43
1 Modeling of technological performance trends using design theory Subarna Basnet Massachusetts Institute of Technology, Department of Mechanical Engineering, 77 Massachusetts Ave, Cambridge, Massachusetts 02139 Christopher L. Magee Massachusetts Institute of Technology, Institute for Data, Systems, and Society, 77 Massachusetts Ave, Cambridge, Massachusetts 02139 Abstract Functional technical performance usually follows an exponential dependence on time but the rate of change (the exponent) varies greatly among technological domains. This paper presents a simple model that provides an explanatory foundation for these phenomena based upon the inventive design process. The model assumes that invention ‐ novel and useful design‐ arises through probabilistic analogical transfers that combine existing knowledge by combining existing individual operational ideas to arrive at new individual operating ideas. The continuing production of individual operating ideas relies upon injection of new basic individual operating ideas that occurs through coupling of science and technology simulations. The individual operational ideas that result from this process are then modeled as being assimilated in components of artifacts characteristic of a technological domain. According to the model, two effects (differences in interactions among components for different domains and differences in scaling laws for different domains) account for the differences found in improvement rates among domains whereas the analogical transfer process is the source of the exponential behavior. The model is supported by a number of known empirical facts: further empirical research is suggested to independently assess further predictions made by the model. Keywords: Modeling, design, combinatorial Invention, technological performance

Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

1

Modelingoftechnologicalperformancetrendsusingdesigntheory

SubarnaBasnet

MassachusettsInstituteofTechnology,DepartmentofMechanicalEngineering,77MassachusettsAve,Cambridge,Massachusetts02139

ChristopherL.Magee

MassachusettsInstituteofTechnology,InstituteforData,Systems,andSociety,77MassachusettsAve,Cambridge,Massachusetts02139

Abstract 

Functionaltechnicalperformanceusuallyfollowsanexponentialdependenceontimebuttherateofchange(theexponent)variesgreatlyamongtechnologicaldomains.Thispaperpresentsasimplemodelthatprovidesanexplanatoryfoundationforthesephenomenabasedupontheinventivedesignprocess.Themodelassumesthatinvention‐novelandusefuldesign‐arisesthroughprobabilisticanalogicaltransfersthatcombineexistingknowledgebycombiningexistingindividualoperationalideastoarriveatnewindividualoperatingideas.Thecontinuingproductionofindividualoperatingideasreliesuponinjectionofnewbasicindividualoperatingideasthatoccursthroughcouplingofscienceandtechnologysimulations.Theindividualoperationalideasthatresultfromthisprocessarethenmodeledasbeingassimilatedincomponentsofartifactscharacteristicofatechnologicaldomain.Accordingtothemodel,twoeffects(differencesininteractionsamongcomponentsfordifferentdomainsanddifferencesinscalinglawsfordifferentdomains)accountforthedifferencesfoundinimprovementratesamongdomainswhereastheanalogicaltransferprocessisthesourceoftheexponentialbehavior.Themodelissupportedbyanumberofknownempiricalfacts:furtherempiricalresearchissuggestedtoindependentlyassessfurtherpredictionsmadebythemodel.Keywords:Modeling,design,combinatorialInvention,technologicalperformance

Page 2: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

2

Nomenclature and terminology QJ=intensiveperformanceofartifactswithinatechnologicaldomain,Jt=timeIOI=individualoperatingideasPIOI=probabilityofcombinationofanytwoIOIIOI0=basicIOI‐IOIthatfirstintroduceanaturalphenomenonintheOperationsregimeIOIC=cumulativenumberofIOIintheOperationsregimeIOIL=maximumnumberofpossibleIOIinOperationsregimeattimetIOISC=IOICsuccessfullyintegratedintoadomainartifactK=annualrateofincreaseinIOIcintheOperationsregimeKJ=annualrate(whentimeisinyears)ofperformanceimprovementmeasuredbytheslopeofaplotoflnQJvs.timefi=fitnessinUnderstandingregimeforascientificfieldiFU=cumulativefitnessofUnderstandingregimedJ=interactionparameteroftechnologicaldomainJdefinedasinteractiveout‐linksfromatypicalcomponenttoothercomponentsinartifactsindomainJsJ=designparameteraffectingtheperformanceofanartifactindomainJAJ=exponentofdesignparameterinpowerlawfordomainJ,relatingperformanceandthedesignparameter

1. Introduction Inventionsaretheoutputsofthedesignprocesswhentheyreachsufficientnoveltyandutilitytoratethatterm:theyareabasicbuildingblockoftechnologicalprogressandthefundamentalunitofthispaper.Inourformulation,technologicaldomainsconsistofdesignedartifactsthatutilizeaspecifiedbodyofknowledgetoachieveaspecificgenericfunction(Mageeet.al.2014).Thus,technologicaldomainsinvolvealargenumberofinter‐relatedinventionsasevensingleartifactscanembodymultipleinventions.Arthur(2006)usedtheterm“technologies”todescribesomethingthatbridgesinventionsandtechnologicaldomains;accordingtoArthur,theseuse“effects”toachievesome“purpose”.Thus,onecanalsosaythateachartifactisamaterialrealizationofitsdesignthatintentionallyembodiestheeffects.Thispaperbringstogetherthreebodiesofresearchthatdonotusuallyinteract.Thefirstisthedesignresearchfield,particularlyitscognitivescientificinsightsonthedesignprocess.Thesecondisthetechnologicalchangefieldwheremostresearchershavebeeneconomistsorbusinessscholars.Thethirdareaisquantitativemodelingofperformanceofartifacts.Theobjectiveoftheworkreportedhereistouseunderstandingofthedesignandinventionprocesstomodelperformance‐howwellaspecificdesignedartifactachievesitsintendedfunctionorpurpose.Inparticular,weexamineperformancetrends‐thetimedependenceofperformanceasrealizedinaseriesofimproveddesignsofartifactsthatarise

Page 3: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

3

overtime.Wedosoinanattempttodevelopanexplanatoryandquantitativepredictivemodelforwhyperformanceimprovesexponentiallyovermultipledesignswithwidelyvaryingratesamongtechnologicaldomains,rangingfrom3to65%annuallyfordomainscharacterizedsofar.Ourresearchquestioniswhetheraquantitativepredictivemodelbaseduponfoundationsandinsightsaboutthedesignprocessleadstoresultsconsistentwiththisexponentialbehaviorandwhethersuchamodelhelpsexplainandpossiblypredictthevariationintherateofimprovement.Wefirstdiscusssomerelevantliteratureineachofthethreeintersectingfields.

2 Background 

2.1 Design, invention and cognitive psychology literature Whatconnectionsbetweentechnologicalchangeanddesignresearchcanbeinferredfromtheexistingliterature?Businessscholarsandeconomistsoftenviewtechnicalchangeasoccurringinsideablackbox,andhaveusuallyavoidedexaminingdesignactivitiesthatarethesourceoftechnologicalchange.AnimportantrecentpublicationthatbeginstobuildabridgebetweenaspectsofdesignresearchandtheeconomicsoftechnologicalchangeisthepaperbyBaldwinandClark(2006).Theseauthors(andLuoetal.2014)pointspecificallytoacentralrolefordesigninachievingeconomicvalue.Inadditiontoeconomicperspectives,anotherviewthatsomewhatignoresdesignisthelinearmodelaccreditedtoVannevarBush(Bush,1945),whichconsiderstechnologicalchangeoccurringthroughapplicationofscience.Asacounterview,inhisseminalbook,TheSciencesoftheArtificial,HerbertSimon(1969,1996)wasthefirsttohighlightthatdesignisanactivitystandingonitsownright,likenaturalsciences,andhasitsownsetoflogic,concepts,andprinciples.Whiletheprimarygoalofnaturalscienceistoproducepredictiveexplanationsofnaturalphenomena,theprimarygoalofdesignistocreateartifacts.Thedesignactivityiscentraltocreationandimprovementofartifactsinalltechnologicaldomainsandinvolvescognitiveactivitiessuchastheuseofknowledge,reasoning,andunderstanding.Theseindisputablecognitiveactivitieshavebeennotedbymanyscholarswhohavestudiedinventionanddesign(Simon1969,Dasgupta1996,GeroandKannengiesser(2004),HatchuelandWeil2009).Inthecontextofrealizinghigherperformancefromsubsequentgenerationsofartifacts,theroleofinvention,asoneoutcomeofthedesignprocess,isacriticalonesinceimprovementinperformanceofartifactsmuststronglyreflecttheinventions.AsVincenti(1990,pg.230)putsit,inventiveactivityisasourceofnewoperationalprinciples,andnormalconfigurationsthatunderliefuturenormalorradicaldesigns.Theoperationalprinciples(Polyani1962,Vincenti1990)ofanartifactdescribehowitscomponentsfulfilltheirspecialfunctionsincombiningtoanoveralloperationtoachievethefunctionoftheartifact.ModelsfoundusefulindescribingthecreativedesignprocessincludetheGeneploremodel(Finke,WardandSmith1996),topologicalstructures(BrahaandReich2003),FBStheory(GeroandKannengiesser2004),CKtheory(HatchuelandWeil2009),infuseddesign(Shaietal.2009),analyticalproductdesign(Frischknechtetal.2009),andothermodelingapproaches.Althoughalloftheseframeworksinclude–tosomedegree‐thekeyideaof

Page 4: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

4

combiningexistingideas(forexample,intheformofconceptualsynthesis,andblendingofmentalmodelsdescribedindiscussionoftheGeneploremodel),theframeworkfoundmosthelpfulinourmodelingofperformancechangesresultingfromacumulativedesignprocessisanalogicaltransfer.AlthoughthisideacanbetracedasbeginningwithPolya(1945)orearlier,theframeworkremainsanactiveareaindesignresearch(Clementetal.1994,HolyoakandThagard1995,Goel1997,GentnerandMarkman1997,LeclerqandHeylighen2002,DahlandMoreau2002,ChristensenandSchunn2007,Linseyetal.2008,Tsengetal.2008,Linseyetal.2012,Fuetal.2013).Scholarsofanalogicaltransfer(GentnerandMarkman1997,Holyoaketal.1995andWeisberg2006)explainanalogicaltransferasinvolvingtheuseofconceptualknowledgefromafamiliardomain(base)andapplyingittocreateknowledgeinadomainwithsimilarstructure(target):analogicaltransferexploitspastknowledgeinboththebaseandtargetdomains.Theanalogiesutilizedcanbelocal,regionalorremote,dependingonsurfaceandstructuralsimilaritiesbetweenobjectsinvolvedinthebaseandtargetdomains.WeisbergdiscussestheexampleoftheWrightbrothersusingseveralanalogicaltransferstofirstrecognizeandsolvetheproblemofflightcontrol.First,theyviewedflyingasbeingsimilartobikinginwhichtheriderhastobeactivelyinvolvedincontrollingthebike,anapplicationofregionalanalogy.Interestingly,manyothersattemptingtodesignartifactsforflyingdidnotaccessthisregionalanalogyandthusdidnotevenidentifythekeycontrolproblem.Second,theWrightbrothersstudiedbirdstoseehowtheycontrolledthemselvesduringflight,andlearnedthattheyadjustedtheirpositionabouttherollingaxisusingtheirwingtips.Fromthisinsight,theyhadtheideaofusingsimilarmovingsurfaces,anotherinstanceofusingregionalanalogy.Lastly,theydevelopedtheideaofwarpingthewings,demonstratedbyusingatwistedcardboardbox,toactlikevanesofwindmillstomaketheairplaneroll.Theuseofthreeanalogicaltransfersincombinationtoseeandsolvetheflightcontrolproblemisaclearcaseofanalogicaltransferbutthereisalsoevidence(citedearlierinthisparagraph)ofmuchwiderapplicability.Therearemoreabstractversionsofcombinatorialanalogicaltransferthathavebeenproposedinthewiderliterature.BasedonanextensivehistoricalstudyofmechanicalinventionsanddrawinginsightsfromGestaltpsychology,Usher(1954)proposedacumulativesynthesisapproachforcreationofinventions.Thenotionofbisociation(Koestler1964,Dasgupta1996)developsthecumulativesynthesisapproachfurtherandsaysthatanewinventiveideaisideatedcombiningdisparateideas.Morerecently,Fleming(2001)andArthur(2006)haverespectivelyusedthesamecombinatorialnotionsofinventioninstudyingtechnologicalchange.Otherresearchinthetechnologicalchangeliteraturealsodiscussesarelatedconceptthatisusuallycalled“spillover”.Rosenberg(1982)showedthatsuchtechnologicalspillovergreatlyimpactedthequantityandqualityoftechnologicalchangeintheUnitedStatesinthe20thcentury–aresultsupportedbyNelsonandWinter(1982)andRuttan(2001).Indeed,arecentpaperbyNemetandJohnson(2012)statesthat“oneofthemostfundamentalconceptsininnovationtheoryisthatimportantinventionsinvolvethetransferofknowledgefromonetechnicalareatoanother”.Wenotethatthesedescriptionsdonotalwaysmakeacleardistinctionregardingwhetherthetransferisoccurringattheidealevelorattheartifactlevel.Theyaresilentregardinghowandfromwheredesignersorinventorsgettheirdisparateideastocombineandregardingdetailsaboutthecomplexitiesoftransferandcombination.

Page 5: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

5

Analogicaltransferofideasasabroadmechanismandexpertiseasthefoundationofideas(Weisberg,2006)providesadequatespecificityformodelingscienceandinventioninthispaper.Weisbergcontendsthatanalogicaltransferisutilizedingenerationofbothscientificandtechnologicalknowledge.Existingknowledgeprovidesthefoundationalbasisforanalogicaltransfertooccur.Asimilarargumenthasbeenappliedtothemoreabstractnotionofcombinations.Usherdescribesacumulativesynthesisapproach‐afourstepsocialprocess(perceptionoftheproblem,settingthestage,theactofinsight,criticalrevision)‐whichbringstogetherinventivestructurestocreatenewinventions.Ruttan(1959),hasarguedthatUsher’sformulationprovidesa“theoryofthesocialprocessesbywhich‘newthings’comeintoexistencethatisbroadenoughtoencompassthewholerangeofactivitiescharacterizedbythetermsscience,invention,andinnovation”.ModelsofbothUnderstandingandOperationsregimeinourpaper(definedinthenextparagraph)utilizetheabstractionthatknowledgeiscreatedbyprobabilistically1combiningexistingknowledgemadeavailablebyanalogicaltransfer.Vincenti(1990),andMokyr(2002)taketheviewthatscientificandtechnologicalknowledgecanbeclassifiedintodescriptive(Understanding)andprescriptive(Operations)knowledge2regimes.TheUnderstandingregimecanbeseenasabodyof‘what’knowledgeandincludesscientificprinciplesandexplanations,naturalregularities,materialsproperties,andphysicalconstants.Aunitofunderstandingisfalsifiable(Popper1959)andenablesexplanationandpredictionaboutspecificphenomena,includingbehaviorofartifacts.TheOperationsregime,ontheotherhand,canbeviewedasabodyof‘designknowledge’,whichsuggestshowtoleveragenatural‘effects’(Arthur,2006,Vincenti,1990))toachieveatechnologicaladvantageorpurpose.Itincludes,operatingprinciples,designmethods,experimentalmethods,andtools(Dasgupta1996,Vincenti1990).Basedonthisdistinction,understandingenablesgenerationofoperationalknowledge,whichultimatelycontributestowardsdesignofsomeartifact.However,operationsisnotentirelybaseduponexistingunderstandingandinfactinnovationsinknow‐howcanandoftendooccurbeforeanyunderstandingofrelatednaturaleffectsisavailable.AnimportantaspectofdesignandinventionisthecooperativeinteractionbetweenUnderstandingandOperationsregimes(Musson,1972,MussonandRobinson1989).Usingahistoricalperspective,Mokyr(2002)hascarefullyobservedthatasynergisticexchangebetweenthetwohasbeenoccurring,whereeachenablestheother.ThecontributionofUnderstandingtoOperationsiswellknown:itprovidesprinciples,andregularitiesofnaturaleffects,includingnewones,intheformofpredictiveequations,anddescriptive1Atapointintime,notallpossiblecombinationsofexistingknowledgeleadtonewknowledge.2Weusetheterms“Understanding”and“Operations”,sinceeachonebringsmoreclaritytothenatureofunderlyingactivity.Understandingreferstoconceptualinsightthatisgeneratedaboutanobjectorenvironment,whereasOperationsreferstotheideaofactingonanobjectorenvironmenttogetsomedesiredeffect,aswellasexperimentalmethodsincludedintheterm‘science’.

Page 6: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

6

facts,suchasmaterialproperties.FlemingandSorenson(2004)provideevidencethatunderstandinghelpsinventorsbyprovidingarichermaptosearchforoperatingideas,whichcanbecombinedtogether.Understandingalsoprovidesinsightaboutwherenewtechnologicalopportunitiesmaybefound(Klevoricetal.1995).Beyondthesecontributions,thereisthemoregeneralview,discussedintheinitialparagraphofthissection,thatnewoperationalideascanbederivedfromnewunderstanding.Whatislessdiscussedisthemulti‐facetedcontributionsofOperationstotheUnderstandingregime.Inhispaper,Sealingwaxandstring,deSollaPrice(1983),aphysicist,andhistorianofscience,highlightedthatinstruments(anoutputoftheOperationsregime)wereadominantforceinenablingscientificrevolutions.Hestates:“changesinparadigmthataccompanygreatandrevolutionarychanges(inscience)werecausedmoreoftenbyapplicationoftechnologytoscience,ratherthanchangesfrom‘puttingonanewthinkingcap’“.Operationsprovidetoolsandinstrumentstomakemeasurements,andtomakenewdiscoveries.Inhisbook,TheScientist:AHistoryofScienceToldThroughtheLivesofitsGreatestInventors,Gribbin(2002),aBritishastrophysicist,andsciencewriter,hasdescribedhowtheabilitytogrindeyeglasslensesmadeitpossibletomakebettertelescopes,andhencepavedthewayforastronomerstomakenewdiscoveries.Neworimprovedobservationaltechniquesarestillamajordriverofprogressinscience.Gribbinhasaptlysummarizedtheenablingexchangebetweenthetworegimes:“newscientificideasleading…toimprovedtechnologyandnewtechnologyprovidingscientistswiththemeanstotestnewideastogreaterandgreateraccuracy”.Additionally,theOperationsregimeprovidesnewproblemsfortheUnderstandingregimetostudy,andhasledtobirthofnewfieldsinUnderstanding(Hunt2010).Basedupontheseinsightsandwithourfocusonexplainingperformanceimprovementarisingfromcontinuingstreamsofinventions,ourmodeltreatsmutualexchangebetweenUnderstandingandOperations.Indesignofartifacts,Simon(1962)introducedthenotionofinteractionsinhisessayonthecomplexityofartifacts.Whenadesignofanartifactischangedfromonestatetoanother(withdifferencesbetweenthetwostatesasdefinedbymultipleattributes,sayD1,D2,andD3)bytakingsomeactions(say,A1,A2,andA3),inmanycases,anyspecificactiontakenmayaffectmorethanoneattribute,thuspotentiallymanifestingasinteractionsoftheattributes.Thesamenotionofinteraction/conflictsiscapturedbytheconceptofcouplingoffunctionalrequirements(Suh2001),ordependenciesbetweencharacteristics(Weber2003),whichcanoccurwhentwoormorefunctionalrequirementsareinfluencedbyadesignparameter.Theoreticallyitseemsidealtohaveonedesignparametercontrollingonefunctionalrequirementtoachieveafullydecomposable(modular)design(Suh2001,BaldwinandClark2000).However,Whitney(1996,2004)hasarguedthat,inreality,howdecomposableadesignofanartifactcanbedependsonthephysicsinvolvedoradditionalconstraints,suchaspermissiblemass.Thesearereflectedascomponent‐to‐component,andcomponent‐to‐systeminteractions,orasaneedtohavemulti‐functionalcomponents.Consequently,Whitneyargues,complexelectro‐mechanical‐optical(CEMO)systems,primarilydesignedtocarrypower,cannotbemadeasdecomposableasVLSIsystemsprimarilydesignedtotransmitandtransforminformation.Forexample,inenergyapplications,theimpedanceoftransmittingandreceivingelementshastobematchedformaximumpowertransfer,thusmakingthetwoelementscoupled.Further,CEMOsystemstypicallyneedtohavemulti‐functionalcomponentsinordertokeeptheartifactsize

Page 7: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

7

reasonable,creatingcouplingofattributesatthecomponentlevel.AnothertypeofinteractionWhitneyhasidentifiedarethesideeffects,suchaswasteheatincomputers,andcorrosionofelectrodesinbatteries‐thatoccurinartifacts,whichinsomeelectro‐mechanicalsystemscanconsumesignificantportionofthedesigneffortfortheirmitigation.Thepresence,andthustheresolution,ofthesedifferentinteractionscausesignificantdelay,consumesignificantengineeringresourcesandpotentiallystopapplicationsofsomeconcepts,thusmakingthelevelofinteractionsofatechnologicaldomainapotentiallystrongfactorinfluencingitsrateofimprovement.BaseduponWhitney’swork,theeffectofinteractionsonratesofimprovementwassuggestedqualitativelybyKohandMagee(2008)andaquantitativemodeloftheeffectwasdevelopedbyMcNerneyetal.(2011)–seesection2.3.Theinfluenceofdesignparametersonartifactperformanceisanessentialpartofdesignknowledge.Manytechnologicaldomainshavecomplexmathematicalequationsrelatingsomeaspectsofperformancewithdesignparameters.Indeed,theso‐calledengineeringscienceliteraturehassuchequationsformanyaspectsaffectingthedesignofartifactsofperhapsalltechnologicaldomains.Simplerrelationshipsconcerningthegeometricalscaleofartifactsarealsoavailableandgenerallygiveperformancemetricsasafunctionofadesignvariableraisedtoapower.Useofpower‐lawrelationshipscanbefoundin:1)Sahal(1985)whostudiedscalinginthreedifferentsetsofartifacts‐airplanes,tractors,andcomputers;and2)BelaGold(1974)whodemonstratedthatdoublingthesizeofablastfurnacereducestheircostbyabout40%.Theconstantpercentchangeperdoublinginsizeresultsfromthepowerlaw(assumedbyGold)betweenperformance/costandgeometricalvariablessuchasvolume.

2.2 Technological change literature Whatdescriptivemodelsandtheorieshelpusunderstandwhytechnologiesimproveandhowtheimprovementpatternsarestructured?Schumpeter(1934)introducedtheideathatentrepreneurs,whoseprimaryroleistoprovideimprovedproductsandservicesthroughinnovation,driveeconomicprogress.Theseinnovations,whichSchumpeterdescribesasindustrialmutations,displacecompetingproductsandservicesfromtheeconomy.However,they,too,aredisplacedbyhigherperforminginnovationsthatfollow,thusperpetuatingthecycleofcreativedestruction.BuildinguponSchumpeter’snotion,Solow(1956)recognizedandincorporatedtechnologicalchangeasthekeyelementinhisquantitativeexplanatorytheoryofeconomicgrowth.Thebasicconclusionthattechnologicalchangeisthefoundationofsustainedeconomicgrowthhasstoodthetestoftime.Latertheoristsofeconomicgrowth(Arrow1962,Romer1990,Acemoglou2002)haveattemptedtodealwiththemorecomplexproblemofembeddingtechnologicalchangewithintheeconomy(endogenoustodifferentdegrees).Althoughthelatertheoriesareimportant,theissuesareoutsidethescopeofthispaperandwillnotbecoveredhere.Arelatedquestionofdemand‐pullandtechnology‐pushdoeshavemorerelevance. Whatdrivestechnologicalinnovation?Someearlyexplanationsemphasizedpuredemandpush(CarterandWilliams(1957,1959),Bakeretal.1967,MyersandMarquis1969,Langrishetal.1972,Utterback1974)wheretheneedsoftheeconomyatagiventime

Page 8: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

8

dictatetechnologicaldirection.MoweryandRosenberg(1979)reanalyzedthedataandmethodologyinthisearlyworkandarrivedatastrongroleforscience/technologypush(thediscoveriesofscientistsandinventorsprimarilydeterminetechnologicaldirection).Takingabalancedview,Dosi(1982)arguedthatbothmarket‐pull(customerneedsandpotentialforprofitability)andtechnology‐push(intheformofpromisingnewtechnology,andtheunderpinningprocedures)areequallyimportantforbeingsourcesofinnovation.TushmanandAnderson(1986)discussdiscontinuitiesashavinglargesocio‐technicaleffectsandnotethatsuchdiscontinuitiesareanessentialelementoftechnologicalchange.Inanotherhighlyreferencedpaper,HendersonandClark(1990)emphasizetheimportanceofarchitecturalchangeofartifacts‐asopposedtocomponentchange‐havinglargeeffectsonthefirm‐levelimpactofchange.Christensen(1996),ontheotherhand,viewstechnologicalchangeoccurringasaseriesofdisruptiveproductinnovationsthatstartinanichemarketcateringtodifferentfunctionalrequirements,butthenrapidlyimprovetowardstherequirementsofmainstreamperformance.Thedisruptivetechnologysurpassesthematuremarketleaders(byachievingthenecessaryperformanceinsmaller,cheaperartifacts),anddisplacesthem.Alloftheconceptsoftechnologicalchangedescribedintheprecedingparagraphs‐atleastimplicitly‐dependuponrelativeratesofchangeofperformance.Thisisthefocusofourmodelingeffortsowewillnowbrieflyreviewconceptsrelatedtotrendsinperformanceofdesignedartifacts,andwhatpatternstheyhavefollowed.Wefirstreviewtwoestablishedframeworks–generalizationsofWright’searlyresearch,andMoore’sLaw‐fordescribingtrendsintechnologicalperformance.In1936TheodorePaulWright(1936)inhisseminalpaper“FactorsaffectingtheCostofAirplanes”forthefirsttimeintroducedtheideaofmeasuringtechnologicalprogressofartifacts.Fromhisempiricalstudyofairplanemanufacturing,hedemonstratedthatlaborcostortotalcostofspecificairplanedesignsdecreasedasapowerlawagainsttheircumulativeproduction.Thisrelationshipisexpressedas: C=C0P‐w (1) WhereC0,andCareunitcostofthefirst,andsubsequentairplanesrespectively,andwherePandwarecumulativeproductionanditsexponentthatrelatesittounitcost.Wrightexplainsthatlaborcostreductionsarerealizedasshopfloorpersonnelgainexperiencewiththemanufacturingprocesses,andmaterialusageandhaveaccesstobetterproductiontools.SinceWright’swork,thisapproachhasbeenusedtostudyproductionofairplanesandshipsduringWorldWarII,andextendedtoprivateenterprises(Yelle,2007).ItshouldbenotedthatWrightdidnotlookatimprovementduetonewdesigns,insteadheonlyconsideredimprovedmanufacturingofafixeddesign.GordonMoore(1965)presentedthesecondapproach‐usingtimeastheindependentvariableandinvestigatingaseriesofnewlydesignedartifacts‐inhisseminalpaperthatdescribesimprovementofintegratedcircuits.Heobservedthatthenumberoftransistorsonadiewasdoublingroughlyevery18months(modifiedto2yearsin1975).This

Page 9: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

9

exponentialrelationshipbetweenthenumberoftransistorsonadieandtime,famouslyknown3asMoore’sLaw,canbemathematicallyexpressedas: QJ(t)=QJ(t0)exp{KJ(t‐t0)} (2)

WhereQJ(t0)andQJ(t)arethenumberoftransistorsperdie(ameasureofperformance)attimet0andtimet,andKJistherateofimprovement(annualiftimeisinyears).Forintegratedcircuits,theexponentialrelationshiphasheldbroadlytrueforfivedecades.Others(Girifalco1991,Nordhaus1996,KohandMagee(2006,2008)andLeinhard2008)utilizedthistemporalapproachtostudyperformanceofdifferenttechnologies,andhavedemonstratedthatmanytechnologiesexhibitexponentialbehaviorwithtime.Morerecently,Mageeetal.(2014)extendedthestudyto73differentperformancemetricsin28differenttechnologydomains.Theperformancecurveshavecontinuedtodemonstrateexponentialbehavior,althoughannualratesvarywidelyacrossdomains.WenotethatMooreandallotherswhousedhisframeworkbasicallycomparedtheperformanceofdifferentdesignsovertimedifferentiatingtheWrightandMooreframeworks.However,itisalsopossibletousetheWrightframeworkfordifferentdesignsbutonlyiftheamountproducedincreasesexponentiallywithtime(Sahal,1979,Nagyetal.2013,Mageeetal2014).Inordertoclarifyforreadersthenatureofempiricalperformancedata,wepresentperformancedatafortwosampledomains,magneticresonanceimaging(MRI)andelectricmotors(Fig.1a),andasummaryofimprovementratesfor28domains(Fig.1bfromMageeetal.2014).Theexponentialtrendforeachdomaincanbedescribedbyequation(2),whereQJ(t)andQJ(t0)aretheintensiveperformanceofanartifactindomainJattimetandt0,andKJistheannualrateofimprovementofthedomaininquestion.

3ThisdesignationwasgiventotherelationshipbyCalTechprofessorCarverMead.

Page 10: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

10

 

  

Fig. 1a: Exponential growth of performance in sample domains – Electric motor and Magnetic resonance imaging (MRI). Adapted from Magee et al. 2014 with permission. 

      KJ(%)

Fig. 1b: Annual rate of performance improvement, KJ, for 28 domains. Adapted from Magee et al. 2014with permission. 

Elec. MotorRate = 3.1 %R² = 0.9657

MRI

Rate = 21.3 %R² = 0.8561

1.E‐05

1.E‐04

1.E‐03

1.E‐02

1.E‐01

1.E+00

1.E+01

1.E+02

1.E+03

1.E+04

1.E+00

1.E+01

1.E+02

1.E+03

1880 1910 1940 1970 2000 2030 2060

MRI (resolution/tim

e)

Electric motor (W

att/liter)

Electric Motor MRI

Page 11: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

11

Arecentpaper(BensonandMagee,2015a)hasempiricallyinvestigatedthevariationoftheimprovementratesinthese28domains.Theworkhasimportantrelationshipstothecurrentworksowedescribeittonotetherelationshipsbuttoalsoclarifythefundamentaldifferences.BensonandMageefoundstrongcorrelationsbetweenspecificmeta‐characteristicsofthepatentsinthe28domains4andtheimprovementrateinthedomains.Theseauthorsfoundthatpatentmeta‐characteristicsreflectingtheimportance(citationsperpatentbyotherpatents),recency(ageofpatentsinadomain)andimmediacy(theaverageovertimeoftheusageofcurrentnewknowledgeinthedomain)areallcorrelatedwiththeimprovementrate.Theyfoundaparticularlystrongcorrelation(r=0.76,p=2.1x10‐6)withametricthatcombinesimmediacyandimportance(theaveragenumberofcitationsthatpatentsinthedomainreceiveintheirfirstthreeyears).Thefindings(andassociatedmultipleregressions)arerobustovertimeandwithdomainselectionandareofpracticalimportanceinpredictingtechnologicalprogressindomainswhereperformancedataisnotavailable(BensonandMagee,2015).Nonetheless,theconceptualbasisforthefindingsisobservedattributesoftheinventiveoutputfromatechnologicalfield(importance,recencyandimmediacyofapatentset)andnottheprocessofinvention,designknowledgeorothertechnicalaspectsofdesignedartifactsinthedomain.Theaimoftheworkreportedinthepresentpaperistodevelopamodelthatyieldsinsightsaboutthepaceofchangewithoutrecoursetoconceptsbaseduponobservationoftheoutputovertime.Iffullysuccessful,wewouldbeabletojudgethepotentialforchangebasedonlyuponthenatureofthedesignknowledgeandwemightevenbeabletofindnewapproachesthatmightachievetechnologicalgoalsatmorerapidimprovementrates.

2.3 Literature on quantitative modeling of technological change Whatresearchhasattemptedtomodelthetechnologicalperformancetrendsthatwejustdiscussed?Muth(1986)andAuerswaldetal.(2000)havedevelopedmodelstoexplainWright’sresultsbyintroducingthenotionofsearchfortechnologicalpossibilities.Eachpaperassumesthatrandomsearch,akeyelementoftechnologicalproblemsolving,forabettertechniqueismadewithinafixedpopulationofpossibilities.Consideringacaseofasinglemanufacturingprocess,Muth(1986)developedamodeltocapturetheideaofsubstitutingmanufacturingsequenceswithbetterones.Hearguesthatshoppersonnelimprovetheprocessbylearningthroughexperienceandmakingrandomsearchfornewtechniques,whichenableimprovementofprocessesleadingtocostreductions.Muthdemonstratedthatthenotionoffixedpossibilitieseasilyleadstofewerandfewerimprovementsthatcanberealizedandhearguesthatthedata(forfixeddesigns)showsalevelingoffandeventualstoppageasthemodelsuggests.BuildingonMuth’sideaofrandomsearchwithinasetoffixeddesignpossibilities,Auerswaldetal.modeledamulti‐processsystem,inwhichdifferentprocessescanbecombinedtocreatediverserecipes,andforthefirsttimeintroducedthenotionofinteractionsbyallowingadjoiningprocessestoaffecteachother’scost.

4Thepatentsarefoundbyanewtechnique‐BensonandMagee2015b

Page 12: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

12

FollowingsimilarreasoningasMuthandAuerswaldetal.,McNerneyetal.(2011)havedevelopedastochasticmodeltoexplainhowthecostreductionofamulti‐componentsystemisinfluencedbycomponentinteractions,whichtheyrefertoasconnectivitybetweencomponents.McNerneyetal.operationalizedthenotionofinteractionsasout‐linksrepresentinginfluenceofacomponentonothercomponents.Whenaspecificcomponentinadomainartifactchangesbyintroducinganewoperationalidea,thechangeaffectsthedesignofallthecomponentsitinfluences.Iftheperformanceoftheartifact(influencingandinfluencedcomponents)asawholeimproves,thenMcNerneyetal.considertheinteractionstoberesolvedandtheoperatingideaisconsideredsuccessful.TheMcNerneyetal.paperdemonstratesthatartifactswithmoreinteractionsimprovemoreslowlythanartifactswithlessinteractions.Usingagent‐basedmodeling,Axtelletal.(2013)havedevelopedacompetitivemicro‐economicmodeloftechnologicalinnovationutilizingthenotionoftechnologicalfitness.AlthoughtheydonotdiscussorciteMoore’slaworhiswork,theyhavedemonstratedthatcumulativetechnologicalfitnessofallagentsincreasesexponentiallyovertime.ThisisdifferentfromotherresearcherswhohavepredominantlybeenfocusedonWright’sframework.Consistently,Axtelletal.considernewdesignsandnotjustprocessoptimization.Usingasimulationapproach,ArthurandPolak(2006)havemodeledhownewgenerationsofartifactsarisebycombiningcurrentlyavailableartifacts.Theartifactsconsideredareelectroniclogicgates.Newdesigns(combinations)aremorecomplexlogicgatesthatcanthenalsobecombinedintoevenmorecomplexlogicgates.Intheirmodel,ArthurandPolakspecifyseveraldesigngoalstowardstowhichthelogicgatesevolve.Theyhavedemonstratedthatdesignswithhigherlevelsofcomplexitycannotbeattainedwithoutrealizingdesignconfigurationswithintermediatelevelsofcomplexity,andnewdesignswithhigherfunctionalitysubstituteforcurrentdesignswithinferiorfunctionality.Thismodelismuchricherthanothermodelsinrepresentingtheartifactpartofthedesignprocess;however,itdoesnotconsiderperformanceimprovement,asdotheothermodels.Itisalsolimitedtodevelopingpre‐specifiedartifactsandisthusaspecificprocess;consequently,itisnotopen‐endedorgeneralwhicharecharacteristicsnecessaryformodelingperformancetrendsforgeneraltechnologicaldomains.Althoughsomearemoreexplicitthanothers,onefeaturecommontoallthesemodelsisthatallutilizethenotionofbuildingupontheperformance(intheformofcost)ordesignsofthepast,akeyfeatureofcumulativeprocessesincludedinthemodelpresentedhere.Ontheotherhand,theydonotconsidertwoaspectswebelieveusefulinansweringourresearchquestion.First,noneofthemdiscussesorincludestheinfluentialroleplayedbyexchangebetweenscienceandtechnology.Inthispaper,wetreatthedesignprocessandtheexchangebetweenscienceandtechnologyasimportantelementsforunderstandingthechangeinperformanceovertimethatinturnisessentialtounderstandingtechnologicalchange.Second,noneconsiderthedesignprocessoroperatingprinciplesandinsteadlookatcombinationsattheartifactlevelinsteadofcombinationofideas.

Page 13: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

13

3. Overview of the model  

3.1 Conceptual basis of model Thedesiredoutputfromtheconstructedmodelareperformanceimprovementrates.Toagreewithknownempiricalresults,performanceshouldincreaseexponentiallywithtime.Weutilizetwosetsofmechanismsfromdesigntoconstructtheoverallmodel.Thefirstset,whichgivesrisetoexponentialtrends,includesgrowthofknowledge‐understandingandoperations‐usingcombinatorialanalogicaltransferaidedwithmutualexchangebetweenthetwo.Thesecondset,whichgivesrisetovariationinimprovementrates,includescomponentinteractionsandscalingofdesignvariables.Sincethegoalofthemodelistodevelopanexplanatoryandquantitativepredictivemodel,whilemodelingthesemechanismswehave,wherenecessary,simplified(removeddetails)andutilizedabstractiontokeepthemodeltractable.TheoverallarchitectureofthemodelisshowninFigure2.BasedontheworkofVincenti(1990)andMokyr(2002)thatwediscussedearlier,weclassifyscientificandtechnicalknowledgeintoUnderstandingandOperationsregimes.WefurthersplittheOperationsregimeintoideaandartifactsub‐regimeswherenon‐physicalrepresentationofartifactsareintheideasub‐regime.Theideasub‐regime,representedasanideaspool,consistsofindividualoperatingideas(IOI).TheIOI(individualoperatingidea)conceptisanabstractionandgeneralizestheideaofoperatingprincipleintroducedbyPolyani(1962)andincludesanyideas,includingoperatingprinciples,inventionclaims,designstructures,componentintegrationtricks,tradesecretsandotherdesignknowledgethatleadtoperformanceimprovementofartifacts.AnIOIisdifferentthanaunitofunderstanding(UOU)whichincludesscientificprinciples,andfactualinformation.Anexampleofaunitofunderstanding(UOU)istheprincipleoftotalinternalreflection,whichdescribeshowabeamoflightundergoesreflectioninsideadensemedium,whentheangleofincidenceisaboveacriticalvalue(seeFig.3).Thisprincipleaccuratelydescribesanaturaleffect,butitdoesnotprescribehowwecanuseittotransmitinformation.Ontheotherhand,apairofparallelsurfaces(orafiber)enclosingadensemediumandutilizingtheprincipleoftotalinternalreflectionprovidesamechanism–anoperatingprinciple‐tomakearayoflighttraveldownthelengthofthemedium(seeFig.3).SuchamechanismisanexampleofanIOI.Unlikeartifacts,whichbelongtoaspecifictechnologicaldomain,wemodelIOIintheideas(IOI)poolasbeingnon‐domainspecificandavailabletoalltechnologicaldomains.Forinstance,theoperatingprincipleoftotalinternalreflectionisutilizedinfiberoptictelecommunications,fluorescentmicroscopy,andfingerprinting,verydistincttechnologicaldomains.Intheideasub‐regime,designers/inventorssourceexistingideas(IOI)usinganalogicaltransferandcombinethemprobabilisticallytocreatenewideas(IOI).OncenewIOIaresuccessfullycreatedthroughprobabilisticcombination,theybecomepartoftheIOIpool,thusenlargingthenumberofideas(IOI)inthepoolforcombination.Itisimportanttoclarifythatmodelconsiderscombinationsattheideaslevelrathercombinationofcomponents,withtheformerbeingfundamentalandallowingcombinationofideasfromdifferentfieldsusinganalogicaltransfer.

Page 14: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

14

Fig.2:ModelofexchangebetweenUnderstandingandOperationsregimesandmodulationofIOIassimilationbyinteraction(dJ)andscaling(AJ)parametersofdomainJ.Exampleofunitofunderstanding(UOU)

Exampleofincrementaloperatingidea(IOI)

Principleoftotalinternalreflection

Totalinternalreflectionbetweenparallelsurfacesenclosingdensemedium:mechanismtomakelighttravellongitudinally(fiberoptics)

Fig.3Examplesofunitofunderstanding(UOU)andincrementaloperatingidea(IOI)

Understanding

regime

IOIpoolIOI C,K

Operationsregime

DomainJ

Interactions

Perf.ScalingAJ

Interactions

QJKJ

FU dJ

Artifact

Page 15: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

15

WemodelgrowthintheexplanatoryreachoftheUnderstandingregimebysimulatingasimilarcombinatorialanalogicaltransferprocess.TheUnderstandingregimeisconceptualizedtoconsistofunitsofunderstanding(UOU).Theunitsofunderstanding(UOU)fromdifferentfieldswithintheunderstandingregimeparticipatetocreateanewunitofunderstanding(UOU)thatpotentially(probabilistically)hasagreaterlevelofexplanatoryandpredictivepower.FollowingthetreatmentinAxtelletal.(2013),wemodeltheexplanatoryandpredictivepowerofafieldofUnderstandingasafitnessparameter,fi.IfthenewUOUhasagreaterfitnessvalue,itreplacestheUOUwiththesmallestfitnessvalue.Sinceourprimaryfocusisonperformance‐theoutputoftheOperationsregime,wesimulatetheUnderstandingregimeonlyatthishigherabstractionlevel.Althoughbothregimes–UnderstandingandOperations–evolveindependently,theycannotdosoindefinitely.WemodelthedeSollaPriceandGribbininsightsbyhavingeachregimeactasa“barrier‐breaker”fortheotherregime.Wheneachregimehitsabarrier,theothercaneventuallyaidinbreakingthebarrier:infusionofunderstandingenablescreationofimportantIOIintheOperationsregime;andinfusionofnewoperationaltoolsenablenewdiscoveriesintheUnderstandingregime.Theperformancesoftheartifactsintechnologicaldomainsareimprovedbyaseriesofdesigns/inventions(IOI)overtime.IOIenabledesignerstochangespecificcomponentsinthedomainartifactleadingtoapotentialimprovement.FollowingMcNerneyetal.’streatment,theIOIinquestionisassimilatedonlyiftheperformanceoftheartifactoverallimproves.Another,andfinal,factorthatwemodelisscaling,apropertyinherentinthephysicsofthedesignoftheartifact.5,6ThesuccessfullyassimilatedIOI,whichwerefertoasIOIS,effectimprovementofthedomainartifactbyenablingfavorablechangeofarelevantdesignparameter.Thedesignparameterisincreasedordecreasedsuchthatitleadstoimprovedperformance7.Scalingreferstohowchangeinadesignparameterrelatestorelativechangeintheperformanceofanartifact.Theformulationweuseinthemodelisthatrelativeperformancechangeisrelatedtodesignparametersraisedtosomepower,inotherwordsscaled.Ascoveredinsection2.1,thisisthemostwidelyusedfunctionalrelationshipwithdecentempiricalsupportandtheoreticaljustificationinsomecases(Barenblatt1996).

5Recallthattheperformanceweconsiderinthispaperisintensive,e.g.,energydensity,w/cm3.6Inrelationstoartifactssuchassoftware,physicsreferstothemathematicsbehindthesoftware.7Taguchi(1992)notedthatsomephenomenatendtoworkbetterwhencarriedoutatasmallerscale(“smallerisbetter”),whileotherarebetteratlargerscale(“largerisbetter”).Integratedcircuits,forexample,performbetterasdimensionsarereduced,sincesmallerdimensionsleadtoshorterdelays,andhigherdensityoftransistors,bothofwhichcontributetowardsimprovedcomputationpervolumeorcost.

Page 16: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

16

3.2 Mathematical summary Aperformance(intensive)metricofadomain,labeledQJ,isafunctionofasetofdesignparameters(s1,s2,s3)ofadomainartifactandtimebutforsimplicityhereweconsideronlyasingleparameter(s).ThedesignparameterischangedbyIOIs(successfullyassimilatedIOIintodomainartifacts),whichinturnareassimilatedfromIOIC(numberofaccumulatedoperatingideasintheIOIpoolshowninFigure2).IOICisafunctionoftime.Equationsdescribingthesenestedvariablesinlogarithmicformare: lnQJ=f1(lns);lns=f2(lnIOISC);lnIOISC=f3(lnIOIC);lnIOIC=f4(t) (3) Assumingthatthefunctionsarecontinuousandalldependenceisthroughthenamedvariables,thechainruleisappliedandyields dlnQJ/dt=dlnQJ/dlns∙dlns/dlnIOISC∙dlnIOISC/dlnIOIC∙dlnIOIC/dt (4) Thefirsttermontherighthandsiderepresentsrelativeimpactofdesignvariablechangeonperformancechange,whichwillbeshowninsection4.5tobeequaltothescalingparameter(AJ)whenQJfollowsapowerlawins:dlnQJ/dlns=AJ.Thesecondtermisthe‘smaller‐is‐better/larger‐is‐better’factor,andcapturesthenotionwhetheradesignvariablehastobeincreasedordecreasedinordertoimproveperformance.Wecapturethisdependenceusinganabstractionandequatedlns/dlnIOIsc=+/‐1. Thus,equation(4)becomes

dlnQJ/dt=AJ∙(±1)∙dlnIOISC/dlnIOIC∙dlnIOIC/dt

(5)

Thethirdtermontherightofequation(5)represents‘difficultyofimplementingideas’inspecificdomains,andthusrelatesthedomainspecificsuccessfulIOISCtotheIOICinthepool:wewillshowinsection4.4‐followingMcNerneyetal.‐thatdlnIOISC/dlnIOIC=1/dJ,wheredJistheinteractionparameterintroducedbyMcNerneyetal.Finally,thefourthtermrepresentstherateofideaproduction.K=dlnIOIC/dtisarrivedatbyasimulationofcombinatorialanalogicaltransferwhichispresentedinthefirst(following)sectionoftheresults.

4. Results 

4.1 Overall IOI simulation  Asnotedinsection3.1,wemodeltheIOIasresultingfromcombiningknowledgefrompriorIOIbyprobabilisticanalogicaltransfer.Fig4aschematicallyrepresentscombinationofIOI,inwhichspecificIOIaandbcombinetocreateIOIdwithaprobability,PIOI.Ifthiscombinationattemptsucceeds,thenewlycreatedIOIdthenisaddedtothepoolofIOI(Fig4b).Insubsequenttimesteps,IOIdcanattempttocombinewithanotherspecificIOIinthepool,suchasIOIc,toprobabilisticallycreateamoreadvancedIOIe.Ascombinationadvances,thecumulativenumberofindividualoperatingideas,IOICgrows.WefurthermakethedistinctionbetweenderivedIOIandbasicIOI,whichwelabelasIOI0.IOI0are

Page 17: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

17

fundamentalIOI,whichfirstintroduceanaturaleffectintoanoperationalprincipletoachievesomepurpose.Theexample(describedinsection3.1)ofapairofcloseparallelsurfaces(orafiber)enclosingadensemediumandutilizingprincipleoftotalinternalreflectiontotransmitabeamoflightlongitudinallycanbeviewedasanexampleofanIOI0.Incontrast,derivedIOI,justasthetermsuggests,areobtainedthroughcombinationoftwoIOI0,orbetweenanIOI0andaderivedIOIorbetweentwoderivedIOI.Inthissense,IOIa,b,andcinthefigurerepresentIOI0andIOIdande,derivedIOI.

Inonerunofthesimulation,westartwiththeinitialnumberofbasicindividualoperatingideas,IOI0.Ateachtimestep,themaximumnumberofcombinationsweallowtobecreatedisequaltohalfthenumberoftotalIOIavailable.Theintentionistoalloweachoperatingideatocombinewithanotheroperatingideaoncepertimesteponaverage.Figure5showsresultsfromasimulationrunstartingwith10basicIOIandaprobabilityofcombination,PIOI,equalto0.25.Figures5aand5bwithtimestepsontheX‐axisandthecumulativenumberofoperatingideas,IOIContheY‐axisshowthatthecumulativenumberofoperatingideas,IOIC,growsexponentiallywithtimeatanimprovementrate(K)of0.116.

Forthissimplifiedcase,therateofgrowthofIOI,K,canbemathematicallyshowntobeequaltoln(1+PIOI/2),=0.118whichcanbeeasilyderivedasfollows:

Atinatimestept,numberofIOInewlycreated=PIOI∙IOIC(t)/2 (6)

Fig.4:Combinationofindividualoperatingideasa)basicandderivedIOIb)accumulationofIOIthroughfeedback

a+b d

PIOI

IOIpool BasicIOI:a,b,cDerivedIOI:d,e

Page 18: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

18

   IOIC     IOIC 

 

Fig.5:GrowthofIOICovertime:initialIOI0=10,probabilityofcombination,PIOI=0.25:(a)linearY‐axis(b)logarithmicY‐axis.

IOIC(t+1)=IOIC(t)+PIOI∙IOIC(t)/2=IOIC(t)∙(1+PIOI/2) (7)

RatioofIOICbetweenconsecutivetimesteps,r=IOIC(t+1)/IOIC(t)=(1+PIOI/2) (8)

Then,ingeneral,IOIc(t)canbewrittenintermsofaninitialIOI0andratio,randtimestep,t;theexpressioncanbestatedinanexponentialform.

IOIC(t)=IOI0rt=IOI0exp{lnr∙t}=IOI0∙exp{ln(1+PIOI/2)∙t}=IOI0∙exp{k∙t} (9)

Where,therateofgrowthofIOIC(t),

K=ln(1+PIOI/2) (10)

ForverysmallvaluesofPIOI,

K≈PIOI/2 (11)

Thesimulationresultstothispointassumethatindefinitelylargenumbersofoperatingideas,IOI,canbecreatedoutoffewbasicIOI.ThisisbecausethemodelassumesthatthesameoperatingideascanberepeatedlyusedtocreatenewIOIwithoutlimit.(Forexample,recombining(a,b)witha,thenwithbwouldgivenewoperatingIOI(((a,b),a),b)andeventuallyanarbitrarilylargenumberofa,bpairs.IndefinitemultipleusesofthesamebasicideatocreateinnumerableIOIdoesnotappeartoberealistic.Inordertobetterreflectthisintuition,weintroduceaconstraintthatanyderivedIOIcanutilizeanIOI0onlyonce.TheconstraintoperationalizesthenotionthatcountingrepetitioususeofbasicIOIasnewdesignsthatpotentiallyimproveperformanceisunrealistic.Accordingtothis

0

200

400

600

800

0 20 40

IOI0 = 10

1.E+00

1.E+01

1.E+02

1.E+03

0 20 40

IOI0 = 10

Page 19: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

19

constraint,derivedIOI((a,b),c)inFigure4wouldbeallowed,butnot((a,b),b).Employingthisconstraint,thesimulationyieldstheresultsinFig.6a,asemi‐loggraph,showingthecumulativenumberofIOIinitiallygrowingexponentiallywithtime.However,lateronthecurvebendsoverandhitsalimit,demonstratingthatallcombinationpossibilitieshavebeenusedup,andthepoolofoperatingideasstagnateswhichisalsoshownonthelinearplot(Figure6b)resemblingawell‐known“Scurve”.

    IOIC 

a)  

   IOIC 

b) 

Fig.6:GrowthofcumulativeIOIC(t)afterimplementingtheconstraintthatIOI0canbeusedonlyoncebyanyspecificderivedIOIs;a)semi‐logplotandb)linearplot.

Themaximumnumberofcombinationpossibilities,whichisafunctionofIOI0inthepool,definesthelimit.Thislimit,ormaximumnumberofcombinationpossibilities,isgivenbyasimplecombinatoricsequation(Cameron1995):

2 1 (12)

Equation12entailsthatthelimitincreasesrapidlyasIOI0increases,duetoitsgeometricdependenceonIOI0.Forexample,forIOI0equalto5,10,15,and20thecorrespondinglimitsare31,1023(Figure6),32767,and1,048575combinationpossibilities.

AnaturalquestionthatarisesfromthisresultiswhatmightdeterminetheIOI0overtime?WepostulatearoleforUnderstandinginthisregardandwefirstbrieflylookathowUnderstandingevolvesovertime.

1.E+00

1.E+01

1.E+02

1.E+03

1.E+04

0 20 40 60 80

IOI0 = 10

0

200

400

600

800

1000

1200

0 20 40 60 80

IOI0 = 10

Page 20: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

20

4.2Combinatoric simulations for Understanding regime JustliketheOperationsregime,wemodeltheUnderstandingregimetoalsogrowthroughaprobabilisticanalogicaltransferprocess,inwhichunitsofunderstandingcombinetocreatenewunitsofunderstanding.Inthismodel,weenvisionthattheUnderstandingregimeiscomposedofmanyfields,witheachfieldhavinganexplanatoryreach.UsingatreatmentsimilartotheoneusedbyAxtelletal.(2013),theexplanatoryreachofafieldmaybeviewedasafitnessvalueofthetheoreticalunderstandingofthatfield,whichwedenotewithfi.FollowingAxtelletal.,whenunitsfromtwofieldswithfitnessvalues,f1andf2,combine,thefitnessoftheresultingunitisrandomlychosenfromatriangulardistributionwiththebaseorX‐axisdenotingthefitnessvaluesrangingfrom0tof1+f2,andtheapexrepresentingthemaximumvalueoftheprobabilitydistributionfunction,givenby2/(f1+f2).SeeFig7a.Iftheresultingfitnessofthenewunderstandingunitishigherthanthefitnessofeitherofthetwocombiningunits,thenewunderstandingunitreplacestheunitwhosefitnessisthesmallestamongthethree.WeassumethecumulativefitnessoftheUnderstandingregime(FU)asawholetobeequaltothesumoftheindividualfitnessvalueofeachfield.Oursimulationassumes10fieldswithstartingfitnessvaluesrangingfrom0to1,whicharerandomlyassigned.Consequently,theaveragecumulativefitness(FU)valueisinitially5.Asthesimulationproceeds,fitnessvaluesofthe10fieldsgrowindependently,andasaresult,thecumulativefitnessoftheUnderstandingregimegrows.Fig.7bshowsresultsfromasimulationrunexhibitingroughlyexponentialgrowthofcumulativefitnessovertime.Thus,asimplemodelforgrowthoftheUnderstandingregimeisalsoexponential.However,aswiththeOperationsregime,unlimitedgrowthbysimplecombinationofscientifictheoriesisnotrealistic.TheUnderstandingregimealsocannotprogressbysimplecombinationofexistingunderstandingbutinsteadexperiencesalimitthatweenvisionasdependinguponavailabilityofoperational(technological)toolsavailablefortestingscientifichypothesesandfordiscoveringneweffects.Weexpressthisdependencethroughanequationwhichexpressesthemaximumcumulativefitnessatanytime,maxFU(t),assimplyproportionaltotheIOIexistingatthattime:maxFU(t)=ZF∙IOIC(t) (13)WhereIOICthusrepresentsanapproximationfortheeffectivenessofavailableoperationaltools,andZFisaconstantofproportionality.ThisequationcapturestheconceptfirstsuggestedbyPricethattheextent(orscope)ofexplanatoryreachoftheUnderstandingregimeisdependentuponwhatexperimentaltoolsareavailableforscientistsandresearchers.Italsorecognizesinthetermsofourmodelthatthesetoolsareessentiallyoperationalartifacts.

Page 21: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

21

a)

b)

Fig. 7: a) Triangular distribution of possible fitness values that can be assumed by a new unit of understanding b) Growth of FU (cumulative fitness of Understanding regime) over time.

 

4.3 Exchanges between Understanding and Operations regimes  Asdiscussedinsection3.1,priorqualitativeworkindicatesthattheinteractionofUnderstandingandOperationsisprobablybestmodeledbyassumingmutualbeneficialinteraction.Inourmodel,wecapturethisenablingexchangefromtheUnderstandingtotheOperationsregimeusingasimplemathematicalcriterion:FU(t)/FU(t_prev)≥cutoff_ratio(R) (14)Where,FU(t)andFU(t_prev)representcumulativefitnessvaluesattimesteptandthemostrecenttimestep,t_prev,atwhichaIOI0hadbeenintroduced.ThiscriterionstatesthatwhencumulativefitnessoftheUnderstandingregimegrowsbysomemultiple(R)fromthetimewhenthelastIOI0wasinvented,understandinghasimprovedenoughtogenerateanewIOI0,whichbecomesavailableforcombinationswithallexistingIOI.Thethresholdratio,R,determinesthefrequencyatwhichIOI0arecreated.

WenowshowresultsfromasimulationincludingtheexchangeandlimitsonIOI0.Inthesimulation,westudyhowsynergisticexchangefromUnderstandinginfluencestherateofgrowthofIOIintheOperationsregime,includingescapefromstagnation.Wefocusparticularlyontwovariables,namely,theinitialnumberofIOI0intheOperationsregime

pdf

a b c

0

FU

Page 22: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

22

andthethresholdratioRforcreationofnewIOI0.Otherpertinentvariablesaretheprobabilityofcombination,PIOI,thenumberofattemptspertimestepandthenumberoftimestepsperyearandarenotvariedinthissetofresults.Forthissimulationstudy,Table(1)presentstheparametervaluesforIOI0(column3)andthethresholdratiosofcumulativefitness(column4)thatareused.Asanexample,5B3RstartswithIOI0of5andanewIOI0iscreatedwhencumulativefitnessgrowsbyafactorof3.BoththeinitialnumberofIOI0andthethresholdratiosofcumulativefitnessaresetat3differentvalues,givingatotalsetof9parametercombinations.Forall9runs,theprobabilityforcombinationiskeptconstantat0.25,andweassumeoneattemptperyearlytimestep.

Table1:Simulationstudy:ParametervaluesofIOI0 andR (thresholdratiosofcumulativefitnessofUnderstanding)forthestudy.Results:KistheslopefittingthesimulationresultstoanexponentialwithR2forthefit(alsoshown).Otherparameters,suchasprobabilityofcombination,PIOI=0.25,arekeptconstant. Simulation

RunInitialIOI0

ThresholdratioR

Simulationavg.K(±2stddev)8

R2 K =ln(1+PIOI/2)

1 5B1.5R 5 1.5 0.123(±0.011) 0.998 0.1182 5B3R 5 3.0 0.055(±0.019) 0.959 0.1183 5B5R 5 5.0 0.039(±0.007) 0.943 0.1184 10B1.5R 10 1.5 0.122(±0.011) 0.997 0.1185 10B3R 10 3.0 0.115(±0.007) 0.998 0.1186 10B5R 10 5.0 0.117(±0.007) 0.983 0.1187 20B1.5R 20 1.5 0.116 (±0.007) 0.998 0.1188 20B3R 20 3.0 0.116(±0.009) 0.998 0.1189 20B5R 20 5.0 0.119(±0.016) 0.998 0.118

8Thestandarddeviationwasestimatedfromsevenrepetitionsforeachsimulationrun.

Page 23: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

23

 

 

Fig. 8: Growth of IOIc; initial IOI0 and R (cumulative fitness ratio) for each run are shown in the legend 

for each run; e.g., 10B5R represents 10 IOI0 and fitness ratio of 5. 

ThesimulationresultsinFig.8showsthetemporalgrowthofIOICintheOperationsregimefortheninerunsshowninTable1.Runs5B3Rand5B5Rclearlystandout:theyhaveabumpygrowthsincetheyencounterperiodsofstagnationmultipletimes,astheyevolve.Moreover,theireffectiveratesofgrowtharemeager,standingonlyat0.055and0.04,whichismuchlowerthan0.118,therategivenbyEquation10{ln(1+PIOI/2)}.Columns5,6,and7listtheK,R2,andKcalculatedusingln(1+PIOI/2)respectively.Thesmalldeviations

1

10

100

1000

10000

0 20 40 60 80 100 120 140 160 180

IOIC

Time

5B1.5R 5B3R 5B5R

10B1.5R 10B3R 10B5R

20B1.5R 20B3R 20B5R

Page 24: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

24

fromequation10foundfortheother7runsarewithinthe2‐sigmaestimatedfrommultiplesimulationrepetitionsforeachrun.Both5B3Rand5B5RstartwithlowinitialIOI0of5andhavehighercumulativefitnessthresholdratios(R)forinfusionofnewIOI0.LowinitialIOI0impliesthattheOperationsregimehasalownumberofcombinatorialpossibilitiesofIOItostartwith.Additionally,sincenewIOI0arenotcomingfastenoughtopushthefrontierofcombinatorialpossibilitiesofIOIfarenough,theOperationsregimequicklyexhauststhepossibilitiesandagainstagnates.Run5B5Rstagnatesforlongerperiodscomparedto5B3Rsinceithasahigherthresholdratio(R)forinfusionofanewIOI0andthusslowerprogress.TheOperationsregimecannotescapethestagnationuntilanotherIOI0iscreatedwithinfusionofnewunderstanding.Itisclearfromthecurvesthatthispatternrepeatsitselftimeaftertime.Othersimulationruns,exceptrun10B5Rgrowexponentiallyandsmoothlyandtheirratesareconsistentwiththetheoreticalvaluecalculatedusingln(1+PIOI/2),0.1178.ThesecurveshaveeitherhighenoughIOI0tostartwithorfastinfusionofIOI0,orboth.Run5B1.5R,forexample,startswithalownumberofIOI0buthasfastinfusionofIOI0,sincethethresholdratioRisonly1.5.Ontheotherhand,run20B5RhasslowinfusionofIOI0(highR),butstartswithhighinitialIOI0.Theserunsdonotexhibitstagnationfortworeasons.ThefirstreasonisthatthefrontierofcombinatorialpossibilitiesforsomerunsisveryfarfromthenumberofrealizedIOIatagiventimestep.Forexample,run20B5Rhasoveramillionpossibilitieswhenitstartswith20IOI0.ThesecondreasonisthatthefrontierofthecombinatorialpossibilitieskeepsonmovingfurtherawayasIOIcincreases.Run5B1.5R,forexample,startswith5IOI0,andyetitneverexperiencesstagnationduetofastinfusionofIOI0(lowR)thatpushthefrontierofcombinatorialpossibilities.ThegrowthofIOICisalsofreeofstagnationforruns(e.g.,suchasRun10B3R)withmediumnumberofinitialIOI0andmediumrateofinfusionofIOI0(mediumR).Thisistruebecausebothfactorsincombinationensurethatfrontierofcombinatorialpossibilitiesisfarenoughtostartwith,andthefrontiercontinuestomoverapidlyenoughwithtime.Run10B5Rexhibitssomewhatunusualbehavior.Althoughitgrowssmoothlyatthebeginningforquitesometime,itexperiencesstagnationlateron.Thisisbecausethefrontierofcombinatorialpossibilitiesisfarenoughawaytosustainsteadygrowthearlyon.Later,theOperationsregimeexhauststhecombinatorialpossibilitiesbeforenewIOI0arrive.However,onceanewIOI0arrives,itjumpstartsagainbutitbrieflyhaltsateachnewlimitdemonstratingthevalueoffrequentinterchangebetweenUnderstandingandOperationsinthissimulation9.9ThesimulationsarebaseduponinfusionofIOI0dependinguponaratio(R)ofgrowthincumulativeunderstanding,butsimilarresultsarefoundwithassumingamodelofdifferenceinFU.

Page 25: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

25

WehaveseenthatacombinatorialprocesscombinedwithsynergisticexchangebetweenUnderstandingandOperationsleadstoanexponentiallygrowingpoolofoperatingideas,IOIC.Thisgrowthisdescribedbyanexponentialfunction:

exp (15A)

(15B)Where,K=theeffectiverateofgrowthofIOIC,IOI0(t0)=thenumberofinitialbasicIOI,t=time,t0=initialtime.Ouroverallmodel(Section3,Figure2)envisagesthatthisexponentiallygrowingpoolofoperatingideas,IOIC,providesthesourcefortheexponentialgrowthofperformanceoftechnologicaldomains.HowdoesthisexponentialgrowthofIOICresultinperformanceimprovementandwhataccountsforthevariationinratesofperformanceimprovementacrosstechnologicaldomains?

4.4 Modeling interaction differences among domains Asexplainedinsection3,twofactorspotentiallyresponsibleformodulatingtheexponentialgrowthofoperatingideasastheyareintegratedintotechnologicaldomainsarethedomaininteractionsandscalingofrelevantdesignvariables.WeconsiderdomaininteractionsfirstfollowingtheworkofMcNerneyetal.(2011)whomodeledhowinteractionsinprocessesaffectunitcost.WebuildontheirmathematicaltreatmenttoanalyzetheeffectofinteractionsbetweencomponentsuponintegratinganIOIintoanartifactinadomain,whichinturnimprovestheartifact’sperformance.Figure9ashowsasimplifiedschematicofanartifactinatechnologicaldomainthathasthreecomponents(1,2,3)withinteractionbeingdepictedbyout‐goingarrows,representinginfluence,fromacomponenttoothercomponents,includingitself.Theoutgoingarrowsarereferredtoasout‐links.Thenumberofout‐links,d,fromacomponentprovidesameasureofitsinteractionlevel,andhasvalueof1orgreaterasMcNerneyetal.assumeeachcomponentatleastaffectsitself.Forsimplicity,Figure9ashowseachcomponentwithtwoout‐links,toitselfandtoanothercomponent.Werepresentaninstanceofanattemptbeingmadetoimprovetheperformanceofcomponent2byanIOIbeinginserted.Sincecomponent2interactswithitselfandanothercomponent,theperformanceoftheinteractingcomponentisalsochangedbytheinsertionbutinafashiondescribedprobabilistically.Theperformanceimprovementattemptisaccepted,onlyiftheperformanceoftheartifactasawholeimproves.Ifthatdoesoccur,wefollowMcNerneyetal.andconsidertheinteractionsbeingsuccessfullyresolvedtoimprovetheperformance.Forasimplifiedartifactwithdnumberofout‐linksforeachcomponent(d=2inFig.9a),McNerneyet.al.’streatment(2011)forunitcostresultsinthefollowingrelationship:dC/dm=‐B∙Cd+1 (16)

Page 26: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

26

Where,C=unitcostnormalizedwithrespecttoinitialcost10,m=numberofattempts,d=numberofout‐links,B=constantThisequationstatesthatthelevelofinteractioninherentinthedomainartifactinfluencestherateofunitcostreduction.Weadaptthisequationforouranalysisinthefollowingmanner.WeinterpretthenumberofattemptsasIOIcsincethenumberofIOIdeterminestheattempts(ateachattemptanIOIisbeingintroducedintoanartifacttomakeadesignchange).Secondly,costreductionisinverselyrelatedtoperformanceimprovement,suchasinatypicalmetrickWh/$.11WiththeseextensionsofMcNerneyetal.equation16canbere‐writtenas:d(Q)/dIOIc=B∙Q‐(d‐1) (17)Where,Q=performance                  

a) 

                  

b)  

Fig. 9: Interactions in an artifact; a) illustration of interactions as out‐links b) sample space of probabilities for unit cost . 

SinceasshowninEquations4and5,successfullyresolvedoperatingideasinadomain,IOISC,arethesourceforitsperformanceimprovement,wereplaceperformanceQofadomainwithIOISC.AnIOIisconsideredasuccessfulattemptiftheinteractionresolution

10Thenormalizedunitcostis1orlesssoincreasesindinequation16resultinlessimprovementperattempt.11Theconceptcanbefurthergeneralizedtoincludeperformancemetricswhichinvolveotherresourceconstraintssuchasvolume,mass,andtime,insteadofcost(e.g.,kWh/m3).

n=3;d=2n=#ofcomponentsd=out‐links(interactions)

d C 

(1,1)

1 2

3

IOI

C1+C2 = C1(t) + C2(t) = C

C1

C

C0

C2

Page 27: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

27

leadstonetperformanceimprovementoftheartifact,andthecountofsuccessfulIOIisdenotedbyIOISC.Themodifiedequationshownbelowstatesthattheinteractionlevel,d,hasaretardingeffectonthegrowthofIOISCinadomain.d(IOISc)/dIOIc=B∙IOISc‐(d‐1) (18)Wesolvethedifferentialequationbyseparatingthevariables(IOISContheleftandIOIContheright),andintegratingbothsidesusingdummyvariables,andexpressIOISCexplicitly.Theintegrationlimitsare:a)fortherightside,0toIOIC,b)fortheleftside,1toIOISC.Theresultis:

∙ ∙ 1 / (19)SinceBanddareclosetounity,andIOIc>>1,wecanignore1inthebrackets.Sinceourgoalistodetermine{dlnIOISC/dlnIOIC},wetakethenaturallogofbothsidesanddifferentiateitwithrespecttolnIOIC,resultinginthefollowingexpressionwhichwillbesubstitutedintoequation5insection4.6:dlnIOIsc/dlnIOIc=1/dJ (20)

4.5 Performance models ‐ scaling of design variables Ourresearchquestionisconcernedwithintensivetechnologicalperformanceofdomainartifacts.Theintensivetechnologicalperformancerepresentsaninnateperformancecharacteristicofanartifact.Weoperationalizethenotionofintensiveperformancebydividingdesirableartifactoutputswithresourceconstraints(e.g.,mass,volume,time,cost).Anintensiveperformancemetricforbatteriesisenergydensity,kWh/m3.Wenowconsiderthreeexamplesofrelationshipsbetweenintensiveperformanceanddesignvariables.

4.5.1 Selected examples Wefirstconsiderblastfurnacesusedinthemanufacturingofsteelasrepresentativeofreactionvesselsofvariouskinds.Widelyusedperformanceattributesforablastfurnacearecapacityandcost,wherecostcanbeconsideredtheresourceconstraint.So,anintensiveperformancemetriccanbedefinedascapacity(outputperhourordaytypically)perunitcost.Thecapacityofareactionvesselisproportionaltoitsvolumewhileitscostisprimarilyproportionaltosurfacearea(Lipseyetal.2005).Thefollowingdimensionalanalysisshowsthatfollowingthesesimplisticassumptions,intensiveperformanceofareactionvesselislinearlyproportionaltosize,s.QRV=capacity/costofreactionvessel=s3/s2=s1 (21)Gold(1974)hasempiricallyshownthatthecostofablastfurnacegoesupby60percentwhenthecapacityisdoubled.IntensiveperformanceQRVusingthisempiricalfindinggoesupby1.25(=2/1.6)whens3doubles,andthussgoesupby1.26(=2.333)closelyagreeingwiththesimplyderivedequation21.

Page 28: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

28

Asecondexampleweconsiderisspecificpoweroutputfrominternalcombustion(andotherheat)engines.Poweroutput(kW)isproportionaltovolumeoccupiedbythecombustionchamberminustheheatlossfromtheengine,whichinturnisproportionaltotheengine’ssurfacearea.Thepower,then,is:power=As3–Bs2;B/A<1 (22)WhereAandBareconstantsforpowergenerationandheatlossrespectively.QIC=specificpowerαpower/volumeofengine;thusspecificpoweris=(As3–Bs2)/s3=A–B/s (23)Equation23indicatesthat,similartoreactionvessels,specificpoweroutputofICenginesincreaseswithsizesobothare“largerisbetter”artifacts”.ForsmallvaluesofB/As,specificpowerincreasesapproximatelylinearlywiths.Forlargervaluesofs,theincreaseislessthanlinearins.Asafinalexample,weconsiderinformationtechnologies,whoseperformanceimprovementranksamongstthehighest.Severalmoderninformationtechnologiesdependuponintegratedcircuit(IC)chips.ElectroniccomputershavebeenimprovingperformancebyreducingthefeaturesizesoftransistorsinICchipsformicroprocessors.Thenumberofcomputationspersecondperunitvolume,anintensivemeasureofperformance,dependsuponfrequencyandthenumberoftransistorsinaunitvolume.Frequencyisinverselyproportionaltothelineardimensionofafeature,s,andthenumberoftransistorsperunitareaisinverselyproportionaltoareaofthefeature.Thus,Computationpersecpercc=1/s∙1/s2=s‐3 (24)Thedimensionalanalysisindicatesthatcomputationspersecondincreasesrapidlyforadecreaseinalineardimensionofafeature.Thisisduetothecubic(orhigher)12dependenceofcomputationspersecondonfeaturesize.Thenegativesigncapturesthefactthatreductionofthedesignvariableincreasesperformance–smallerisbetterforthisartifact.

4.5.2 Generalization of scaling of design variables Thethreeexampleswehavepresentedillustratethenotionthatintensiveperformanceimprovedbydifferentdegreesdependinghowthedesignvariablesarescaled.Inthefirsttwocases,a10percentincreaseinadesignvariablewillimproveperformanceby10percentorless.However,inthecaseofcomputations,forthesame10percentchangeindesignvariable(featuresize),theperformancewouldimprovebyover33percent.Thisdependenceismodeledasapower‐law13:

12Iftheverticaldimensionalsodecreasesovertimeasthefeaturesizedecreases,ahigherpower‐perhapsapproaching4‐wouldapply.13Theengineexampledemonstratesthatthisisanapproximationinmanycases.

Page 29: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

29

(25) lnQJ=AJlns (26)

dlnQJ/dlns=AJ (27)

Where,AJisthescalingfactorfordomainJ,sisthedesignvariable. 

4.6 Bringing all elements together WenowbringtheresultsforrateofIOISCgrowthandinfluenceofinteractionandscalingtogether.Forthereader’sconvenience,wereproduceequation4here,andsubstitutetheresultsforthefourfactors: dlnQJ/dt=dlnQJ/dlns∙dlns/dlnIOISC∙dlnIOISC/dlnIOIC∙dlnIOIC/dt (4)Substitutingtheresultsfromequations27,20,and15Bforthefirst,thirdandfourthterms,±1forthesecondterm,andthenrearranging,weget:

dln

∓1 1

(28)

Equation28representstheoverallmodeloftheannualrateofimprovementfordomainJ.Accordingtothisequation,KJ,theannualrateofimprovementofdomainJdependsuponK,theexponentialrateatwhichtheIOICpoolincreasesinsize.Kisthenmodulatedbydomainspecificparameters,dJ(interaction)inverselyandAJ(scaling)proportionallytoresultinadomainspecificrateof improvementKJ.TheminussignisconvertedintopositiveonebynegativesignofAJ(forthosecaseswheresmallerisbetter).OneobservationtonoteisthatAJand dJ are constants for a given domain, thus resulting in a time invariant rate (or asimpleexponential)foradomain.

5. Discussion 

Thegoalofthispaperwastodevelopamathematicalmodelthatutilizesmechanismsinthedesign/inventionprocesstoexaminethenatureoftechnologicalperformanceimprovementtrends.TheexplorationhasutilizedsimulationtogaininsightintoacombinatorialprocessbaseduponanalogicaltransferandUnderstanding/Operationsexchangeandquantitativelymodeledinteractionsandscaling.Inthissection,wefirstbrieflyreviewtheconsistenciesofthemodelwithempiricalresults(andwhatisknownabouttechnologicalchange).Allempiricalresultsweareawareofarefoundtosupportthemodel.Wethenconsidertheasyetuntestedpredictionsfromthemodelaswellastheassumptionsmadeinthemodel.Accordingtothemodel,theexponentialnatureofperformanceimprovementforalltechnologicaldomainsarisesintheidearealmoftheoperationalknowledgeregime,where

Page 30: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

30

newinventiveideasarecreatedusingcombinatorialanalogicaltransferofexistingideas,which,inturn,becomethebuildingblocksforfutureinventiveideas.Weemphasizethatthecombinationsmodeledareoccurringattheidealevel,althoughcombinationscanalsotakeplacebetweencomponents.Asnotedinsection3.1,wemakethisdistinctionastheformerismuchmorepervasiveandallowscombinationofideasfromdifferentfields;however,itislikelythatsomeideascannotbecombinedandthisistreatedprobabilisticallysincemanycombinationattemptsfail.ThemodeldemonstratesthisincessantcumulativecombinatorialaspectofknowledgeinboththeUnderstandingandtheOperationsregimesmanifestsasexponentialtrends.ThecombinatorialmodelissimplebutitleadsnaturallytotheexponentialbehaviorwithtimethathasonlybeenobtainedpreviouslybyAxtelletal.inamodelthatwentbeyondperformancetodiffusionoverasetofagents.Sincesuchexponentialbehaviorwithtimeisoneofthemostwidelynotedbehaviorsoftechnicalperformance(Moore1965,KohandMagee2006,2008,Nagyetal.2013,Mageeet.a.2014),thecombinatoricmodelenactinganalogicaltransferthatwasdevelopedinthecurrentpaperisclearlysupportedbywhatisknownempiricallyaboutperformancetrendswithtime.TheOperationsandtheUnderstandingregimescanimproveindependentlyinthemodelbutnotindefinitely.HowlongtheOperationsregimecanimprovedependsinthemodeluponthesizeofthetechnologicalpossibilityspace,whichaccordingtothemodelisdependentonthenumberofbasicIOI,fundamentaloperationalprinciples,existing.TheUnderstandingregimecanalsoexperiencestagnation,butthishappenswhentheoperationaltoolsthatscientistsandresearchersusefordiscoveryandtestinghypothesesarenotadequate.TheOperationsregimecomestoitsrescuebyprovidingtheseoperationaltoolsinformofempiricalmethods,toolsandinstruments(increasednumbersofindividualoperatingideas),whichgreatlyenhancesthescientistsabilitytodiscoverandtest,andthusfurtherpushthelimitsofunderstandinginthemannersuggestedbyPrice(1983),Gribbin(2002)andinthefollowingquotefromToynbee(1962).

Physical Science and Industrialism may be conceived as a pair of dancers both of whom know their steps and have an ear for the rhythm of the music. If the partner who has been leading chooses to change parts and to follow instead there is perhaps no reason to expect that he will dance less correctly than before.

Inthissense,theOperationsregimeandtheUnderstandingregimeareliketwoindependentneighborswhointeractformutualbenefit.Inthemodel,theirfrequencyofinteractionhoweverinfluencestheireffectiverateofgrowth.Ourmodelisaspecificrealizationthatachievesthismutualinteractionthathaspreviouslybeenwidelynotedfromdeepqualitativeresearch.TheresultsinFigure8aresummarizedasasurfaceplotinFigure10.K,theeffectiverateofgrowthofIOICwasdeterminedbytheinitialIOI0,andthefrequencyofinteraction(α1/lnR).Theformerdeterminedtheenvelopeoftechnologicalpossibilityspace.WhenIOI0arehigh,theeffectiverateofgrowthKisclosetothetheoreticalcombinatorialratedeterminedbyEquation10{ln(1+PIOI/2)},irrespectiveofwhethertherewasfrequentexchange.However,whentheIOI0arelow,thelimitishitrepeatedly,translatinginto

Page 31: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

31

haltingandareducedeffectiverateofgrowth.ThevalueofKinthiscasewasdeterminedbythefrequencyofenablingexchangefromtheUnderstandingregime,withhigherfrequency(lowR)leadingtohighereffectiverate.Withsufficientlyhighfrequency,evenwithlowinitialIOI0,theeffectiverateKeventuallyapproachesthetheoreticalrate.

Page 32: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

32

Fig.10:VariationofKasafunctionofinitialIOI0andR.LowerRreferstohigherfrequencyofinteractionwiththeUnderstandingregime.Detailedhistoricalstudiesoftechnologicalchange(Mokyr2002)notecenturiesofslow,haltingprogressthateventuallybecomesmuchmorerapidandsustainedstartinginthelate18thcenturyintheUK.Aninterestingconsistencyoftheseobservationswithourmodelisseensinceourmodelattributesthetransitiontosustainedhigherimprovementratetothecombinatorialgrowthofindividualideasthatareabletoreinforceoneanotherbytheanalogicaltransfermechanism.ThatourmodelpartiallyaccomplishesthisthroughthesynergisticexchangebetweenUnderstandingandOperationsisalsoconsistentwiththedetailedhistoricalstudiesasinterpretedbymanyobservers(Schofield1963,Musson1972,RosenbergandBirdzell1986,MussonandRobinson1989,Mokyr2002,Lipseyetal.2005).TheKJvaluesfoundempiricallyvarybyapproximatelyafactorof22(from0.03to0.65accordingtoMageeetal.(2014).Equation28statesthatannualimprovementrateforadomainisdeterminedbytheproductofKtimesthescalingparameter,AJ,andthereciprocaloftheinteractionparameter,dJ.Accordingtothisresult,thelasttwoparametersproducethevariationofimprovementratesacrossdomains.Duringtheembodimentprocess,interactionsprevalentinthedomainartifactsinfluencehowmanyinventiveideascanbeabsorbed.ThepercentincreaseinsuccessfullyabsorbedideasbyadomainartifactisinverselyproportionaltotheaverageinteractionparameterofthedomaindJ.By

5

10

150

0.04

0.08

0.12

0.16

1.5

3

5

K

R

0.12‐0.16

0.08‐0.12

0.04‐0.08

0‐0.04

Page 33: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

33

definition,theminimumvalueofdis1andthemaximummightbehigherbutavalueof6appearsreasonable.Theotherfactorthatispredictedtodifferentiatedomainsisperformancescaling.Inventiveideasaffectartifactperformancebymodifyingthedesignparametersindomainartifacts.ThemodelindicatesthattherelativeimprovementofperformanceforagivennumberofabsorbednewoperatingideasisgovernedbythescalingparameterAJ.Theexamplespresentedinsection4.5illustratedthatthevalueofAJcanvaryacrossdomains.Inparticular,fortheICdomain(wheresmallerisbetter),AJisapparently3to4timeslargerthanfortypicallarger‐is‐betterdomainssuchascombustionengines.Thus,therangeofKJempiricallyobservedispotentiallyexplainablebychangesindJandAJ,butmuchmoreempiricalworkisneededtofullysupportthesequantitativeimplicationsofEquation28aswillbediscussedfurtherbelow.TheempiricalfindingsofBensonandMagee(2015a)alsosupportthemodel.Inparticular,theyfoundnocorrelationofratesindomainswitheffortinadomain(measuredbynumberofpatentsorpatentingrate)orwiththeamountofoutsideknowledgeusedbyadomain(thisisverylargeforalldomains).Theyinterpretedtheirfindingsbya“risingseametaphor”thatrepresentsallinventionsandscientificoutputbeingequallyavailabletoalldomainsbutthatfundamentalsinthedomainsdeterminetherateofperformanceimprovement.OveralleffortinUnderstanding(science)andinventionincreasetheratesinalldomainsbutthedifferencesamongratesofimprovementareduetodifferencesinfundamentalcharacteristicsamongthedomains.Themodelinthispaperidentifiesinteractionsandscalingastwosuchfundamentalsandequation28isspecificaboutthevariationexpectedduetothesetwofundamentalcharacteristics.Thus,ourmodelissupportedbywhatisknownempiricallyincludingexponentialdependenceofperformanceontime;slow,haltingprogressintheearlystagesoftechnologicaldevelopment;aroleforscienceinenablingtechnologicalperformanceimprovement;therangeofvariationinperformanceimprovementacrossdomains;andtheimportanceofdomainfundamentalstovariationinperformance.However,towhatextentdoesitachievetheideallevelofunderstandingmentionedinsection2whendiscussingtherelatedBensonandMageeresearch?Itis‐asdesired‐baseduponwhatisknownaboutthedesign/inventiveprocessanddoesnotrelyuponcharacteristicsonlydeterminedbyobservationofoutputinadomain.Moreover,itprovidesexplanationsofexistingempiricalresultsnotmadebypriormodels.However,doesitmakeanynewpredictions;doitsassumptionsappearreasonable;andwhatnewavenuesofdesignresearch,ifany,doesitopenupforfurtherexploration?Weconsidertheseissuesintheremainderofthediscussion.TherearethreenewpredictionsmadebythemodelasinstantiatedinEquation28.Theseare:1)thatthenoiseinestimatingKJshouldvarywithKJlinearlyratherthanforexamplebeindependentofKJ;2)thatperformanceimprovementcomparisonsacrossdomainsvaryas1/dJwheredistheinteractionparameter;and3)thatperformanceimprovementacrossdomainsvaryasAJ.ThefirstpredictionfollowsfromthefactthatthemodelascribesallvariationintheprocesstotheprobabilisticanalogicaltransferprocessthatcreatesIOIandthusanynoisegeneratedintheprocessisamplifiedbythesamefactorsthatdetermineKJ(namely1/dJandAJ.).Veryrecentworkappearstoconfirmthefirstprediction.Inacareful

Page 34: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

34

studyoftheobservednoiseinawidevarietyofdomains,FarmerandLafondhavefindthatthevariationinKJisproportionaltoKJofferingempiricalsupporttotheformofEquation28.ThisispotentiallyanimportantconfirmationofapredictionofthemodelbutthecarefulworkbyFarmerandLafondhaspotentialdatalimitations(detailedintheirpaper)andfurtherworkofthiskindishighlydesirable.Prediction2isthatcomponentinteractions(dJ),whichcharacterizethedomains,influenceimprovementratebymodulatingtheimplementationofIOIinthedomainartifacts.ThispredictioncanbetestedbystudyoftheperformanceimprovementratesoveravarietyofdomainswhereanindependentassessmentofdJismade.Theauthorshaveperformedsuchatestusingpatentdata(BasnetandMagee2016)andtheresults,whichdemonstratepositivecorrelationbetweenimprovementrates(KJ)andinteractionparameter(dJ),offersupportfortheanalysisofMcNerneyetal.thatweuseinourmodel.Prediction3isthatrelativeimprovementamongdomainsvariesproportionallytothescalingparameterforthedomaindesignparameters,aconsequenceofperformancefollowingapowerlawwiththedesignparameters.Ifscalinglawswerefound(orderived)foravarietyofdomainswhoserateofprogressisknown,prediction3canalsobetested.Inthispaper,weshowedthatthefactorAisatleast3timeslargerforIntegratedcircuitsthanforcombustionengines.WhilethisprovidespreliminarysupportforthemodelsinceIntegratedcircuitsimproveabout7timesfasterthancombustionengines(Mageeetal,2014),twopointsdonotachievearigoroustest.Onewouldneedtohavereliablescalingfactorsforatleast10domainswithvaryingKJtodeterminewhetherthispartofthemodelisempiricallysupported.Afundamentalaspectoftheoverallmodelisthatitdifferentiatesbetweentheidea/knowledgeandartifactaspectsofdesignandinvention.Suchdecompositionisanessentialstepinarrivingatourkeyresult(equation28throughequation5).Itisnotclearthatthisassumptionistestablesoitmustremainanunverifiedassumptionordefinitionbutwedonotethatitappearstoaccordwithrealityinthatinventors/designersspendsignificantamountoftimeworkingwithideasandrepresentationsofartifacts,forexampleintheformofsketchesanddrawings,wellbeforetheybuildartifacts.Othershavenotedthehigherleverageofanalogicaltransferbetweenideasasopposedtodesignedartifacts(Weisberg2006).Apotentiallyimportantandnon‐obviousassumptionmadeinthemodelisthatinventiveeffortincreasesasthecumulativenumberofindividualoperatingideas‐IOIC‐increases.ThisassumptionisintroducedwhenweassumethateveryexistingIOIundergoesacombinationattemptineachtimestep.AsIOICincreases,thismeansthatmoreinventionsareattemptedineachsuccessivetimestep.ThisassumptioniscriticaltoobtainingtheexponentialtimedependenceforIOICandthusforQbecausethegrowthofIOICwouldbechokedoffifinventiveattemptsdidnotincreaseovertime.Althougharigoroustestofthisassumptionissuggestedforfurtherwork,wedonotesupportfortheassumptionintheexponentialgrowthofpatentsovertime(Younetal.2014,PackalenandBhattachayra,2015)14.ApproximatesupportisalsogivenbytheroughlyexponentialgrowthofR&D

14Bothofthesepapersshowmorerapidexponentialincreasesbefore1870andslowerbutstillexponentialincreasesovertimefrom1870tothepresentinthenumberofUSpatents.

Page 35: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

35

spendingovertime(NSF,2014)andbytheroughlyexponentialgrowthofgraduateengineersglobally15overtime(NSF,2014)ThemodelassumesasimpleexchangebetweenUnderstanding(largelyscience)andOperations(largelytechnology)asdescribedbyEquations13and14.Thedetailsofthismechanismarenottestablebutinouropinionnotcriticalbecauseotherformalisms(basedupondifferencesratherthanratiosandbaseduponcountofunitsofunderstandingratherthanourchoiceofexplanatoryreach)leadtoresultscloselysimilartothosereportedhere.Therefore,thisassumptionremainsunverifiedbutisnotcriticaltoourconclusions.Similarly,theinitialvalueofIOI0choseninthesimulation(andtheexchangefrequencywithUnderstanding(α1/lnR))isessentialtoourfindingofhaltingslowgrowththatcantransitiontosustainedandmorerapidgrowth.Althoughthisfindingisconsistentwithdetailedobservationasnotedaboveandtheinitialnumberofusefulideasmustbesmall,thereisnoindependentmeansofassessingIOI0.Moreover,wehavemadeanumberofassumptionsinparametervaluestoconstructasimpleandoperationalsimulation.Thevaluesforparametersinthesimulation,suchasPIOI,numberoftimesteps,numberofscientificfields,R,fitnessvaluesarechosentokeepthecomputationalcostreasonable,withoutsacrificingtheessentialaspects.Simulationsshowthatresultsarerobusttodifferentcombinationsofparametervalueswithrespecttoexponentialtrendsandvariationinrates.Therefore,thesechoicesandsimplificationsdonotundercuttheexplanatoryorpredictivecapabilitiesofthemodelbutdolimitthepotentialfornon‐calibratedcalculationof,forexample,theimprovementrateforadomainsinceKisonlyapproximatelyknown.Tomakethemodeltractable,wehavemadenumberofsimplifyingabstractions,introducingseveralotherlimitationstothemodel.Sincethemodelisnotagent‐based,itdoesnotdistinguishbetweenorganizationsnorbetweeninventors.Sinceourgoalistoexplainthepatternsatthedomainlevel,weconsiderthedomainasoneentity.Forthisreason,variationsamongorganizationsoramonginventorswithinadomainarenottakenintoaccount,andhencethemodelisnotusefultounderstandorganizationalorindividualinventoreffectivenessinitscurrentformandanysystematicdifferencesamonginventorcapabilityacrossdomainsisignored.Second,onceIOIarecreatedbyanyinventor,themodelassumestheyareinstantlyavailableforcombinatorialanalogicaltransferacrossthepoolunderlyingalldomains.Thus,themodeldoesnottakeintoaccounttimedelaythatcanresultdueto,forexample,geography,secrecyandgovernmentalregulations,andhenceisnotusefulforstudyingsuchfactors’influenceintechnologicalchange.Third,themodelassumesthat2pre‐existingideasaresufficient(probabilistically)tocreateanotherideawhereasinventionsalsoresultfrombringingmorethan2pre‐existingideastogether.However,addingsuchcomplicationstothemodelandsimulationdoesnotchangethefundamentalfindingssincethecreationofnewideaswouldstillincreaseasthenumberofpre‐existingideasincreaseaslongaswestillassumeanincreasinginventioneffort.Fourth,althoughconceptuallythenotionoffitnessofscientificfieldsmakessense,howthe

15OthersupportingevidenceisalsopossibletoseeintheNSFmaterialathttp://www.nsf.gov/statistics/seind14/index.cfm/overview/c0s1.htm#s2

Page 36: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

36

fitnesscanbemeasured,andwhomeasuresitforascientificfieldarecontested,especiallyforrapidlygrowingfields.ThisanalysisofthepredictionspointsoutthatsomekeyaspectsofEquation28havethepotentialtobeempiricallytestedandthusareclearfutureresearchactivitiessuggestedbythemodel.Amongthesefutureresearchactivities,oneimportantissuetodiscussistheextensionspossibletodesignresearchpotentiallyopenedupbythecurrentwork.Themodelinthispaperexplicitlyconsidersdesignchangesinsucceedingartifactsinaseriestobethecentralelementintechnologicalchangeovertime.Thus,itaddstothefewotherpapers(BaldwinandClark2006,Luoetal.2014)thathaveconnectedthesetwolargefieldsofresearch‐technologicalchangeanddesigntheory.Thispaperinparticularconnectsdesignconceptuallyandquantitativelytochangesinperformanceovertime.Sincethereissignificantdataofthistype(Moore,1965,Girifalco1991,Nordhaus1996,KohandMagee(2006,2008)andLeinhard2008),thispaperpointsthewayforfurtherquantitativecomparisonsofmodelsbasedupondesigntheorywithdata.Anotherlineofresearchthatthismodelsuggestsismoreexplicitconsiderationofinteractionsandscalingaspartofdesigntheories.Thecurrentmodelexploressimplemodelsforbothofthesethatarecapableofpredictingdifferencesintimedependenceofperformanceindifferingdomains.Designofartifactscouldconceptuallybechangedsothatthepotentialforimprovementwithongoingredesignisenhancedpossiblythroughreducedinteractionsormoreintensivescalingrelationships.Thus,thecurrentpapersuggeststhepotentialimportanceoffurtherresearchonspecificdifferencesindesignapproacheswithdifferentscalinglawsandwithdifferentlevelofinteractions.

6. Concluding remarks Themodelandsimulationsoftheimprovementsinperformanceduetoaseriesofinventions(newdesigns)overtimepresentedinthisworkarebaseduponasimpleversionofanalogicaltransferasacombinatorialprocessamongpre‐existingoperational/inventiveideas.Themodelissupportedbyanumberofempiricallyknownaspectsoftechnologicalchangeincluding:1. Thetransitionfromslow,hesitanttechnologicalchangetomoresustainedtechnological

progressastechnologicalideasaccumulate;2. Arolefortheemergenceofthescientificprocessinstimulatingthetransitioninpoint1;3. Theexponentialincreaseofperformancewithtime(generalizedMoore’sLaw)seen

quitewidelyempirically;4. Thatstochasticnoiseintheslopesofthelogperformancevs.timecurvesis

proportionaltotheslope;5. Thelevelofeffortindomainsisnotimportantintherateofprogress.

Themodelalsoindicatesthat:6. Therateofperformanceincreaseinatechnologicaldomainisatleastpartly(and

possiblylargely)duetofundamentaltechnicalreasons(componentinteractionsand

Page 37: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

37

scalingofdesignvariables),ratherthancontextualreasons(suchasinvestmentinR&D,scientificandengineeringtalent,ororganizationalaspects).

Numerousmodelingassumptionsweremadeindevelopingthemodelbutonlysomeofthesearecriticaltotheconclusionsjustlisted.Furtherspecificresearchissuggestedtomovesomecriticalassumptionsintothetestablecategory,andtoconsiderinteractionsandscalingparametersinnewdesignapproaches.Thesearediscussedinthepaperparticularlyfortheassumptionsunderlyingpoint6above.Thetestsinvolvedetailedstudiesoftheinteractionandscalingparametersinavarietyofdomains.Allofthisfutureresearchcouldsupportorleadtomodificationofpoint6.

Acknowledgement 

TheauthorsaregratefultotheInternationalDesignCenterofMITandtheSingaporeUniversityofTechnologyandDesign(SUTD)foritsgeneroussupportofthisresearch.WewouldalsoliketothankDr.JamesMcNerneyforhelpfuldiscussionaboutartifactinteractions.WewanttoalsoacknowledgevaluableinputonanearlierversionofthispaperbyDr.JamesMcNerneyandDr.Daniel.E.Whitney.

 

Page 38: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

38

Reference 

Acemoglu, D. (2002). Directed Technical Change. TheReviewofEconomicStudies, 69(4),781–809.http://doi.org/10.2307/1556722

Arrow, K. J. (1962). The economic implications of learning by doing’. The Review ofEconomicStudies,29(3).

Arthur, W. B. (2007). The structure of invention. Research Policy, 36(2), 274–287.http://doi.org/10.1016/j.respol.2006.11.005

Arthur,W.B.,&Polak,W. (2006). TheEvolution of Technologywith a Simple ComputerModel.Complexity,11(5),23–31.http://doi.org/10.1002/cplx

Auerswald, P., Kau, S., Lobo, H., & Shell, K. (2000). The production recipes approach tomodeling technological innovation : An application to learning by doing. JournalofEconomicDynamics&Control,24,389–450.

Axtell, R. L., Casstevens, R., Hendrey, M., Kennedy, W., & Litsch, W. (2013). CompetitiveInnovation and theEmergence ofTechnologicalEpochsClassification : Social SciencesShort title : Competitive Innovation Author contributions : Retrieved fromhttp://www.css.gmu.edu/~axtell/Rob/Research/Pages/Technology_files/TechEpochs.pdf

Baker,N.R.,Siegman J.,R.A.H. (1967).TheEffectsofPerceivedNeedsandMeanson theGeneration of Ideas for Industrial Research and Development Projects. IEEETransactionsonEngineeringManagement,(December).

Baldwin,C.Y.,&Clark,K.B.(2006).Between“Knowledge”and“TheEconomy”:TheNotesontheScientificStudyofDesigns.InB.Kahin&D.Foray(Eds.),AdvancingKnowledgeandTheKnowledgeEconomy(pp.298–328).Cambridge,MA:TheMITPress.

Baldwin,CarlissY.,Clark,K.B. (2000).DesignRules:ThePowerofModularity. Cambridge,MA:MITPress.

Barenblatt, G. I. (1996). Scaling,Self‐similarity,andIntermediateAsymptotics:DimensionalAnalysis and Intermediate Asymptotics. New York, New York, USA: CambridgeUniversityPress.

Basnet, S., & Magee, C. L. (2016). Dependence of technological improvement on artifactinteractions.Retrievedfromhttp://arxiv.org/abs/1601.02677

Benson, C. L., & Magee, C. L. (2015a). Quantitative Determination of TechnologicalImprovement from Patent Data. PloS One, (April).http://doi.org/DOI:10.1371/journal.pone.0121635April15,2015

Benson,C.L.,&Magee,C.L.(2015b).Technologystructuralimplicationsfromtheextensionof a patent search method. Scientometrics, 102(3), 1965–1985.http://doi.org/10.1007/s11192‐014‐1493‐2

Braha, D., & Reich, Y. (2003). Topological structures for modeling engineering designprocesses. Research in Engineering Design, 14(4), 185–199.http://doi.org/10.1007/s00163‐003‐0035‐3

Cameron, P. J. (1995). Combinatorics:Topics,Techniques,Algorithms (1st ed.). New York,

Page 39: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

39

NewYork,USA:CambridgeUniversityPress.

Carter, C.F. andWilliams, B.R. (1959). Carter,C.F.andWilliams,B.R.,1959. Investment inInnovation.(London:OxfordUniversityPress.

Carter, C.F.,Williams,B. R. (1957). IndustryandTechnicalProgress:FactorsGoverningtheSpeedofApplicationofSciencetoIndustry.London:OxfordUniversityPress.

Christensen, B. T., & Schunn, C. D. (2007). The relationship of analogical distance toanalogicalfunctionandpreinventivestructure:thecaseofengineeringdesign.Memory&Cognition,35(1),29–38.http://doi.org/10.3758/BF03195939

Christensen, C.M.,&Bower, J. L. (1996). CustomerPower, Strategic Investment, and theFailure of Leading Firms. Strategic Management Journal, 17(3), 197–218.http://doi.org/10.1002/(SICI)1097‐0266(199603)17:3<197::AID‐SMJ804>3.0.CO;2‐U

Clement,C.a,Mawby,R.,&Giles,D.E.(1994).TheEffectsofManifestRelationalSimilarityon Analog Retrieval. Journal of Memory and Language.http://doi.org/10.1006/jmla.1994.1019

Dahl,D.W.,&Moreau,P. (2002).The InfluenceandValueofAnalogicalThinkingDuringNewProductIdeation.JournalofMarketingResearch,39(1),47–60.

Dasgupta,S.(1996).CreativityandTechnology.OxfordUniversityPress.

deSollaPrice,D.J.(1986).Sealingwaxandstring.InLittleScience,BigScienceandbeyond.NewYork,NewYork,USA:ColumbiaUniversityPress.

Dosi, G. (1982). Technological paradigms and technological trajectories. ResearchPolicy,11(3),147–162.http://doi.org/10.1016/0048‐7333(82)90016‐6

Farmer,J.D.,&Lafond,F.(2015).Howpredictableistechnologicalprogress ?Retrievedfromhttp://arxiv.org/abs/1502.05274

Fehrenbacker,K. (2012).WecanthankMoore’sLawfor theVCcleantechbust.Retrievedfrom http://gigaom.com/2012/02/01/we‐can‐thank‐moores‐law‐for‐the‐vc‐cleantech‐bust/

Finke, R. A.,Ward, T. B., & Smith, S.M. (1996).CreativeCognition:Theory,Research,andApplications.Cambridge,MA:MITPress.

Fleming,L.(2001).RecombinantUncertaintyinTechnologicalSearch.ManagementScience,47(1),117–132.http://doi.org/10.1287/mnsc.47.1.117.10671

Fleming, L., & Sorenson, O. (2004). Science as a map in technological search. StrategicManagementJournal,25(89),909–928.http://doi.org/10.1002/smj.384

Frischknecht, B., Gonzalez, R., Papalambros, P. Y., & Reid, T. (2009). A design scienceapproachtoanalyticalproductdesign.InternationalConferenceonEngineeringDesign,DesignSociety,PaloAlto,CA,(August),35–46.

Fu,K.,Chan,J.,Cagan,J.,Kotovsky,K.,Schunn,C.,&Wood,K.(2013).TheMeaningof“Near”and “Far”:The Impact of StructuringDesignDatabases and theEffect ofDistanceofAnalogy on Design Output. Journal of Mechanical Design, 135(2), 021007.http://doi.org/10.1115/1.4023158

Gentner, D., & Markman, A. B. (1997). Structure mapping in analogy and similarity.

Page 40: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

40

AmericanPsychologist,52(1),45–56.

Gero, J. S., & Kannengiesser, U. (2004). The situated function‐behaviour‐structureframework. Design Studies, 25(4), 373–391.http://doi.org/10.1016/j.destud.2003.10.010

Girifalco. (1991). Dynamics of Technological Change. New York, New York, USA: VanNostrandReinhold.

Goel,A.K.(1997).Design,analogy,andcreativity.IEEEExpert,12(3).

Gold,B.,The,S.,Economics, I.,&Sep,N.(1974).EvaluatingScaleEconomies :TheCaseofJapaneseBlastFurnaces.TheJournalofIndustrialEconomics,23(1),1–18.

Gribbin, J. (2002).TheScientists:AHistoryofScienceToldThroughtheLivesofItsGreatestInventors.NewYork,NewYork,USA:RandomHouse.

Hatchuel, A.,&Weil, B. (2009). C‐Kdesign theory:An advanced formulation.ResearchinEngineeringDesign,19(4),181–192.http://doi.org/10.1007/s00163‐008‐0043‐4

Henderson,R.M.,&Clark,K.B. (1990).Architectural Innovation :TheReconfigurationofExistingProductTech‐nologies and theFailureofEstablishedFirms.AdministrativeScienceQuarterly,35(1),9–30.

Holyoak, K. J., & Thagard, P. R. (1995). Mental Leaps: Analogy in Creative Thought.Cambridge,MA:MITPress.

Hunt, B. J. (2010). PursuingPower and Light. Baltimore, MD: Johns Hopkins UniversityPress.

Klevorick, A. K., Levin, R. C., Nelson, R. R., & Winter, S. G. (1995). On the sources andsignificance of interindustry differences in technological opportunities. ResearchPolicy,24(2),185–205.http://doi.org/10.1016/0048‐7333(93)00762‐I

Koestler,A.(1964).TheActofCreation.London:Hutchinson&Co.

Koh,H.,&Magee,C.L.(2006).Afunctionalapproachforstudyingtechnologicalprogress:Application to information technology. TechnologicalForecastingandSocialChange,73(9),1061–1083.http://doi.org/10.1016/j.techfore.2006.06.001

Koh,H.,&Magee,C.L.(2008a).Afunctionalapproachforstudyingtechnologicalprogress :Extension to energy technology☆. TechnologicalForecastingandSocialChange, 75,735–758.http://doi.org/10.1016/j.techfore.2007.05.007

Koh,H.,&Magee,C.L.(2008b).Afunctionalapproachforstudyingtechnologicalprogress:Extension to energy technology. TechnologicalForecastingandSocialChange, 75(6),735–758.http://doi.org/10.1016/j.techfore.2007.05.007

LangrishJ.,GibbonsM.,EvansW.G.,Jevons,F.R.(1972).WealthfromKnowledge:AStudyofInnovationinIndustry.NewYork,NewYork,USA:Halsted/JohnWiley.

Leclercq, P., & Heylighen, A. (2002). Analogies Per Hour. In J. S. Gero (Ed.), ArtificialIntelligenceinDesign’02(pp.285–303).Dordrecht:KluwarAcademicPublishers.

Lienhard, J. H. (2008). How Invention Begins: Echoes of Old Voices in the Rise of NewMachines.NewYork,NewYork,USA:TheOxfordUniversityPress,UK.

Page 41: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

41

Linsey, J. S., Markman, A. B., & Wood, K. L. (2012). Design by Analogy: A Study of theWordTree Method for Problem Re‐Representation. Journal of Mechanical Design,134(4).

Linsey,J.S.,Wood,K.L.,&Markman,A.B.(2008).Modalityandrepresentationinanalogy.ArtificialIntelligence forEngineeringDesign,AnalysisandManufacturing, 22, 85–100.http://doi.org/10.1017/S0890060408000061

Lipsey, Richard G., Carlaw, Kenneth I., Bekar, C. T. (2006). Economic Transformations:GeneralPurposeTechnologiesandLongTermEconomicGrowth.NewYork,NewYork,USA:TheOxfordUniversityPress.

Luo, J.,Olechowski,A.L.,&Magee,C.L.(2014).Technology‐baseddesignandsustainableeconomic growth. Technovation, 34(11), 663–677.http://doi.org/10.1016/j.technovation.2012.06.005

Magee,C.L.,Basnet,S.,Funk, J.L.,&Benosn,C.L. (2014).Quantitativeempiricaltrendsintechnical performance (No. ESD‐WP‐2014‐22). Cambridge, MA. Retrieved fromhttp://esd.mit.edu/WPS/2014/esd‐wp‐2014‐22.pdf

McNerney,J.,Farmer,J.D.,Redner,S.,&Trancik,J.E.(2011).Roleofdesigncomplexityintechnology improvement. PNAS, 108(38), 9008–9013.http://doi.org/10.1073/pnas.1017298108/‐/DCSupplemental.www.pnas.org/cgi/doi/10.1073/pnas.1017298108

Meyers,S.,Marquis,D..(1969).SuccessfulIndustrialinnovation.Washington,D.C.:NationalScienceFoundation.

Mokyr, J. (2002). The Gifts of Athena: Historical Origins of the Knowledge Economy.Princeton:PrincetonUniversityPress.

Moore, G. E. (1965). Cramming more components onto integrated circuits. Electronics,38(8),1–4.

Mowery,D.,&Rosenberg,N. (1979).The influenceofmarketdemandupon innovation:acritical review of some recent empirical studies. Research Policy, 8(2), 102–153.http://doi.org/10.1016/0048‐7333(79)90019‐2

Musson,A.E.(1972).Science,technologyandeconomicgrowthintheeighteenthcentury.(A.E.Musson,Ed.)(1sted.).Routledge.

Musson, A. E., & Robinson, E. (1989). ScienceandTechnologyintheIndustrialRevolution.GordonandBreachSciencePublishers.

Muth, J.F. (1986).SearchTheoryand theManufacturingProgressFunction.ManagementScience,32(8),948–962.http://doi.org/10.1287/mnsc.32.8.948

Nagy, B., Farmer, J. D., Bui, Q. M., & Trancik, J. E. (2013). Statistical basis for predictingtechnological progress. PloS One, 8(2), e52669.http://doi.org/10.1371/journal.pone.0052669

Nelson, Richard R., Winter, S. G. (1982). An Evolutionary Theory of Economic Change.Cambridge,MA:HarvardUniversityPress.

Nemet,G.,Johnson,E.(2012).Doimportantinventionsbenefitfromknowledgeoriginating

Page 42: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

42

inothertechnologicaldomains?ResearchPolicy,41(1).

Nordhaus,W. D. (1996). Do Real‐Output and Real‐WageMeasures Capture Reality? TheHistory of Lighting Suggests Not. In The Economics of New Goods (pp. 27–70).Retrievedfromhttp://www.nber.org/chapters/c6064.pdf

Polanyi, M. (1962). PersonalKnowledge:Towards aPost‐CriticalPhilosophy. Chicago, IL:UniversityofChicagoPress.

Polya,G.(1945).HowtoSolveIt:ANewAspectofMathematicalMethod(1sted.).Princeton,NJ:PrincetonUniversityPress.

Popper,K.(1959).LogicofScientificDiscovery(1sted.).Hutchinson&Co.

Romer,P.M.(1990).EndogenousTechnologicalChange.JournalofPoliticalEconomy,98(5).

Rosenberg, N. (1982). Inside theBlackBox:Technology andEconomics. Cambridge, MA:CambridgeUniversityPress.

Rosenberg, N., & Birdzell, L. E., J. (1986). How the West Grew Rich: The EconomicTransformationoftheIndustrialWorld.US:BasicBooks.

Ruttan,V.W.(1959).UsherandSchumpeteronInvention,Innovation,andTechnologicalChangeAuthor ( s ): VernonW .RuttanReviewedwork ( s ): Publishedby :OxfordUniversityPress.TheQuarterleyJournalofEconomics,73(4),596–606.

Ruttan, V. W. (2001). Technology, Growth, and Development: An Induced InnovationPerspective.NewYork,NewYork,USA:OxfordUniversityPress.

Sahal, D. (1979). A Theory of Progress Functions. AIIE Transactions, 11(1), 23–29.Retrieved fromhttp://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:A+I+I+E+Transactions#8

Sahal,D.(1985).Technologicalguidepostsandinnovationavenues.ResearchPolicy,14(2),61–82.http://doi.org/10.1016/0048‐7333(85)90015‐0

Schofield,R.(1963).TheLunarSocietyofBirmingham:ASocialHistoryofProvincialScienceandIndustryinEighteenth‐CenturyEngland.Clarendon.

Schumpeter, J. A. (1934).TheTheoryofEconomicDevelopment. Cambridge, MA: HarvardUniversityPress.

Shai,O.,Reich,Y.,&Rubin,D.(2009).Creativeconceptualdesign:Extendingthescopebyinfused design. CAD Computer Aided Design, 41(3), 117–135.http://doi.org/10.1016/j.cad.2007.11.004

Simon, H. A. (1962). The Architecture of Complexity. Proceedings of the AmericanPhilosophicalSociety,26(6),467–482.http://doi.org/10.1016/S0016‐0032(38)92229‐X

Simon,H.A.(1969).TheSciencesoftheArtificial(1sted.).Cambridge,MA:TheMITPress.

Simon,H.A.(1996).TheSciencesoftheArtificial(3rded.).Cambridge,MA:TheMITPress.

Solow, R. M. (1956). A Contribution to the Theory of Economic Growth. TheQuarterlyJournalofEconomics,70(1),65–94.http://doi.org/10.2307/1884513

Page 43: Modeling of technological performance trends design theoryweb.mit.edu/~cmagee/www/documents/46-PerformancemodelingDe… · Massachusetts Institute of Technology, Institute for Data,

43

Suh, N. P. (2001).AxiomaticDesign:AdvancesandApplications (1st ed.). New York, NewYork,USA:TheOxfordUniversityPress,UK.

Taguchi, G. (1992). Taguchi on Robust Technology Development: Bringing QualityEngineeringUpstream.AsmePressSeries.

Toynbee,A. J. (1962). Introduction:TheGenesesofCivilizations. InAStudyofHistory,12Vol.NewYork,NewYork,USA.

Tseng, I., Moss, J., Cagan, J., & Kotovsky, K. (2008). The role of timing and analogicalsimilarity in thestimulationof ideageneration indesign.DesignStudies,29(3),203–221.http://doi.org/10.1016/j.destud.2008.01.003

Tushman,M. L.,&Anderson, P. (1986). TechnologicalDiscontinuities andOrganizationalEnvironmentslifecycles.AdministrativeScienceQuarterly,31,439–465.

Usher,A.P.(1954).AHistoryofMechanicalInventions(1sted.).NewYork,NewYork,USA:BeaconPress,BeaconHill,MA.

Utterback,J.M.(1974).Innovationinindustryandthediffusionoftechnology.Science(NewYork,N.Y.),183(4125),620–626.http://doi.org/10.1126/science.183.4125.620

Vincenti, W. (1990).WhatEngineersKnow,andHowTheyKnow It. Baltimore, MD: JohnHopkinsUniversityPress.

Weber, C., & Deubel, T. (2003). NEW THEORY‐BASED CONCEPTS FOR PDM AND PLMProperty‐DrivenDevelopment/Design(PDD),1–10.

Weisberg,R.W.(2006).Creativity.InCreativity(1sted.,pp.153–2007).Hoboken,NJ:JohnWiley&Sons,Inc.

Whitney,D.E.(1996).WhyMechanicalDesignWillNeverbeLikeVLSIdesign.ResearchinEngineeringDesign,8,125–138.

Whitney, D. E. (2004). Physical limits to modularity. InMITEngineeringSystemDivisionInternal Symposium. Retrieved fromhttps://esd.mit.edu/symposium/pdfs/papers/whitney.pdf

Wright,T.P.(1936).FactorsAffectingtheCostofAirplanes. JournalofAero.Science,122–138.

Yelle, L. E. (2007). The learning curve: historical review and comprehensive survey.DecisionSciences.

Youn,H.,Bettencourt,L.M.a.,Strumsky,D.,&Lobo,J.(2014).InventionasaCombinatorialProcess: Evidence from U.S. Patents. Physics Society, June, 1–22. Retrieved fromhttp://arxiv.org/abs/1406.2938v1