利用人工智能設(shè)計(jì)教育指南

DesigningforEducationwith

ArtificialIntelligence:

AnEssentialGuideforDevelopers

July2024

DesigningforEducationwithArtificialIntelligence:AnEssentialGuideforDevelopers

MiguelA.Cardona,Ed.D.

Secretary,U.S.DepartmentofEducation

RobertoJ.Rodriguez

AssistantSecretary,OfficeofPlanning,EvaluationandPolicyDevelopmentJuly2024

ExamplesAreNotEndorsements

Thisdocumentcontainsexamplesandresourcematerialsthatareprovidedfortheuser's

convenience.Theinclusionofanymaterialisnotintendedtoreflectitsimportance,norisitintendedtoendorseanyviewsexpressedorproductsorservicesoffered.Thesematerialsmay

containtheviewsandrecommendationsofvarioussubjectmatterexpertsaswellascontact

addresses,websites,andhypertextlinkstoinformationcreatedandmaintainedbyotherpublicandprivateorganizations.TheopinionsexpressedinanyofthesematerialsdonotnecessarilyreflectthepositionsorpoliciesoftheU.S.DepartmentofEducation.TheU.S.DepartmentofEducationdoes

notcontrolorguaranteetheaccuracy,relevance,timeliness,orcompletenessofanyinformation

fromothersourcesthatareincludedinthesematerials.Otherthanstatutoryandregulatory

requirementsincludedinthedocument,thecontentsofthisguidedonothavetheforceandeffectoflawandarenotmeanttobindthepublic.

ContractsandProcurement

Thisdocumentisnotintendedtoprovidelegaladviceorapprovalofanypotentialfederalcontractor'sbusinessdecisionorstrategyinrelationtoanycurrentorfuturefederalprocurementand/orcontract.Further,thisdocumentisnotaninvitationforbid,requestforproposal,orothersolicitation.

LicensingandAvailability

ThisguideisinthepublicdomainandavailableonaU.S.DepartmentofEducationwebsiteat

.

RequestsforalternateformatdocumentssuchasBrailleorlargeprintshouldbesubmittedtotheAlternateFormatCenterbycalling1-202-260-0852orbycontactingthe504coordinatorvia

email

atom_eeos@.

NoticetoPersonswithLimitedEnglishProficiency

IfyouhavedifficultyunderstandingEnglish,youmayrequestlanguageassistanceservicesforU.S.DepartmentofEducationinformationthatisavailabletothepublic.Theselanguageassistance

servicesareavailablefreeofcharge.Ifyouneedmoreinformationaboutinterpretationor

translationservices,pleasecall1-800-USA-LEARN(1-800-872-5327)(TTY:1-800-437-0833);emailusat

Ed.Language.Assistance@;

orwritetoU.S.DepartmentofEducation,Information

ResourceCenter,LBJEducationBuilding,400MarylandAve.SW,Washington,DC20202.

HowtoCite

Whilepermissiontoreprintthispublicationisnotnecessary,thesuggestedcitationisasfollows:

U.S.DepartmentofEducation,OfficeofEducationalTechnology,DesigningforEducationwithArtificialIntelligence:AnEssentialGuideforDevelopers,Washington,D.C.,2024.

Acknowledgements

ProjectTeam

DesigningforEducationwithArtificialIntelligence:AnEssentialGuideforDeveloperswasdevelopedundertheleadershipandguidanceofRobertoJ.Rodríguez,AssistantSecretaryfortheOfficeofPlanning,EvaluationandPolicyDevelopment,AnilHurkadli,InterimDeputyDirectorfortheOfficeofEducationalTechnology,BernadetteAdams,SeniorPolicyAdvisorfortheOfficeof

EducationalTechnology,andKevinJohnstun,EducationProgramSpecialistfortheOfficeofEducationalTechnology.

ThisworkwasdevelopedwithsupportfromDigitalPromiseundercontract(91990019C0076)andledbyJeremyRoschellewithAnthonyBaker,PatiRuiz,EricNentrup,GabrielleLueandSarahMartin.

ContributingMembersoftheDeveloperCommunity

●KristenDiCerbo,KhanAcademy

●TeddyHartman,GoGuardian

●NeilHeffernan,WorcesterPolytechnicInstitute

●KarlRectanus,EDSAFEAIAlliance

●SteveRitter,CarnegieLearning

●SharadSundararajan,MerlynMind

●AlyssaVanCamp,TeachFx

●JuliaWinter,AlchemieSolutions,Inc.

Wealsothankthemanydevelopers,industryassociations,andnonprofitorganizationsthatattendedourlisteningsessionsandcontributedtheirideasfortranslatingtheDepartment’srecommendationsforArtificialIntelligenceineducationintopracticalguidelines.

TableofContents

Introduction 1

RespondingtotheOctober2023ExecutiveOrder 2

Defining“ArtificialIntelligence”and“EdTech”Broadly 3

KeyMessage:SharedResponsibilityforBuildingTrust 4

APathwayfromDesigningforEducationtoEarningTrust 10

Recommendation1:DesigningforTeachingandLearning 12

WhattoKnow 12

QuestionstoAsk 16

DirectionstoPursue 16

Resources 17

Recommendation2:ProvidingEvidenceforRationaleandImpact 18

WhattoKnow 19

QuestionsToAsk 23

DirectionstoPursue 23

Resources 24

Recommendation3:AdvancingEquityandProtectingCivilRights 26

WhattoKnow 27

QuestionsToAsk 31

DirectionstoPursue 32

Resources 32

Recommendation4:EnsuringSafetyandSecurity 33

WhattoKnow 33

QuestionstoAsk 38

DirectionstoPursue 38

Resources 39

Recommendation5:PromotingTransparencyandEarningTrust 40

WhattoKnow 40

QuestionsToAsk 44

DirectionstoPursue 44

Resources 44

Conclusion 45

OFFICEOFEducationalTechnology1

DesigningforEducationwithArtificialIntelligence:AnEssentialGuideforDevelopers

Introduction

Todayandinthefuture,agrowingarrayofArtificialIntelligence(AI)modelsandcapabilitieswillbeincorporatedintotheproductsthatspecificallyserveeducationalsettings.TheU.S.DepartmentofEducation(Department)iscommittedtoencouraginginnovativeadvancesineducational

technology(edtech)toimproveteachingandlearningacrossthenation’seducationsystemsandtosupportingdevelopersastheycreateproductsandservicesusingAIfortheeducationalmarket.

BuildingontheDepartment’spriorreport,

ArtificialIntelligenceandtheFutureofTeachingand

Learning:InsightsandRecommendations

(2023AIReport),thisguideseekstoinformproductleads

andtheirteamsofinnovators,designers,developers,customer-facingstaff,andlegalteamsastheyworktowardsafety,security,andtrustwhilecreatingAIproductsandservicesforuseineducation.Thislandscapeisbroaderthanthosebuildinglargelanguagemodels(LLMs)ordeployingchatbots;

itincludesallthewaysexistingandemergingAIcapabilitiescanbeusedtofurthersharededucationalgoals.

Ourinsightshereareintendedtosupportpeoplewhoaremanagingteamsinthedesignand

developmentofproductsthatleverageAItoimproveteachingandlearning.Wehaveattemptedtoaddresstopicsthatwillberelevantacrossthecontinuumofedtechdevelopers,whichincludes

establishedfirmsandnewcomers,aswellasdevelopersacrossresearch,nonprofit,andfor-profit

organizations.Weaddressnotonlydevelopersofproductsforformaleducationsettings—

includingelementaryandsecondaryschools,colleges,anduniversities—butalsoforeducationalusesathome,community,andotherinformalsettings.

Tothisend,eachsectionofthisdocumentisbuiltaroundacorerecommendationandincludesasetofdiscussionquestionsthatleadersinorganizationscanusetofosterconversation,nextstepstopromoterobustdevelopmentprocesses,andresourcesthatcanprovideadditionalsupport.Please

notethattheDepartmentandotherfederalagenciesareactivelyconsideringnextstepsto

promotethesafeandresponsibleuseofAI.Thus,thisdocumentsuggests“questionstoask”and“directionstopursue”todevelopersthataredeliberatelyopen-ended.

Thisguideprovidesnon-regulatory,education-specificguidancethatisalignedwithfederal

guidelinesandguardrails.Thisguide’scoverageofexistingfederalguidelinesandguardrailsisnotcomprehensiveorexhaustive.Itisnotintendedtoanddoesnotenableadevelopertoestablishitscompliancewithregulations.Also,itisnotintendedtoanddoesnotintroduceanynew

requirements.Whereexamplesaregiven,includinglinkstonon-U.S.Governmentwebsites,theyareintendedtobeillustrativeandnottorestricttheapplicationofthisguidetoadditionalformsofAIastheybecomeavailableforuseineducation.Weareprovidingtheseexternallinksbecause

theycontainadditionalinformationrelevanttothetopic(s)discussedinthisdocumentorthat

OFFICEOFEducationalTechnology2

otherwisemaybeusefultothereader.Wecannotattesttotheaccuracyofinformationprovidedonthecitedthird-partywebsitesoranyotherlinkedthird-partysite.Weareprovidingtheselinksforreferenceonly;linkingtoexternalresourcesdoesnotconstituteanendorsementbythe

Department.Developerscanusethisguidetoincreasetheirunderstandingofessentialfederal

guidelinesandguardrailstoguidetheirworkastheycreateAIapplicationsforeducationalsettings.

RespondingtotheOctober2023ExecutiveOrder

ThisguideisresponsivetoPresidentJoeBiden’sOctober30,2023,

ExecutiveOrderontheSafe,

Secure,andTrustworthyDevelopmentandUseofArtificialIntelligence

(ExecutiveOrderon

AI),whichstatesthefollowing:

TohelpensuretheresponsibledevelopmentanddeploymentofAIintheeducationsector,theSecretaryofEducationshall,within365daysofthedateofthisorder,develop

resources,policies,andguidanceregardingAI.Theseresourcesshalladdresssafe,

responsible,andnondiscriminatoryusesofAIineducation,includingtheimpactAIsystemshaveonvulnerableandunderservedcommunities,andshallbedevelopedinconsultationwithstakeholdersasappropriate.

Thisguideisinformedbyanextensiveseriesofpubliclisteningsessionswithstudents,parents,

andeducatorsalongwithdevelopers,industryassociations,andnonprofitorganizations.This

includedacrosssectionofdevelopersrepresentingavarietyofcompanysizes,fundingmodels,

andorganizationtypes(for-profit/nonprofit).Sessionparticipantssharedtheircurrentapproachestosafetyandsecurity,theriskstheyandtheirusersface,suggestionsonsupportsandresources,

andthoughtsaboutopportunitiestobuildtrustinthefuture.Severaladditionallisteningsessionsoccurredwithasmallersetofdevelopers(listedinthecontributingmemberssectionabove)drawnfromthosewhoparticipatedinitially.Wherethisguidereferstolisteningsessions,itincludesall

theseopportunitiestohearfromconstituents.

ThisguidedrawsonagrowingseriesoffederalpublicationsonAI,whichincludestheseexamples:

●

OfficeofScienceandTechnologyPolicyBlueprintforanAIBillofRights:

The2022WhiteHousewhitepaperthatbroadlyshapesastrategyinformingandprotecting

citizensfromAIandrelatedtechnologies

●

NationalInstituteofStandardsandTechnology(NIST)AIRiskManagement

Framework:

Aseven-stepframeworktoprotectdeveloperandenduserinterestsalike,particularlyrelevanttoedtechcompaniesusingAIcomponentsandforthcoming

emergingtechnologies

●

FACTSHEET:Biden-HarrisAdministrationSecuresVoluntaryCommitmentsfrom

EightAdditionalArtificialIntelligenceCompaniestoManagetheRisksPosedbyAI|

TheWhiteHouse:

TheWhiteHouse’sengagementwithtechnologycompanies

highlightssharedresponsibilityamongthefederalgovernment,developerorganizations,andotherconstituents.

Defining“ArtificialIntelligence”and“EdTech”Broadly

TheDepartmenttakesabroadviewoftheterms“AI”and“edtech.”Thisdocument’sguidanceappliesbroadlyacrossthemanytypesofAIthatdevelopersmayintegrateandthemanywaystheirproductsmaybeusedineducationalsettings.

BoxA:ArtificialIntelligenceasdefinedinthe

ExecutiveOrderonAI.

Theterm“artificialintelligence”meansamachine-basedsystemthatcan,foragivensetofhuman-definedobjectives,makepredictions,recommendations,ordecisionsinfluencingrealorvirtual

environments.Artificialintelligencesystemsusemachineandhuman-basedinputstoperformthefollowing:

●Perceiverealandvirtualenvironments

●Abstractsuchperceptionsintomodelsthroughanalysisinanautomatedmanner

●Usemodelinferencetoformulateoptionsforinformationoraction

Asastartingpointfordefining“AI,”wenotethestatutorydefinitionofAI(seeBoxAabove),

whichalsoappearsintheExecutiveOrderonAI.AswenotedintheDepartment’s

2023AI

Report,

AIisanumbrellatermformanysubfieldsofresearchandinnovation.Followingthe

2023

AIIndexReport,

fromtheNationalArtificialIntelligenceResearchResourceTaskForce,wecontinuetoobservetheextremelyrapidevolutionofAIcapabilitiesinmanydomains,suchasspeech,vision,robotics,andtext.

BoxB:DefiningEducationalTechnology(edtech)

AsdefinedintheDepartment’s

2023AIReport,

edtechincludesthefollowing:

A.Technologiesspecificallydesignedforeducationaluse

B.Generaltechnologiesthatarewidelyusedwithineducationalsettings

Likewise,wedefine“edtech”broadly(BoxB).Inan

independentmarketresearchfirm’sreport,

theglobaledtechmarketforbothK-12andhighereducationisvaluedat$123billionasof2022.Thismarketincludesstartups,smallbusinesses,nonprofitorganizations,andlargercorporations—

companieswithamissionspecifictoeducationandotherswithproductsthatareusedineducationandothersectors.Thereisabroadspectrumofproductsrangingfrominfrastructuretostudent

informationsystemstolearningmanagementsystemstospecificend-userapplicationsandmore.

Whereasamarketisdefinedbybuyersandsellers,wedefinetheedtech“ecosystem”broadlytoincludethemanydifferentpeopleandorganizationsworkingtogethertodesignandrefinenewproductsandservices.Withinthisecosystem,discussionsoftheExecutiveOrderconceptsof

“safety,security,andtrust”willbeinfusedwithconceptsmorespecifictoeducation(e.g.,evidence,fairness,dataprivacy)aselaboratedbelow.Ecosystemparticipantsinclude

OFFICEOFEducationalTechnology

3

OFFICEOFEducationalTechnology4

educationalprocurementdepartments,additionaleducationaldecisionmakers,educators,

parentsandguardians,students,nonprofits,postsecondaryeducationinstitutions,andbroadercommunitymembers.TheecosystemincludespeoplewhodirectlycreateoruseAIproducts

andpeopleinvolvedineducationalsystemswhoareaffectedbyAIproducts.Shared

responsibilityforbuildingtrustgoeswellbeyondbuyingandselling;itthrivesonopenandengagedcommunicationacrosstheecosystem.

KeyMessage:SharedResponsibilityforBuildingTrust

Throughlisteningsessions,theDepartmentlearnedthatonekeyrecommendationinthe

2023

AIReport,

“PrioritizeStrengtheningTrust,”resonatedwithdevelopersasacalltoaction.

Developersrecognizethefundamentalimportanceof“trust”intheedtechecosysteminwhich

theyparticipate.Trustimprovestheco-designprocessbetweendevelopersandeducatorssothattogethertheycancreateandscaleinnovativeproducts.Consequently,developerscan

benefitgreatlyfromunderstandinghowtheycanworkwithothersintheecosystemtostrengthentrust.Specifickeymessagesfollow:

1.Trustisasharedresponsibility.

ThePresident’sExecutiveOrderonAImakesclearbeginningwithitstitlethatthefederal

governmenthasaresponsibilitytopromote“safe,secure,andtrustworthydevelopment”and

articulatesastanceofsharedresponsibility(seequotebelow).Consequently,developerswillfind

informationwithinthisdocumentonwheretolocatefederallawsandotherfederalresourcesthataredirectlyapplicabletotheirworkineducation.Becausethetechnologyisevolvingrapidly,

developersmayfinditvaluabletogobeyondattendingtoandcomplyingwithtoday’sfederal

guidelinesandguardrailstoearntrust.OneimportantexampleistheSoftwareandInformationIndustryAssociation’s

PrinciplesfortheFutureofAIinEducation,

whicharticulatesseven

principles(e.g.,theirevaluationprinciplecallsforcontinuallyassessingtheimpactofAI).TeachAIhasalsodeveloped

PrinciplesforAIinEducation

aspartofitsAIGuidanceforSchoolsToolkittoguidetheeffectivedevelopmentandapplicationofAIinteachingandlearning.Asathird

example,manydevelopershaveparticipatedintheEDSAFEAIAlliance,whichhasproducedacommonframework,the

SAFEBenchmarks.

“HarnessingAIforgoodandrealizingitsmyriadbenefitsrequiresmitigatingitssubstantialrisks.Thisendeavordemandsasociety-wideeffortthatincludes

government,theprivatesector,academia,andcivilsociety.”

—ExecutiveOrderonAI

Othermembersoftheedtechecosystemalsoareacceptingresponsibilityanddeveloping

conditionsfortrust.Educationalleadersatalllevels,includingstate,district,andbuilding-levelleaders,arewritingtheirownguidance(BoxC).Developerscanlooktotheseresourcesto

understandthestepsthateducatorsaretakingtobuildunderstandingandcapacity;to

strengthenprocurementprocesses;toprotectprivacy,security,andfairness;andtomanageotherformsofrisk.Further,inaforthcoming

ToolkitforEducationalLeaders

developed

pursuanttotheExecutiveOrderonAI,theDepartmentwillbeshapingthesharedresponsibility

conversationamongeducators.Manynonprofitsarealsodevelopinghelpfultoolkitsand

resources(seeBoxD).Askeyparticipantsintheedtechecosystem,developersareencouragedtointeractresponsiblywiththeecosystemtodeveloptrust.

BoxC:StateguidanceresourcesaboutAIineducation,asofJune2024

Arizona

Kentucky

Oregon

California

Mississippi

Utah

Connecticut

NorthCarolina

Virginia

Hawaii

Ohio

WashingtonState

Indiana

Oklahoma

WestVirginia

BoxD:PartiallistingofnonprofitsofferingguidanceresourcesaboutAIineducation

●

TheConsortiumforSchoolNetworking

(CoSN)

GenAIMaturityTool

●

EDSAFEAIAllianceSAFEBenchmarksFramework

●

InternationalSocietyforTechnologyandEducation’

sArtificialIntelligenceinEducation

ResourceList

●

Software&InformationIndustryAssociation’s

PrinciplesforAIinEducation

●

TeachAI'sAIGuidanceforSchoolsToolkit

●

All4Ed’sFutureReadySchoolsEmergingPracticesGuide

2.TrustrequiresactivelymanagingAIriskssothatwecanseizeitsbenefits.

Throughitsconversationswithdevelopers,theDepartmentobservedanimportantshiftinhowdevelopersareengagingwithothersaroundtheirwork.Whereasithasbeencommonto

present“solutions''—howtechnologiescanimproveteaching,learning,andothereducationalprocesses—developersarenowalsoopenlydiscussinghowtheyaremanagingrisks.Some

developershavepubliclyshareddetailsontheprocesstheywentthroughtoidentify,prioritize,andmanagerisks.Asdevelopersopenlydiscussriskmanagement,theDepartmentsuggests

attendingtotwokindsofprocesses:(a)technicaldevelopmentprocessesthatresultintrustworthysystemsand(b)engagementstrategiesthatbuildtrustingrelationshipsamongdevelopersandotherecosystemmembers.

OFFICEOFEducationalTechnology

5

Thus,theDepartmentunderstandstheimportanceofdiscussingbothopportunitiesandrisksinaresponsiblemanner.BoxElistsexamplesofsalientcategoriesofrisk.1,2,3Bothrisksand

opportunitieswillbediscussedinmoredepthlaterinthisguide.

BoxE:TypesofrisksofAI,orderedalphabetically,notbypriority

●AI“Race-to-Release”Risks

●BiasandFairnessRisks

●DataPrivacyandSecurityRisks

●HarmfulContentRisks

●IneffectiveSystemRisks

●MaliciousUseRisks

●MisinformationManagementRisks(including“hallucinations”)

●TransparencyandExplainabilityRisks

●UnderpreparedUserRisks

Inconsideringrisks,itisimportanttonotethatAIisevolvingrapidly.Forexample,justas

educatorsaregainingfamiliaritywithtext-orientedchatbots,industryisadvancingand

releasingmultimodalcapabilitiesthataddnewlayerstothepotentialrisks.Hence,thisguideoutlinesrisksbroadlyandasksdeveloperstoadoptriskmitigationprocessesthataddressboththerisksthatareforeseeabletodayandthosethatwillnewlyemerge.

Risksarenotonlyintrinsictothetechnology;risksalsoemergeattheinterfaceoftechnologyandhumanactivity.AspeopleuseAI,bothforeseeableandunforeseenriskswillarise.Inits

2023AIReport,

theDepartmentrecommended“humansintheloop”andyet,askingan

educatortorevieweveryuseofAIoreveryAI-basedoutputisneitherpracticalnorfair.

Developersshareresponsibilitytobe“intheloop”toreviewusesandoutputsofAI,bothduring

thedevelopmentprocessandasaproductisusedinthefield.BuildingonBoxE,weillustratechallengingscenarioswherebothdevelopersandeducatorswilllikelyneedtoattendto

emergentrisks,withadivisionofresponsibilitythatisyettobedetermined:

1

PotentialRisksofArtificialIntelligenceIntegrationintoSchoolEducation:ASystematicReview.

2GettingToKnow—andManage—YourBiggestAIRisks.

3

ThePromisesandPerilsofGenerativeAIinEducation:TFA'sEvolvingPerspective.

OFFICEOFEducationalTechnology

6

OFFICEOFEducationalTechnology7

●AsteachersgeneratepersonalizedlessonplanswithAIservices,whowillreviewandrevisetheoutputstoeliminatefalseinformationgeneratedbylargelanguagemodelsandconfirmthecontentisaccurateandalignedtoeducationalobjectives?

●AscurriculumcoordinatorsengageAItosupporttheirworkincuratinginstructional

resourcesandformativeassessmentsforuseintheirschools,whowillweighevidence

fortheefficacyoftheresourcesandthevalidityoftheassessments?Whowillverifythatresourcesaddresstheneedsofunderservedandvulnerablepopulations?

●AsguidancecounselorsuseAI-assistedtoolstorecommendcollegeandcareerpathways,whowilldetectandcounterunfairnessintherecommendationsduetobiasesin

historicaldatasetsthatwereusedtodeveloptheAImodelandwhichcouldharmvulnerablepopulations?

●AseducatorsuseAItosimplifytheirworkofwritingemailsorothercorrespondenceabouttheirstudents’work,whoisresponsibleforsafeguardsagainstdisclosinga

student’sprivateinformationtounintendedrecipients,includingthedeveloperoftheAImodel?

●Asadministratorsandschoolleadersprocureearlywarningsystemstoidentifystudentswhomaybe“atrisk,”whohassufficientknowledgeandtimetoevaluatewhethertheAIdeveloperadheredtoscientific,legal,andprivacystandardsnecessarytosafeguard

students’civilrights?

●Aseducatorsdeployanti-plagiarismdetectorstoidentifyastudent’sinappropriateuseofedtech,whohasresponsibilityforrecognizingweaknessesandbiasesinAI-based

detectorsthatcouldleadtodiscipliningstudentsunfairlyorunequally?Whoensuresthatunderservedandvulnerablepopulationsarenotunfairlytargeted?

Ineachchallengingcaseabove,theDepartmentrespectstheroleofeducatorsinoverseeing

educationaldecisions.However,AIdevelopersshouldassumesignificantresponsibilitybecauseitisunreasonabletoaskeducatorstobeprimaryreviewersofthedataandmethodsusedto

developAImodelsandrelatedsoftware.BuildingonanimportantemphasisoftheExecutiveOrderonAI,theDepartmentcallsondeveloperstopayspecialattentiontoidentifyingandmitigatingpotentialharmstounderservedandvulnerablepopulationsfromAI.

3.CoordinatingInnovationandResponsibilityThroughoutDevelopment:A“DualStack”

Developersusethevisualmetaphorofastacktodescribehowproductsarebuiltfromlayeredorconnectedcomponents.Adevelopmentstackisapowerfulwaytocoordinatetheworkof

manyinnovatorsinacomplexorganization;itcoordinateshowtheoverallproductorservicewillbeproducedanddeliveredtomarket.TheDepartmentstronglyencouragesdevelopmentorganizationstonowdefineaparallelstrengthviaacoordinated“responsibility”stack.This

stackwouldestablishhowpeopleinacomplexedtechdevelopmentorganizationworktogethertoearnthetrustofeducationalusersoftheirproducts.SeeFigure1.

OFFICEOFEducationalTechnology8

Figure1:DevelopersshouldintegrateaResponsibilityStackwiththeirInnovationStack.

Thespecificelementsofeachstackareillustrative,notcomprehensive.Eachdeveloperorganizationshouldelaboratethestackstofitthespecificsoftheirefforts.

Indeed,whereasdevelopersmayhavepreviouslyemphasizedoneparticularriskmanagementofficeorroleintheirorganization,suchasadataprivacyrole,asinglepersonwhoattendsto

safetywillnotbeenoughinanageofAI.Whiletheseexistingriskmanagementrolesshould

continue,additionalrolesthataddressrepresentationintrainingdata,measureandmitigate

algorithmicbias,rectifyerrorsandmisinformationinoutputs,andtackleotherconcernsare

alsoneeded.Specificriskswillemergeinparticularcomponents(suchasaLargeLanguage

Modelcomponentvs.aComputerVisioncomponent),ininteractionsbetweencomponents,

andthroughthedifferentstagesthattakeplacewhilebuildingandintegratingAIsystems(suchasincollectingdatasetsfortraining,buildingAImodelsbasedonthedata,anddesigninguserinteractionbasedonthosemodels).Educationaldeveloperscanexaminehowtheirinternal

rolesandprocessesareorganizedtoachieveconsistentattentiontoresponsibleAI,andmanyarealreadydoingthis(BoxF).

BoxF:ExamplesofDevelopers’PublicDocumentsaboutResponsibility

●DuoLingoreleasedits

ResponsibleAIStandardsforitsEnglishLanguage