【W(wǎng)EF安永微軟】2024人工智能的影響力：社會創(chuàng)新中負責任人工智能的PRISM框架白皮書327mb

AIfor

Impact:ThePRISMFrameworkforResponsibleAIinSocial

InnovationWHI

T

E PAPE

RJU

N

E 202

4Incollaboration

withEYand

MicrosoftContentsImages:Dimagi,Getty

ImagesForewordExecutive

summaryIntroductionPathwaysforadoption:ThePRISM

FrameworkThePRISM

FrameworkCasestudiesalongtheadoptionpathways2Capabilitiesand

risksEthicsDataBusinessand

organizationTechnologyCostsand

metricsConclusionContributorsEndnotes3458912151618202225282931DisclaimerThisdocumentispublishedbytheSchwabFoundationforSocial

EntrepreneurshipinpartnershipwiththeWorldEconomicForum.Thefindings,interpretationsandconclusionsexpressedhereinarearesultofacollaborativeprocessfacilitatedandendorsedbytheSchwab

FoundationforSocialEntrepreneurshiportheWorldEconomicForumbutwhoseresultsdonotnecessarilyrepresenttheviewsoftheSchwabFoundationandWorld

EconomicForum,northeentiretyofitsMembers,Partnersorother

stakeholders.?2024SchwabFoundationforSocialEntrepreneurshipandtheWorldEconomicForum.Allrightsreserved.Nopart

ofthispublicationmaybereproducedortransmittedinanyformorbyanymeans,includingphotocopyingandrecording,orbyanyinformationstorageandretrieval

system.AIforImpact:ThePRISMFrameworkforResponsibleAIinSocial

Innovation2ForewordNowmorethanever,businessesandorganizationsfindthemselvesatapivotaljuncture–onethatcompelsthemtoprioritizetheircommunitiesandtheenvironmentatthecoreoftheir

operations.Thetumultuouslandscapeofongoingglobalcrisessignalsthatbusinessasusualisnolongeranoption.Againstthisbackdropofchangeandchallenge,socialinnovatorsaredemonstratingglobally,acrosssectorsandregionsandintheircommunities,whatthecapitalismofthefuturecanlooklike–amodelwherepurposeandprofitnotonlycoexistbutreinforceeachother,thrivingwith

profit.Yet,asthe2030deadlinefortheSustainableDevelopmentGoalslooms,significantgapsremainintheimpacttargetssetbybusinessesandgovernmentstoensureajust,equitableandsustainableworldforall.Achievingthesegoalsdemandsnothinglessthancollectiveactionandacertainleapfroggingtoacceleratethespeed

atwhichthetargetsarebeingmet.Theemergenceofartificialintelligence(AI)presentsatransformativeopportunityforsocialinnovators,non-profits,businessesandgovernmentsalike.TheAIforImpact:TheRoleofArtificialIntelligenceinSocialInnovationreportreleasedinApril2024examinesthelandscapeofsocialinnovatorswhoarecurrentlydeployingAIworldwide.Thekeyinsightsfromthisreportunderscoretheimmensepotentialof

AIasatoolfordrivingsocialimpact,particularlyincriticaldomainssuchashealthcare,educationandeconomic

empowerment.Torealizeimpactonagrandscale,widespreadAIawarenessandthestrategic,responsibledeploymentofthetechnologyisnecessary–whetherintegratedintooperatingmodelsorasaproductdeployedasameanstoeffect

change.Yet,internalandexternalbarriersremainthatpreventsocialinnovatorsfromadoptingAIat

scale,hinderingtheirabilitytoharnessitsfullpotential.Forsocialinnovators,challengessuchastechnicalcomplexity,skillsgaps,accesstoAItechnologyalongwithbalancingsocialandbusinessobjectivescanbeprohibitivetotheirunderstandinganddeploymentofthetechnology.Ontheotherhand,technologyorganizationsanddevelopershavearesponsibilitytodevelopAIcapabilitiesthatareethicalandfair,addressglobaldisparitiesanddonotexacerbateexistingbiases.Thisnecessitatesthecreationoftwo-waydialoguebetweentechnologyorganizationsandsocialinnovatorsthatintegratesthevoicesofsocialinnovators,enablingthemtoinfluencethetechnologyroadmapandensuringthatAIcapabilitiesaretrulyfitforpurposeforimpact

enterprises.ThispaperbuildsonearliereffortstopresentastrategicroadmapforsocialinnovatorsassessinghowtoincludeAIintheiroperationsormodels.ItdrawsinspirationfromthoughtleadershippublishedbytheWorldEconomicForum’sAIGovernanceAllianceandisinformedbytheexpertiseofanetworkofsocialinnovatorsandintermediariesonthefrontlineofchange.SpecialthanksgotothoseinvolvedintheworkinggroupsoftheGlobalAllianceforSocialEntrepreneurshipAIforSocialInnovationworkstream,whosetimeandexpertisehavecontributedtothedevelopmentofthispaperandwhoselivedexperiencescanbefoundasexamplesthroughoutthe

paper.Thispaperwillformthebasisforworkwithintermediarypartnersandotherecosystemcollaboratorsonthefurtherdevelopmentoftrainingandskill-buildingtoolkitsforsocial

enterprises.Ourhopeisthattheframeworksandroadmapspresentedherein,alongwithresourcesfromcorporatecollaborators,canhelpsocialinnovatorsrealizethepromiseofAI-drivensocial

change.Jean-PhilippeCourtoisExecutiveVice-PresidentandPresident,NationalTransformation

Partnerships,MicrosoftAIforImpact:ThePRISM

FrameworkforResponsibleAIinSocial

InnovationJune

2024Cathy

LiHead,AI,DataandMetaverse;Memberofthe

Executive

Committee,

WorldEconomic

ForumGillian

HindeEYGlobalCorporateResponsibilityLeader,

EYGSAIforImpact:ThePRISMFrameworkforResponsibleAIinSocial

Innovation3Executive

summaryThePRISMFrameworkenablesorganizationstoadoptartificialintelligenceforimpactandhighlightstechnologyshortcomingsforethical

implementation.Therapiddevelopmentanddisseminationofartificialintelligence(AI)capabilitiespresentsaviablesolutionforsocialinnovatorstoleapfrogstructuralgapsandscaletheirsolutions.Thefirst

reportinthisseries,AIforImpact:TheRoleofArtificialIntelligenceinSocialInnovation,waspublishedbytheSchwabFoundation’sGlobalAllianceforSocialEntrepreneurshipincollaborationwith

EYandMicrosoft.ItoffersacomprehensiveanalysisoftheroleofAIinsocialinnovation,withinsightsfromover300socialinnovators.Itdemonstrateshowsocialinnovatorsalreadyembracethetechnologyfortangibleoutcomes,mostlyinareas

suchashealthcare,environmentalaction,economicempowermentor

education.ThissecondpapertakesadeeplookintobestpracticesintheprocessofapplyingAI.ItfeaturessocialinnovatorswhoarespearheadingthedeploymentofAIforimpact,thosewhoareimplementingitasanenablerforscaleandthosewhoarejuststartingtoconsidertheuseofAIfortheiroperations.Datashowsthatmorethan50%ofsocialinnovatorscurrentlyapplyAItoenhancecoreproductsorservices.AlmostathirdleverageAItodevelopentirelynewsolutionsand

another20%useittoenhanceinternalefficienciesorenablescalefortheir

organization.Buildingonthisdataset,researchandover20in-depthinterviews,thispaperintroducesthePRISMFrameworktoguidesocialinnovatorsandotherorganizationsthroughthenuancedlandscapeofAIintegration.Theframeworkencouragesorganizationstostartwith

low-risk,low-costAIapplicationsandstressestheimportanceoforganizationalreadinessovermeretechnologicalcapability.ItalignswithestablishedmethodologiesliketheAIGovernanceAlliance’sPresidioFramework.Itadvocatesan

iterativebutstrategicallyalignedapproachwithspecificevaluationgatesfortheimplementationof

AI.ThePRISMFrameworkoutlinesthreelayersofAIimplementation:impactmissionand

strategy,adoptionpathways,andcapabilitiesandrisks.Eachlayeraddressesdifferentelementsofreadinessandpotentialapplications,fromsimple,internalusecasestocomplex,externallyfacing

deployments.ThemodularapproachallowsorganizationstotailortheirAIstrategiestotheirspecificneedsandcapacities,ensuringthatAIadoptionis

bothimpactfulandsustainable.Richcasestudies,suchasSASBrasil’scervicalcancerscreeningandHighResolvesAI-enhancededucationalinitiativesillustratetheframework’s

use.Theframeworkalsohighlightsrisksandtechnologicalshortcomingsthatneedtobeaddressedforanequitableimplementation.Thisincludes,forexample,databiasesortechnologystructuresthatarenotfittoallowforexplainabilityofAIdecisions–akeyelementtobuildtrustamongthecommunitiesthatsocialinnovatorsaretryingtoserve.Therefore,thepapercalls

foractiveengagementbetweenAIforsocialinnovationecosystem,technologyleadersandsocialinnovatorstojointlyenabletheethicaladoptionofAIforpositive

impact.Image

credit:LifeBankAIforImpact:ThePRISM

FrameworkResponsibleAIinSocialInnovation 4Artificialintelligence(AI)offerstransformativepotentialinsolvingcomplexsocietalchallengesandisdeployedacrosssectorsfromhealthcaretoagriculture.InApril2024,theSchwabFoundation’sGlobalAllianceforSocialEntrepreneurship’spublishedtheAIforImpact:TheRoleofArtificialIntelligenceinSocialInnovationreportincollaborationwithEYandMicrosoft.It

analysesAIimplementationsbyover300social

innovatorsandrevealsthatthereisnogapinimplementationbetweenlow-/middle-incomeandhigh-incomecountriesbutastrongdiversityinthesectorsofimplementation:healthcareandenvironmenthadthehighestshareofAIimplementationsacrosstheboard.However,80%ofeconomicempowermentsolutionsweredeployedbysocialinnovatorsinlow-/middle-income

countries.IntroductionArtificialintelligencehasthepotentialtoscaleimpactinseveraldomainsbutrequirescollaborationtohelpsocialinnovatorsrealizeitsmaximum

potential.Geographicalsummaryofthesocialinnovator

datasetFIGURE

177UnitedStatesIndia31United

Kingdom20Kenya18Nigeria12Brazil12Canada10Australia10Singapore10Mexico8New

Zealand81Number

of

social

innovators77NorthAmerica,

29%Asia,20%Africa,18%Europe,

16%LatinAmerica,11%Oceania,6%Despitebroadimplementation,thereportidentifiesgapshinderingwideradoption.Only13%ofAIinitiativesfocusoneducationaltoolkits,primarilyintheGlobalNorth.Genderdisparitiespersist,withjust25%ofwomen-ledsocialenterprises

usingAIcomparedtohalfinthebroadersector.Ethicalandequitychallengesalsoemerged.Forexample,mostcommerciallyavailablemodelsaretrainedondatafromhigh-incomecountries,deliveringsub-parresultsforlow-andmiddle-income

countries.AIforImpact:ThePRISMFrameworkforResponsibleAIinSocial

Innovation5SocialinnovatorsapplyingAIandimpact

domainsFIGURE

22%Healthcare28%Environment21%Economicempowerment16%Technology8%Securityandjustice5%Publicandsocial

sector5%Equalityandinclusion2%Education8%Informationverificationand

validation2%Infrastructure5%Crisisresponse1%Socialimpact

domainSocialinnovatorsdeployingAIby

genderAIforImpact:ThePRISMFrameworkforResponsibleAIinSocial

Innovation6FIGURE

333%31%

26%26%18%15%0%

5%

10%15%20%25%30%35%OceaniaAsiaAfricaNorth

AmericaEuropeLatin

AmericaWomanfounderorchiefexecutive

officerOverallaverageofwomen-led

enterprisesAIinitiativesby

typeFIGURE

4Partners,

networksand

acceleratorsConferenceOpenplatform,researchand

dataAI

educationToolkitGrant/rewardprogramme0%10%20%30%50%40%

43%19%18%7%6%6%ThispaperbuildsonpreviousinsightsandframeworkssuchasthePresidioFrameworkbytheWorldEconomicForum’sAIGovernanceAlliancetohighlightavenuesforbroaderAIadoptionamongsocialinnovators.Itcombinesdatafrom

previousstudieswithinsightsfrom22in-depthinterviewsandinteractiveworkshopswithsocialinnovatorsandtechleaders,highlightingthenuanceddecision-makingprocessesinAIdeploymentandtheongoingchallengesfacedbysocial

innovators.Image

credit:SAS

BrasilAIforImpact:ThePRISMFrameworkforResponsibleAIinSocial

Innovation7Pathwaysfor

adoption:ThePRISM

Framework1Buildingonexistingconcepts,thePRISMFrameworkguidestheresponsibleadoptionofAIfor

impact.TheApril2024AIforImpact:TheRoleofArtificialIntelligenceinSocialInnovationreportemphasizeshowsocialinnovatorsareleadinginAIadoption,notablyinprojectslikeSASBrasil’scervicalcancerscreeningandHighResolves’sinitiativesinAI-aidededucationdespiteconnectivitychallenges.ThissecondpaperusestheirexperiencestoillustratehowAIcandrivesocialandenvironmentalbenefits.Itsynthesizesinsightsfrom300socialinnovatorsandexpertinterviewsintothePRISM

Framework,aguidetoimplementingAIiteratively,responsiblyand

effectively.Thefindingsencouragesocialinnovatorstostartsmallandexecutelow-cost/low-riskimplementationwhenfirstapproachingAIusecases.ItstressesthatsocialinnovatorsandconventionalcompaniesalikeneedtoconsidertheirorganizationalreadinesswhendesigningAIimplementation,as

internalpreparednessoftenoutweighstechnologicalanddataconsiderations.Theframeworkhighlightsadoptionpathwaysbasedonorganizationalreadinessandrelevantcapabilitiesandriskslinkedtothesepathways.TheframeworkalignswithotherconceptslikethePresidioFrameworkbytheWorldEconomicForum’sAIGovernanceAlliance,which

supportsorganizationsinresponsiblyenhancingproductivityandredefiningindustriesthroughAI.ThePresidioFrameworkstartsbyidentifyingandfilteringthebestgenerativeAIusecasesalignedwithstrategicobjectivesandfunnelsthemthroughthreecommonevaluationgates–businessimpact,operationalreadinessandinvestmentstrategy.Theseevaluationgatesaredesignedtobeappliedinanysequenceand

iteratively.ThePresidioFramework:Funnellingusecasesthroughevaluation

gatesAIforImpact:ThePRISMFrameworkforResponsibleAIinSocial

Innovation8FIGURE

5FilterthebestIterateontherestIdentify

generativeAIuse

casesBusinessimpactOperationalreadinessInvestmentstrategyFunnelthroughevaluation

gatesScale

andtransform123Funnelingusecasesthroughevaluation

gatesThePRISMFrameworkbuildsonthePresidioframework.Ithighlightshowdifferentadoptionpathwaysareinterdependentwithdifferentlevelsoforganizational,technologicalanddata

readiness.Italsoemphasizestheneedforalignmentwithanorganizationsimpactambitionateverystage(seeFigure6).Ithighlightsthatsocial

innovatorsare,bynature,impact-firstorganizationsandtheirconsiderationofstrategicalignment(asoutlinedintheframeworkbytheAIGovernanceAlliance)goesbeyondbusinessalignment.Their

considerationsprioritizeimpactandalsoincludecorevaluesandethicalprincipleswhenapplyingAI(suchasprioritizingcommunityvoicesinthedesignofsolutions).Thatmaynotalwaysbeanextensive,highly

designedandresource-intensestrategyprocessbutratheraconstantreconciliationofplannedusecaseswithimpactobjectives(e.g.earlyengagement

ofcommunitiesinproduct/servicedevelopment,dataownershipamongbeneficiariesortheprohibitionofdecision-makingwithouthuman

involvement).1.1 ThePRISM

FrameworkThePRISM

FrameworkFIGURE

6Image

credit:

EducationforEmploymentImpactmissionLayer

1:Impactmissionand

strategyLayer

2:Adoption

pathwayLayer

3:Capabilitiesand

risksEthicsDataBusinessand

organizationTechnologyMetricsand

costs1Tinkering2Value3

Efficiency4Delivery5Coreasset6

AI-firstAIforImpact:ThePRISMFrameworkforResponsibleAIinSocial

Innovation9Thesecondlayerhighlightsthe

different

adoption organization’smaturity,startingwithlow

investmentpathways.Innovatorsmay

choose

different andlowriskimplementation.Theexamplesbelowpathwaysdependingontheirreadiness

for

AI, illustratethesepathwaysbasedonconcretecasewhichwillevolveovertime.It

showcases

different studiesfoundinthedatasetandinterviews.approachestoAIimplementationdependingon

theTABLE

1Socialinnovatorexamplesofadoption

pathwaysAdoption

pathwayFocusExampleConscious

tinkeringSimple,low-riskimplementationleveragingoff-the-shelf

solutionsInternalAshoka:TheorganizationstarteditsadoptionofAIwithexperiments,leveragingopen-sourceAIcapabilities,todevelopdifferentinternaltoolswhileinitiallykeepingtheriskprofilelow.Forexample,theorganizationanalysesthetranscriptsofitsboardmeetingsthroughamulti-queryprocesswithChatGPT-4Turbotoproviderecommendations

for

process

improvements

and

implementation

of

new

AI

tools.Proofof

valueImplementationoftoolswithsignificantefficiencygainsornewinsights,demonstratingthevalueofAItothe

organizationInternalAshoka:Forpromisingapplications,AshokasetsouttoproveAIvaluetotheboardandstafftoencouragewidespreadinternaluse.Forexample,AshokahasleveragedAItoanalysequalitativeinsightsfromitsinternaldatabaseofsocialinnovators(changemakers).Thatway,itwasabletohighlightopportunitiesfornewpartnershipsbetweenthemandcollectiveactionasacommunityatanypointintime.ItalsocreatedthefoundationforamorediverseAIportfolio,whichnowcounts21

applications.Efficiency

enhancementStructuredadoptionofAItoolswithintheorganizationtostreamlinecore

processesInternalEducationforEmployment(EFE)promotesacultureofAIusebyemployeestosavetimeandimproveinternalefficiencyindepartmentssuchasmarketingandgrantwriting.AItoolssuchasMicrosoftCopilothasenabledteammemberstosaveanestimated10-20%(about1to3hours)intimeversusiftheyhadnotusedAIatall.Inotherinstances,theorganizationhasdeployedAIwebscrapingtoolstopredictjobtrendsandinforminternal

strategy.Delivery

improvementsImplementAIwithexternallyfacingprocessesandstaff,improvingproduct/serviceexperienceand

impactExternalLifeBank’sadoptionofAIhasenabledthecompanytoleapfrogbarrierspresentedbysupplychaincapitallimitations.TheenterprisedevelopeditsproprietaryAItoolsforsupplychainmanagementandtomanagelast-milepayments.It’sAI-poweredbrainbox,Light,communicateswithhealthcareworkersatthelast-mile,improvingsupplychainefficiencybyreducinghumanerrorsin

decision-making.AIasacore

assetLeveragingAItodevelopnewsolutionsandservicestoachievetheimpact

objectiveExternalSASBrasilisseekingtodevelopanAI-enableddiagnostictooldrawingonlarge-scalecervicalcancerdatasetstoimprovehealthcareoutcomesforwomeninvulnerable

communities.AI-first

organizationOrganizationaltransformationtobecomedata-driven

andAI-enabledinalmosteveryofferingand

serviceExternalGeekiemadethedecisiontobecomeatechnologyorganizationtofullyleverageAI

foreducation.Ittransformeditsbasicofferingtobefree-of-chargetobeabletogatherdataatscaleandbuildAImodels.Ittransformeditselftoembedtechnologyintoeveryfacetoftheorganization–fromtrainingplanstosuggestionsforremedial

work.AIforImpact:ThePRISMFrameworkforResponsibleAIinSocialInnovation 10userid:549683,docid:171458,date:2024-08-14,AIusecasesamongsocial

innovatorsFIGURE

754%30%12%5%0%10%40%50%60%AddingAIintocoreproduct/servicesNewAIproducts/servicesandbusiness

modelsUsingAItodeliverimpactnotpreviously

scalableUsingAItoimproveinternaloperationsandprocessesinternallyAIimplementations

insights20% 30%Amongthe300AIinsocialinnovationusecases,mostinnovators(54%)leveragedAItoenhanceproductofferings(3.efficiencyor4.delivery),30%createdentirelynewproductsorservicesand12%leveragedAItounlockscalefortheirsolutions(5.coreasset),5%leverageditforinternalprocesses(1.tinkeringand3.

efficiency).WhenimplementingAIacrosstheadoptionpathways,differentcapabilitiesandrisksareconsideredwithvaryingintensity.ThisisdescribedinthethirdlayerofthePRISM

framework:1.Ethics:Beyondtheconsiderationofgeneralvalues,principlesandtheimpactmissioninlayer1,thiscapabilityandriskconsiderationincludeselementssuchasconsideringbiasinAImodels,ensuringtransparencyofdecision-making,accountabilityforthedecisionsmadeandintegratingethicalAIasabusinessprincipleintotheorganizationatalaterstageoftheorganizational

deployment.2.Data:DatastrategiesmaybeappliedindifferentfacetsduringtheimplementationofAI,includingananalysisofexistingdataassets,thecleaningandpreparationofdata,managingdataintakeaswellasdataprivacyconsiderationswhenusingdata–especiallyinhighlysensitiveimpactareassuchas

healthcare.3.Businessandorganization:OrganizationalreadinessisakeycomponentofimplementingAI.ThismayincludesettingaclearvisionandleadershipfortheuseofAI(especiallywhenleveragingAIasacoreassetortransformingtheorganizationtoanAI-firstorganization),acontinuouslearningstrategy,thedevelopmentofskillsandtalentaswellaschangemanagementandthedevelopmentofanAIordataculture.Forsocialinnovators,stakeholdermanagementisparticularlyimportanttocreatethenecessarybuy-inandtrustamongkeyconstituents.4.Technology:Manysocialinnovatorsstrugglewiththeidentificationoftherighttechnologyfortheirusecases.Inearlyadoptioncases,socialinnovatorsmaysimplyapplyoff-the-shelfsolutionswithlowimplementationcomplexityandcosts.Lateron,modelcomplexity,theneedforreinforcementtraining(especiallyforusecaseswiththeneedforcontextual

data),interoperabilitywithecosystemactorsorinternetaccessissuesmaysurfaceas

well.5.Costsandmetrics:GiventhebroadavailabilityofAItechnology,costsarenotmuchofanissueinearlystagesofAIadoption.Thenumberofrequestswithanapplicationprogramminginterface(API)remainslow,datavolumestendtoberestrictedandorganizationalreadinessmaynotneedtobeconsidered.Butinlater-stageimplementations,variablecostsquicklyrampupandindirectexpensesmounttoensurethatAIimplementationsremainimpactfulandethicallyresponsible.Socialinnovatorsmayneedtoconsiderandmonitorcoststhat

arenotlimitedtosoftwareorlicensingfeesbutalsoincludedataacquisition,datapreparationorhardwarecosts.Thefollowingsectionexplorestheabove-mentionedcapabilitiesandrisksinmoredetail.ThePRISMFrameworkoffersamodularapproachtoconsideringtheimplementationofAIfor

impact.Ononeendofthespectrum,socialinnovatorswithstrongorganizationalreadiness–solidinternalAIskillsandcapabilities,ahealthyenterprise-wideattitudetothetechnologyandacleararticulationandunderstandingoftheirdataassetsandethicalconcerns–mayapplycapabilitiesandmanagerisksacrosstheentiretyoftheframework.Organizationsthathavenotbuiltsuchinternalcapabilitiesmayleveragethetechnologytotinkerand

thereforeusealow-riskentrancepointtoimplementthetechnologyatlowcostandwithminimalimpactonthebusinessstrategicgoalsandethical

concerns.AIforImpact:ThePRISMFrameworkforResponsibleAI