版權(quán)說明:本文檔由用戶提供并上傳,收益歸屬內(nèi)容提供方,若內(nèi)容存在侵權(quán),請進(jìn)行舉報(bào)或認(rèn)領(lǐng)
文檔簡介
GenAIDoesn’tJustIncreaseProductivity.ItExpands
Capabilities.
SEPTEMBER05,2024
By
DanielSack
,
LisaKrayer
,EmmaWiles,MohamedAbbadi,UrviAwasthi,RyanKennedy,CristiánArnolds,andFran?oisCandelon
READINGTIME:12MIN
Thisisthesecondmajor?eldexperimentledbytheBCGHendersonInstitutedesignedtohelpbusinessleadersunderstandhowhumansandGenAIshouldcollaborateintheworkplace.Ourpreviousstudy
assessedthe
valuecreated—anddestroyed—byGenAI
whenusedbyworkersfortaskstheyhadthe
?2024BostonConsultingGroup1
capabilitiestocompleteontheirown.OurlatestexperimenttestshowworkerscanuseGenAItocompletetasksthatarebeyondtheircurrentcapabilities.
Anewtypeofknowledgeworkerisenteringtheglobaltalentpool.Thisemployee,augmentedwith
generativeAI
,canwritecodefaster,createpersonalizedmarketingcontentwithasingleprompt,andsummarizehundredsofdocumentsinseconds.
Theseareimpressiveproductivitygains.Butasthenatureofmanyjobsandtheskillsrequiredtodothemevolve,workerswillneedtoexpandtheircurrentcapabilities.CanGenAIbeasolutionthereaswell?
Basedontheresultsof
anewexperiment
conductedbytheBCGHendersonInstituteandscholarsfromBostonUniversityandOpenAI’sEconomicImpactsresearchteam,theanswerisan
unequivocalyes.We’venowfoundthatit’spossibleforemployeeswhodidn’thavethefullknow-howtoperformaparticulartaskyesterdaytouseGenAItocompletethesametasktoday.
“
Employeeswhodidn’thavethefullknow-howtoperformaparticulartaskyesterdaycanuseGenAItocompletethesametasktoday.
Withthatinmind,leadersshouldembraceGenAInotonlyasatoolforincreasingproductivity,butas
a
technology
thatequipstheworkforcetomeetthechangingjobdemandsoftoday,tomorrow,andbeyond.TheyshouldconsidergenerativeAIanexoskeleton:atoolthatempowersworkersto
performbetter,anddomore,thaneitherthehumanorGenAIcanontheirown.
Ofcourse,thereareimportantcaveats—forexample,employeesmaynothavetherequisite
knowledgetochecktheirwork,andthereforemaynotknowwhenthetoolhasgottenitwrong.Ortheymaybecomelessattentiveinsituationswheretheyshouldbemorediscriminating.
Butleaderswhoe?ectivelymanagetheriskscanreapsigni?cantrewards.TheabilitytorapidlytakeonnewtypesofworkwithGenAI—particularlytasksthattraditionallyrequirenicheskillsthatare
harderto?nd,suchasdatascience—canbeagame-changerforindividualsandcompaniesalike.
HowGenAICanEquipKnowledgeWorkers
Inthepreviousexperiment,wemeasuredperformanceontasksthatwerewithintherealmoftheparticipants’
capabilities.1
(SeetoprowofExhibit1.)FortaskswhereGenAIishighlycapable,we
?2024BostonConsultingGroup2
foundthataugmentedworkersperformsigni?cantlybetterthanhumansworkingwithoutthe
technology.However,whenthetechnologyisnotcapableofperformingthetaskatexpertlevel,
humanstendtoover-relyonGenAIandperformworsethaniftheyhadcompletedthetaskontheirown.
:
Butwhathappenswhen,insteadofusingGenAItoimproveperformancewithintheircurrentskillset,peopleuseGenAItocompletetasksthatareoutsidetheirowncapabilities?DoesbeingaugmentedwithGenAIexpandthebreadthoftaskspeoplecanperform?
Forourlatestexperiment,morethan480BCGconsultantsperformedthreeshorttasksthatmimicacommondata-sciencepipeline:writingPythoncodetomergeandcleantwodatasets;buildinga
predictivemodelforsportsinvestingusinganalyticsbestpractices(e.g.machinelearning);and
validatingandcorrectingstatisticalanalysisoutputsgeneratedbyChatGPTandapplyingstatistical
2
metricstodetermineifreported?ndingsweremeaningful.
Whilethesetasksdon’tcapturetheentiretyofadvanceddatascientists’workload,theyare
su?cientlyrepresentative.Theyweredesignedtopresentasigni?cantchallengeforanyconsultant
3
andcouldnotbefullyautomatedbytheGenAItool.
TohelpevaluatetheperformanceimpactofGenAI,onlyhalfoftheparticipantsweregivenaccesstotheGenAItool,andwecomparedtheirresultstothoseof44datascientistswhoworkedwithouttheassistanceofGenAI.Whenwedivedeeperintotheresults,threecritical?ndingsemerge.
TheImmediateAptitude-ExpansionE?ect
WhenusingGenAI,theconsultantsinourstudywereabletoinstantlyexpandtheiraptitudefornewtasks.Evenwhentheyhadnoexperienceincodingorstatistics,consultantswithaccesstoGenAI
wereabletowritecode,appropriatelyapplymachinelearningmodels,andcorrecterroneous
4
statisticalprocesses.(SeeExhibit2.)
?2024BostonConsultingGroup3
:
Weobservedthebiggestaptitude-expansione?ectforcoding,ataskatwhichGenAIishighlyadept.
Participantswereaskedtowritecodethatwouldcleantwosalesdatasetsbycorrectingmissingorinvaliddatapoints,mergingthedatasets,and?lteringtoidentifythetop?vecustomersina
speci?edmonth.
ParticipantswhousedGenAIachievedanaveragescoreequivalentto86%ofthebenchmarksetbydatascientists.Thisisa49-percentage-pointimprovementoverparticipantsnotusingGenAI.The
GenAI-augmentedgroupalso?nishedthetaskroughly10%fasterthanthedatascientists.
Eventhoseconsultantswhohadneverwrittencodebeforereached84%ofthedatascientists’
benchmarkwhenusingGenAI.Oneparticipantwhohadnocodingexperiencetoldus:“IfeelthatI’vebecomeacodernowandIdon’tknowhowtocode!Yet,IcanreachanoutcomethatIwouldn’thave
beenabletootherwise”ThoseworkingwithoutGenAI,ontheotherhand,o代endidnotgetmuch
furtherthanopeningthe?lesandcleaningupthe?rst“messy”data?elds;theyachievedjust29%ofthedata-scientistbenchmark.
It’simportanttonotethatmostconsultantsareexpectedtoknowthebasicsofdatacleaningando代enperformdata-cleaningtasksusingno-codetoolssuchasAlteryx.Therefore,whiletheydidnothaveexperiencedoingthecodingtaskinPython,theyknewwhattoexpectfromacorrectoutput.
ThisiscriticalforanyGenAI-augmentedworker—iftheydon’thaveenoughknowledgetosupervisetheoutputofthetool,theywillnotknowwhenitismakingobviouserrors.
APowerfulBrainstormingPartner
Forthetaskthatinvolvedpredictiveanalytics,ourparticipantsfacedachallengingscenario:neithertheynortheGenAItoolwerehighlyadeptatthattask.Here,thetechnologywasstillvaluableasabrainstormingpartner.
?2024BostonConsultingGroup4
WhileallthetasksinourexperimentweredesignedsuchthattheGenAIcouldnotindependently
solvethem,thepredictive-analyticstaskrequiredthemostengagementfromparticipants.Theywereaskedtocreateapredictivemodel,usinghistoricaldataoninternationalsoccermatches,todevelopaninvestmentstrategy.Theirultimategoalwastoassesshowpredictable,orreliable,theirmodel
wouldbeformakinginvestmentdecisions.
“
ManyparticipantsusedGenAItobrainstorm,combiningtheirknowledgewiththetool’sknowledgetodiscovernewmodelingandproblem-
solvingtechniques.
AsshowninExhibit2,thiswasthetaskonwhichtheGenAI-augmentedconsultantwasleastlikelytoperformonparwithadatascientist,regardlessofpreviousexperienceincodingorstatistics.ThisisbecausetheGenAItoolislikelytomisunderstandtheultimategoalofthepromptiftheentiretaskiscopiedandpasteddirectlyintothetoolwithoutbreakingthequestionintopartsorclarifyingthe
goals.Asaresult,participantswithaccesstoGenAIweremorelikelytobeledastraythantheirnonaugmentedcounterparts.
Evenso,wefoundthat,withthesupportofGenAI,manyparticipantswereabletostepoutsidetheircomfortzone.Theybrainstormedwiththetool,combiningtheirknowledgewithGenAI’sknowledge
todiscovernewmodelingtechniquesandidentifythecorrectstepstosolvetheproblemsuccessfully.TheGenAI-augmentedparticipantswere15percentagepointsmorelikelytoselectandappropriatelyapplymachine-learningmethodsthantheircounterpartswhodidnothaveaccesstoGenAI.
Reskilled,butOnlyWhenAugmented
Participants’aptitudeforcompletingnewandchallengingtaskswasimmediatelyboostedwhen
usingGenAI,butweretheyreskilled?Reskillingisde?nedasanindividualgainingnewcapabilitiesorknowledgethatenableshimorhertomoveintoanewjoborindustry.Wefoundinourstudythat
GenAI-augmentedworkerswereinasense“reskilled,”inthattheygainednewcapabilitiesthatwerebeyondwhateitherthehumanorGenAIcoulddoontheirown.ButGenAIwasonlyanexoskeleton;theparticipantswerenotintrinsicallyreskilled,because“doing”withGenAIdoesnotimmediately
norinherentlymean“l(fā)earningtodo”
Whileeachparticipantwasassignedjusttwoofthethreetasksintheexperiment,wegaveeveryonea?nalassessmentwithquestionsrelatedtoallthreetaskstotesthowmuchtheyactuallylearned.Forexample,weaskedacodingsyntaxquestioneventhoughnoteveryonedidthecodingtask—andthereforenoteveryonewouldhavehadachanceto“l(fā)earn”syntax.Yetthepeoplewhoparticipated
?2024BostonConsultingGroup5
inthecodingtaskscoredthesameontheassessmentaspeoplewhodidn’tdothecodingtask.
Performingthedata-sciencetasksinourexperimentthusdidnotincreaseparticipants’knowledge.
Ofcourse,participantsonlyhad90minutestocompletethetask.Withrepetition,morelearningmighthaveoccurred.Wealsodidn’tinformparticipantsthattheywouldbetestedattheend,soincentivizinglearningmightalsohavehelped.Thisisimportant,becausewefoundthathavingatleastsomebackgroundknowledgeofagivensubjectmatters.
“
WefoundthatcodingexperienceisakeysuccessfactorforworkerswhouseGenAI—evenfortasksthatdon’tinvolvecoding.
GenAI-augmentedparticipantswithmoderatecodingexperienceperformed10to20percentage
pointsbetteronallthreetasksthantheirpeerswhoself-identi?edasnovices,evenwhencodingwas
5
notinvolved.Infact,thosewithmoderatecodingexperiencewerefullyonparwithdatascientistsfortwoofthethreetasks—oneofwhichhadzerocodinginvolved.
Basedonthis,wepositthatitistheengineeringmindsetthatcodinghelpsdevelop—forexample,havingtheabilitytobreakaproblemdownintosubcomponentsthatcanbee?ectivelycheckedandcorrected—thatultimatelymatters,moresothanthecodingexperienceitself.
Theriskoffullyautomatingcode,then,isthatpeopledon’tformthismindset—becausehowdoyoumaintainthisskillwhenthesourceofitsdevelopmentisnolongerneeded?Thisispartofalarger
discussion:Whatotherseeminglyautomatableskillshavesuchimportance?WilltheseskillsbecomethenewLatin,taughtmostlytocultivateaparticularmindset?
?2024BostonConsultingGroup6
:
ManagingtheTransition
Whilewehaveuseddatascienceasacasestudy,webelievethatour?nding—thataugmentedworkerscanskillfullyperformnewtasks—canbeappliedtoany?eldthatiswithinthetool’s
capabilities.We’veidenti?ed?vecoreimplicationsforcompanyleaders.(SeeExhibit3.)
TalentAcquisitionandInternalMobility.Theresultsacrossourworkforceexperimentshave
shownthatwhatanindividualcanperformonhisorherownbynomeansapproacheswhatcanbeaccomplishedwhenaugmentedbyGenAI.Thissuggeststhatthetalentpoolforskilledknowledge
workisexpanding.
RecruitersshouldthereforeincorporateGenAIintotheinterviewprocesstogetamorecompletepictureofwhataprospectiveemployeemightbecapableofwhenaugmentedbythetechnology.
Leadersmayalso?ndthatanunlikelypersoninsidetheirorganizationcan?llanopenrole.We’re
notsuggestingthatnontechnicalgeneralistscanimmediatelybecomedatascientists.Butageneralistmarketercould,forexample,takeonmarketinganalysttasksorroles.
LearningandDevelopment.Whatdoesthismeanforemployeesseekingpathstoseniorroles
and/orleadership?HowshouldmembersoftheGenAI-augmentedworkforce,whocan?exiblytakeonvariousroles,cultivatetherightskillsforcareeradvancement—andwhatarethemostimportantskillsforthemtoretainlongterm?
?2024BostonConsultingGroup7
WhileGenAIhasanimmediateaptitude-expansione?ect,learninganddevelopmentremainthemostimportleverforcultivatingadvancedskillsandsupportingeachemployee’sprofessional
trajectory.Leadersthereforemustensurethatemployeeshaveincentivizedandprotectedtimetolearn.
Otherresearch
hasshownthatwhenspeci?callyusedforlearning(and,unlikeour
participants,peoplearegenerallyincentivizedtolearnintheirjobs),GenAIisane?ectivepersonalizedtrainingtool.
“
LeadersshouldensurethatfutureimplementationsofGenAItoolsincludethefunctionalitytoinformtheuserifataskisoutsidethetechnology’s
capabilityset.
Ouranalysisalsosuggeststhatdevelopingsometechnicalskillsleadstogreaterperformance,evenfornontechnicalworkers.Regardlessofthetrainingemployeesreceive,companyleadersshould
ensuretheirfutureimplementationsofGenAItoolsincludethefunctionalitytoinformtheuserifataskisoutsidethetechnology’scapabilityset—informationthatshouldbecompiledfromregularbenchmarking.
Companiesarelikelyto?ndcompetitiveadvantagefromdevelopingtoolsandprocessesthat
preciselyassessthecapabilitiesofGenAImodelsfortheirusecases.AsshowninExhibit1,howaworkershoulduseGenAIgreatlydependsonunderstandingwhereatasklieswithintheirownskillsetandwithinthecapabilitiesofthetechnology.
TeamingandPerformanceManagement.Althoughourresultsshowitispossibleforageneralisttotakeonmorecomplexknowledgework,itwillbecrucialtomanagetheirperformanceandensurethequalityoftheiroutput.Thiscouldmeandesigningcross-functionalteamstoprovidegeneralistswitheasyaccesstoanexpertwhentheyneedhelpandestablishingregularoutput-review
checkpoints—becauseanovercon?dentgeneralistmaynotalwaysknowwhentoaskforsupport.
Leaderswillneedtorunpilotstoensuretheirteamingcon?gurationsleadtothebestoutcomes.Thismaybeanopportunitytobreaksilosandintegrateteamsofgeneralistswithexpertsfromvarious
centersofexcellence.
StrategicWorkforcePlanning.Giventheimplicationsfortalentandteaming,howshould
organizationsthinkaboutspecializedexperttracksandthestructureoftheirworkforce?Whatdoesstrategicworkforceplanningforknowledgeworkmeaninaworldofconstantjobtransformation
andtechnologicaladvancement?Wedon’thavealltheanswers.Butwedoseethattheskillsneeded
?2024BostonConsultingGroup8
foragivenroleareblurring,andworkforceplanningwillnolongerbesolelyfocusedon?ndingacertainnumberofpeoplewithaspeci?cknowledgeskill,suchascoding.
Instead,planningshouldincludeafocusonbehavioralskillsandenablersthatwillsupportamore?exibleworkforce.WhileknowledgeworkersmaybetechnicallycapableoftakingonnewroleswiththehelpofGenAI,noteveryoneisequallyadeptatembracingchange.
ProfessionalIdentity.TheimpactofGenAIonprofessionalidentityisanimportantandcontentioustopic.Buta
recentsurvey
suggeststhatnegativeimpactscanbemitigatedwhenemployeesfeelsupportedbytheiremployers.
Infact,inourstudy,wefoundthat82%ofconsultantswhoregularlyuseGenAIforworkagreewiththestatements“GenerativeAIhelpsmefeelcon?dentinmyrole”and“IthinkmycoworkersenjoyusingGenAIfortheirwork,”comparedto67%ofworkerswhodon’tuseitonaweeklybasis.Morethan80%ofparticipantsagreedthatGenAIenhancestheirproblem-solvingskillsandhelpsthemachievefasteroutputs.
Thissuggeststhathighlyskilledknowledgeworkersgenuinelyenjoyusingthetoolwhenitallows
themtofeelmorecon?dentintheirrole—whichalignswithourprevious?ndingsthat
mandating
theuseofAI
canactuallyimproveemployeeperceptionofAI.However,thisisonlytrueif
employees
believethatAIisbeingdeployedtotheirbene?t
.
WeareonlyatthebeginningoftheGenAItransformationjourney,andthetechnology’scapabilitieswillcontinuetoexpand.Executivesneedtobethinkingcriticallyabouthowtoplanforthisfuture,
includinghowtorede?neexpertiseandwhatskillstoretaininthelongterm.
Buttheyarenotalone:
Skilldevelopment
isacollaborativee?ortthatincludeseducationsystems,corporatee?orts,andenablementplatformssuchasUdemyandCoursera.Eventheprovidersof
GenAImodelsshouldbethinkingabouthowtheirtoolscanfurtherenablelearningand
development.PreparingfortheGenAI-augmentedworkforcemustbeacollectiveendeavor—becauseourcollectivefuturedependsonit.
?BCGHENDERSONINSTTTUTB
TheBCGHendersonInstituteisBostonConsultingGroup’sstrategythinktank,dedicatedto
exploringanddevelopingvaluablenewinsightsfrombusiness,technology,andsciencebyembracing
thepowerfultechnologyofideas.TheInstituteengagesleadersinprovocativediscussionandexperimentationtoexpandtheboundariesofbusinesstheoryandpracticeandtotranslate
innovativeideasfromwithinandbeyondbusiness.FormoreideasandinspirationfromtheInstitute,pleasevisitour
website
andfollowuson
and
X(formerlyTwitter)
.
?2024BostonConsultingGroup9
Authors
DanielSack
MANAGINGDIRECTOR&PARTNER
Stockholm
LisaKrayer
PRINCIPAL
Washington,DC
EmmaWiles
ASSISTANTPROFESSOROFINFORMATIONSYSTEMS,
BOSTONUNIVERSITY’SQUESTROMSCHOOLOFBUSINESS
MohamedAbbadi
CONSULTANT
Washington,DC
UrviAwasthi
DATASCIENTIST
NewYork
RyanKennedy
AIENGINEER
Boston
CristiánArnolds
CONSULTANT
NewYork
Fran?oisCandelon
ALUMNUS
?2024BostonConsultingGroup10
1Thatexperimentwasconductedusingthe?rstversionofGPT-4.
2Oftheconsultantswhooriginallysigneduptoparticipate,480
completedtheexperiment.Participantswererandomlysplitintoa
controlgroupthatwasnotallowedtouseGenAIforthetasksanda
“treatment”groupthatwasaskedtouseGenAI.Each
溫馨提示
- 1. 本站所有資源如無特殊說明,都需要本地電腦安裝OFFICE2007和PDF閱讀器。圖紙軟件為CAD,CAXA,PROE,UG,SolidWorks等.壓縮文件請下載最新的WinRAR軟件解壓。
- 2. 本站的文檔不包含任何第三方提供的附件圖紙等,如果需要附件,請聯(lián)系上傳者。文件的所有權(quán)益歸上傳用戶所有。
- 3. 本站RAR壓縮包中若帶圖紙,網(wǎng)頁內(nèi)容里面會有圖紙預(yù)覽,若沒有圖紙預(yù)覽就沒有圖紙。
- 4. 未經(jīng)權(quán)益所有人同意不得將文件中的內(nèi)容挪作商業(yè)或盈利用途。
- 5. 人人文庫網(wǎng)僅提供信息存儲空間,僅對用戶上傳內(nèi)容的表現(xiàn)方式做保護(hù)處理,對用戶上傳分享的文檔內(nèi)容本身不做任何修改或編輯,并不能對任何下載內(nèi)容負(fù)責(zé)。
- 6. 下載文件中如有侵權(quán)或不適當(dāng)內(nèi)容,請與我們聯(lián)系,我們立即糾正。
- 7. 本站不保證下載資源的準(zhǔn)確性、安全性和完整性, 同時(shí)也不承擔(dān)用戶因使用這些下載資源對自己和他人造成任何形式的傷害或損失。
最新文檔
- 商業(yè)停車位租賃合同范本
- 小學(xué)一年級數(shù)學(xué)兩位數(shù)加減一位數(shù)水平自測模擬題大全附答案
- 人工智能研發(fā)項(xiàng)目股權(quán)協(xié)議書
- 【物理課件】運(yùn)動的快慢課件
- 【大學(xué)課件】發(fā)展職業(yè)生涯要立足本人實(shí)際
- 《廣告與公共關(guān)係》課件
- 次北固山下(教學(xué)課件)
- 《現(xiàn)場改善培訓(xùn)資》課件
- 【物理課件】自然界的水課件
- 《蒸汽污染及防止》課件
- 燙傷的防治與護(hù)理
- 2024年全國職業(yè)院校技能大賽高職組(護(hù)理技能賽項(xiàng))備賽試題庫(含答案)
- 駕駛員三年內(nèi)工作總結(jié)
- 青年你為什么要入團(tuán)-團(tuán)員教育主題班會-熱點(diǎn)主題班會課件
- 司法鑒定工作應(yīng)急預(yù)案
- 《竹結(jié)構(gòu)建筑技術(shù)規(guī)程》
- 大一中國近代史綱要期末考試試題及答案
- (完整版)鋼筋加工棚驗(yàn)算
- 安徽省合肥市廬陽區(qū)2023-2024學(xué)年三年級上學(xué)期期末數(shù)學(xué)試卷
- 概念方案模板
- 西南交大畢業(yè)設(shè)計(jì)-地鐵車站主體結(jié)構(gòu)設(shè)計(jì)
評論
0/150
提交評論