BCG-工业分销商价值创造者2023（英）_市场营销策划_重点报告202301202_doc.docx

GenerativeAPsRoleintheFactoryoftheFutureDECEMBER08,2023ByDanielKupper,KristianKuhlmann,MonikaSaundersjJohnKnapp,Kai-FredericSeitzjJuIianEnglberger,TilmanBuchner,andMartinKleinhansREADINGTIME:15MINGenerativeAlisoneoftoday,shottestbusinesstopics,withcompaniesexploringitspotentialapplicationsandbenefitsacrossindustriesandfunctions,includingmanufacturing.Butdespitetherecentbuzz,manufacturersshouldrecognizethatsimplyapplyingtoolslikeChatGPTontheirownwillnotrevolutionizefactoryoperations.InsteadofreplacingtraditionalAl,GenAIofferscomplementaryusecasesintheareasofassistance,recommendations,andautonomythatpavethewaytothefactoryofthefuture.Itdoessothroughitscapacitytogeneratecontent,suchastextandimages,tailoredtospecifictasksorinquiries.(SeeuHowGenAIWorks.n)Howgenaiworks-TodiscusstheapplicationsofGenAI,itisessentialtofirstdefinehowitdiffersfromttclassica,machinelearning(ML).ClassicalMLalgorithmsdiscernpatternswithinobserveddata,enablingthemtogeneralizetheseinsightstonew,previouslyunseendata.Forinstance,anMLmodelmightbetrainedusingspecifictextfragmentssuchasoperatorincidentreportsinwhichmachinebreakdowndescriptionsareclassifiedintospecificrootcausessuchas,endoftoolinglife"or"operatorerror."Basedonthistraining,themodelcanprocesspreviouslyunseentextfragmentsofincidentreportsandjudgewhatcausedtheincident.Thebasisforsuchmodelsmaybedeepneuralnetworks,supportvectormachines,orothermethods.GenAItakesthisapproachfurther.Beyondmerelyclassifyingexistingtext,itcangeneratenewtextbasedonspecifiedcriteria-suchasoperatorinstructionsthatoutlineaprocesstoresolveaparticularrootcauseofamachinebreakdown.AlthoughtheprogressionfromclassicalMLtoGenAImightseemincremental,itposesafundamentaltechnicalchallenge.InclassicalML,themodelmerelyneedssuffcienttrainingtoconfidentlycategorizeatextfragment.Incontrast,GenAImustconstructatextfragmentfromindividualwordsandletters,ensuringthatitisgrammaticallycorrect,comprehensible,andaccuratelyrepresentstheprocess.ThenumberofpotentialoutputsfromGenAIisvirtuallylimitless.Consideringthatthereareroughly170,000Englishincurrentuse,amerefive-wordtexthasmorethan140septillionpotentialcombinations.Ontheotherhand,onlyafractionofthemwouldbegrammaticallycorrectandunderstandable.Amongthose,anevensmallerfractionwouldaccuratelydescribeagivenprocesstofixtherootcauseofamachinebreakdown.Consequently,themarginforerrorinGenAImodelsisincrediblynarrow,necessitatingextremelyprecisemodels.Toattainthisprecision,GenAImustuseufoudationalmodels5*insteadofbeingtrainedonlyoncontext-specificdata.Foundationalmodelsaretrainedonextensivedatasets,suchasallavailabletextorimagesonline,andaresubsequentlyfine-tunedforspecificapplications.Thesemodelscanbelargelanguagemodels(suchasOpenAsGPT-4orAmazonQ)orimageorspeechmodels.Theyseemtogainanunderstandingofrealityfromtheextensivedatasets.However,foundationalmodelsareobservationallearnersthatdonotapplylogicorreasonashumansdo.ThismeansthatthereisnoguaranteeofplausibleoraccurateresultsfromGenAI.Inouroperatorinstructionexample,thefoundationalmodelfirstlearnswhatconstitutescomprehensibleandaccuratetext,withprocessdescriptionsbeingasmallsubset.Next5themodelisfine-tunedbylearningwhatoperatorinstructionslooklikeandhowtheycorrelatewithgivenmachinebreakdownrootcauses.However,thereisnoassurancethatthemodelwillcreatecorrectorhigh-qualityoperatorinstructions.Ergonomicsillustratestheproblem.BecausetheGenAImodellacksinsightintotheprocessthatfixestherootcauseandthepeoplewhoaretheoperators,itmightoverlookpotentiallimitations,suchasinfeasiblemovementsorinaccessiblespaces.Asaresult,aqualityassessmentisalwaysrequiredtoensurethattherecommendedremediationispracticalfromanergonomicperspective.GenAsgreatertechnicalcomplexityelevatestheimportanceofestablishingarobusttechnologicalfoundationtoharnessitscapabilitieseffectively.Withanumberofarchetypespossible,manufacturersmustunderstandthefactorsthatdetermineanoptimalchoice.TheycanapplythisknowledgetointegrateGenAIintofactoryoperations,consideringthevalue-addingapplications,change-andpeople-relatedinitiatives,andrequiredtechnologicalinfrastructure.ManufacturersArePrioritizingGenAIforitsDisruptivePotentialBCGrecentlysurveyedmanufacturerstounderstandtheirperspectiveontechnologydevelopments.(See"AbouttheSurvey/*)Regardlessoftheiraffinityfordigitaltechnology,manufacturingexecutivesrankedAl(includingGenAI)firstamongtechnologiesthatcouldpositivelydisrupttheiroperations.(SeeExhibit1.)ThepotentialROIwarrantstheirenthusiasm.ABCGanalysisfoundthattheuseofAlcouldenhanceshop-floorproductivitybymorethan20%.ABOUTTHESURVEYBCGconductedaglobalsurveyfromJanuarythroughMarch2023toassessthelatesttechnologiesinthemanufacturingindustry.Approximately1,800respondentsfrom15countriesspanningNorthandSouthAmerica,Europe,andAsiatookpartinthestudy,witheachcountrycontributingmorethan100completedsurveys.Theparticipantsrepresentabroadarrayofproductionindustries,includingautomotive,capitalgoods,consumergoods,energy,IT,healthcare,andmaterials.Exhibit1-ManufacturingExec