healthcare,housingservices,lightbulbs,computers, gasoline, and other goods and services that were available in 1993 at 1993

prices?Orwouldyouratherbuytheequivalent2013basketofservicesat2013

prices? Bananasoragallonofgasolinehavenotreallychangedqualitativelysince

1993,sotheonlydifferencetoconsideristheirprice.Ifthatweretheonly

difference, inflation would be easy to calculate, and the “would you rather” comparison would be a lot easier, too. For other goods, though, especially second machine age goods like online information and mobile phone capabilities,therehavebeenbigchangesinquality,sotherealquality-adjusted pricemayhavefallenevenifthenominalstickerpricehasincreased.What’s more,therearealotofnewgoodsthatdidn’texistbefore,especiallydigital goods.Therearealsosomeoldergoodsandservicesthathavebeendiscontinued ordegraded.It’shardtofindagoodhorsehiderazorstropthesedays, 7 ora1993 vintagepersonalcomputer,oragasstationwheretheattendantsroutinelywash yourwindshieldfornocharge,liketheyoncedid. Onceyoupickwhichcatalogyoulikebetter,thenextstepaskshowmuch money I would have to pay you to make you indifferent between the two catalogs. If I have to pay you 20 percent more to make you just as happy shoppingfromthenewcatalogasyouwouldbeshoppingfromtheoldcatalog,

thentheoverallpriceindexhasincreasedby20percent.Andifyourincomehas

notchanged,thenthaterosionofpurchasingpowertranslatestoanequivalent fallinyourstandardofliving.Similarly,ifyourincomeincreasesfasterthanthe priceindex,thenyourstandardoflivingisincreasing. Thisapproachmakessenseconceptually,andit’sthebasisforthewaymost moderngovernmentscalculatechangesinthestandardofliving.Forinstance, thecostoflivingadjustmentsusedtoindexSocialSecuritypaymentsarebased onthiskindofanalysis. 8 Butthedatausedforthesecalculationsarealmost alwaysdrawn,understandably,frommarkettransactionswheremoneychanges hands.Thefreeeconomyisnotfactoredin.

ConsumerSurplus:HowMuchWouldYouPayIfYouHad

To?

Analternativeapproachmeasurestheconsumersurplusgeneratedbygoodsand services.Consumersurpluscomparestheamountaconsumerwouldhavebeen willingtopayforsomethingtotheamounttheyactuallyhavetopay.Ifyou wouldhappilypayonedollartoreadthemorningnewspaperbutinsteadyouget itforfree,thenyou’vejustgainedonedollarofconsumersurplus.However,as notedabove,replacingapaidnewspaperwithanequivalentfreenewservice woulddecreaseGDPeventhoughitincreasedconsumersurplus. 9 Inthiscase, consumersurpluswouldbeabettermeasureofoureconomicwell-being.Yetas appealingasconsumersurplusisasaconcept,itisalsoextremelydifficultto measure. Thedifficultyinmeasuringtheconsumersurplus,however,hasnotstoppeda

numberofresearchersfromtryingtoekeoutsomeestimates.In1993,Erik

wroteapapercalculatingthattherapidlygrowingconsumersurplusfromprice

declinesincomputersincreasedeconomicwelfarebyabout$50billioneach

year.* 10 Ofcourse,whentheproductbeingstudiedisalreadyfree,lookingatprice declines doesn’t work. Recent research that Erik did with Joo Hee Oh, a postdoctoralstudentatMIT,tookadifferentapproach.Theystartedwiththe observationthatevenwhenpeopledon’tpaywithmoney,theystillgiveup somethingvaluablewhenevertheyusetheirInternet:theirtime. 11 Nomatter howrichorpoorweare,eachofusgetstwenty-fourhoursinaday.Inorderto consume YouTube, Facebook, or e-mail, we must ‘pay’ attention. In fact, AmericansnearlydoubledtheamountofleisuretimetheyspentonInternet

between2000and2011.Thisimpliesthattheyvalueditmorethantheother

waystheycouldspendtheirtime.Byconsideringthevalueofusers’timeand comparingleisuretimespentontheInternettotimespentinotherways,Erik

andJooHeeestimatedthattheInternetcreatedabout$2,600ofvalueperuser

eachyear.NoneofthisshowedupintheGDPstatisticsbutifithad,GDP growth—and thus productivity growth—would have been about 0.3 percent

highereachyear.Inotherwords,insteadofthereported1.2percentproductivity

growthfor2012,itwouldhavebeen1.5percent.

Incontrasttoleisure,wheremoretimeisagoodthing,valueatworkis created by saving time. Hal Varian, the chief economist at Google, looked specificallyattimesavingsgainedfromGooglesearches. 12 Heandhisteam gatheredarandomsampleofGooglequeries,suchas:“Inmakingcookies,does theuseofbutterormargarineaffectthesizeofthecookie?”Theteamthendid their best job to answer the questions without using Google—by looking

answersupinthelibrary,forinstance.Onaverageittookabouttwenty-two minutes to answer a query without Google (not counting travel time to the library!)butonlysevenminutestoanswerthesamequerywithGoogle.Google savedanaverageoffifteenminutesperquery.Whenyoumultiplythattime differenceoutacrossallthequeriesthattheaverageAmericanmakesusingthe

averagehourlywageofAmericans($22),thatworksouttoabout$500peradult

workerperyear. As anyone who has been caught up in the pleasures of surfing the Web (perhaps while ‘doing research’ for a book) can attest, though, the strict distinctionbetweenworkandplayorinputandoutputthateconomistsmakeis notalwayssoclear.Thebillionsofhoursthatpeoplespenduploading,tagging, andcommentingonphotosonsocialmediasiteslikeFacebookunquestionably createsvaluefortheirfriends,family,andevenstrangers.Yetatthesametime thesehoursareuncompensated,sopresumablythepeopledoingthis‘work’find itmoreintrinsicallyrewardingthanthenextbestuseoftheirtime.Togetasense onthescaleofthiseffort,considerthatlastyearuserscollectivelyspentabout

200millionhourseachdayjustonFacebook,muchofitcreatingcontentfor

otheruserstoconsume. 13 That’stentimesasmanyperson-hoursaswereneeded tobuildtheentirePanamaCanal. 14 NoneofthisiscountedinourGDPstatistics aseitherinputoroutput,butthesekindsofzero-wageandzero-priceactivities stillcontributetowelfare.ResearcherslikeLuisvonAhnatCarnegieMellonare workingonwaysofmotivatingandorganizingmillionsofpeopletocreatevalue viacollectiveprojectsontheInternet. 15

NewGoodsandServices

Intheearlydaysofthe1990sInternetboom,venturecapitalistsusedtojokethat

therewereonlytwonumbersintheneweconomy:infinityandzero.Certainly,a

bigpartofthevalueintheneweconomyhascomefromthereductioninthe

priceofmanygoodstozero.Butwhatabouttheotherendofthatspectrum,

pricedropsfrominfinitydowntosomefinitenumber?SupposeWarnerBros.

makesanewmovieandyoucanwatchitforninedollars.Hasyourwelfare

increased?Beforethemoviewasconceived,cast,filmed,anddistributed,you

couldn’tbuyitatanyprice,eveninfinity.Inasense,payingjustninebucksisa

prettylargepricereductionfrominfinity,orwhateverthemaximumpricewas

thatyouwouldhavebeenwillingtopay.Similarly,wenowhaveaccesstoall

sortsofnewservicesthatneverexistedbefore,someofwhichwesawinearlier

chapters.Muchoftheincreaseinourwelfareoverthepastcenturycomesnot justfrommakingexistinggoodsmorecheaplybutfromexpandingtherangeof goodsandservicesavailable. Seventy-sevenpercentofsoftwarecompaniesreporttheintroductionofnew productseachyear,andInternetretailinghasvastlyexpandedthesetofgoods availabletomostconsumers. 16 Withafewclicks,overtwomillionbookscanbe foundandpurchasedatAmazon.com.Bycontrast,thetypicalphysicalbookstore

hasabout40,000titlesandeventhelargestBarnes&NoblestoreinNewYork

Citystocksonly250,000titles.AsdocumentedinaresearchpaperthatErik

wrotewithMichaelSmithandJeffreyHu,therehavebeensimilarincreasesin theonlineselectionforothercategoriessuchasvideos,music,electronics,and collectibles.Everytimeanewproductismadeavailable,itincreasesconsumer surplus. Onewaytothinkofthevaluecreatedistoimaginethatthenewproduct alwaysexisted,butonlyatsuchahighpricethatnoonecouldbuyit.Makingit availableislikeloweringthepricetoamorereasonablelevel.Therehaveeven beensubstantialincreasesinthenumberofstockkeepingunits(SKUs)inmost physicalstoresascomputerizedinventorymanagementsystems,supplychains, and manufacturing have become more efficient and flexible. For the overall economy,theofficialGDPnumbersmissthevalueofnewgoodsandservices

addedtothetuneofabout0.4percentofadditionalgrowtheachyear,according

toeconomistRobertGordon.*Rememberthatproductivitygrowthhasbeenin the neighborhood of 2 percent per year for most of the past century, so contributionofnewgoodsisnotatrivialportion.

ReputationsandRecommendations

Digitizationalsobringsarelatedbutsubtlerbenefittothevastarrayofgoods andservicesthatalreadyexistintheeconomy.Lowersearchandtransaction costsmeanfasterandeasieraccessandincreasedefficiencyandconvenience. Forexample,theratingsiteYelpcollectsmillionsofcustomerreviewstohelp dinersfindnearbyrestaurantsinthequalityandpricerangestheyseek,even whentheyarevisitingnewcities.ThereservationserviceOpenTablethenlets thembookatablewithjustafewmouseclicks. Inaggregate,digitaltoolslikethesemakealargedifference.Inthepast, ignoranceprotectedinefficientorlower-qualitysellersfrombeingunmaskedby unsuspecting consumers, while geography limited competition from other

sellers. With the introduction of structured comparison sites like FindTheBest.com and Kayak, airline travel, banking, insurance, car sales, motionpictures,andmanyotherindustriesarebeingtransformedbyconsumers’ abilitytosearchforandcomparecompetingsellers.Nolongercanasellerof substandard services expect to feed on a continuing stream of naïve or ill- informed consumers. No longer can the seller expect to be insulated from competitorsinotherlocationswhocandeliverabetterserviceforless.Research by Michael Luca of Harvard Business School has found that the increased transparencyhashelpedsmallerindependentrestaurantscompetewithbigger chainsbecausecustomerscanmorequicklyfindqualityfoodviaratingservices like Yelp, reducing their reliance on brand names’ expensive marketing campaigns. 17 Theintangiblebenefitsdeliveredbythegrowingsharingeconomy—better matches,timeliness,customerservice,andincreasedconvenience—areexactly

thetypesofbenefitsidentifiedbythe1996BoskinCommissionasbeingpoorly

measuredinourofficialpriceandGDPstatistics. 18 Thisisanotherwayinwhich ourtruegrowthisgreaterthanthestandarddatasuggest.

IntangibleAssets

Justasfreegoodsratherthanphysicalproductsareanincreasinglyimportant share of consumption, intangibles also make up a growing share of the economy’scapitalassets.Productioninthesecondmachineagedependslesson physicalequipmentandstructuresandmoreonthefourcategoriesofintangible assets:intellectualproperty,organizationalcapital,user-generatedcontent,and humancapital. Intellectualpropertyincludespatentsandcopyrights.Therateofpatentingby Americaninventorshasbeenincreasingrapidlysincethe1980s, 19 andother typesofintellectualassetshavealsogrown. 20 Inaddition,alotofresearchand development(R&D)isneverformalizedasintellectualpropertybutisstillvery valuable. The second—and even larger—category of intangibles is organizational capital like new business processes, techniques of production, organizational forms, and business models. Effective uses of the new technologies of the secondmachineagealmostinvariablyrequirechangesintheorganizationof work. For instance, when companies spend millions of dollars on computer hardware and software for a new enterprise resource planning system, they

typicallyalsoincludeprocesschangesthatarethreetofivetimesascostlyasthe

original investments in hardware and software. Yet, while the hardware and softwarespendinggenerallyshowsupasadditionstothenation’scapitalstock, thenewbusinessprocesses,whichoftenoutlastthehardware,aregenerallynot

countedascapital.Ourresearchsuggeststhatacorrectaccountingforcomputer-

relatedintangibleassetswouldaddover$2trilliontotheofficialestimatesofthe

capitalassetsintheUnitedStateseconomy. 21 User-generatedcontentisasmallerbutrapidlygrowingthirdcategoryof intangibleassets.UsersofFacebook,YouTube,Twitter,Instagram,Pinterest,and othertypesofonlinecontentnotonlyconsumethisfreecontentandgainthe consumersurplusdiscussedabovebutalsoproducemostofthecontent.There are43,200hoursofnewYouTubevideoscreatedeachday, 22 aswellas250 millionnewphotosuploadedeachdayonFacebook. 23 Usersalsocontribute valuablebutunmeasuredcontentintheformofreviewsonsiteslikeAmazon, TripAdvisor,andYelp.Inaddition,user-generatedcontentincludesthesimple binaryinformationusedtosortreviewsandpresentthebestcontentfirst(e.g., whenAmazonasks“Wasthisreviewhelpfultoyou?”).Hardwareandsoftware companiesnowcompetetoimprovetheproductivityofuser-generatedcontent activities.Forexample,smartphonesandappsforsmartphonesnowincludeeasy orautomatictoolsforpostingphotosonFacebook.Thiscontenthasvalueto otherusersandcanbethoughtofasyetanothertypeofintangiblecapitalasset thatisbeingaddedtoourcollectivewealth. Thefourthandbiggestcategoryisthevalueofhumancapital.Themany years that we all spend in schools learning skills like reading, writing, and arithmetic—aswellastheadditionallearningthathappensonthejobandonour own—makesusmoreproductiveand,insomecases,isintrinsicallyrewarding. It is also a contribution to the nation’s capital stock. According to Dale JorgensonandBarbaraFraumeni,thevalueofhumancapitalintheUnitedStates isfivetotentimeslargerthanthevalueofallthephysicalcapitalintheUnited States. 24 Humancapitalhasnotalwaysbeenthisimportanttotheeconomy.The greateconomistAdamSmithunderstoodthatoneofthegreatdrawbacksofthe

firstmachineagewasthewayitforcedworkerstodorepetitivetasks.In1776,

he noted, “The man whose whole life is spent in performing a few simple operations,ofwhichtheeffectsareperhapsalwaysthesame,orverynearlythe same,hasnooccasiontoexerthisunderstanding.” 25 Aswe’lldiscussfurther laterinthebook,investmentsinhumancapitalwillbeincreasinglyimportantas routinetasksbecomeautomatedandtheneedforhumancreativityincreases.

Importantastheseintangibleassetsare,theofficialGDPignoresthem.User-

generated content, for example, involves unmeasured labor creating an unmeasuredassetthatisconsumedinunmeasuredwaystocreateunmeasured consumersurplus.Inrecentyears,however,therehavebeensomeeffortsto createexperimental‘satelliteaccounts.’Theytracksomeofthesecategoriesof intangibleassetsintheU.S.economy.Forinstance,thenewsatelliteaccounts createdbytheBureauofEconomicAnalysisestimatethatinvestmentinR&D

capitalaccountedforabout2.9percentofGDPandhasincreasedeconomic

growthbyabout0.2percentperyearbetween1995and2004. 26 It’shardtosayexactlyhowlargethebiasisfrommiscountingallthetypesof intangibleassets,butwearereasonablyconfidenttheofficialdataunderestimate theircontribution.*

NewMetricsAreNeededfortheSecondMachineAge

It’s a fundamental principle of management: what gets measured gets done. Modern GDP accounting was certainly a huge step forward for economic progress.AsPaulSamuelsonandBillNordhausputit,“WhiletheGDPandthe restofthenationalincomeaccountsmayseemtobearcaneconcepts,theyare trulyamongthegreatinventionsofthetwentiethcentury.” 27 Buttheriseindigitalbusinessinnovationmeansweneedinnovationinour economicmetrics.Ifwearelookingatthewronggauges,wewillmakethe wrongdecisionsandgetthewrongoutputs.Ifwemeasureonlytangibles,then wewon’tcatchtheintangiblesthatwillmakeusbetteroff.Ifwedon’tmeasure pollutionandinnovation,thenwewillgettoomuchpollutionandnotenough innovation.Noteverythingthatcountscanbecounted,andnoteverythingthat canbecounted,counts. AsNobelPrizewinnerJoeStiglitzputit:

ThefactthatGDPmaybeapoormeasureofwell-being,orevenofmarketactivity,has,of course, long been recognized. But changes in society and the economy may have heightened the problems, at the same time that advances in economics and statistical

techniquesmayhaveprovidedopportunitiestoimproveourmetrics.28

Thenewmetricswilldifferbothinconceptionandexecution.Wecanbuildon someoftheexistingsurveysandtechniquesresearchershavebeenusing.For instance,thehumandevelopmentindexuseshealthandeducationstatisticstofill insomeofthegapsinofficialGDPstatistics 29 ;themultidimensionalpoverty indexusestendifferentindicators—suchasnutrition,sanitation,andaccessto

water—toassesswell-beingindevelopingcountries. 30 Childhooddeathrates andotherhealthindicatorsarerecordedinperiodichouseholdsurveyslikethe DemographicandHealthSurveys. 31 Thereareseveralpromisingprojectsinthisarea.JoeStiglitz,AmartyaSen, and Jean-Paul Fitoussi have created a detailed guide for how we can do a comprehensiveoverhaulofoureconomicstatistics. 32 Anotherpromisingproject istheSocialProgressIndexthatMichaelPorter,ScottStern,RobertoLoria,and theircolleaguesaredeveloping. 33 InBhutan,they’vebegunmeasuring“Gross National Happiness.” There is also a long-running poll behind the Gallup- HealthwaysWell-BeingIndex. 34 Theseareallimportantimprovements,andweheartilysupportthem.Butthe biggestopportunityisinusingthetoolsofthesecondmachineageitself:the extraordinaryvolume,variety, andtimelinessof dataavailabledigitally. The Internet,mobilephones,embeddedsensorsinequipment,andaplethoraofother sourcesaredeliveringdatacontinuously.Forinstance,RobertoRigobonand AlbertoCavallomeasureonlinepricesfromaroundtheworldonadailybasisto createaninflationindexthatisfartimelierand,inmanycases,morereliable, thanofficialdatagatheredviamonthlysurveyswithmuchsmallersamples. 35 Othereconomistsareusingsatellitemappingofnighttimeartificiallightsources toestimateeconomicgrowthindifferentpartsoftheworld,andassessingthe frequency of Google searches to understand changes in unemployment and housing. 36 Harnessing this information will produce a quantum leap in our understanding of the economy, just as it has already changed marketing, manufacturing,finance,retailing,andvirtuallyeveryotheraspectofbusiness decision-making. Asmoredatabecomeavailableandastheeconomycontinuestochange,the abilitytoasktherightquestionswillbecomeevenmorevital.Nomatterhow brightthelightis,youwon’tfindyourkeysbysearchingunderalamppostif that’snotwhereyoulostthem.Wemustthinkhardaboutwhatitiswereally value,whatwewantmoreof,andwhatwewantlessof.GDPandproductivity growth are important, but they are means to an end, not ends in and of themselves.Dowewanttoincreaseconsumersurplus?Thenlowerpricesor moreleisuremightbesignsofprogress,eveniftheyresultinalowerGDP.And, of course, many of our goals are nonmonetary. We shouldn’t ignore the economicmetrics,butneithershouldweletthemcrowdoutourothervalues simplybecausetheyaremoremeasurable. Inthemeantime,weneedtobearinmindthattheGDPandproductivity

statisticsoverlookmuchofwhatwevalue,evenwhenusinganarroweconomic

lens.What’smore,thegapbetweenwhatwemeasureandwhatwevaluegrows

everytimewegainaccesstoanewgoodorservicethatneverexistedbefore,or

whenexistinggoodsbecomefreeastheysooftendowhentheyaredigitized.

*

*Therehavebeenanumberofrelatedfindingssincethen.Lastyear,theeconomistsJeremy Greenwood and Karen Kopecky applied a similar approach and found a similar growth contributionforpersonalcomputersalone.ShaneGreensteinandRyanMcDevitt,anotherpairof economists,askedhowmuchconsumersurpluswascreatedbythespreadofbroadbandInternet access.Theylookedathowtherealpriceofbroadbandhaddeclinedovertimeandhowadoption oftheservicehadincreased.Fromthat,theyestimatehowmuchpeoplewouldhavebeenwilling topaycomparedtowhattheyactuallypaid,andthusarriveattheconsumersurplus.Aresearch team at McKinsey took a more direct approach. The team asked 3,360 consumers what they wouldhavebeenwillingtopayforsixteenspecificservicesavailableviatheInternet.Theaverage willingnesstopayaddeduptofiftydollarspermonth.Basedonthis,theteamestimatedthat

Americansreceivedover$35billionworthofconsumersurplusfromthefreeInternet.Thebiggest

singlecategorywase-mail,withsocialnetworkslikeFacebookclosebehind.

* Yes, our long-time friend, the same Robert J. Gordon we discussed in chapter 6. See

http://faculty-web.at.northwestern.edu/economics/gordon/p376_ipm_final_060313.pdf.

*Unlikeunmeasuredintangibleconsumptiongoods,thebadmeasuresofintangiblecapitalgoods don’tautomaticallybiasofficialproductivitystatistics.Ononehand,likeallintangibles,intangible capital goods make the output numbers bigger. But at the same time, they are also used for productionandthusmaketheinputnumbersbigger.Inasteadystatewhereboththeinputand outputnumbersaregrowingatthesamerate,thesetwoeffectscancelout,sothereisnobiasin theproductivitynumbers,definedasoutput/input.Steadygrowthhasbeenroughlytrueforsome typesofintangibles,suchasthehumancapitalassetsthatarecreatedbyeducation.Butother categories—likecomputer-relatedorganizationalcapitalortheuser-generatedcapitalondigital contentsites—appearstohavebeengrowingrapidly.Forthesecategoriesofintangibleassets, theofficialproductivitynumbersunderstatethetruegrowthoftheeconomy.

“Animbalancebetweenrichandpooristheoldestandmostfatalailmentofall republics.” —Plutarch

“Animbalancebetweenrichandpooristheoldestandmostfatalailmentofall

republics.”

—Plutarch

OFTHE 3.5 TRILLION photosthathavebeensnappedsincethefirstimageofa busyParisianstreetin1838,fully10percentweretakeninthelastyear. 1 Until recently, most photos were analog, created using silver halide and other chemicals.Butanalogphotographypeakedin2000. 2 Today,over2.5billion peoplehavedigitalcamerasandthevastmajorityofphotosaredigital. 3 The effectsareastonishing:ithasbeenestimatedthatmorephotosarenowtaken everytwominutesthaninallofthenineteenthcentury. 4 Wenowrecordthe peopleandeventsofourliveswithunprecedenteddetailandfrequency,and sharethemmorewidelyandeasilythaneverbefore. Whiledigitizationhasobviouslyincreasedthequantityandconvenienceof photography, it has also profoundly changed the economics of photography productionanddistribution.AteamofjustfifteenpeopleatInstagramcreateda

simpleappthatover130millioncustomersusetosharesomesixteenbillion

photos(andcounting). 5 Withinfifteenmonthsofitsfounding,thecompanywas

soldforover$1billiontoFacebook.Inturn,Facebookitselfreachedonebillion

usersin2012.Ithadabout4,600employees 6 includingbarely1,000engineers. 7 Contrastthesefigureswithpre-digitalbehemothKodak,whichalsohelped customers share billions of photos. Kodak employed 145,300 people at one point,one-thirdoftheminRochester,NewYork,whileindirectlyemploying thousandsmoreviatheextensivesupplychainandretaildistributionchannels requiredbycompaniesinthefirstmachineage.Kodakmadeitsfounder,George Eastman,arichman,butitalsoprovidedmiddle-classjobsforgenerationsof people and created a substantial share of the wealth created in the city of

Rochesteraftercompany’sfoundingin1880.But132yearslater,afewmonths

before Instagram was sold to Facebook, Kodak filed for bankruptcy. 8 Photographyhasneverbeenmorepopular.Today,seventybillionphotosare uploadedtoFacebookeachyear,andmanytimesmorearesharedviaother digitalserviceslikeFlickratnearlyzerocost.Thesephotosarealldigital,so hundreds of thousands of people who used to work making photography chemicalsandpaperarenolongerneeded.Inadigitalage,theyneedtofind someotherwaytosupportthemselves. Theevolutionofphotographyillustratesthebountyofthesecondmachine age, the first great economic consequence of the exponential, digital, combinatorialprogresstakingplaceatpresent.Thesecondone,spread,means

therearelargeandgrowingdifferencesamongpeopleinincome,wealth,and otherimportantcircumstancesoflife.We’vecreatedacornucopiaofimages, sharingnearlyfourhundredbillion“Kodakmoments”eachyearwithafew clicks of a mouse or taps on a screen. But companies like Instagram and Facebook employ a tiny fraction of the people that were needed at Kodak. Nonetheless,FacebookhasamarketvalueseveraltimesgreaterthanKodakever didandhascreatedatleastsevenbillionairessofar,eachofwhomhasanet worth ten times greater than George Eastman did. The shift from analog to digitalhasdeliveredabountyofdigitalphotosandothergoods,butithasalso contributedtoanincomedistributionthatisfarmorespreadoutthanbefore. Photographyisnotanisolatedexampleofthisshift.Similarstorieshavebeen andwillbetoldinmusicandmedia;infinanceandpublishing;inretailing, distribution,services,andmanufacturing.Inalmosteveryindustry,technological progresswillbringunprecedentedbounty.Morewealthwillbecreatedwithless work.Butatleastinourcurrenteconomicsystem,thisprogresswillalsohave enormouseffectsonthedistributionincomeandwealth.Iftheworkaperson producesinonehourcaninsteadbeproducedbyamachineforonedollar,thena profit-maximizingemployerwon’tofferawageforthatjobofmorethanone dollar.Inafree-marketsystem,eitherthatworkermustacceptawageofone dollaranhourorfindsomenewwaytomakealiving.Conversely,ifaperson findsanewwaytoleverageinsights,talents,orskillsacrossonemillionnew customersusingdigitaltechnologies,thenheorshemightearnonemilliontimes asmuchaswouldbepossibleotherwise.Boththeoryanddatasuggestthatthis combinationofbountyandspreadisnotacoincidence.Advancesintechnology, especially digital technologies, are driving an unprecedented reallocation of wealthandincome.Digitaltechnologiescanreplicatevaluableideas,insights, andinnovationsatverylowcost.Thiscreatesbountyforsocietyandwealthfor innovators,butdiminishesthedemandforpreviouslyimportanttypesoflabor, whichcanleavemanypeoplewithreducedincomes. The combination of bounty and spread challenges two common though contradictoryworldviews.Onecommonviewisthatadvancesintechnology alwaysboostincomes.Theotheristhatautomationhurtsworkers’wagesas peoplearereplacedbymachines.Bothofthesehaveakerneloftruth,butthe reality is more subtle. Rapid advances in our digital tools are creating unprecedentedwealth,butthereisnoeconomiclawthatsaysallworkers,or evenamajorityofworkers,willbenefitfromtheseadvances. Foralmosttwohundredyears,wagesdidincreasealongsideproductivity.

Thiscreatedasenseofinevitabilitythattechnologyhelped(almost)everyone. But more recently, median wages have stopped tracking productivity, underscoringthefactthatsuchadecouplingisnotonlyatheoreticalpossibility butalsoanempiricalfactinourcurrenteconomy.

How’stheMedianWorkerDoing?

Let’sreviewsomebasicfacts. Agoodplacetostartismedianincome—theincomeofthepersonatthe

fiftiethpercentileofthetotaldistribution.Theyear1999wasthepeakyearfor

the real (inflation-adjusted) income of the median American household. It

reached$54,932thatyear,butthenstartedfalling.By2011,ithadfallennearly

10percentto$50,054,evenasoverallGDPhitarecordhigh.Inparticular,

wagesofunskilledworkersintheUnitedStatesandotheradvancedcountries havetrendeddownward. Meanwhile,forthefirsttimesincebeforetheGreatDepression,overhalfthe

totalincomeintheUnitedStateswenttothetop10percentofAmericansin

2012.Thetop1percentearnedover22percentofincome,morethandoubling

theirsharesincetheearly1980s.Theshareofincomegoingtothetophundredth

ofonepercentofAmericans,afewthousandpeoplewithannualincomesover

$11million,isnowat5.5percent,afterincreasingmorebetween2011and2012

thananyyearsince1927–28. 9 Several other metrics have also been increasingly unequal. For instance, whileoveralllifeexpectancycontinuestorise,lifeexpectanciesforsomegroups havestartedtofall.AccordingtoastudybyS.JayOlshanskyandhiscolleagues publishedinHealthAffairs,theaverageAmericanwhitewomanwithoutahigh

schooldiplomahadalifeexpectancyof73.5yearsin2008,comparedto78.5

yearsin1990.Lifeexpectancyforwhitemenwithoutahighschooleducation

fellbythreeyearsduringthisperiod. 10 It’s no wonder that protests broke out across America even as it was beginningtorecoverfromtheGreatRecession.TheTeaPartymovementonthe rightandtheOccupymovementonthelefteachchanneledtheangerofthe millionsofAmericanswhofelttheeconomywasnotworkingforthem.One group emphasized government mismanagement and the other abuses in the financialservicessector.

HowTechnologyIsChangingEconomics

Whileundoubtedlybothoftheseproblemsareimportant,themorefundamental challengeisdeepandstructural,andistheresultofthediffusiontothesecond machineagetechnologiesthatincreasinglydrivetheeconomy. Recentlyweoverheardabusinessmanspeakingloudly(andcheerfully)into hismobilephone:“Noway.Idon’tuseanH&RBlocktaxprepareranymore. I’veswitchedtoTurboTaxsoftware.It’sonlyforty-ninedollars,andit’smuch quickerandmoreaccurate.Iloveit!”Thebusinessmanwasbetteroff.Hehada betterserviceatalowerprice.Multipliedbymillionsofcustomers,TurboTax hascreatedagreatdealofvalueforitsusers,notallofwhichevenshowsupin the GDP statistics. The creators of TurboTax are also better off—one is a billionaire. But tens of thousands of tax preparers now find their jobs and incomesthreatened. Thebusinessman’sexperienceholdsamirrortothebroaderchangesinthe economy. Consumers are better off and enormous wealth is created, but a relativelysmallgroupofpeopleoftenearnsmostoftheincomefromthenew productsorservices.Likethechemistswhousedsilverhalidetocreatecamera film in the 1990s, human tax preparers have a hard time competing with machines.Theycan bemadeworse offbyadvances intechnology,notjust relativetothewinners,butalsorelativetotheirincomewhentheywereworking withtheoldertechnologies. Thecrucialrealityfromthestandpointofeconomicsisthatittakesonlya relatively small number of designers and engineers to create and update a program like TurboTax. As we saw in chapter 4, once the algorithms are digitizedtheycanbereplicatedanddeliveredtomillionsofusersatalmostzero cost.Assoftwaremovestothecoreofeveryindustry,thistypeofproduction processandthistypeofcompanyincreasinglypopulatestheeconomy.

ASmallerSliceofaBiggerPie

Whathappenswhenyouscaleupthesetypesofexamplestoawholeeconomy? Istheresomethingbiggergoingon?Thedatasayyes.

Between1983and2009,Americansbecamevastlywealthieroverallasthe

totalvalueoftheirassetsincreased.However,asnotedbyeconomistsEdWolff

andSylviaAllegretto,thebottom80percentoftheincomedistributionactually

sawanetdecreaseintheirwealth. 11 Takenasagroup,thetop20percentgotnot

100percentoftheincrease,butmorethan100percent.Theirgainsincludednot

onlythetrillionsofdollarsofwealthnewlycreatedintheeconomybutalso

someadditionalwealththatwasshiftedintheirdirectionfromthebottom80

percent.Thedistributionwasalsohighlyskewedevenamongrelativelywealthy

people.Thetop5percentgot80percentofthenation’swealthincrease;thetop

1percentgotoverhalfofthat,andsoonforever-finersubdivisionsofthe

wealth distribution. In an oft-cited example, by 2010 the six heirs of Sam Walton’sfortune,earnedwhenhecreatedWalmart,hadmorenetwealththanthe bottom40percentoftheincomedistributioninAmerica. 12 Inpart,thisreflects thefactthatthirteenmillionfamilieshadanegativenetworth.

Alongwithwealth,theincomedistributionhasalsoshifted.Thetop1percent

increasedtheirearningsby278percentbetween1979and2007,comparedtoan

increaseofjust35percentforthoseinthemiddleoftheincomedistribution.The

top1percentearnedover65percentofincomeintheUnitedStatesbetween

2002and2007.AccordingtoForbes,thecollectivenetworthofthewealthiest

fourhundredAmericansreachedarecordtwotrilliondollarsin2013,morethan

doublingsince2003. 13

INSHORT,medianincomehasincreasedverylittlesince1979,andithasactually

fallen since 1999. But that’s not because growth of overall income or productivity in America has stagnated; as we saw in chapter 7, GDP and productivityhavebeenonimpressivetrajectories.Instead,thetrendreflectsa significantreallocationofwhoiscapturingthebenefitsofthisgrowth,andwho isn’t. Thisisperhapseasiesttoseeifonecomparesaverageincomewithmedian income.Normally,changesintheaverageincome(totalincomedividedbythe totalnumberofpeople)arenotverydifferentfromchangesinmedianincome (incomeofthepersonexactlyinthemiddleoftheincomedistribution—half earnmoreandhalfearnless).However,inrecentyears,thetrendshavediverged

significantly,asshowninfigure9.1.

How is this possible? Consider a simple example. Ten bank tellers are

drinkingbeersatabar.Eachofthemmakes$30,000ayear,soboththemean

andmedianincomeofthisgroupis$30,000.InwalkstheCEOandordersa

beer.Nowtheaverageincomeofthegrouphasskyrocketed,butthemedian

hasn’tchangedatall.Ingeneral,themoreskewedtheincomes,themorethe

meantendstodivergefromthemedian.Thisiswhathashappenednotonlyin

ourhypotheticalbarbutalsoinAmericaasawhole.

Overall,between1973and2011,themedianhourlywagebarelychanged,

growingbyjust0.1percentperyear.Incontrast,asdiscussedinchapter7,

productivitygrewatanaverageof1.56percentperyearduringthisperiod,

acceleratingabitto1.88percentperyearfrom2000to2011.Mostofthegrowth

inproductivitydirectlytranslatedintocomparablegrowthinaverageincome. The reason why median income growth was so much lower was primarily becauseofincreasesininequality. 14

FIGURE9.1RealGDPvs.MedianIncomeperCapita

14 FIGURE9.1 RealGDPvs.MedianIncomeperCapita TheThreePairsofWinnersandLosers

TheThreePairsofWinnersandLosers

Inthepastcoupleofdecades,we’veseenchangesintaxpolicy,greateroverseas competition,ongoinggovernmentwaste,andWallStreetshenanigans.Butwhen welookatthedataandresearch,weconcludethatnoneofthesearetheprimary driverofgrowinginequality.Instead,themaindriverisexponential,digital,and combinatorialchangeinthetechnologythatundergirdsoureconomicsystem. Thisconclusionisbolsteredbythefactthatsimilartrendsareapparentinmost advancedcountries.Forinstance,inSweden,Finland,andGermany,income inequalityhasactuallygrownmorequicklyoverthepasttwentytothirtyyears thanintheUnitedStates. 15 Becausethesecountriesstartedwithlessinequality intheirincomedistributions,theycontinuedtobelessunequalthantheUnited States,buttheunderlyingtrendissimilarworldwideacrosssometimesmarkedly differentinstitutions,governmentpolicies,andcultures. As we discussed in our earlier book Race Against the Machine, these

structuraleconomicchangeshavecreatedthreeoverlappingpairsofwinnersand

losers.Asaresult,noteveryone’sshareoftheeconomicpieisgrowing.Thefirst

twosetsofwinnersarethosewhohaveaccumulatedsignificantquantitiesofthe

rightcapitalassets.Thesecanbeeithernonhumancapital(suchasequipment,

structures,intellectualproperty,orfinancialassets),orhumancapital(suchas

training,education,experience,andskills).Likeotherformsofcapital,human

capitalisanassetthatcangenerateastreamofincome.Awell-trainedplumber

canearnmoreeachyearthananunskilledworker,eveniftheybothworkthe

samenumberofhours.Thethirdgroupofwinnersismadeupofthesuperstars

amonguswhohavespecialtalents—orluck.

Ineachgroup,digitaltechnologiestendtoincreasetheeconomicpayoffto

winnerswhileothersbecomelessessential,andhencelesswellrewarded.The

overallgainstothewinnershavebeenlargerthantotallossesforeveryoneelse.

Thatsimplyreflectsthefactwediscussedearlier:productivityandtotalincome

havegrownintheoveralleconomy.Thisgoodnewsofferslittleconsolationto

thosewhoarefallingbehind.Insomecasesthegains,howeverlarge,havebeen

concentratedamongarelativelysmallgroupofwinners,leavingthemajorityof

peopleworseoffthanbefore.

Skill-BiasedTechnicalChange

Themostbasicmodeleconomistsusetoexplaintechnology’simpacttreatsitas asimplemultiplieroneverythingelse,increasingoverallproductivityevenlyfor everyone. 16 Thismodelcanbedescribedinmathematicalequations.Itisusedin most introductory economics classes and provides the foundation for the common—and until recently, very sensible—intuition that a rising tide of technical progress will lift all boats, that it will make all workers more productive and hence more valuable. With technology as a multiplier, an economy is able to produce more output each year with the same inputs, including labor. And in the basic model all labor is affected equally by technology,meaningeveryhourworkedproducesmorevaluethanitusedto. Aslightlymorecomplexmodelallowsforthepossibilitythattechnologymay notaffectallinputsequally,butrathermaybe‘biased’towardsomeandagainst others. In particular, in recent years, technologies like payroll processing software, factory automation, computer-controlled machines, automated inventorycontrol,andwordprocessinghavebeendeployedforroutinework, substitutingforworkersinclericaltasks,onthefactoryfloor,anddoingrote

informationprocessing. By contrast, technologies like big data and analytics, high-speed communications,andrapidprototypinghaveaugmentedthecontributionsmade bymoreabstractanddata-drivenreasoning,andinturnhaveincreasedthevalue ofpeoplewiththerightengineering,creative,ordesignskills.Theneteffecthas beentodecreasedemandforlessskilledlaborwhileincreasingthedemandfor skilled labor. Economists including David Autor, Lawrence Katz and Alan Krueger,FrankLevyandRichardMurnane,DaronAcemoglu,andmanyothers havedocumentedthistrendindozensofcarefulstudies. 17 Theycallitskill- biased technical change. By definition, skill-biased technical change favors peoplewithmorehumancapital.

FIGURE9.2WagesforFull-Time,Full-YearMaleU.S.Workers,1963–2008

WagesforFull-Time,Full-YearMaleU.S.Workers,1963–2008

Theeffectsofskill-biasedtechnicalchangecanbevividlyseeninfigure9.2,

whichisbasedondatafromapaperbyMITeconomistsDaronAcemogluand DavidAutor. 18 Thelinestellastoryaboutthedivergingpathsofmillionsof

workersoverrecentgenerations.Before1973,Americanworkersallenjoyed

briskwagegrowth.Therisingtideofproductivityincreasedeveryone’sincomes, regardless of their educational levels. Then came the massive oil shock and

recessionofthe1970s,whichreversedthegainsforallgroups.However,after

that,webegantoseeagrowingspreadofincomes.Bytheearly1980s,those

withcollegedegreesstartedtoseetheirwagesgrowingagain.Workerswith graduate degrees did particularly well. Meanwhile, workers without college degreeswereconfrontedwithamuchlessattractivelabormarket.Theirwages stagnated or, if they were high school dropouts, actually fell. It’s not a

coincidencethatthepersonalcomputerrevolutionstartedintheearly1980s;the

PCwasactuallyTimemagazine’s“machineoftheyear”in1982.

Theeconomicsofthestorybecomeevenmorestrikingwhenoneconsiders thatthenumberofcollegegraduatesgrewveryrapidlyduringthisperiod.The number of people enrolled in college more than doubled between 1960 and 1980,from758,000to1,589,000. 19 Inotherwords,therewasalargeincreasein thesupplyofeducatedlabor.Normally,greatersupplyleadstolowerprices.In thiscase,thefloodofgraduatesfromcollegeandgraduateschoolshouldhave pusheddowntheirrelativewages,butitdidn’t. Thecombinationofhigherpaydespitegrowingsupplycanonlymeanthatthe relativedemandforskilledlaborincreasedevenfasterthansupply.Andatthe same time, the demand for tasks that could be completed by high school dropoutsfellsorapidlythattherewasaglutofthistypeofworker,eventhough

theirrankswerethinning.Thelackofdemandforunskilledworkersmeantever-

lowerwagesforthosewhocontinuedtocompeteforlow-skilljobs.Andbecause

mostofthepeoplewiththeleasteducationalreadyhadthelowestwages,this

changeincreasedoverallincomeinequality.

OrganizationalCoinvention

Whileaone-for-onesubstitutionofmachinesforpeoplesometimesoccurs,a broader reorganization in business culture may have been an even more importantpathforskill-biasedchange.WorkthatErikdidwithStanford’sTim Bresnahan, Wharton’s Lorin Hitt, and MIT’s Shinkyu Yang found that companies used digital technologies to reorganize decision-making authority, incentivessystems,informationflows,hiringsystems,andotheraspectsoftheir managementandorganizationalprocesses. 20 Thiscoinventionoforganization and technology not only significantly increased productivity but tended to requiremoreeducatedworkersandreducedemandforless-skilledworkers.This reorganizationofproductionaffectedthosewhoworkeddirectlywithcomputers aswellasworkerswho,atfirstglance,seemedtobefarfromthetechnology. Forinstance,adesignerwithaknackforstylemightfindherselfingreater

demand at a company with flexible equipment in distant factories that can quicklyadapttothelatestfashions,whileanairportticketagentmightfind himselfreplacedbyanInternetwebsiteheneverknewexisted,letaloneworked with. Amongtheindustriesinthestudy,eachdollarofcomputercapitalwasoften the catalyst for more than ten dollars of complementary investments in “organizationalcapital,”orinvestmentsintraining,hiring,andbusinessprocess redesign. 21 Thereorganizationofteneliminatesalotofroutinework,suchas repetitiveorderentry,leavingbehindaresidualsetoftasksthatrequirerelatively morejudgment,skills,andtraining. Companies with the biggest IT investments typically made the biggest organizationalchanges,usuallywithalagoffivetosevenyearsbeforeseeing thefullperformancebenefits. 22 Thesecompanieshadthebiggestincreaseinthe demandforskilledworkrelativetounskilledwork. 23 Thelagsreflectedthetime that it takes for managers and workers to figure out new ways to use the technology.Aswesawinourearlierdiscussionofelectrificationandfactory design,businessesrarelygetsignificantperformancegainsfromsimply“paving thecowpaths”asopposedtorethinkinghowthebusinesscanberedesignedto takeadvantageofnewtechnologies. 24 Creativityandorganizationalredesignare crucialtoinvestmentsindigitaltechnologies.* Thismeansthatthebestwaytousenewtechnologiesisusuallynottomakea literalsubstitutionofamachineforeachhumanworker,buttorestructurethe process. Nonetheless, some workers (usually the less skilled ones) are still eliminatedfromtheproductionprocessandothersareaugmented(usuallythose with more education and training), with predictable effects on the wage structure. Compared to simply automating existing tasks, this kind of organizationalcoinventionrequiresmorecreativityonthepartofentrepreneurs, managers,andworkers,andforthatreasonittendstotaketimetoimplementthe changesaftertheinitialinventionandintroductionofnewtechnologies.But oncethechangesareinplace,theygeneratethelion’sshareofproductivity improvements.

TheSkillSetAffectedbyComputerizationIsEvolving

Ifwelookmorecloselyatthejobseliminatedascompaniesreorganized,skill-

biasedtechnicalchangecanbeabitofamisleadingmoniker.Inparticular,it

wouldbeamistaketoassumethatall‘college-leveltasks’arehardtoautomate

while ‘kindergarten tasks’are easy. In recent years, ‘low-skill tasks’haven’t always been the ones being automated; more often it has been ‘tasks that machinescandobetterthanhumans.’Ofcourse,that’sabitofatautology,buta usefultautologynonetheless.Repetitiveworkonanassemblylineiseasierto automate than the work of a janitor. Routine clerical work like processing paymentsiseasiertoautomatethanhandlingcustomers’questions.Atpresent, machinesarenotverygoodatwalkingupstairs,pickingupapaperclipfromthe floor,orreadingtheemotionalcuesofafrustratedcustomer. Tocapturethesedistinctions,workbyourMITcolleaguesDaronAcemoglu andDavidAutorsuggeststhatworkcanbedividedintoatwo-by-twomatrix:

cognitiveversusmanualandroutineversusnonroutine. 25 Theyfoundthatthe demandforworkhasbeenfallingmostdramaticallyforroutinetasks,regardless of whether they are cognitive or manual. This leads to job polarization: a collapseindemand formiddle-incomejobs, whilenonroutinecognitive jobs (suchasfinancialanalysis)andnonroutinemanualjobs(likehairdressing)have helduprelativelywell. BuildingonAcemogluandAutor’swork,economistsNirJaimovichofDuke UniversityandHenrySiuoftheUniversityofBritishColumbiafoundalink betweenjobpolarizationandthejoblessrecoveriesthathavedefinedthelast threerecessions.Formostofthenineteenthandtwentiethcenturies,employment

usuallyreboundedstronglyaftereachrecession,butsincethe1990semployment

didn’trecoverbrisklyafterrecessions.Again,it’snotacoincidencethatasthe computerization of the economy advanced, post-recession hiring patterns

changed.WhenJaimovichandSiucomparedthe1980s,1990s,and2000s,they

foundthatthedemandforroutinecognitivetaskssuchascashiers,mailclerks, andbanktellersandroutinemanualtaskssuchasmachineoperators,cement masons,anddressmakerswasnotonlyfalling,butfallingatanacceleratingrate.

Thesejobsfellby5.6percentbetween1981and1991,6.6percentbetween1991

and2001,and11percentbetween2001and2011. 26 Incontrast,bothnonroutine cognitiveworkandnonroutinemanualworkgrewinallthreedecades. CONVERSATIONSWITH seniorexecutiveshelpexplainthispatterninthedata.A fewyearsago,wehadaverycandiddiscussionwithoneCEO,andheexplained thatheknewforoveradecadethatadvancesininformationtechnologyhad rendered many routine information-processing jobs superfluous. At the same time,whenprofitsandrevenuesareontherise,itcanbehardtoeliminatejobs. When the recession came, business as usual obviously was not sustainable, whichmadeiteasiertoimplementaroundofpainfulstreamliningandlayoffs.

Astherecessionendedandprofitsanddemandreturned,thejobsdoingroutine work were not restored. Like so many other companies in recent years, his organizationfounditcouldusetechnologytoscaleupwithouttheseworkers.

Aswesawinchapter2,thisreflectsMoravac’sparadox,theinsightthatthe

sensory and motor skills we use in our everyday lives require enormous computationandsophistication. 27 Overmillionsofyears,evolutionhasendowed uswithbillionsofneuronsdevotedtothesubtletiesofrecognizingafriend’s face,distinguishingdifferenttypesofsounds,andusingfinemotorcontrol.In contrast, the abstract reasoning that we associate with ‘higher thought’like arithmetic or logic is a relatively recent skill, developed over only a few thousandyears.Itoftenrequiressimplersoftwareandlesscomputerpowerto mimicorevenexceedhumancapabilitiesonthesetypesoftasks. Ofcourse,aswe’veseenthroughoutthisbook,thesetoftasksmachinescan doisnotfixed.Itisconstantlyevolving,justasouruseoftheword“computer” itselfhasevolvedfromreferringtoajobthathumansdotoreferringtoapieceof equipment.

Intheearly1950s,machinesweretaughthowtoplaycheckersandcould

soonbeatrespectableamateurs. 28 InJanuary1956,HerbertSimonreturnedto teaching his class and told his students, “Over Christmas, Al Newell and I invented a thinking machine.” Three years later, they created a computer programmodestlycalledthe“GeneralProblemSolver,”whichwasdesignedto solve,inprinciple,anylogicproblemthatcouldbedescribedbyasetofformal rules.ItworkedwellonsimpleproblemslikeTic-Tac-Toeortheslightlyharder TowerofHanoipuzzle,althoughitdidn’tscaleuptomostreal-worldproblems becauseofthecombinatorialexplosionofpossibleoptionstoconsider. Cheered by their early successes and those of other artificial intelligence pioneerslikeMarvinMinsky,JohnMcCarthyandClaudeShannon,andSimon andNewellwerequiteoptimisticabouthowrapidlymachineswouldmaster

humanskills,predictingin1958thatadigitalcomputerwouldbetheworld

chesschampionby1968. 29 In1965,Simonwentsofarastopredict,“machines willbecapable,withintwentyyears,ofdoinganyworkamancando.” 30

SimonwontheNobelPrizeinEconomicsin1978,buthewaswrongabout

chess,nottomentionalltheothertasksthathumanscando.Hismistakemay havebeenmoreaboutthetimingthantheultimateoutcome.AfterSimonmade hisprediction,computerchessprogramsimprovedbyaboutfortypointsperyear on the official Elo chess rating system. On May 11, 1997, forty years after Simon’sprediction,anIBMcomputercalledDeepBluebeattheworldchess

champion,GaryKasparov,afterasix-gamematch.Today,nohumancanbeat evenamid-tiercomputerchessprogram.Infact,softwareandhardwarehave

progressedsorapidlythatby2009,chessprogramsrunningonordinarypersonal

computers,andevenmobilephones,haveachievedgrandmasterlevelswithElo ratingsof2,898andhavewontournamentsagainstthetophumanplayers. 31

LaborandCapital

Technologyisnotonlycreatingwinnersandlosersamongthosewithdiffering amountsofhumancapital,itisalsochangingthewaynationalincomeisdivided betweentheownersofphysicalcapitalandlabor(peoplelikefactoryownersand factoryworkers)—thetwoclassicalinputstoproduction. WhenTerryGou,thefounderofFoxconn,purchasedthirtythousandrobotsto workinthecompany’sfactoriesinChina,hewassubstitutingcapitalforlabor. 32 Similarly, when an automated voice-response system usurps some of the functionsofhumancallcenteroperators,theproductionprocesshasmorecapital andlesslabor.Entrepreneursandmanagersareconstantlymakingthesetypesof decisions,weighingtherelativecostsofeachtypeofinput,aswellastheeffects onthequality,reliability,andvarietyofoutputthatcanbeproduced.

RodBrooksestimatesthattheBaxterrobotwemetinchapter2worksforthe

equivalentofaboutfourdollarsperhour,includingallcosts. 33 Aswediscussed atthestartofthischapter,totheextentthatafactoryownerpreviouslyemployed ahumantodothesametaskthatBaxtercoulddo,theeconomicincentivewould betosubstitutecapital(Baxter)forlaboraslongasthehumanwaspaidmore thanfourdollarsperhour.Ifoutputstaysthesame,andassumingnonewhires aremadeinengineering,management,orsalesatthecompany,itwouldincrease theratioofcapitaltolaborinput.* Compensationoftheremainingworkerscouldgoupordowninthewakeof Baxter’sarrival.Iftheirworkisaclosesubstitutefortherobots’,thentherewill bedownwardpressureonhumanwages.ThatwillgrowevenworseifMoore’s LawandotheradvancesallowfutureversionsofBaxtertoworkfortwodollars perhour,andthenonedollarperhour,andsoon,whilehandlinganincreasing varietyandcomplexityoftasks.However,economictheoryalsoholdsopenthe possibilitythattheremainingworkerswouldseeanincreaseinpay.Inparticular, iftheirworkcomplementsthetechnology,thendemandfortheirserviceswill increase. In addition, as technical advances increase labor productivity, employers can afford to pay more for each worker. In some cases, this is

reflecteddirectlyinhigherwagesandbenefits.Inothercases,thepricesof productsandservicesfall,sotherealwageofworkersincreasesastheyareable tobuymorewitheachdollar.Asproductivityimproves,totalamountofoutput perpersonwouldincreasebuttheamountearnedbyhumanworkerscouldeither fallorrise,withtheremaindergoingtocapitalowners. Ofcourse,almosteveryeconomyhasbeenusingtechnologytosubstitute capitalforlaborfordecades,ifnotcenturies.Automaticthreshingmachines

replacedafull30percentoftheagriculturallaborforceinthemiddleofthe

nineteenthcentury,andindustrializationcontinuedatabriskpacethroughoutthe twentieth century. Nineteenth-century economists like Karl Marx and David Ricardopredictedthatthemechanizationoftheeconomywouldworsenthefate ofworkers,ultimatelydrivingthemtoasubsistencewage. 34 What has actually happened to the relative share of capital and labor? Historically,despitechangesinthetechnologyofproduction,theshareofoverall GDPgoingtolaborhasbeensurprisinglystable,atleastuntilrecently.Asa result,wagesandlivingstandardshavegrowndramatically,roughlyinlinewith the dramatic increases in productivity. In part, this reflects the increases in humancapitalthathaveparalleledthemorevisibleincreasesinequipmentand buildingsintheeconomy.DaleJorgensonandhiscolleagueshaveestimatedthat theoverallmagnitudeofthehumancapitalintheU.S.economy,asmeasuredby itseconomicvalue,isasmuchastentimesthevalueofthephysicalcapital. 35 As a result, labor compensation has grown along with payments to owners of physicalcapitalviaprofits,dividends,andcapitalgains.

Figure9.3showsthatinthepastdecade,therelativelyconsistentdivision

betweenthesharesofincomegoingtolaborandphysicalcapitalseemstobe comingtoanend.AsnotedbySusanFleck,JohnGlaser,andShawnSpraguein

theMonthlyLaborReview:“Laborshareaveraged64.3percentfrom1947to

2000.IntheUnitedStates,theshareofGDPgoingtolaborhasdeclinedoverthe

past decade, falling to its lowest point in the third quarter of 2010, 57.8 percent.” 36 What’s more, this is a global phenomenon. Economists Loukas KarabarbounisandBrentNeimanoftheUniversityofChicagofindthat“the global labor share has significantly declined since the early 1980s, with the declineoccurringwithinthelargemajorityofcountriesandindustries.” 37 They arguethatthisdeclineislikelyduetothetechnologiesoftheinformationage.

FIGURE9.3WageShareofGDPvs.CorporateProfitShareofGDP

Thefallinlabor’sshareisinparttheconsequenceoftwotrendswehave

Thefallinlabor’sshareisinparttheconsequenceoftwotrendswehave alreadynoted:fewerpeopleareworking,andwagesforthosewhoareworking arelowerthanbefore.Asaresult,whilelaborcompensationandproductivityin thepastroseintandem,inrecentyearsagrowinggaphasopened. Ifproductivityisgrowingandlaborasawholeisn’tcapturingthevalue,who is?Ownersofphysicalcapital,toalargeextent.Whiletheeconomyremained miredinaslump,profitsreachedhistorichighslastyear,bothinabsoluteterms

($1.6trillion)andasashareofGDP(26.2percentin2010,upfromthe1960–

2007averageof20.5percent). 38 Meanwhile,realspendingoncapitalequipment

andsoftwarehassoaredby26percentwhilepayrollshaveremainedessentially

flat,asnotedbyKathleenMadigan. 39 What’s more, the collapse in the share of GDP going to labor actually understateshowthesituationhasdeterioratedforthetypicalworker.Theofficial measureoflaborcompensationincludessoaringwagesforasmallnumberof superstarsinmedia,finance,sports,andcorporatepositions.Furthermore,itis debatablethatallofthecompensationgoingtoCEOsandothertopexecutivesis solelyduetotheir‘labor’income.Itmayalsoreflecttheirbargainingpower,as suggestedbyHarvardLawProfessorLucianBebchukandothers. 40 Inthissense, itmightmakesensetothinkofCEOs’incomeasduetotheircontrolofcapital, notlabor,atleastinpart. Whiletheshareofnationalincometocapitalhasbeengrowingattheexpense oflabor,economictheorydoesnotnecessarilypredictthatthiswillcontinue, evenifrobotsandothermachinestakeovermoreandmorework.Thethreatto

capital’ssharecomesnot(just)fromthebargainingpowerofvarioustypesof humanlabor,fromCEOsorlaborunionsbut,ironically,fromothercapital.Ina free market, the biggest premiums go to the scarcest inputs needed for production.Inaworldwherecapitalcanbereplicatedatarelativelylowcost (thinkofcomputerchipsorevensoftware),themarginalvalueofcapitalwill tendtofall,evenifmorecapitalisusedoverall.Thevalueofexistingcapital willactuallybedrivendownasnewcapitalisaddedcheaplyatthemargin. Thus,therewardsearnedbycapitalistsmaynotautomaticallygrowrelativeto labor.Insteadthe shareswilldepend ontheexactdetailsofthe production, distribution,andgovernancesystems. Mostofall,thepayoffwilldependonwhichinputstoproductionarescarcest. Ifdigitaltechnologiescreatecheapsubstitutesforlabor,thenit’snotagoodtime tobealaborer.Butifdigitaltechnologiesalsocanincreasinglysubstitutefor capital,thencapitalownersshouldn’texpecttoearnhighreturnseither.What will be the scarcest, and hence the most valuable, resource in the second machineage?Thisquestionbringsustoournextsetofwinnersandlosers:

superstarsversuseveryoneelse.

superstarsversuseveryoneelse. *

*Thisechoestheproductivityeffectsofelectricitydiscussedearlier.Aswithdigitaltechnologies,

thebiggestgainsdidnotoccuruntilfactorieswereredesigned,andevenworkerswhodidn’twork

directlywiththenewmachinesweresignificantlyaffected.

* The effect on the economy overall would depend on how other companies reacted. Output

wouldlikelyincreaseatcompaniesthatdesignandbuildrobotsand,dependingonhowcapital-

intensive they are, the net ratio of capital to labor in the overall economy could increase,

decrease,orstaythesame.We’lldiscusstheseeffectsinmoredetailinchapter12.

“Onemachinecandotheworkoffiftyordinarymen.Nomachinecandotheworkofone extraordinaryman.” —ElbertHubbard

“Onemachinecandotheworkoffiftyordinarymen.Nomachinecandotheworkofone

extraordinaryman.”

—ElbertHubbard

WEVE SEEN THAT SKILL-BIASED technical change has increased the relative demandforhighlyeducatedworkerswhilereducingdemandforlesseducated workerswhosejobsfrequentlyinvolveroutinecognitiveandmanualtasks.In addition, capital-biased technological changes that encourage substitution of physicalcapitalforlaborhaveincreasedtheprofitsearnedbycapitalownersand reducedtheshareofincomegoingtolabor.Ineachcase,historicamountsof wealth have been created. In each case, we also have seen increases in the earningsofthewinnersrelativetothelosers.Butthebiggestchangesofallare drivenbyathirdgapbetweenwinnersandlosers:thegapbetweenthesuperstars inafieldandeveryoneelse.

MindtheGap

Call it talent-biased technical change.* In many industries, the difference in payoutbetweennumberoneandsecond-besthaswidenedintoacanyon.Asa controversialNikeadnoted,youdon’twinsilver,youlosegold. 1 When‘winner- take-all’marketsbecomemoreimportant,incomeinequalitywillrisebecause payattheverytoppullsawayfrompayinthemiddle. 2 The growing gaps in wages between people with and without college education,andbetweencapitalownersandworkers,havebeendwarfedbyeven

biggerchangesattheverytop.Asnotedearlier,between2002and2007,thetop

1percentgottwo-thirdsofalltheprofitsfromthegrowthintheU.S.economy.

Butwhoarethe1percent?Theyaren’tallonWallStreet.UniversityofChicago

economist Steve Kaplan found that most of them are in other industries: in mediaandentertainment,sports,andlaw—ortheyareentrepreneursandsenior executives.

Ifthetop1percentarestarsofasort,theycanlookuptosuperstarswhohave

seenevenbiggerincreases.Whilethetop1percentearnedabout19percentof

allincomeintheUnitedStates,thetop1percentofthe1percent(orthetop0.01

percent)—sawtheirshareofnationalincomedoublefrom3percentto6percent

between1995and2007.Thisisnearlysixtimesasmuchasthe0.01percent

earnedbetweenWorldWarIIandthelate1970s.Inotherwords,thetop0.01

percentnowgetabiggershareofthetop1percentofincomethanthetop1

percentgetofthewholeeconomy.Becauseitishardtomaintainanonymity

whenreportingdataforsmallnumbersofpeople,itishardtogetreliabledataat

incomelevelshigherthanthetop0.01percent.Afterall,whilethereareover

1.35millionhouseholdsinthetop1percentwithanaverageincomeof$1.12

million,the0.01percentrepresentsjust14,588familieseachwithincomesover

$11,477,000.* 3 Buttheevidencesuggeststhatthespreadofincomescontinuesat highlevelsofincomewithafractal-likequality,witheachsubsetofsuperstars watchinganevensmallergroupofsuper-duper-starspullingaway.

HowSuperstarsThriveintheWinner-Take-AllEconomy

In the previous chapter, we saw Intuit’s TurboTax automate the job of tax preparation,allowingamachinetodothejobsofhundredsofthousandsof human tax preparers. That’s an example of technology automating routine information-processing jobs, and also an example of capital substituting for labor.Butmostimportantly,it’sanexampleofthesuperstareconomyinaction. Intuit’s CEO made $4 million last year and Scott Cook, the founder, is a billionaire. 4 Likewise,thefifteenpeoplewhocreatedInstagramdidn’tneedalot ofunskilledhumanhelpersanddidleveragesomevaluablephysicalcapital.But mostofall,theybenefittedfromtheirtalent,timing,andtiestotherightpeople. Topperformersinotherindustrieshavealsoseentheirfortunesrise.J.K. Rowling,authoroftheHarryPotterseries,istheworld’sfirstbillionaireauthor in an industry not known for minting the super wealthy. As George Mason University’sAlexTabarroknotesofRowling’ssuccess:

Homer,ShakespeareandTolkienallearnedmuchless.Why?ConsiderHomer,hetoldgreat

storiesbuthecouldearnnomoreinanightthansay50peoplemightpayforanevening’s

entertainment.Shakespearedidalittlebetter.TheGlobetheatercouldhold3000andunlike

Homer,Shakespearedidn’thavetobeatthetheatertoearn.Shakespeare’swordswere

leveraged.5

J.R.R.Tolkien’swordswereleveragedfurther.Bysellingbooks,Tolkien

couldselltohundredsofthousands,evenmillionsofbuyersinayear—more

thanhaveeverseenaShakespeareplayinfourhundredyears.Andbookswere

cheapertoproducethanactors,whichmeantthatTolkiencouldearnagreater

shareoftherevenuesthandidShakespeare.

TechnologyhassuperchargedtheabilityofauthorslikeRowlingtoleverage

theirtalentsviadigitizationandglobalization.Rowling’sstoriescanbecaptured

inmoviesandvideogamesaswellastext,buteachofthoseformats,including

theoriginalbooks,canbetransmittedgloballyattrivialcost.Sheandother

superstar storytellers now reach billions of customers through a variety of channelsandformats. Moreoftenthannot,whenimprovementsindigitaltechnologiesmakeitmore andmoreattractivetodigitizesomething,superstarsinvariousmarketsseea boostintheirincomeswhilesecond-bestshaveahardertimecompeting.Thetop performers in music, sports, and other areas have also seen their reach and incomesgrowsincethe1980s. 6 Atthesametime,othersworkinginthecontentandentertainmentindustries have not seen a big increase. Only 4 percent of software developers in the burgeoningappeconomyhavemadeoveramilliondollars. 7 Three-quartersof themmadelessthanthirtythousanddollars.Whileahandfulofwriters,actors, orbaseballplayerscanbecomemillionaires,manyothersstruggletomakeends meet. Agold-medal winner at the Olympics can earn millions of dollars in endorsements,whilethesilvermedalwinner—letalonethepersonwhoplaced tenthorthirtieth—isquicklyforgotten,evenifthedifferenceismeasuredin tenthsofasecondandcouldhaveresultedfromagustofwindoraluckybounce oftheball. Eventopexecutiveshavestartedearningrock-starcompensation.Theratioof

CEOpaytoaverageworkerpayincreasedfromseventyin1990tothreehundred

in 2005. Much of this growth is linked to the greater use of information technology,accordingtoresearchthatErikcompletedwithhisstudentHeekyung Kim. 8 One rationale for this increase in executive pay is that technology increases the reach, scale, or monitoring capacity of a decision-maker. If executivesusedigitaltechnologiestoobserveactivitiesinfactoriesthroughout theworld,togivespecificinstructionsforchangingaprocess,andtomakesure

instructionsarecarriedoutwithhighfidelity,thenthevalueofthosedecision-

makers increases. Direct management via digital technologies makes a good

managermorevaluablethaninearliertimeswhenmanagershaddiffusecontrol vialongchainsofsubordinates,orwhentheycouldonlyaffectasmallerscaleof activities. Directdigitaloversightalsomakeshiringthebestcandidateratherthanthe second-bestthatmuchmoreimportant.Companiesarereadytopayapremium forexecutiveswhomtheyperceivetobethebest,reasoningthatevenasmall differenceinqualitycanhavehugeconsequencesforshareholders.Thebigger themarketvalueofacompany,themorecompellingtheargumentfortryingto gettheverybestexecutive. 9 Asingledecisionthatincreasesvaluebyamodest1

percentisworth$100milliontoaten-billion-dollarcompany.

Inacompetitivemarket,evenasmalldifferenceintheperceivedtalentsof

CEOcandidatescanleadtofairlylargedifferencesintheircompensation.As

economistsRobertFrankandPhilipCooknoteintheirbook,TheWinner-Take-

AllSociety,“Whenasergeantmakesamistakeonlytheplatoonsuffers,but whenageneralmakesamistakethewholearmysuffers.” 10

WhenRelativeAdvantageLeadstoAbsoluteDomination

Theeconomicsofsuperstarswasfirstformallyanalyzedin1981byeconomist

SherwinRosen. 11 Inmanymarkets,buyerswithachoiceamongproductsor services will prefer the one with the best quality. When there are capacity constraintsorsignificanttransportationcosts,thenthebestsellerwillonlybe

abletosatisfyasmallfractionoftheglobalmarket(forinstance,inthe1800s,

eventhebestsingersandactorsmightperformforatmostafewthousandpeople eachyear).Otherinferiorsellerswillalsohaveamarketfortheirproducts.But whatifatechnologyarisesthatletseachsellercheaplyreplicatehisorher servicesanddeliverthemgloballyatlittleornocost?Suddenlythetop-quality providercancapturethewholemarket.Thenext-bestprovidermightbealmost asgood,butitwillnotmatter.Eachtimeamarketbecomesmoredigital,these winner-take-alleconomicsbecomealittlemorecompelling.

Winner-take-allmarketswerejustcomingtotheforeinthe1990s,when

FrankandCookwrotetheirremarkablyprescientbook.Theycomparedthese winner-take-all markets, where the compensation was mainly determined by relative performance, to traditional markets, where revenues more closely trackedabsoluteperformance.Tounderstandthedistinction,supposethebest, hardest-workingconstructionworkercouldlayonethousandbricksinaday while the tenth-best laid nine hundred bricks per day. In a well-functioning market,paywouldreflectthisdifferenceproportionately,whetheritcouldbe attributedtomoreefficiencyandskill,orsimplytomorehoursofwork.Ina

traditionalmarket,someonewhois90percentasskilledorworks90percentas

hardcreates90percentasmuchvalueandthuscanearn90percentasmuch

money.That’sabsoluteperformance. Bycontrast,asoftwareprogrammerwhowritesaslightlybettermapping application—onethatloadsalittlefaster,hasslightlymorecompletedata,or prettier icons—might completely dominate a market. There would likely be little,ifany,demandforthetenth-bestmappingapplication,evenitgotthejob donealmostaswell.Thisisrelativeperformance.Peoplewillnotspendtimeor

effortonthetenth-bestproductwhentheyhaveaccesstothebest.Andthisisnot acasewherequantitycanmakeupforquality:tenmediocremappingtoolsare nosubstituteforonegoodone.Whenconsumerscaremostlyaboutrelative performance,evenasmalldifferenceinskilloreffortorluckcanleadtoa thousand-foldormillion-folddifferenceinearnings.Therewerealotoftraffic

appsinthemarketplacein2013,butGoogleonlyjudgedone,Waze,worth

buyingforoveronebilliondollars. 12

WhyWinner-Take-AllIsWinning

Whyarewinner-take-allmarketsmorecommonnow?Shiftsinthetechnology

forproductionanddistribution,particularlythesethreechanges:

a) the digitization of more and more information, goods, and services, b) the vast improvements in telecommunications and, to a lesser extent,transportation,and c)theincreasedimportanceofnetworksandstandards.

AlbertEinsteinoncesaidthatblackholesarewhereGoddividedbyzero,and thatcreatedsomestrangephysics.Whilethemarginalcostsofdigitalgoodsdo notquiteapproachzero,theyarecloseenoughtocreatesomeprettystrange

economics.Asdiscussedinchapter3,digitalgoodshavemuchlowermarginal

costsofproductionthanphysicalgoods.Bitsarecheaperthanatoms,notto mentionhumanlabor. Digitizationcreateswinner-take-allmarketsbecause,asnotedabove,with digital goods capacity constraints become increasingly irrelevant. A single producerwithawebsitecan,inprinciple,fillthedemandfrommillionsoreven billionsofcustomers.JennaMarbles’shomemadeYouTubevideo“Howtotrick people into thinking you’re good looking,” to take one wildly successful example,garnered5.3millionviewstheweeksheposteditinJuly2010. 13 She’s nowearnedmillionsofdollarsfromoveronebillionviewingsofhervideos aroundtheworld.Everydigitalappdeveloper,nomatterhowhumbleitsoffices or how small its staff, almost automatically becomes a micro-multinational, reachingglobalaudienceswithaspeedthatwouldhavebeeninconceivablein thefirstmachineage. Incontrast,theeconomicsofpersonalservices(nursing)orphysicalwork

(gardening)areverydifferent,sinceeachprovider,nomatterhowskilledor hard-working, can only fulfill a tiny fraction of the overall market demand. Whenanactivitytransitionsfromthesecondcategorytothefirstthewaytax preparationdid,theeconomicsshifttowardwinner-take-alloutcomes.What’s more,loweringprices,thetraditionalrefugeforsecond-tierproducts,isoflittle benefit for anyone whose quality is not already at or near the world’s best. Digitalgoodshaveenormouseconomiesofscale,givingthemarketleadera hugecostadvantageandroomtobeatthepriceofanycompetitorwhilestill makingagoodprofit. 14 Oncetheirfixedcostsarecovered,eachmarginalunit producedcostsverylittletodeliver. 15

ImprovementsinTelecommunications:ReachOutand

TouchMorePeople

Secondly, winner-take-all markets have also been boosted by technological improvementsintelecommunicationsandtransportationthatalsoexpandthe marketindividualsandcompaniescanreach.Whentherearemanysmalllocal markets,therecanbea‘best’providerineach,andtheselocalheroesfrequently canallearnagoodincome.Ifthesemarketsmergeintoasingleglobalmarket, topperformershaveanopportunitytowinmorecustomers,whilethenext-best performers face harsher competition from all directions. Asimilar dynamic comes into play when technologies like Google or even Amazon’s recommendation engine reduce search costs. Suddenly second-rate producers cannolongercountonconsumerignoranceorgeographicbarrierstoprotect theirmargins. Digitaltechnologieshaveaidedthetransitiontowinner-take-allmarkets,even for products we wouldn’t think would have superstar status. In a traditional camerastore,camerastypicallyarenotrankednumberoneversusnumberten. Butonlineretailersmakeiteasytolistproductsinrankorderbycustomer ratings, or to filter results to include only products with every conceivable desirablefeature.Productswithlowerrankingsoronlynineoutoftendesirable featuresreceivedisproportionatelylowersalesfromevensmalldifferencesin quality,convenience,orpricingperformance. 16 Digital ranking and filtering create disproportional returns even in labor markets for workaday, non-superstar careers. Companies have digitized their hiringprocessesanduseautomatedfilterstowinnowthefloodofapplicants.For example,companiescanreadilycullallthecandidatesthatdon’thaveacollege

degreeasasimpleexpedientevenifthejobdoesnotactuallyrequireacollege

education. 17 Thiscanamplifyatrickleofskill-biasedtechnicalchangeintoa torrentofstardomforaluckyfew.Similarly,jobcandidateresumesthatmissthe

buzzwordrequirementsmightdropfromconsiderationevenifthe90-percent-

qualifiedcandidatemightotherwisebeastellaremployee.

NetworksandStandards:TheValueofScale

Thirdly,theincreasedimportanceofnetworks(liketheInternetorcreditcard networks) and interoperable products (like computer components) can also createwinner-take-allmarkets.Justaslowmarginalcostscreateeconomiesof scaleontheproductionside,networkscancreate‘demandsideeconomiesof scale’thateconomistssometimescallnetworkeffects. We see them at work whenuserspreferproductsorservicesthatotherpeopleareflockingto.Ifyour friendskeepintouchviaFacebook,thatmakesFacebookmoreattractivetoyou, too.IfyouthenjoinFacebook,thesitebecomesmorevaluabletoyourfriendsas well. Sometimesnetworkeffectsareindirect.Youcanmakeaphonecallequally welltosomeoneusinganiPhoneoranAndroidphone.Butthetotalnumberof usersonagivenplatforminfluencesappdevelopers:thebiggernetworkofusers willtendtoattractmoredevelopers,orencourageappdeveloperstoinvestmore inagivenplatform.Themoreappsavailableforagivenphone,thegreaterits appealtousers.Thus,yourbenefitsfrombuyingoneortheotherwillbeaffected bythenumberofotheruserswhobuythesameproduct.WhenApple’sapp ecosystemisstrong,buyerswillwanttobuyintothatplatform,attractingeven moredevelopers.Buttheoppositedynamiccanunraveladominantstandard,as it almost did for the Apple Macintosh platform in the mid-1990s. Like low marginalcosts,networkeffectscancreatebothwinner-take-allmarketsandhigh turbulence. 18

TheSocialAcceptabilityofSuperstars

In addition to the technical changes that have increased digitization, telecommunication,networks,andotherfactorsthatcreatesuperstarproducts and companies, there are more aspects at work in boosting superstar compensationforindividuals.Insomecases,culturalbarrierstoverylargepay packages have fallen. CEOs, financial executives, actors, and professional

athletes may be more willing to demand seven- or even eight-figure compensationdeals.Asmorepeoplegetthosedeals,apositivefeedbackloop emerges:itbecomeseasierforotherstomakesimilarrequests. Infact,theconcentrationofwealthitselfcancreatewhatFrankandCookcall “deeppocket”winner-take-allmarkets.AsthegreateconomistAlfredMarshall noted,“arichclientwhosereputationorfortuneorbothareatstakewillscarcely countanypricetoohightosecuretheservicesofthebestmanhecanget.” 19 If mass-marketmediaenablesanathletelikeO.J.Simpsontoearnmillions,then hecanaffordtopayalawyerlikeAlanDershowitzmillionstodefendhimin court,evenifDershowitz’sservicesarenotreplicatedtomillionsofpeoplelike Simpson’sare.Inasense,Dershowitzisasuperstarbyproxy:hebenefitsfrom theabilityofhissuperstarclientswhoselaborhasbeenmoredirectlyleveraged bydigitizationandnetworks.* Lawsandinstitutionshavealsochangedinwaysthatoftenboosttheincomes

ofsuperstars.Thetopmarginaltaxratewasashighas90percentduringthe

Eisenhoweryearsandover50percentearlyinRonaldReagan’sadministration,

butfellto35percentin2002,whereitremainedthrough2012.Whilethisshift

obviouslyboostedtheafter-taxincomeoftopearners,researchsuggestsitcan also affect reported pre-tax income by motivating people to work harder (becausetheykeepmoreofeachdollartheyearn)andreportmoreoftheiractual income,ratherthanseekwaystohideorshelterit(becausethecostsofreporting totaxauthoritiesaren’tashighasbefore). Restrictionsontradehavealsodecreased.Likecheapertelecommunications and transportation, this makes markets more global, allowing international superstarstomoreeasilycompetewith,anddriveout,localproducers.When

KiapoachedPeterSchreyerfromAudiin2006,itwasasignalthatthemarket

fortalentedautomobiledesignerswasincreasinglyglobal,notlocal.

Althoughthetop1percentand0.01percenthaveseenrecordincreasesin

theirearnings,thesuperstareconomyhasfacedafewheadwinds.Perhapsthe most important among these is the growth of the long tail—the increased availabilityofnicheproductsandservices.Technologyhasnotjustlowered marginalcosts;inmanycasesithasalsoloweredfixedcosts,inventorycosts, andthecostsofsearching.Eachofthesechangesmakesitmoreattractiveto offer a greater variety of products and services, filling small niches that previouslywentunfilled. Instead of going head-to-head with a superstar, some individuals and businessesareinsteadfindingwaystodifferentiatetheirproducts,tofindor

createanalternativenichewheretheycanbetheworld’sbest.J.K.Rowlingisa

billion-dollarauthor,buttherearealsomillionsofotherauthorswhonowhavea

chancetopublishformorespecializedaudiencesofafewthousandorevena

fewhundredreaders.Amazonwillstocktheirbooksandmakethemaccessible

topeopleacrosstheplanet.ThatwillbeprofitableforAmazonevenifitwould

havebeenunprofitableforanyphysicalbookstore,withamuchsmallersetof

customers,tostockthebook.Evenasthetechnologydestroysgeography—a

barrierthatusedtoprotectauthorsfromworldwidecompetition—itopensup

specializationasasourceofdifferentiation.

Insteadofbeingthethousandth-bestchildren’sbookauthorintheworld,it maybemoreprofitabletobethenumber-oneauthorinScience-BasedAdvice forEcologicalEntrepreneurs,orFootballClockManagement. 20 Followingthis principle,developershavecreatedoversevenhundredthousandappsforthe iPhoneandAndroid,whileAmazonoffersovertwenty-fivemillionsongs.An evenlargernumberofblogposts,Facebookstories,andYouTubevideoshave beencreatedinthesharingeconomy,creatingeconomicvalueifnotnecessarily directincomefortheircreators.Aswe’veseen,however,opportunitiestocreate

newproductsdon’tnecessarilycomewithbigpaychecks.Asuperstarorlong-

taileconomywithlowbarrierstoentryisstillonewithfarmoreinequality.

ThePowerCurveNation

Aneconomydominatedbywinner-take-allmarketshasverydifferentdynamics thantheindustrialeconomytowhichweareaccustomed.Aswediscussedatthe beginningofthechapter,theearningsofbricklayerswillvaryalotlessthanthe winner-take-allearningsofappdevelopers,butthat’snottheonlydifference. Instead of stable market shares, where revenues and income correspond proportionallytodifferencesintalentandeffort,competitioninwinner-take-all marketswillbemuchmoreunstableandasymmetrical.Thegreateconomist JosephSchumpeterwroteof“creativedestruction,”whereeachinnovationnot onlycreatedvalueforconsumersbutalsowipedoutthepreviousincumbent. Thewinnersscaledupanddominatedtheirmarkets,butwereinturnvulnerable to the next generation of innovators. Schumpeter’s observation describes marketsinsoftware,media,andtheInternetmuchbetterthantraditionalmarkets in manufacturing and services. But as more and more industries become increasinglydigitizedandnetworked,wecanexpecttheSchumpeteriandynamic tospread. 21

Inasuperstareconomy,thedistributionofincomeisn’tjustmorespreadout; ithasaverydifferentshape.It’snotjustthatasmallgroupatthetopseesbig increases.It’salsoachangeinthefundamentalstructureofthedistribution. When revenues are roughly proportional to absolute performance, as in the exampleofthebricklayer,theearningsdistributionislikelytorou