ContentslistsavailableatScienceDirect
Structural Change and Economic Dynamics
j o ur na l ho me p a g e:w w w . e l s e v i e r . c o m / l o c a t e / s c e d
Structural learning: Embedding discoveries and the dynamics of production
夽Antonio Andreoni
∗CentreforScienceTechnologyandInnovationPolicy,InstituteforManufacturing,DepartmentofEngineering,UniversityofCambridge, UnitedKingdom
a rt i c l e i n f o
Articlehistory:
ReceivedJanuary2012
ReceivedinrevisedformJuly2013 AcceptedSeptember2013 Available online xxx
JELclassification:
D20 D83 L23 O33 Keywords:
Productionprocess Learning
Structuraldynamics
a b s t ra c t
Productionandlearningofproductiveknowledgeareprofoundlyintertwinedprocesses astheactivationofeitherprocesstriggerstheother,veryoftenimplyinginterdependent transformations.Thepaperaimstoopenthe‘productionblackbox’byproposingtheana- lyticalmapofproductionasatoolfordisentanglingthesetofinterdependentrelationships amongcapabilities,tasksandmaterials.Theconceptofstructurallearningisintroducedto identifythecontinuousprocessofstructuraladjustmenttriggeredandorientedbyexisting productivestructuresateachpointintime.Structurallearningtrajectoriesallowforthe transformationofstructuralconstraintssuchasbottlenecksandtechnicalimbalancesinto structuralopportunities.Complementarities,similaritiesandindivisibilitiesareessential focusingdevicesforactivatingcompulsivesequencesoftechnologicalchangeaswellas discoveringstructurallyembeddedopportunities.Thepapertheninvestigatesthetension betweenstructureandagencypresentinstructurallearningtrajectories,andexaminesthe formittakesindifferentproductiveorganisations.
© 2013 Elsevier B.V. All rights reserved.
1. Introduction
Productionandlearningofproductiveknowledgeare profoundly intertwined processes as the activation of either process triggers the other, very often implying interdependent transformations. Production theory has conventionallyexplainedproductionprocessesasrelation-
夽 Thispaper developsalineofresearchinitiallypresentedatthe DIMEworkshop“ProductionTheory-Process,TechnologyandOrganisa- tion:TowardsausefulTheoryofProduction”(LEM,ScuolaSuperioreSant’
Anna,Pisa8–9November2010).IamgratefultoPatrizioBianchi,Ha-Joon Chang,MikeGregory,PrueKerr,MichaelLandesmann,EoinO’Sullivanand RobertoScazzieriforcommentsanddiscussions,andtoGuidoBuenstorf forcommentsonanearlierdraftofthispaper.Finally,Iamthankfultotwo anonymousrefereesfortheirvaluablecommentstothemanuscriptand theirconstructivesuggestions.Theusualcaveatapplies.TheCentrefor Science,TechnologyandInnovationPolicygratefullyacknowledgesthe supportoftheGatsbyCharitableFoundation.
∗ Tel.:+441223339738.
E-mailaddress:aa508@cam.ac.uk
shipsbetweencombinations of productivefactors –i.e.
inputquantities– and certain quantitiesofoutputs. By assumingthatproducers‘knowhow’certaininputsmay becombinedandtransformedtoobtaincertainoutputs, productionfunctionsdonotmakeanyexplicitreference tothecapabilitiesneededtoperformrealproductionpro- cesses.Thus, instandard production theory,thereis no productionprocessstrictly speaking(Loasby,1999).Not onlyistheproductionprocesstreatedasablackbox,also thelearningdynamicsoccurringingivenproductionstruc- turesarefundamentallyignored.Indeed,economistsoften treatlearningasacostlessandautomaticprocessfunction- allydependentoncumulativeoutput,time,orinvestment, whosemaineffectistoreduceaverageproductioncosts.
Averyinfluentialattempttocopewiththefundamen- tallimitations of more conventional production models can be found in the capability theory of the firm, an approach that emerged at the intersection of various researchfields,specificallyorganisationalstudies(March andSimon,1993;Penrose,1959;Richardson,1960,1972;
0954-349X/$–seefrontmatter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.strueco.2013.09.003
Teece, 1980; Langlois, 1992;Morroni, 2006; Pitelis and Teece,2009;JacobidesandWinter,2012),andinstitutional and evolutionaryeconomics (Nelsonand Winter, 1982;
Cohen and Levinthal, 1990; Lundvall, 1992; Dosi et al., 2000),andempiricalworkindevelopmenteconomics(Bell, 1982;Lall, 1992).Witha particular focus onthetrans- formationofcognitivecontentsandevolvingcapabilities, thesecontributions have shown how theknowledge of productivepossibilities–i.e.inputcombinations–hasto becomplementedbytheavailabilityofrelevantcapabil- itiesforproductivetasksbeingperformed.Mostnotably evolutionaryeconomicshashighlightedthecomplexcogni- tivedynamicsunderlyinglearningprocesses.Ithasdrawn attentiontothemultifacetednatureofknowledge,itstacit componentsaswellasthecomplexitiesconnectedtoits creation,diffusion,adaptation,adoptionandaccumulation inorganisational‘routines’.
Byintegratingtheabovementionedresearchstreams withstructuraltheoriesdealingwiththecomplex‘archi- tecture of production’, this paper analyses production structures in transformation, examining the embedded constraintsand opportunitieswhich are responsiblefor learning dynamics. From this perspective, learning is understoodasadynamicprocesstriggeredandconstrained byexistingproductionstructures.Thismeansthatproduc- tionstructuressetthestagefor learningdynamics,that is,theypreparehumanmindsfortheintuitivediscovery ofnewproductivepossibilities.Thepaperalsorecognises thatstructuresofcognitionandstructuresofproduction arelinkedbyabundleofbidirectionaltransformativerela- tionships.
Thegoalofthepresentpaperistwo-fold.Firstly,the paperembedsdifferentformsoflearningsuchas‘learning by doing’ and ‘learning by using’ in production struc- tures. The paper therefore proposesan ‘analytical map of production’ as a stylised representation of the sys- temofinterrelatedtasksthroughwhichtransformations of materials are performed according to different pat- terns of capacities/capabilities coordination, subject to certainscaleandtimeconstraints(Section2).Withinthis newanalyticalframework,thesecondcontributionofthe paperistointroducetheconceptof‘structurallearning’.
Inconventionalapproaches learningissimplydescribed asacognitive/behaviouraldynamicinvolvingproduction agents.Incontrast,inouranalyticalframework,learning isunderstoodasaprocessthroughwhich‘structuralcon- straints’inproductionsuchasbottlenecksandtechnical imbalancesaretransformedinto‘structuralopportunities’.
In this context, static and dynamic complementarities, as well as similarities and indivisibilities, are essential focusingdevices for triggeringcompulsivesequencesof technologicalchangewhichpermitthediscoveryofnew
‘worldsofproduction’(Section3).Productivepossibilities havetobe‘seen’,thatisdiscoveredand‘actualised’bypro- ductiveorganisations,forstructurallearningtobefeasible.
Theconceptofstructurallearninghighlightsafundamen- talanalytical tension betweenstructure and agencyor, morespecifically,betweenproductivestructuresandpro- ductiveagents(thelatterincludingbothindividualsand collectivities).Giventhesameproductivestructures,struc- turallearningmayfollowdifferentpatternsaccordingto
differentformsofproductiveorganisation(Section4).The analytical accountof specifichistorical casesis adopted as main heuristic for disentangling structural learning dynamics.
2. Embeddinglearninginproductiondynamics 2.1. Learninginproduction:ataxonomy
Intheircriticalreviewoflearningcurvestudies,1Adler andClark(1991,p.270)proposedafundamentaldistinc- tionbetweenfirst-orderandsecond-orderlearning.
First-orderlearningreferstothose‘learningbydoing’
processesdirectlyexperiencedbyworkersviarepetition ofproductivetasksandtheresultingincrementaldevel- opmentofexpertise.Here,learningisbothanindividual and collective process as interactions among workers within the firm are integral parts of their learning by doing. The concept of ‘learning by doing’ expressed in KennethArrow’s(1962)seminalcontributioncapturesthe Smithian intuition that the accumulation of production experienceincreasesworkers’productivity.Inparticular, Smithmentionsthree‘differentcircumstances’responsi- bleforthisincreaseinlabourproductivity:‘theincreaseof dexterityineveryparticularworkman’,‘thesavingofthe timewhichiscommonlylostinpassingfromonespecies ofworktoanother’,and‘theinventionofagreatnumber ofmachineswhichfacilitateand abridgelabour’(Smith, 1976[1776],p.17).
Conventional learning models based on ‘learning by doing’ and learning curves have been mainly used for explainingproductivitygrowthatthesectoralandmacro level (Malerba,1992,p.846;Thompson,2010).Inthese models, production is treated as a timeless black box andheroicassumptionsaremadeconcerningproducers’
knowledgeoftheentirespectrumofproductionpossibil- itiesaswellastheavailabilityofappropriateproductive capabilities.2Onthecontrary,astheliteratureonlocalised technicalchange(AtkinsonandStiglitz,1969)hasshown, given thelocal andcumulative characterof knowledge, producersareonlyawareofa limitednumberoffactors compositionlaws–i.e.proximateproductionpossibilities.
Moreover,asshowninthecapabilityliterature,production
“hastobeundertakenbyhumanorganisationsembodying specificallyappropriateexperienceandskills”(Richardson, 1972,p.888).3
Second-order learning refers to those managerial or engineering actions purposefully aimed at changing the internalstructureofproductionbyintroducingnewtech- nologies,newequipmentsorinvestinginworkerstraining.
Learningdynamicsofthissecondkindtendtobetriggered byaseriesoffactorswhicharebothinternalandexternal tothefirm(Malerba,1992).Interms oftheformer, not
1Thelongtraditioninlearningcurvestudiesisusuallyassociatedwith theempiricalanalysisof‘learningbydoing’effectsonproductivityand wasinitiatedbyWright(1936)andhisworkintheaircraftindustry.
2ThestochasticmodelbyJovanovicandNyarko(1995)isanexception inprovidingamicrofoundationofArrow’s...‘learningbydoing’.
3Theanalyticalandtechnicallimitationsoftheproductionfunction modelsarediscussedinGeorgescu-Roegen(1970),Scazzieri(1993).
onlymaythefirmspecificallyinvestinsearchactivities andproduction/technologyresearchaimedatexpanding itsknowledgebase(NelsonandWinter,1982)butitmay alsoattempttoincreaseitslearningandabsorptivecapaci- tiesthemselves(Stiglitz,1988;CohenandLevinthal,1990).
Asforthelatter,insomecases,triggersoflearningdynam- icsareexternaltothefirmandmayinvolveusersoffinal goods(Rosenberg,1982;Rheeetal.,1984),otherproducers ofintermediateorfinalgoodsinthesameorindifferent industries(Lundvall,1992)orpossiblyotheractors,typi- callythoseinvolvedinscientificandtechnologicalresearch (KlineandRosenberg,1986).
Rosenberg’s(1982,p.122)conceptof‘learningbyusing’
arises from the recognition that “in an economy with complexnewtechnologies,thereareessentialaspectsof learningthatareafunctionnotoftheexperienceinvolved in producing the product but of its utilisation by final users...theperformancecharacteristicsofadurablecap- italgoodoftencannotbeunderstooduntilafterprolonged experience with it”.4 The related ideas of ‘learning by exporting’(Rheeetal.,1984)and‘learningbyinteracting’
(Lundvall,1992)withupstreamanddownstreamproduc- ersdevelopRosenberg’sfundamentalintuitionand,thus, maybeconsideredassub-categoriesof‘learningbyusing’.
InLundvall’s(1992)frameworklearningbyinteractingis acriticalfeatureofsocieties.Capabilitiesarecollectively developedthroughsocialinteractionsmainlybyobserv- ingandimitatingothers’actionsaswellasbymirroring theirattitudes.Thisiswhytheorganisationaldesignofpro- ductionprocessesaswellasfirms’underlyingrelational structurescanaffectpeople’sdispositiontowardsmutual learning andknowledge discovery.Historically,learning by interacting has taken various forms from more co- operativetomore competitiveonessuchas learningby exportingand,thus,thoughupgradingproductscharacter- istics,butalsoimportingandcopyingmachines,recruiting foreignskilledworkersandtechnicianexchange,poolingof technology,organisationofexpos(industrialexhibitions) andindustrialespionage(Chang,2002;Poni,2009).
In all these cases, learning dynamics are initially triggeredbyfactorsexternalorinternaltothefirm that eventually result in the reconfiguration of the firm’s internal production structure. Of course this reconfigu- ration mayor maynot happen dependingon how the firminquestionreactstotheinternalorexternalstimuli.
Alloftheabove suggeststhatin ordertoanalysethese compulsivesequencesoftransformationitisnecessaryto embedlearningdynamicsinproductionstructuresandto understandinwhichwaysthesedynamicsareconstrained, butalsotriggered,byexistingproductionstructures.Fig.1 provides a taxonomyof the differentforms of learning reviewedabove.
2.2. Theanalyticalmapofproduction
Inmainstreameconomics,productionfunctionsrepre- sent complete sets of feasibleinput combinations fora
4Mukoyama(2006)developsastochasticmodeloflearningbasedon thisidea.
givenoutput;inanisomorphicway,utilityfunctionsestab- lisharelationshipbetweencombinationsofconsumption goodsand the satisfaction that theyprovide – i.e. util- ity.Bothproductionandutilityfunctionsaredesignedto showintheuniverseofrationalchoiceandequilibrium allocations how the combinations chosen (respectively ofinputsandconsumptiongoods)reflectrelativeprices.
Given that conventional production theory does not provideanyanalyticalrepresentationoftheinternalstruc- tureofproductionprocesses,qualitativetransformations generatedbyinnovationsandchangesinthetechnology and structures of production remain completely unex- plored. In other words,conventional economics adopts an‘outsidetheproduction machineperspective’and, as aresult,productionand,thus,learningdynamicsremain blackboxes.
In contrast, the analysis of the internal structure of productioncombinedwithastrongemphasisontherepre- sentationofthecomplexsystemofinterrelatedproduction processesindifferentsectorswasatthecentreoftheclassi- caltheoriesofproduction.Classicaleconomistsfocusedon thelimitedavailabilityofnon-produciblegoods,theuti- lisationproblemandthevariousconstraintsdetermined bytheproductionscaleanditstimestructure.Thereare fourmaincomponentsoftheClassicaltheoreticalframe- work.FrancoisQuesnay’searlyformulationoftheconcept ofproductiveinterdependenciescalledattentiontothe‘cir- cularflow’ of wealthproduction and reproduction (see alsoLeontief, 1928).AdamSmith’s analysisoftheinter- nalstructureofthepinfactoryrevealedthemicroeconomic advantagesofthedivisionoflabourandthemacroeconomic conditions onwhich it is based–i.e. stock of circulating capitalflows.CharlesBabbage’sfocus‘onthecausesand consequencesoflargefactories’ledtotheformulationof thelawofmultiplesand,thus,tothediscoveryofdiffer- entpatternsofproportionalutilisationandmaintenanceof indivisibleinputs.Finally,KarlMarx’sanalysisofdifferent arrangementsofproductionprocesseshighlightedthemain featuresofthemodernfactorysystemandthusthework- ingofthesocalled‘collectivemachine’(Landesmann,1986, 1988;Scazzieri,1993).
In this line, more recently,WassilyLeontief’s (1947) input–output analysis and Nicholas Georgescu-Roegen’s (1970,1971,1990)fund-flowmodeldevelopedthebuilding blocks for a series of structural approaches to produc- tion(Landesmann,1986;Scazzieri, 1981,1993;Bianchi, 1984; Morroni, 1992; Landesmannand Scazzieri, 1996;
Buenstorf,2004,2007).Thesecontributionsviewagiven productionprocessPr(r=1,...,k)asaparticularsystem ofinterrelatedtasksthroughwhichasequenceoftransfor- mationsofmaterialsareperformedaccordingtodifferent combinationsofflowinputs(suchasproductiveagentsand mechanicalartefacts)andfundinputs(suchasfuel,chem- icalcatalystsandelectricity),subjecttocertainscaleand timeconstraints.
Approachingproductionfromthepointofviewofstruc- turaleconomicsimpliesananalyticalfocusonthefollow- ingsetofbothquantitativeandqualitativecoordination problems:(i)howtosynchroniseandarrangethesystem ofinterrelatedtasksintime;(ii)howtoarrangethepro- ductionprocessgiventhespecificpropertiesofmaterialsin
Triggers Learning dynamic
Internal to the firm External to the firm
First-order learning
Producer – production Learning by doing Second-order learning
Producer – research Learning by searching Learning to learn
Producers – science Learning from science Producers – users
Producers – producers
Learning by using
- Learning by interacting - Learning by exporting Fig.1.Learningdynamics:ataxonomy.
transformation;(iii)howtoorganiseandactivatethepro- ductionprocessbycombiningdifferentfundinputseach ofthemendowedwithcertaincapabilitiesorcapacities.5 Interdependenciesamongthesecoordinationproblemsare pervasiveinthesensethatforexampletasksarrangement dependonboththepropertiesofmaterialsintransforma- tionandfirm’savailabilityofcapabilities/capacities.6
Tasksrefertothoseproductionoperationsthatarepur- posefullyperformedina givenproductionprocess.Each taskTj(withj=1,2,...,j,...,J)canbedecomposedintoele- mentaryoperationsorclusteredingroupsoftasks.They canbearrangedsimultaneouslyorsequentiallyinvarious stagesoffabricationj(withj=1,2,...,j,...,J),sometimes in a discrete way but sometimes in a continuous way, thatis,withorwithoutinterruptions.7Thislastdistinction provestobeveryrelevantassoonasweconsiderhowdif- ferentformsofproductionorganisationhavehistorically developeddifferenttechniquesforinventoryandstorage capacities management (Rosenberg, 1994; Landesmann andScazzieri,1996,chap.8).
Materialsrefertowhatistransformedinthefabrication stagesofaproductionprocess.8Therelationshipbetween materialsandstagesoffabricationcanberepresentedby adescriptivematrixM=[mij]inwhichanyelementrefers tothemateriali(withi=1,2,...,n)thathasbeentrans- formedinthefabricationstagejthroughtheexecutionof thetaskTj.Ateachfabricationstage,givenacertainstock ofmaterialsavailabletotheproductiveorganisation,only
5 Mechanicalartefactspresentcertainproductioncapacities,whileeach productiveagentischaracterisedbyacertainsetofproductioncapabilities.
6 Richardson(1972,p.885)stresseshow“thehabitofworkingwith modelswhichassumeafixedlistofgoodsmayhavetheunfortunateresult ofcausingustothinkofcoordinationmerelyintermsofthebalancingof quantitiesofinputsandoutputsandthusleavetheneedforqualitative coordinationoutofaccount”.
7 Asforthetimestructure,thematerialtransformationprocessescanbe visualisedasasystemofpipelines(Landesmann,1986;seealsoMorroni, 1992).
8 Inthecaseof‘immaterialproduction’–e.g.serviceactivities–‘mate- rialsinprocesscannotbeidentified,atleastintheusualsense,andthe productionprocessgenerallytakestheformofacloseinteractionamong fundagents,inthecourseofwhichsomeofthecharacteristicsofsuch agents(andsometimestheircapabilitiesaswell)maygettransformed (LandesmannandScazzieri,1996,pp.252–253).
somematerialswillbeutilisedandthustransformed.This impliesthatforeachproductionprocesswewillobserve acertain‘realised’matrixM*=[mij],whoseinternalstruc- turerepresentsallthematerialsinuseinthestagesofthat productionprocess.
In order to perform a certain systemof interrelated tasksthroughwhichmaterialsaretransformedintofinal commodities,theproductionprocesshastobe‘activated’
bya seriesofinputs suchasfuel,chemicalcatalysts,but also machines and productive agents, that is, workers.
Flowinputssuchasfuel,chemicalcatalysts,electricityand fertilisersareutilisedincertainstagesofmaterialtrans- formationbuttheydonotmateriallyconstitutethefinal outputoftheprocessasthematerialsinusedo.Flowinputs used in a certain production process can be described throughadescriptivematrixF=[fij]inwhichanyelement referstotheflowinputithathasbeenconsumedinthe fabricationstagejthroughtheexecutionofthetaskTj.For eachproductionprocesswewillobserveacertainrealised matrixF*=[fij].
M=
⎛
⎜ ⎜
⎜ ⎝
m11 m1J
··· mij
···
mn1 mnj
⎞
⎟ ⎟
⎟ ⎠
F=⎛
⎜ ⎜
⎜ ⎝
f11 f1J
··· fij
···
fn1 fnj
⎞
⎟ ⎟
⎟ ⎠
Incontrast,fundinputsarebothmechanical artefacts suchas machines, toolsand equipment and productive agents(i.e.workers,supervisors,engineersandmanagers).
Fund inputs maintain theircharacteristics substantially unaltered duringthe production process, provided that certaintolerancethresholdsarenotviolated(Georgescu- Roegen, 1970;Landesmann,1986).Mechanical artefacts presentacertainproductioncapacity,whileeachproductive agentischaracterisedbyasetofcomplementaryproductive capabilities.Byactivatingsomeofthesecapabilities,each productiveagentisabletoperformasingletaskorasetof similartasks(i.e.thosetaskswhichrequiretheutilisationof thesamesetofcomplementarycapabilities).Althoughpro- ductiveagentsmaylearntoperformdifferenttasks,their capabilitiesarelimitedsotheycannotswitchbetweenall productive tasks,especiallywhen complexproductsare considered.
Justaswedidwithmaterialsandflowinputs,wecan distinguishbetweenthebundleofcapacities/capabilities embodied in a certain set of fundinputs (i.e. potential capacities/capabilities)andthecapacities/capabilitiesactu- allyutilisedbytheproductiveorganisationinperforming a certain setof tasks (i.e.capacities/capabilities inuse).9 Theformeris describedbythematrixC=[cij]whilethe latterbythematrixC*=[cij],wherecijdenotestherela- tionship between thecapacity/capability i and thetask Tjperformedatthestageoffabricationj.Thedistinction betweenthematrixC=[cij]and C*=[cij] illustrateshow thesameproductionprocessPr(r=1,...,k)canbeper- formedbyusingdifferentbundlesofmechanicalartefacts andproductiveagents,thatis,differentcombinationsof productioncapacityandproductivecapabilities.However, evenwhentwodifferentproductiveorganisationsperform thesameprocessPrbycombiningthesamebundleofpro- ductivecapacities/capabilities,thelattercanbeemployed indifferentproportions.Forexample,twofirmsFirm1and Firm2canperformthesameproductionprocessbyusing thesametwofundagents–i.e.workerswandmachines m–butindifferentcombinations–forexample,onefund- inputcombinationmaybemorelabourintensivethanthe other.
C1∗=
5w 7w1m 1m
C2∗=
1w 2w3m 3m
Thus,bycomparingthetwomatricesC1*andC2*we can discoverspecificfeatures of theproduction process Pr performedbyFirm1 andFirm2.Inparticular,thetwo matricesexpressdifferentrelationshipsofcomplementar- ityamongthetwofundinputsconsidered(machinesand workersrespectively).Inourcase,thefirststageT1ofthe productionprocessPrcanbeperformedeitherbycombin- ingonemachinewithfiveworkersorthreemachinesand oneworker(seeabove).Giventheserelationshipsofcom- plementaritybetweenfundinputsandalsothefactthat machinestendtobetasks-specificandonlypartiallyflex- ible, thekindof combinationsoffundinputs thatfirms canselect fromthespaceC=[cij] for performingPr are limited.Moreover,scalinguptheproductionprocessnot onlyrequires theconsiderationof theserelationshipsof complementarity but also that a lawof proportionality amongallthecomponentsoftheprocessissatisfied(see below).
BasedonCartwright(1989),ithasbeennotedhowvery oftenthecapabilities(andcapacities)offundinputscanbe expressedinaquantitativeform,sothatwecanassume theyarecomparableincardinalspace(Landesmannand Scazzieri,1996,p.197).Onepossiblequantitativespeci- ficationofthematrixC=[cij]reliesontheconsideration ofthetimestructureoftheproductionprocessinrelation tothecapacity/capabilitiesinuse.ThematrixC*=[cij]can
9Ashasbeenstressed,thisdistinctionleadsustointerprettheemer- genceofnewproductivestructureswithinthespaceofvirtualpracticesas
“theoutcomeofaclusteringprocessthatbringsaboutarearrangementof theprimitiveelementsofproductiveactivity;[thus]structuralchange maybeconsideredasacaseofvariationwithinaspectrumofvirtual possibilities”(Scazzieri,1999,p.230).
betransformedintoamatrixofcapacities/capabilitiesuse– times*=[ij]wherethegenericijrepresentstheuse- timeofthecapacity/capabilitycijintheproductionprocess Pr.
C=
⎛
⎜ ⎜
⎜ ⎝
c11 c1J
··· cij
···
cn1 cnj
⎞
⎟ ⎟
⎟ ⎠
∗=⎛
⎜ ⎜
⎜ ⎝
11 1J
···
ij
···
n1 nj
⎞
⎟ ⎟
⎟ ⎠
Taking the case in which two different productive organisationsperformthesameprocessPrwiththesame combination of fund inputs described by C*=[cij], we can compare the matrices * to discover if one time arrangementofPrismoretimewastingthananother.For example,thereconfigurationofthetimestructure ofPr
fromoneinlinetooneinparallelcanreducetheamount ofidletimeoffundinputsacrossfabricationstages(see below). Given appropriate transformations such as the oneproposedabove,assoonastheproductivecapabili- tiesandcapacitiesbecomecomparableincardinalspace, thecapacity–capabilityratioscanbecalculated foreach productivetask (orgroups oftasks) and organisedin a matrix.Thiswillelucidatetheinterdependenciesbetween differentkindsoffundinputs.However,thesetofinter- dependenciescharacterisingeachproductionprocessdoes notsimplyinvolveonesubsetofitscomponents(here,fund inputs).Instead,eachproductionprocessrequiresthecoor- dinationofallitscomponents(namelytasks,materialsand flowinputsaswellasfundinputs).
Interdependencies among components can be visu- alised by mapping the relationships between capac- ity/capabilities,tasksandmaterials.10Theentirespectrum ofpossiblecombinatoricsisrepresentedthroughtheana- lytical mapof productionrelationships (see Fig.2).11 The mappingfromthecapacity/capabilityspaceCtothetask spaceT(i.e.jobspecificationprogramme)canbedetermined following different criteria (Landesmann and Scazzieri, 1996).Forexampledifferentcombinationsoffundinputs mayberelatively moreor less adequatefor theexecu- tionofonetask(orclusteroftasks)thananother.Also, areconfigurationofthejobspecificationprogramme(i.e.
differentmappingfromthespaceCtothespaceT)may allowtheactivationofpreviouslyunusedfundinputsor theachievementofhigherefficiencyintheutilisationof thecapacity/capabilitiesinuse.
The network of relationships and interdependencies among the spaces C, T, M hasto besynchronised over timeand according to specificscale requirementsdeter- minedbytheexistenceofprocessindivisibilitiesaswell
10 Forclaritytheflowagentsaretakenoutofthepicture.Thedecisionto privilegetheotherthreedimensionsmatricesC,TandMisduetothefact thatcommonlytherearehigherdegreesoffreedomintheircombinatorics andtheuseofflowagentsisstrictlydependentontheutilisationoffund agents.
11 Theconceptof‘analyticalmapofthetrue[interpersonal]relations’is proposedinGeorgescu-Roegen(1976,p.205)asonepossiblerealisation ofthe‘entirespectrumofpeasantinstitutions’.
Fig.2.Theanalyticalmapofproductionrelationships.
asindivisiblefundinputs.12Asforthetimestructure,syn- chronisationhastobepursued atthree different levels (coordinationintheutilisationoffundinputs,arrangement ofinterdependenttasksovertime,andtransformationof materialsovertime).Thedifficultyofmatchingthe‘time sequencingrequirements’ofthesethreedimensionsmakes perfectsynchronisationacrossthethreeabovementioned levelsimpossible and explains the co-existence of pat- ternsofsimultaneityorsequentiality.Timegapsandidle timeinproductionprocessesarethuslargelystructurally determinedand,withinthegivenstructure,onlypartially reduciblethroughvariousformsoflearning(Landesmann andScazzieri,1996).
Movingontoindivisibilities,processesareindivisible whentheyarenot‘indifferenttosize’.Asinthebiological world,allindividualproductionprocesses“followexactly thesamepattern:beyonda certainscale somecollapse, othersexplode,ormelt,orfreeze.Inaword,theycease to work at all. Below another scale, they do not even exist”(GeorgescuRoegen,1976,p.288).Thefactthatpro- cessesare ‘scale-specific’ (in otherwords that theyare characterisedbyupperandlowerbounds)impliesthatcon- ductingaprocessonasmalleroralargerscalecanonly bedoneifalawofproportionalityamongthecomponents oftheprocessissatisfied.Thisideawasoriginallyformu- latedbyCharlesBabbage’slawofmultiple.InBabbage’s view:‘[w]henthenumberofprocessesintowhichitismost advantageoustodivide[theproductionprocess],andthe numberofindividualstobeemployedinit,areascertained, thenallfactorieswhichdonotemployadirectmultipleof thislatternumber,willproducethearticleatagreatercost’
(Babbage,1835,p.211).
At the level of the components, limitations in the bundling and unbundling of fund inputs are extremely stringent,whileflowinputsaswellasmaterialsintrans- formationaremoreoftendivisible.Asfarasfundinputs areconcerned,theexistenceofindivisiblefundsofcapaci-
12 Foracomprehensivediscussionoftheroleoftimeandscaleinpro- ductionseeLandesmann(1986),Bianchi(1984),Morroni(1992),Scazzieri (1993),LandesmannandScazzieri(1996).
ties(suchasamachinetoolwithcertainscale-determined technicalcharacteristicsandspecifications)aswellasindi- visiblefundsofcapabilities(thatis,productiveagentssuch asworkersandengineersattheshopfloor)meanthat,for afundinputtobefullyutilised,aspecificscaleofproduc- tionhastobeachieved.Forsmallscalesofproduction,fund inputswouldinevitablybeunderutilised.However,iffund inputsarenottoospecialisedintheexecutionofsomepro- ductivetasks,productiveorganisationscanovercomescale constraintsbyutilisingthesameindivisiblefundinputsfor theproductionofothercommodities.Neverthelessthese new commoditiesgenerally possess a certaindegree of similarityasfundinputsareendowedwithonlyalimited setofcomplementarycapacitiesorcapabilities.
2.3. Embeddinglearningdynamics
Thestructural representationof production provided abovenowallowsustoseesomeofthemanylimitations arisingfromtheunderstandingoflearningdynamicsasa disembedded process,asis thecase withtoday’smain- streameconomics(seeSection2.1).Togiveoneexample ofthemainformoffirst-orderlearning,Arrow’sconceptof
‘learningbydoing’referstoaprocessinvolvingonesub- setofthespaceC(i.e.capabilitiesoffundinputssuchas workers,engineersandmanagers).Inthiscase,‘learning bydoing’isnothingmorethananincreaseinproductive capabilities,whichgenerallyresultinareductionofcapa- bilitiesuse-times.Inotherwordstheexecutionofthesame productivetaskwillrequirelesstimeduetoaccumulated experienceandasaresulttheoverallproductivityofthe productiveorganisationwillbeincreased.However,aswe shallseebelow,ouranalyticalmapofproductionshows how‘learningbydoing’doesnotalwaysimplysuchpro- ductivityincreasesanditmightevenleadtotheemergence ofbottlenecksandimbalances.13
13Infactinhisseminalworkon‘learningbydoing’,KennethArrow recogniseshow“learningassociatedwithrepetitionofessentiallythe sameproblemissubjecttosharplydecreasingreturns”and,thus,that learningmainlyconsistsoffindingnewsolutionstoemerging‘stimulus
Thereasonforthisbecomesclearassoonaswevisu- alise interdependencies among components, i.e. tasks, materials,flowinputsandfundinputs (andtheircapac- ity/capability).Thedevelopmentofincreasingcapabilities intheexecutionofacertainsetofproductivetasksgener- allyimpliesthatcertainstagesoffabricationwillrequire lesstime,whileotherstagesremaininvariantasaresultof constrainingfactorssuchasfixedtimesformaterialtrans- formation(e.g.timeneededforcertainchemicalreactions) or the scale of otherexisting fund inputs, in particular machinesandequipment.Theselatterstagesoffabrication, giventheirinvariantproperties,willappearasbottlenecks intheproductionprocessand mayend upaffectingthe entirejobspecificationprogramme,potentiallyevenneu- tralising or counteracting the productivity increases of
‘learningbydoing’.
Togiveasecondexampleoftheimportanceofunder- standinglearningasembeddedwecanlookatRosenberg’s conceptof‘learning byusing’,thelatterbeingthemain formofsecond-orderlearningdiscussedabove.Thecon- ceptoflearning byusingwasdevelopedwithreference to ‘products involvingcomplex interdependent compo- nentsormaterials’.Asaresultoftheparticularindustry he focused on, that is, aircraft,14 Rosenberg underlined thefactthat‘learningbyusing’impliesa“feedbackloop in thedevelopmentstagewhich, inturn,increaseseffi- ciencyand/orrequireschangesinproductivetechniques”
(1982, p. 123).15 Rosenberg distinguishes between two kindsofusefulknowledgearisingfrom‘learningbyusing’
inbothproductsandprocesses.Embodiedknowledgeis that which requires ‘appropriate design modifications’, whiledisembodiedknowledge“leadstocertainalterations in usethat require no(oronly trivial) modifications in hardware design”, although even the latter still “leads to new practices that increase the productivity of the hardware” – for example modification in maintenance practices intheaerospaceindustry(Rosenberg,1982,p.
124).
Ofcourse,thesetwoforms of‘learning byusing’are intertwined. By generating new embodied knowledge,
‘learningbyusing’infactfacilitatesthediscoveryofnew formsofdisembodiedknowledgeand evenmakesthem necessary.Whatisimplicitlysuggestedhereisthat‘learn- ing byusing’ maytriggerthe rearrangementof thejob specificationprogramme.Thisoccursbecauseofthenew productivetasks andfundinputs requiredtocopewith designmodifications(embodiedknowledge),orasaresult ofalterationsinproductivepracticeswhoseperformances dependontherearrangementoffundinputsavailable(dis- embodiedknowledge).
situations’(Arrow,1962,p.155).However,theeffectsof‘evolvingstimuli’
inthetransformationofproductivestructuresarenotanalysedgiventhe lackofananalyticalmapofproduction.
14Eventodayaircraftareamongthemostcomplexproducts,composed ofalmost6millionparts(bywayofcomparisonacaristypicallycomposed ofjust6thousandsparts).
15Inthisrespect,seealsoHippelvon(1988)whosecontributionlinksthe learningbyusingdynamicstoproductdiversificationpatterns.Also,Kline andRosenberg(1986)presentsa‘chain-linkedmodel’wherefeedback loopsintheinnovationprocessarerecognisedaskeyfactors.
Thefirstcaseismoreclearlydetectableasitrequires definitedesignmodifications(i.e.technologicalimprove- ments) whilethesecondset oftransformationstend to be under-estimated as they do not call for the intro- ductionof any new fundinputs. The analytical map of production allows us to understand how a production process maybe qualitativelytransformed even without equipping the productive organisation with new fund inputsorwithouttransformingtheexistingones.Instead, theproduction process maybe transformedjust by re- arrangingfundinputsamongthesystemoftaskswhich have to be performed or by synchronising tasks in a differentway over time. In fact there arevarious ways of combining elementary operations into new tasks or clustering existing tasks in new ways. Once again the extenttowhichthiscanbedonedependsonthecapac- ities/capabilities embedded in funds inputs and their degree of utilisation, as well as on the properties of the materials in transformation and on time arrange- ments.
Learningprocessesareintrinsicallyheterogeneousand occurthrough time at several nested levels of produc- tion,thelatterbeingstructurallydeterminedbyproductive interdependencies.16Assoonasweattemptarestructu- ringof‘learningbydoing’or‘learningbyusing’,itbecomes obviousthatthemajorityofexisting studiesfocustheir attentiononwhat triggersthelearning processorwhat itsoutputis.Theprocess persenot discussed.In other words,theconventionalanalysisoflearningendsexactly wherethelearningprocessstarts.Evenwhenthere isa moredetailedinvestigationoflearningdynamicsinpro- duction,asintheworkofeconomichistorians(Rosenberg, 1969,1976,1979,1982;Noble,1986;PioreandSabel,1984;
MoweryandRosenberg,1998;Mokyr,2002;Landes,2000;
Poni,2009),productionstructuresaregenerallyseenonly asconstraintsthatproductiveagents overcomethrough problem-solvingactivitiesandchangesinproductivetech- niques.Forexample,Mokyr(1990,p.9italicsadded)argues that‘[t]echnologicalchangeinvolvesanattackbyanindi- vidual on a constraint that everyone else has taken as given’.
However,asweshallseebelow,ananalyticalaccountof anumberofhistoricalcasesallowsustounderstandhow existingand evolvingproductionstructuresare notjust constraints.Instead,existingproductionstructuresorien- tateproductiveagentstowardscertainlearningtrajectories andallowthemtodiscoverstructurallyembeddedoppor- tunities.As shown by Hicks (1969),the adoption of an analyticalapproachtoeconomichistorycanbeavehicle
16 Anotheraspectthatconventionalapproachestendtoforgetisthat
‘learningintime’canproceedatdifferentspeedsaccordingtothetime requiredforreconfiguringtheproductionstructureoraccordingtothe timeknowledgerequirestoflow(i.e.bedisseminatedandabsorbed) throughouttheproductionorganisation orattheinter-firmlevel.In otherwordstheproblemisnotonly‘whattolearn’or‘howtolearnto learn’butalso‘howtolearnfaster’.AsshownbyDodgson(1991),the differentialabilityinlearningquicklyabouttechnologicalopportunities is acrucialdeterminant especiallyinthose sectors(e.g. biotechnol- ogy) characterised by an uncertain and generallyrapid process of transformation.
fordevelopinga‘quasi-theory’,thatistosayastylisedrep- resentationofeconomicfactsthroughwhichtheoriescan bedeveloped.
3. Structurallearning:ananalyticalframework
3.1. Learninginastructuredspace:ananalyticalaccount ofhistoricalcases
The analytical map of production relationship eluci- datesfirstlythe‘architectureofcomplexity’inproduction (Simon, 1962; Buenstorf, 2005) and secondly, the fact thatlearningdynamicsarerealisedina‘structuredspace’
over time. This means that learning in production is not simply a process occurring as a result of cognitive dynamics; rather it is also a process triggeredand ori- entatedby structural dynamics.The latter openup the possibility of transforming ‘structural constraints’ such asbottlenecks and technical imbalancesinto‘structural opportunities’.Ashighlightedinstructuralanalysesofpro- duction(seeabove), coordinationproblemsinthespace ofcapacity/capabilities,materialsandtasksmaybesolved inmultiple(albeitinterdependent)ways.Inotherwords, thereare‘worldsofproduction’,thatis,avarietyofproduc- tionarrangements(‘worldofpossibilities’)thatarefeasible evenundersamesetsofcontextualconditions(Salaisand Storper,1997).
‘Worldsofpossibilities’permitthetransformationofpro- ductionprocesses andtheiroutcomes– i.e.processand productinnovations(Sabel andZeitlin,1997).Ofcourse saying that there are multiple possibilities should not blindus to thefact that bottlenecks, materials proper- tiesortechnicalimbalancesareallpervasiveconstraints.
Infact,discoveringthesepossibilities,givencertainstruc- tural constraints,is the very essence ofwhat I call the structuralprocess of learning. The concept of structural learning is introduced here to identify the continuous process of structural adjustment and transformation of production ‘triggered’ and ‘orientated by’ existing and evolvingproductionstructures.Staticanddynamiccom- plementarities,aswellassimilaritiesandindivisibilities, are essentially focusing devices for activating compul- sive sequences of technological change and discovering new production possibilities at the firm and inter-firm level.
We will now move to an analytical account of his- torical cases in which “[c]omplex technologies create internalcompulsionsandpressureswhich,inturn,initiate exploratoryactivityinparticulardirections”(Rosenberg, 1969,p.4).Thishistoricalanalysisisthefirststeptowards disentanglingthosestructuraldynamicsthatpreparethe setting for learning and those specific factors trigger- ing learning processes in production. The second step istoidentifya number ofstructurallearning trajectories andillustratethemwithananalyticalmapofproduction relationships. The third step will be to re-link dynam- ics occurringat thelevel of production structures with those occurringat thelevel of thestructures of cogni- tioninproductiveorganisations.Asweshallsee,thisthird stepwillallowustoshowtheanalyticaltensionbetween structure and agency in learning dynamics and also
elaboratehowstructureandagencyarelinkedbyabun- dleofbidirectionaltransformativerelationships(Bourdieu, 1972).17
Rosenberg (1969) identifies three main ‘induce- ment mechanisms’of learning, namelytechnicalimbal- ancesorbottlenecks,labour-saving/uncertainty-reducing machinesandsubstitutesoralternativesourcesofsupply offundandflowinputsormaterials.Anumberofhistori- calexampleswillhelptoillustratethispoint.In1900the machinetoolindustrywasrevolutionisedbytheintroduc- tionofhigh-speedsteelwhichallowedanincreaseinthe hardnessofcuttingtools.However“itwasimpossibleto takeadvantageofhighercuttingspeedswithmachinetools designedfortheoldercarbonsteelcuttingtoolsbecause theycouldnotwithstandthestressesandstrainsorprovide sufficientlyhighspeedsin theothercomponentsofthe machinetool”(Rosenberg,1969,p.7;seealsoAndreoni andGregory,2013).Asaconsequence,transmissions,con- trolelementsandothermachinetoolcomponentshadto beredesignedandthischange“inturn,enlargedconsider- ablythescopeoftheirpracticaloperationsandfacilitated theirintroductionintonewuses”(Rosenberg,1969,p.8).
Thisisatypicalexampleofatechnicalimbalanceleading tochangesincomplementaryprocessesaswellascompo- nents,thatis,tasks,materialsandcapabilities/capacities.
Ithighlightshowatechnicalconstraintcanactuallyacti- vateaprocessofexplorationandsearchinginwhich“the size of thediscovery need bear nosystematic relation- shiptothesizeoftheinitialstimulus”(Rosenberg,1969, p.9).
Indeed, the initial technical imbalance in a certain industrymaytriggerstructurallearningprocessesinother industriesandsectors.IntheearlynineteenthcenturyUS agriculturalsector,beforetractorswereintroduced,John Deererevolutionisedagriculturalproductionbyinventing thesteelplow(Andreoni,2011a,b).Abiologicalconstraint triggeredtheintroductionofsteelplows,butalsoofother complementary tools made with the same or different materialsaccordingtospecifictaskrequirements.Tradi- tional wood plows could not plow therich soil of the Middle-West without breaking. At that time given the scarcityofsteelandtheneedtoimportitfromGreatBritain, JohnDeeremadehisfirstplowoutofanoldbladesaw.
Afteraseriesoftestsondifferenttypesofsoil,thenew steelplowwasreadytobeabsorbedintothe‘crop-growing technique’adoptedatthattime.Inturn,theintroduction ofthesteelplowtriggerednewcomplementarydiscover- iesaswellastheapplicationofthesamenewmaterialto otherequipmentsrequiringthesamehardness.Infact,as recognisedbyRosenberg(1979,p.37),‘thesubstitutionof newmaterials(e.g.aluminiumandrust-resistantsteels) for old ones, improvedtechniques of friction reduction (lubrication and roller bearings) have led to a consid- erable extension of the useful life of a wide range of capitalequipment’aswellastoother‘cumulativeimprove- ments’.
17AtthecentreofBourdieu’sanalysisthereisthedialecticbetween
‘externalisingtheinternal’and‘internalisingtheexternal’whichattempts togobeyondpreciselythesametension.
Thus, the process of structural learning in a given sector may develop an intersectoralcharacter. In other words,complementarities(aswellasinnovationssuitable for similar tasks performed in other productive activi- ties)mayspreadfromonesectorofaneconomicsystem toanothersectortriggeringaspecificformofstructural learning which we labelhere intersectorallearning. The latterexpressionidentifiesadynamicprocessofinterlock- ing and mutual reinforcing technological developments whichlinks theinnovativepatternsoftwoormoresec- torsinarelationshipofcomplementarityand/orsimilarity.
In sum,technicalcomplementarities amongfundinputs ortheapplicationofaninnovation–e.g.anewmaterial withcertainproperties–intheexecutionofabroadset ofsimilarproductivetasksarethefundamentaldynamics underlyingthelearning processesconsideredbyRosen- berg.
It is notjust constrained posedbyexisting technical processes thatcanworkastriggersforstructural learn- ingdynamics:socialprocesscanfunctioninthiswayas well.InthePovertyofPhilosophyKarlMarxobservedhow
“aftereachnewstrikeofanyimportancethereappeareda newmachine”(n.d.:134;firstsourceRosenberg,1969).The threatofstrikesintroducesacriticalelementofuncertainty tothesupplyoflabourandstronglyaffectsthedelicatetime structure ofaproduction process,therebypromptlythe inventionofnewlabour-savingmachines.Socialchanges inducetheinventionordiscoveryofnewmachinesandthis inturnsetsoffafurthertrainofchanges.
Robert’s self-acting mule, the Jacquard punching machineandtheintroductionbytheBritishGovernment ofthe‘AmericanSystemofManufacturing’inthegunmak- ingindustryin1854,areallcasesinwhichtheinvention oracquisitionofanewmorepowerfulmachineisjustthe first stepin a subsequentprocess ofstructural learning (Rosenberg,1969;seealsoChang,2002).Allthesecases highlighthowwhenanewmachinebecomesavailablepro- ductionthatwastechnicallyfeasiblebutnoteconomically convenientbecomespossible.Thispossibilitydependson increasing the scale of complementary machines or in the re-arrangement of workers in the production unit, providedthattheycanperformacertainsetofsimilarpro- ductivetasks.
Togetherwiththeabovementionedinducementmech- anisms identified by Rosenberg, the need/opportunity of increasing the scale of production is another factor triggering processes of structural learning. For example complementaryinnovationssuchasrefrigerators,railways and steamships affectedthereduction oftransportation costs,increasedthedegreeofregionalspecialisationand openedtheopportunitytobenefitfromscaletechnology expansionsandfromspecialisationinalimitedsetofpro- ductive tasks performedat highproductivitystandards.
Indeed, as soon as the scale of production increases ‘a shiftingsuccessionofbottlenecks’willemerge.Focusing onthem,engineerswillstartexploringnewpossiblecon- figurationsoftheproductionprocess,whichmayleadto serendipitouslydiscovering‘singletontechniques’(Mokyr, 2002).Problemsrelatedtoscaleconstraints,whicharise both from indivisible fund inputs and indivisible pro- cesses,maytriggerthediscoveryofinnovative(structural
as well as organisational) configurations of production processes(AndreoniandScazzieri,2013).Agoodexam- pleofthisisthetypicalproblemsfacedbysmallfarmers and small firmswhen trying togain access toindivisi- blefundinputssuchasmachinesandotherequipment.
Historically fund inputs indivisibilities as well as scale invariant processes have triggered institutional innova- tionssuchas‘renting/sharing’solutionsimplementedby producercooperatives(Lissoni, 2005)as wellasforcing productive agents to rearrange job specification pro- grammes.
3.2. Structurallearningtrajectories
Thehistoricalcasesdocumenthowinducementmecha- nismsoflearningdynamics,andtheresulting‘compulsive sequences’oftransformations,areembeddedinandtrig- gered by existing production structures at each point in time. Specifically complementarities and similarities amongtasksormaterials,aswellasfundinputsindivis- ibilities,have beencrucial focusingdevicesinstructural learningdynamics.Theanalyticalaccountofthesehistor- icalcasesleadstotheidentificationofthreefundamental structurallearningtrajectories.Givenacertainbottleneck ortechnicalimbalanceinproduction,thefirsttwostruc- turallearningtrajectoriesaretriggeredbytheexistenceof similaritiesandofcomplementarityamongmaterials,tasks andfundinputs.Thethirdstructurallearning’strajectoryis triggeredbytheexistenceofindivisibilitiesinproduction.
Thefundamentalintuitionbehindthefirsttwostruc- turallearning trajectoriesmaybefoundinRichardson’s (1960, 1972, 2003) observation that different forms of inter-firmcooperationweseearisefromdifferentpatterns of similarity and complementarity among productive activities.18 Richardson breaks down the production of eachfinalcommodityintovariousstagesoractivities,each ofthemexecutablebydifferenttypesoffirms.“Activities whichrequirethesamecapabilityfortheirundertaking”
arecalledsimilaractivities(Richardson,1972,p.888).On theotherhand,activitiesarecomplementary“whenthey representdifferentphasesofaprocessofproductionand requirein somewayoranotherto becoordinated[...]
bothquantitativelyandqualitatively”(Richardson,1972, pp. 889–890). Building on this dichotomy, Richardson explains how the complex and interlocking clusters, groups and alliances offirms we observe are in reality differentresponsestothesameproblem:theneedtocoor- dinate“closelycomplementarybutdissimilaractivities”.19 Asfirmscannotaccumulateallthecapabilitiesrequired for performing a broad setof dissimilar activities,they willspecialiseinafewactivitiesandcooperatewiththose firms specialised in closely complementary activities.
Principlesofsimilarityandcomplementarityalsooperate atthefirmlevelandareresponsiblefordistinctstructural learningtrajectories.
18 SeealsoMénard(2004),GibbonsandRoberts(2011),Garnseyand McGlade(2006).
19 ThisanalyticalpointisdevelopedinSection4withrespecttothe differentformsofproductionorganisation.
3.2.1. Structurallearningtrajectorytriggeredby similarities
Overcominga productiveconstraintbyintroducinga newsetoftasks,capabilitiesormaterialsmayinducethe sameorotherfirmstoadoptthesamesetoftasks,capa- bilitiesormaterialsforovercomingasimilarconstraint,in thesameorotherkindofproductiveprocesses.Asdocu- mentedbyRosenberg(1963,pp.422–423,italicsadded)
“industrialisationwascharacterisedbytheintroductionof arelativelysmallnumberofbroadlysimilarproductivepro- cessestoalargenumberofindustries.Thisfollowsfrom thefamiliarfactthat industrialisationin thenineteenth centuryinvolvedthegrowingadoptionofametal-using technologyemploying decentralised sources of power”.
Furthermore,discoveringanewwayofperformingacertain taskaffectsallthoseproductiveprocessesinwhichsimi- lartasksareperformed.Thisexplainswhy“manyofthe benefitsofincreasedproductivityflowingfromaninnova- tionarecapturedinindustriesotherthantheoneinwhich theinnovationwasmade”(Rosenberg,1979,p.41;seealso Usher,1954).20
Many examples might be provided which highlight theexistenceoftechnologicallinkagesamongapparently uncorrelated products such as guns, sewing machines, bicycles,motorcycles,andautomobiles.Amongthemany historical examples ‘the development of the universal millingmachine by Brown and Sharpeis, perhaps, the mostoutstandingexampleofa machinewhich wasini- tially developed as a solution to a narrow and specific rangeof problems andwhich eventually had enormous unintended ramifications as the technique was applied to similar productive processes over a wide range of metal-usingindustries’ (Rosenberg, 1963, p. 432, italics added).
Inthespecificcaseoffirmswhoseproductionprocess consistsofasystemofsimilartasks,thediscoveryofanew wayofperformingacertaintaskortheintroductionofa newmaterialimplies a complete reconfigurationof the entireprocess.However,asinthis specificcase produc- tiveagentswouldalreadybeendowedwithsimilarkinds andamountsofcapabilities,theywillbesubstitutableand canbearrangedinmanydifferentwaysacrosstime.The productionprocessofmorecomplexproducts(orcompo- nents)tendstoassumetheformofasystemofdissimilar tasks.Indeed,complexproductsaredefinedasthose“com- posedofmanysubsystemsthatinteractincomplexways”
(Rosenberg,1982,p.136).Inthecaseofcomplexproducts requiringtheperformanceofcloselyinterdependentdis- similartasks,intraandinter-firmcomplementaritieswill bepervasive.
3.2.2. Structurallearningtrajectorytriggeredby complementarities
“[I]nnovations hardly ever function in isolation”
(Rosenberg,1979,p. 26).Thetheoreticalframework we have constructed allows us to analytically specify and explain this intuitive insight of Rosenberg’s. Innovation
20 ThisanalyticalpointwillformthebasisofourdiscussionintheThird Essayoftheconceptofintersectorallearning.
occurs in this bunched fashion because of the utilisa- tion and theproductivity of fund inputs (i.e. machines withcertaincapacitiesorproductiveagentswithcertain capabilities)both criticallydepend onthesimultaneous availability ofcomplementaryfundinputs. Complemen- tarities among fund inputs may trigger direct learning dynamics,orlearningdynamicsovertime.Directlearning dynamics occur when one fund input makes the func- tioningofanotherfundinputpossibleormoreefficient.
Learningdynamicsovertimeoccurwhenonefundinput makesthefunctioningorintroductionofotherfundinputs possibleovertime.
Inthespecificcaseofaproductionprocessconstituted bya systemof dissimilartasks,fundinputs performinga specifictaskinonestageoffabricationarecombinedwith othersperformingothertasksinotherstagesoffabrication inarelationshipofcomplementarityratherthanofsubsti- tutability.Nowiftasksareverydissimilar andcomplex, productive agents (or even entire productive organisa- tions)havetospecialiseintheexecutionofonlyonetask, oreveninperformingelementaryoperationsofmorecom- plextasks.Inthiscase,anumberofprocessesofthesame typecanbeorganisedinseries(alsocalledinsequence)so thatspecialisedproductiveagents(ororganisations)can performthetaskinwhichtheyarespecialisedwithoutlong periodsofinactivity.Discoveringthispossibilityandapply- ingittotheproductionprocessallowsfirmstoreducetime wastageasproductiveagentswillshiftovertimefromone processtoanother.
Additionally,accordingtothedegreeofdecomposabil- ity of a given production process, firms may decideto adoptamodularisationstrategy(Langlois,2002;Buenstorf, 2005).Interestingly,inthecase ofproductive processes composedofcloselycomplementarybutdissimilartasks, modularisation may guarantee static efficiency at the costof dynamicefficiency.Thisproblemoccursbecause modularisationtendstoreducethenumberoflearningtra- jectoriestriggeredbycomplementarities.
3.2.3. Structurallearningtrajectorytriggeredby indivisibilities
Indivisiblefundinputsandmaterialsaswellasscale- invariant tasks (or processes) impose a proportionality path on alltransformations of the internal structure of production(seeabovethereferencetoBabbage’slawof multiples).Thismeans,forexample,thatifacertainindi- visiblefundinput(e.g.anewmachine)isadopted,then,the firmhastoreconfigurethejobspecificationprogrammein suchawaythatscaleeconomiesgeneratedbytheuseof thenewmachineareexploitedandpotentialbottlenecks andtimeormaterialwastesareavoided.
The existence of indivisibility might also trigger incremental innovations both at the technological and organisational levels. For example adopters of thenew indivisible input (or scale-invariant task) “could invent around thenew machine and remove those technolog- ical constraints that limit their ex ante or ex post size.
[...]Alternativelytheycouldattackitdirectlybyfinding thewaytosplitthedifferentfunctionsthattheoriginal innovationperformsjointly,thusdecomposingthelatter intoa few(possiblycompatible)modules,each ofthem
