解鎖協(xié)作人工智能的潛力-關(guān)于聯(lián)邦機(jī)器學(xué)習(xí)的社會(huì)技術(shù)挑戰(zhàn) UNLOCKING THE POTENTIAL OF COLLABORATIVE AI – ON THE SOCIO-TECHNICAL CHALLENGES OF FEDERATED MACHINE LEARNING

Thirty-firstEuropeanConferenceonInformationSystems(ECIS2023),Kristiansand,Norway1

UNLOCKINGTHEPOTENTIALOFCOLLABORATIVEAI–

ONTHESOCIO-TECHNICALCHALLENGESOF

FEDERATEDMACHINELEARNING

ResearchPaper

TobiasMüller,TechnicalUniversityofMunich,SchoolofComputation,Informationand

Technology,DepartmentofComputerScience,GermanyandSAPSE,Germany,tobias1.mueller@tum.de

MilenaZahn,TechnicalUniversityofMunich,SchoolofComputation,Informationand

Technology,DepartmentofComputerScience,GermanyandSAPSE,Germany,milena.zahn@tum.de

FlorianMatthes,TechnicalUniversityofMunich,SchoolofComputation,Informationand

Technology,DepartmentofComputerScience,Germany,matthes@tum.de

Abstract

ThedisruptivepotentialofAIsystemsrootsintheemergenceofbigdata.Yet,asignificantportionisscatteredandlockedindatasilos,leavingitspotentialuntapped.FederatedMachineLearningisanovelAIparadigmenablingthecreationofAImodelsfromdecentralized,potentiallysiloeddata.Hence,FederatedMachineLearningcouldtechnicallyopendatasilosandthereforeunlockeconomicpotential.However,thisrequirescollaborationbetweenmultiplepartiesowningdatasilos.Settingupcollaborativebusinessmodelsiscomplexandoftenareasonforfailure.CurrentliteraturelacksguidelinesonwhichaspectsmustbeconsideredtosuccessfullyrealizecollaborativeAIprojects.ThisresearchinvestigatesthechallengesofprevailingcollaborativebusinessmodelsanddistinctaspectsofFederatedMachineLearning.Throughasystematicliteraturereview,focusgroup,andexpertinterviews,weprovideasystemizedcollectionofsocio-technicalchallengesandanextendedBusinessModelCanvasfortheinitialviabilityassessmentofcollaborativeAIprojects.

Keywords:FederatedMachineLearning,CollaborativeDataProcessing,BusinessModel,Alliances

1Introduction

ArtificialIntelligence(AI)hadanimmenseeconomicimpactinthelastcoupleofyears.In2021alone,themarketofAI-basedservicesincludingsoftware,hardwareandservicesexceeded500$billionwithafive-yearcompoundannualgrowthrateof17.5%(ForradellasandGallastegui,2021).Thepotentialprofitabilityraiseiscurrentlyestimatedbyanaverageof38%,whichimpliesaneconomicimpactof$14trillionuntil20351.Unmistakably,theusageofAIenablesnew,unprecedentedbusinessmodelswithamonumentalimpactontheindustry.Themainenablerforthisdisruptivenewmarketistheemergenceofbigdata,whichformsthefundamentalbasisforAIsystems.Eventhoughvastamountsofdataisfreelyavailable,aconsiderableamountoftheworld’sdataisscattered,storedandlockedupindecentralizedIoTdevicesanddatasilos.Naturally,thesiloeddataishardlyaccessible,leavingalargeportionofalreadygenerateddata,andthereforeeconomicpotential,largelyuntapped.Theemergence

1/fr-fr/_acnmedia/36dc7f76eab444cab6a7f44017cc3997.pdf

Thirty-firstEuropeanConferenceonInformationSystems(ECIS2023),Kristiansand,Norway2

Socio-TechnicalChallengesofCollaborativeAI

ofdatasilosisstrengthenedbydataprotectionlawsandregulationssuchastheGeneralDataProtectionRegulation(GDPR),CaliforniaConsumerPrivacyAct,CyberSecurityLawandtheGeneralPrinciplesoftheCivilLaw.Theseregulationsjustifiablyaimtoprotecttheprivacyofindividualsandthereforerestrictdirectdatasharingbetweendifferentparties(Lietal.,2022).Thisprotectionofprivacyisanimportantpursuitbutleadstomoredatasilosandthereforeunusedeconomicpotential.

FederatedMachineLearning(FedML)introducedbyMcMahanetal.(2016)isanovelmachinelearning(ML)technologywiththepotentialofbuildingpredictionmodelsofdecentralizedandthereforesiloeddatasets.Incontrasttotraditional,centralizedML,FedMLsystemsinitiallytrainaglobalMLmodelwhichisthendistributedtoallparticipants.Then,eachparticipantindividuallytrainsthemodellocallyontheirowndataset.Theclientssolelyreturntheupdategradientresultingfromthelocaltraining.Throughthismodel-to-dataapproach,thedataneverleavestheclient’sdevice,butstillenablesthedevelopmentofajointMLmodel.Thus,FedMLenablestappingthepotentialofbigdatawithoutprivacyleakage.

FedMLtechnicallyhasthepotentialtoleveragesiloeddatawhilestillpreservingtheintellectualproperty(IP)andprivacyofeachindividuals’dataset.Hence,FedMLenablestheusageofcurrentlyuntappeddataandthereforebringsthepotentialtobethecatalystfornovel,disruptivebusinessmodelinnovationandlockingunprecedentedvaluefromsiloeddata.However,thisrequiresthecollaborationofmultiplepartieswhichownthesedatasilos.Hence,acollaborativebusinessmodelisneededasaframeworkforhowvaluecanbecreated,anddifferentpartiescanbeincentivizedforparticipatinginsuchacollaborativenetwork.Settingupcollaborativebusinessmodelsiscomplexandapotentialreasonforfailure.Thecurrentliteraturelacksguidelinesfordecision-makersonwhichaspectsmustbeconsideredforthesuccessfulrealizationofcollaborativeAIprojects.

Thisworkaimstowardsclosingthisknowledgegap.Morespecifically,weinvestigatethechallengesofprevailingcollaborativebusinessmodelsthroughasystematicliteraturereviewandidentifydistinctaspectsofcollaborativeFedMLprojectsbyconductingafocusgroupinterviewandmultipleexpertinterviews.Weworktowardsasystemizedcollectionofsocio-technicalchallengesandaneasilyconsumablebusinessmodelcanvas(BMC)toaiddecision-makersintheinitialviabilityassessmentofcollaborativeAIprojects.Summarized,weaimtoanswerthefollowingresearchquestions(RQs):RQ1:Whatarethegeneralchallengesofcollaborativebusinessmodels?

RQ2:Whataretheaspectsofinter-organizationalFedMLbusinessmodelsinrelationtoprevailingcollaborativebusinessmodels?

RQ3:Whichaspectsandattributesshouldbeconsideredforinter-organizationalFedMLprojectsandhowcanthesebestructuredintoanextendedBMC?

Toaddresstheseresearchquestions,wefirstdescribethetheoreticalbackgroundofourstudybyintroducingFederatedMachineLearningandprovidingbackgroundinformationoncollaborativebusinessmodels(section2).Following,weelaborateonourtripartiteresearchmethodology,whichconsistsofasystematicliteraturereview,in-depthfocusgroupinterviews,andsemi-structuredexpertinterviews(section3).Subsequently,wepresenttheresultsofourresearchincludingasystemizedoverviewofchallengesforcollaborativebusinessmodels,astructuredlistofdistinctsocio-technicalaspectsforFedMLprojectsandaproposalforacorrespondingextendedBMC(section4).Finally,wediscussourworkbyreflectingtheunderlyingresearchproblemandresearchgaps.Thediscussionisfollowedbyasummaryofourcontributions,answerstotheRQsandlimitationsofourwork.Ourstudyconcludeswithanoutlineoffutureresearch(section5).

2TheoreticalBackground

Thefollowingsectionpresentsthetheoreticalbackgroundofourstudy.Wefirstdescribethemotivation,terminologies,andthebasicconceptofFedMLasoriginallyproposedbyMcMahanetal.(2016).Subsequently,weprovidegeneralbackgroundinformationonbusinessmodelstoestablishacommon

Thirty-firstEuropeanConferenceonInformationSystems(ECIS2023),Kristiansand,Norway3

Socio-TechnicalChallengesofCollaborativeAI

understandingforthisstudy.Finally,weelaborateoncollaborativebusinessmodelsandcorrespondingextensionsoftheBMCbyOsterwalderandPigneur(2010).

2.1FederatedMachineLearning

AclassicMLapproachrequiresthecollaboratingparticipantstoassembletheirdatasetsinacentrallocationandtrainauniqueMLmodelMSUM,exposingthedatatoeachotherandthecentralserver.TheparticipantstherebyrisklosingtheirdatasovereigntyandIP,whichinhibitscompaniestocollaborateandsharedata(Schomakersetal.,2020).IntroducedbyMcMahanetal.(2016),FedMLcounteractstheneedofsharingdatasetsthroughamodel-to-dataapproach.AsillustratedinFigure1,aglobalMLmodelischosen,whichisdistributedamongstallclients.Theclientstrainthemodellocallyontheirindividualdataset.Theupdategradientsaresentbacktotheserverandusedtoimprovetheglobalmodel.Thereby,FedMLenablesdataownerstotrainajointmodelMFEDwithouttheneedtodisclosetheirdata.

Figure1.OneiterationoftheFederatedMachineLearningprocess(source:ownwork).

IntheoriginalFedAVGimplementationbyMcMahanetal.(2016)themodelislearnedthrough

stochasticgradientdescent(SGD),whereeachpartykcomputestheaveragegradientgk=7Fk(wt)

Thepartysubmitsthegradientstothecentralserver,whichaggregatestheupdatesfromallpartiesas:

onitslocaldatankatthecurrentmodelwtanditeratesmultipletimesovertheupdatewk←wk?刀gk.

wt+1←wt?/=1w1

w1←w?刀gk,?k

WhileclassicFedMLoperatesonaclient-serverarchitecture,alternativesthatdonotrelyonacentralorchestratingserverarealsopossible.Forinstance,partiescanexchangemodelupdatesbyestablishing

apeer-to-peernetwork,increasingthesecurityoftheprocessattheexpenseofconsumingmorebandwidthandresourcesforencryption(Royetal.,2019).

Moreover,thedistributionoffeaturesandsamplesacrossdatasetsmaynotbehomogeneous.Horizontal

FederatedLearning(HFL)referstothesetupinwhichalldatasets{D1}fromtheKpartiescontain

differentsamplesthatsharethesamefeaturespace.Ifinstead,thesamesamplesarepresentinall

datasets,butfeaturespacesaredisjoint,thesetupisknownasVerticalFederatedLearning(VFL).

Consideringthehighheterogeneityofdata,especiallyifspreadacrossdifferentorganizations,some

authorshaveproposedtoovercometheproblemofsparseoverlappingdatasetsthroughFederated

TransferLearning(FTL)(Liuetal.,2020).Inthisscheme,partiesmayselectsamplesfortrainingthat

minimizesthedistancebetweentheirdistributions(instance-basedFTL)orlearnacommonfeature

spacecollaboratively(feature-basedFTL).Alternatively,partiesmaystartbyusingpre-trainedmodels

orbylearningmodelsfromalignedsamplestoinfermissingfeaturesandlabels(model-basedFTL).

Finally,itisimportanttonotethattheperformancesvSUMandvFEDoftherespectivecentralizedand

vFED<6andwillbestronglydependentonthecharacteristicsoftheparticularapplication.

federatedmodels,mightdifferconsiderably.Thisperformancegap6ischaracterizedbyvSUM?

Thirty-firstEuropeanConferenceonInformationSystems(ECIS2023),Kristiansand,Norway4

Socio-TechnicalChallengesofCollaborativeAI

Consequently,FedMLintroducesapotentialtrade-offbetweenthelossofperformancerespecttothecentralizedsetupandtheprivacyguaranteesprovidedbythedistributedapproach(Yangetal.,2019).

2.2CollaborativeBusinessModels

Abusinessmodeldescribesessentialaspectsofanorganization,explaininghowtheorganizationcreates,delivers,andcapturesvalue(OsterwalderandPigneur,2010).Intheacademicliterature,thedefinitionofthetermisfragmented,andnoconsistentboundariesareestablished.Nevertheless,itcanbestatedthatabusinessmodelprovidesanorganizationalandstrategicdesignforimplementingabusinessopportunity(George,2011).

Inaddition,OsterwalderandPigneur(2010)arguethatasharedunderstandingofthebusinessmodeliscrucialtoitscreationandsuccess.Therefore,creatinganddiscussingabusinessmodelrequiresasimple,relevant,andintuitivelyunderstandableconceptwithoutoversimplifyingthecomplexityofhowtheorganizationworks.TheBMCbyOsterwalderandPigneur(2010)isatooloftenusedinpracticetopresentabusinessmodelstructuredinninecomponents.

Businessmodelsarenotonlyusedforasinglecompanybutcanalsosupportassessingthefeasibilityandprofitabilityofcollaborationsacrosscompanies(KristensenandUcler,2016).Thetrendofaninterconnectedanddynamicenvironmentencouragesorganizationstocollaborateinter-organizationallyandco-createvalue(DiirrandCappelli,2018).Inliterature,nounifiedframeworkexistsforcollaborations.Still,someapproachesutilizeOsterwalderandPigneur(2010)generalapproachofabusinessmodelasabasisandcustomizeittosetahigherfocusonspecifics(KristensenandUcler,2016).Forexample,EppingerandKamprath(2011)highlighttheimportanceofapartnerandcustomernetworkinpersonalizedmedicinebymodifyingthecanvascomponentsandaddingnewones,likeintellectualpropertystrategy.Theapproachesintheliteraturereachfrommodificationsofbusinessmodelcomponents(e.g.,EppingerandKamprath(2011)orKristensenandUcler(2016)),toconfigurationoptionsofthebusinessmodel(e.g.,Curtis(2021)orManandLuvison(2019)).However,thecustomizationsaremainlyapplication-oriented,tailoredtotheprojecttosuittheneedsandcaptureuniquefeaturesinfluencingthebusinessmodelandthusdecisivefortheproject'ssuccess.

3Methodology

Thisresearchwasstructuredintothreedistinctparts.Afterasystematicliteraturereview(SLR)togainanoverviewofthechallengesofcollaborativebusinessmodels,weorganizedanin-depthfocusgroupinterviewtoexplorethenovelfieldofinter-organizationalFedMLbusinessmodels.Bythis,weaimedtoaugmentthefindingsfromtheSLRandidentifydistinctchallengesofbusinessmodelsforcollaborativeFedMLprojects.Sincefocusgroupsarecharacterizedbytheirhomogeneousgroupdemographic,wepursuedmoregenericallyapplicableresultsbyconductingadditionalsemi-structuredexpertinterviews.Theresearchtimelineisdisplayedinfigure2.Thefollowingsubsectionswillgointomoredetailabouttheusedresearchmethodologies.

Figure2.ResearchTimeline

Thirty-firstEuropeanConferenceonInformationSystems(ECIS2023),Kristiansand,Norway5

Socio-TechnicalChallengesofCollaborativeAI

3.1SystematicLiteratureReview

Toassessandidentifythechallengesofprevailingcollaborativebusinessmodels,weconductedasystematicliteraturereview.Bythis,weintendtoextractfundamentalsandcriticalattributesofbusinessmodelsforinter-organizationalcollaborations,whichwillbecollected,structured,andsummarized.WefollowedasearchstrategybyZhangetal.(2011)toidentifythemostrelevantliterature.Hence,thesearchisdividedintobaseliteraturesearch,mainsearch,andbackwardsearch.

TheBaseLiteratureconsistsof13papersincludingfivepublicationsonbusinessmodeltheoryandeightoncollaborativeprojectswhichwereknownbytheauthorspriortothesearch.Basedontheinitialliteraturecorpus,wefocusedonfindingkeywordsrelatedtointer-organizationalbusinessmodels.TheresultingsearchstringSisasfollows:

String

Query

S1

(collaborate*ORfederatedORinterorganization*ORinter-organization*ORintercompanyOR

cross-companyORmulti-partyORcross-industr*ORmulti-institution*ORsharedORsharing

ORallianceORnetworked)

S2

(“businessmodel*”O(jiān)R“modelcanvas”O(jiān)R“businessvaluemodel*”)

S

S1ANDS2

Table1.Compiledsearchstringforthedatabasesearch.

TheMainSearchwasconductedfromApril2022toJune2022.ThelistofsearcheddatabasescomprisesIEEEXplore,ACMDigitalLibrary,ScienceDirect,WileyInterScienceandSCOPUS.Weonlyincludedpeer-reviewedEnglishandGermanpublicationswithfull-textaccess.WiththedefinedsearchstringS,databases,andcriteriawecollected262distinctpublications.Weonlyaimtoincludeworkinthefieldofcomputerscienceandtechnology(coarsefocus)aswellasliteratureregardinginter-organizationalcollaborationsandbusinessmodels(narrowfocus).Successively,thecorpusconsistingof262distinctpublicationshasbeenfilteredsolelybytitle,abstractandfulltextregardingthedefinedcoarseandnarrowfocus.Bythis,18publicationsremained.

FortheBackwardSearch,wescannedthereferencesoftheresulting18publicationsfromthemainsearch.Again,thesereferencedpublicationswerefilteredbytitle,abstractandbodyaccordingtotheinclusionandexclusioncriteria.Aftereliminatingduplicates,weaddedonefurtherstudyresultinginatotalof19publications.

Finally,relevantinformationwasextractedandsynthesizedfromthefinalliteraturecorpus.Thestructuredandconsolidatedoutputyieldedasetofcriticalattributesandchallengesofbusinessmodelsforinter-organizationalcollaborationsinthetechnologicalsector.InthefollowingeveryinsightfromtheSLRisreferencedasusualwiththecorrespondingpublication.

3.2In-depthGroupInterview

BasedontheidentifiedchallengesofprevailingcollaborativebusinessmodelsfromtheSLR,weaimedtoexplorethedistinctaspectsofcollaborativebusinessmodelsforinter-organizationalFedMLprojects.Duetothenoveltyofthetopicandtheneedforexploration,weorganizedanin-depthfocusgroupinterview(DilshadandLatif,2013)tostudythebusinessrequirementsofcollaborativeMLprojectsbasedonthefindingsfromtheSLR.

Thefocusgroupconsistedoffiveparticipantsandtwomoderators,whereonemoderatorensuressmoothprogressandtheotherensuresthatalltopicsarecovered.AllparticipantsworkedonaprojectinvolvingtheadoptionofFedMLinacross-companyusecase.TheparticipantswerebriefedaboutcollaborativebusinessmodelsandweregivenanoverviewofthefindingsfromtheSLR.Afterwards,thegroupwas

Thirty-firstEuropeanConferenceonInformationSystems(ECIS2023),Kristiansand,Norway6

Socio-TechnicalChallengesofCollaborativeAI

askedaboutthecriticalattributesandchallengesofbusinessmodelsrelatedtotheircollaborativeFedMLprojectfollowedbyareflectionandlivelydiscussion.Throughthis,wewereabletoidentifyfurtherchallengesbasedontheirreal-worldexperiences.TheemergingdatawascodedbytworesearchersandincorporatedintotheresultsoftheSLR.Inthefollowing,everyinsightwhichwasgainedthroughthein-depthfocusgroupinterviewisreferencedviatheindex(FG).

3.3Semi-StructuredExpertInterviews

Eventhoughthefocusgrouphelpedexplorethesocio-technicalchallengesofFedMLcollaborations,theinsightsmightbehighlybiasedduetothehomogeneousdemographicsoftheparticipants.Togainamoregenericallyapplicableunderstanding,weaimedtodrawfromtheexperiencesoffurtherexpertsworkinginthefieldofappliedAI,especiallywithexperienceinFedMLprojects.Fortheseexpertinterviews,wedrawfromtheGroundedTheorymethodology(Hodaetal.,2011).Hence,weconfrontedtheintervieweeswithasetofpre-definedquestionsandrecordedaswellastranscribedtheinterviews.Wesuccessivelyconductedandcomparedtheresultsofeachinterview.After5interviewstheoreticalsaturationwasreachedandconsequently,theinterviewstudywasclosed.Thesetofintervieweesrepresentedamorediversesetofexpertsfromdifferentorganizationsanddomains.Table1presentsacodifiedtableofoursample.WedevelopedaninterviewguidebasedontheresearchquestionsandfindingsfromtheSLRaswellasthefocusgroupinterviewincludingopenquestionsaboutpotentiallymissingattributes,challenges,andfurtherinsights.Theseinterviewsallowedustogomorein-depthandidentifymissingaspectsandgainmoredetailed,in-depthindividualunderstandingtodeveloptheguidelinequestionnairefurther.

Theintervieweesallowedthefindingstobepublishedinananonymizedmannerbutdidnotagreetodisclosethefulltranscriptions.Therefore,thefulltranscriptsarenotincluded.Thefindingsfromthesemi-structuredinterviewsarereferencedinthefollowingwiththeparticipantIDaslistedintable2.

ParticipantID

Position

Organization

Duration

E1

AIBusinessDeveloper

LargeGermansoftwareenterprise

52

E2

AIProjectLead

LargeGermansoftwareenterprise

44

E3

PrincipalDataScientist

LargeGermansoftwareenterprise

45

E4

AppliedResearcher

Medium-sizedinnovationcompany

35

E5

ScientificResearcher

Researchinstituteforsoftwaredevelopment

59

Table2.InterviewStudyParticipants

4Socio-TechnicalChallengesofInterorganizationalFederatedMachineLearning

Applyingcollaborativemodelscanbechallengingindifferentdomains,especiallywhenseveralcompaniesareinvolved.Whenthebusinessisoperationalized,complexityincreasessignificantlybecausethegeneralbusinessmodelideaneedstobalancetheinterestsofallparticipants(Paunaetal.,2021).Collaborationswithmultipleparticipantsarecomplexinnature,andcollaborationfailureratesarehigh,leavingmuchrevenueatriskandunrealizedvalue(ManandLuvison,2019).Moreover,aligningthebusinessmodelwithoperationalandgovernance-relatedaspectsissuggestedtohelppositiontheorganizationtodeliveronitsvaluepropositionforasuccessfulimplementationofthebusinessmodel(Curtis,2021).Hence,earlyidentificationofthecollaborationchallengesiscriticalforthesuccessfulcreationofthecollaborativebusinessmodel.

Tobetterunderstandwhichspecificcollaborationchallengesshouldbeconsidered,wefirstinvestigatethechallengesofprevailinginter-organizationalbusinessmodelsand,secondly,whichFedML-related

Thirty-firstEuropeanConferenceonInformationSystems(ECIS2023),Kristiansand,Norway7

Socio-TechnicalChallengesofCollaborativeAI

socio-technicalaspectsarecriticalforsuccessfulimplementationandthereforeshouldbeconsideredinacorrespondingcollaborativebusinessmodel.

4.1ChallengesofCollaborativeBusinessModels

Jointworkofdifferentorganizationsiscomplex,andorganizationsshouldbepreparedtofacechallengesarisingfromcooperation.Inthefollowing,wegiveanoverviewofthesystematizedresultsoftheSLRonthechallengesofinter-organizationalbusinessmodels.WepresentourkeyresultsinastructuredmannerbasedontheworkofDiirrandCappelli(2018).

DiirrandCappelli(2018)dividethechallengesofinter-organizationalcollaborationsintothreecategories:external,internal,andnetwork-relatedchallenges.Externalandinternalchallengesaredetachedfrominter-organizationalcollaboration.Externalchallengesrelatetoenvironmentalchallenges,e.g.,naturalevents,andinternalchallengesarisefrominsidetheproject,forexample,infrastructureproblems.Network-relatedchallengesfocusontherelationshipsandinteractionsbetweenorganizationsandcanbefurthersubdividedintomanagement,businessprocess,andcollaborationchallenges(DiirrandCappelli,2018).BasedonthesecategoriesinconjunctionwiththefindingsoftheSLRwederivedthefollowingnetwork-relatedchallengesaslistedintable3.

Category

Description

Aspects

Management

Challenges

the

Includehoworganizationscreateandestablishcollaboration,compromisingthefollowingaspects

Selectionofsuitableparticipatingactors(Paunaetal.,2021).

Changemanagementfordynamiccollaboration(Caridàetal.,2015;Redlichetal.,2014).

Cooperationestablishment:lackofcommitmentfromparticipatingorganizations(ProulxandGardoni,2020);buildingandexpandingtrustbetweentheparties(Blejaetal.,2020;DiirrandCappelli,2018;Redlichetal.,2014).

Decision-makingandcoordinationslownesswithinthecollaboration(Blejaetal.,2020;Caridàetal.,2015;DiirrandCappelli,2018;Redlichetal.,2014).

Communicationwithgovernmentauthoritiesrequiresadifferentapproachduetomultipleparties'interactions(Paunaetal.,2021).

BusinessProcessChallenges

Addresseswayorganization’sstructureanddesignpartnershipoperations

Definitionofamutualbusinessgoalofthecollaboration(DiirrandCappelli,2018).

Co-CreationManagementfordeliveringthevalueproposition:

?Distributionoffinancials,investment(Paunaetal.,2021),costs,andrevenues(Blejaetal.,2020;Caridàetal.,2015;Paunaetal.,2021).

?Riskallocation(DiirrandCappelli,2018).

?Ownershipstructure(DiirrandCappelli,2018;Kujalaetal.,2020).

?Responsibilityassignment(DiirrandCappelli,2018).

?Alignonqualityofco-creationproduct(DiirrandCappelli,2018).

?IntellectualpropertyManagement(EppingerandKamprath,2011).

Thirty-firstEuropeanConferenceonInformationSystems(ECIS2023),Kristiansand,Norway8

Socio-TechnicalChallengesofCollaborativeAI

Slowerbusinessstrategyandprocessidentification(Berkersetal.,2020;DiirrandCappelli,2018).

Alignmentofthestructuresofheterogeneousorganizationswithdistinctcharacteristics(DiirrandCappelli,2018).

Infrastructureformanagingrelationshipsbetweenmultiplecollaborationactors(Caridàetal.,2015;DiirrandCappelli,2018).

Collaboration

Challenges

Describe

how

organizationsjointlyworktogethertoachievethegoalofcollaboration

Agreementoncollaborationandalignmentwiththeorganizations'ownobjectives(Blejaetal.,2020;CostaandDaCunha,2015;DiirrandCappelli,2018;ManandLuvison,2019;Paunaetal.,2021).

Alignmentofdifferentorganizations:cultureandcommonethics(Blejaetal.,2020;DiirrandCappelli,2018;Kujalaetal.,2020).

Riskofopportunismofparticipantsandconsequencesofaction(DiirrandCappelli,2018).

Table3.OverviewofChallengesforInter-OrganizationalCollaborations

ThisoverviewofchallengesisaconsolidationoftheselectedacademicsourcesoftheSLRandaimstoprovideageneralunderstandingofthedifficultiesofsuchcollaborations.Itisimportanttonotethatnaturally,thislistmightnotbecomprehensiveandthatcertain,potentiallyimportant,aspectsmightbemissing.

4.2AspectsofInterorganizationalFedMLBusinessModels

Toprovideinitialguidanceinthecreatio