人工智能對學習、教學和教育的影響

TheImpactofArtificialIntelligenceonLearning,Teaching,andEducation

Policiesforthefuture

Author:Tuomi,Ilkka

Editors:Cabrera,Marcelino;Vuorikari,Riina;Punie,Yves

EUR29442EN

PAGE\*roman

ii

ThispublicationisaScienceforPolicyreportbytheJointResearchCentre(JRC),theEuropeanCommission’sscienceandknowledgeservice.Itaimstoprovideevidence-basedscientificsupporttotheEuropeanpolicymakingprocess.ThescientificoutputexpresseddoesnotimplyapolicypositionoftheEuropeanCommission.NeithertheEuropeanCommissionnoranypersonactingonbehalfoftheCommissionisresponsiblefortheusethatmightbemadeofthispublication.

Contactinformation

Address:EdificioExpo.C/IncaGarcilaso3,E-41092Seville(Spain)Email:

marcelino.cabrera@ec.europa.eu

Tel.:+34954488246

EUScienceHubhttps://ec.europa.eu/jrc

JRC113226EUR29442EN

PDF ISBN978-92-79-97257-7 ISSN1831-9424 doi:10.2760/12297

Luxembourg:PublicationsOfficeoftheEuropeanUnion,2018

?EuropeanUnion,2018

ThereusepolicyoftheEuropeanCommissionisimplementedbyCommissionDecision2011/833/EUof12December2011onthereuseofCommissiondocuments(OJL330,14.12.2011,p.39).Reuseisauthorised,providedthesourceofthedocumentisacknowledgedanditsoriginalmeaningormessageisnotdistorted.TheEuropeanCommissionshallnotbeliableforanyconsequencestemmingfromthereuse.ForanyuseorreproductionofphotosorothermaterialthatisnotownedbytheEU,permissionmustbesoughtdirectlyfromthecopyrightholders.

Allimages?EuropeanUnionunlessotherwisespecified

Howtocitethisreport:Tuomi,I.TheImpactofArtificialIntelligenceonLearning,Teaching,andEducation.Policiesforthefuture,Eds.Cabrera,M.,Vuorikari,R&Punie,Y.,EUR29442EN,PublicationsOfficeoftheEuropeanUnion,Luxembourg,2018,ISBN978-92-79-97257-7,doi:10.2760/12297,JRC113226.

Title:TheImpactofArtificialIntelligenceonLearning,Teaching,andEducation

Abstract

Thisreportdescribesthecurrentstateoftheartinartificialintelligence(AI)anditspotentialimpactforlearning,teaching,andeducation.Itprovidesconceptualfoundationsforwell-informedpolicy-orientedwork,research,andforward-lookingactivitiesthataddresstheopportunitiesandchallengescreatedbyrecentdevelopmentsinAI.Thereportisaimedforpolicydevelopers,butitalsomakescontributionsthatareofinterestforAItechnologydevelopersandresearchersstudyingtheimpactofAIoneconomy,society,andthefutureofeducationandlearning.

Contents

Preface 1

Executivesummary 2

Introduction 5

WhatisArtificialIntelligence? 7

Athree-levelmodelofactionforanalysingAIanditsimpact 7

ThreetypesofAI 10

Data-basedneuralAI 10

Logic-andknowledge-basedAI 12

RecentandfuturedevelopmentsinAI 13

Modelsoflearningindata-basedAI 15

Towardsthefuture 16

AIimpactonskillandcompetencedemand 17

SkillsineconomicstudiesofAIimpact 18

Skill-biasedandtask-biasedmodelsoftechnologyimpact 20

AIcapabilitiesandtasksubstitutioninthethree-levelmodel 21

Trendsandtransitions 22

NeuralAIasdata-biasedtechnologicalchange 23

Educationasacreatorofcapabilityplatforms 23

DirectAIimpactonadvanceddigitalskillsdemand 25

Impactonlearning,teaching,andeducation 27

Currentdevelopments 27

“NoAIwithoutUI” 28

TheimpactofAIonlearning 28

Impactoncognitivedevelopment 30

TheimpactofAIonteaching 31

AI-generatedstudentmodelsandnewpedagogicalopportunities 31

Theneedforfuture-orientedvisionregardingAI 32

Re-thinkingtheroleofeducationinsociety 32

Policychallenges 34

References 37

PAGE

11

Preface

ArtificialIntelligence(AI)iscurrentlyhighonthepoliticalandresearchagendasaroundtheworld.Withtheemergenceofeverynewtechnology,thereisalwaysbothalotofhypeandscepticismarounditsimplicationsforsocietyandtheeconomy.AlthoughacknowledgingthatthefoundationsforAIhavebeenalreadyaroundforseveraldecades,recenttechnologicalbreakthroughsareacceleratingwhatAIcoulddo.Thisstudylooksatwhatthiscouldmeanforlearning,teaching,andeducation.ItaimstoprovideacriticalreviewandprospectiveangleonrelevantAIdevelopmentsasabasisforwell-informedpolicy-orienteddiscussionsaboutthefutureofthesedomains.

ThisreportisacontributiontotheDigitalEducationActionPlan1whichforeseespolicyresearchandguidanceontheimpactandpotentialofdigitaltechnologiesineducation.ItisdoneonbehalfoftheDirectorate-GeneralforEducation,Youth,SportandCulture,authoredbyIlkkaTuomiandeditedbytheJRC.Anotherreport,appraisingAIfromdifferentperspectives,entitled"ArtificialIntelligence:AEuropeanperspective",willbereleasedsoonunderthelabelofJRCflagshipreports,providinganoverallassessmentofopportunitiesandchallengesofAIfromaEuropeanoutlook,andsupportingthedevelopmentofEuropeanactionintheglobalAIcontext.

TheJRChascarriedoutresearchonLearningandSkillsfortheDigitalErasince2005.Itaimstoprovideevidence-basedpolicysupporttotheEuropeanCommissionanditsMemberStatesonhowtoharnessthepotentialofdigitaltechnologiestoencourageinnovationineducationandtrainingpractices;improveaccesstolifelonglearning;andimpartthenew(digital)skillsandcompetencesneededforemployment,personaldevelopmentandsocialinclusion.Morethan20majorstudieshavebeenundertakenontheseissues,resultinginmorethan120differentpublications.

Recentworkhasfocusedonthedevelopmentofdigitalcompetenceframeworksforcitizens(DigComp),educators(DigCompEdu),educationalorganisations(DigCompOrg)andconsumers(DigCompConsumers).Aframeworkforopeninguphighereducationinstitutions(OpenEdu)wasalsopublishedin2016,alongwithacompetenceframeworkforentrepreneurship(EntreComp).Someoftheseframeworksareaccompaniedby(self-)assessmentinstruments.TheJRCisalsoentrustedtodevelopafutureframeworkforpersonalandsocialdevelopment,includinglearningtolearn.AdditionalresearchhasbeenundertakenonLearningAnalytics,MOOCs(MOOCKnowledge,MOOCs4inclusion),Computationalthinking(Computhink)andpoliciesfortheintegrationandinnovativeuseofdigitaltechnologiesineducation(DigEduPol).

MoreinformationonallourstudiescanbefoundontheJRCSciencehub:https://ec.europa.eu/jrc/en/research-topic/learning-and-skills.

1CommunicationfromtheCommissiontotheEuropeanParliament,theCouncil,theEuropeanEconomicandSocialCommitteeandtheCommitteeoftheRegionsontheDigitalEducationActionPlan(COM(2018)237final).

Executivesummary

AttheNovember2017GothenburgSummit,theCommissionpresentedtheCommunication'StrengtheningEuropeanIdentitythroughEducationandCulture',thatsetoutavisionforaEuropeanEducationAreaandannouncedadedicatedDigitalEducationActionPlan2,whichaimstofosterdigitalskillsandcompetencesforallcitizens.TheActionPlanfocusesonimplementationandtheneedtostimulate,supportandscaleuppurposefuluseofdigitalandinnovativeeducationpractices.Ithasthreepriorities:makingbetteruseofdigitaltechnologyforteachingandlearning;developingrelevantdigitalcompetencesandskillsforthedigitaltransformation;andimprovingeducationthroughbetterdataanalysisandforesight.ArtificialIntelligence(AI)willhaveanimpactonallthese,andinthelastprioritytheCommunicationspecificallyinvitestoexploreitsimpactineducationandtrainingthroughpilots.ThispolicyforesightreportsuggeststhatinthenextyearsAIwillchangelearning,teaching,andeducation.Thespeedoftechnologicalchangewillbeveryfast,anditwillcreatehighpressuretotransformeducationalpractices,institutions,andpolicies.ItisthereforeimportanttounderstandthepotentialimpactofAIonlearning,teaching,andeducation,aswellasonpolicydevelopment.

AIiscurrentlyhighonthepoliticalagendasaroundtheworld.SeveralEUMemberStateshavedeclareditasapoliticalpriority.InfluentialstudiesnowsuggestthatperhapsoneintwooccupationsintheindustrializedcountriesislikelytobecomeautomatedusingalreadyexistingAItechnologies.PolicymakersattheEuropeanParliamenthavehighlightedtheimportanceoftheissue,andtheEuropeanCommission,inits2018annualworkprogramme,setsitswishtomakethemostofAI,whichwillincreasinglyplayaroleinoureconomiesandsocieties3.AIisnowoftencalled“thenextelectricity.”Thetransformativeimpactofgeneralpurposetechnologies,likeAI,however,becomesvisibleonlygradually,whensocietiesandeconomiesreinventthemselvesasusersofnewtechnologies.Technologicalchangebringssocialandculturalchangethatisreflectedinlifestyles,norms,policies,socialinstitutions,skills,andthecontentandformsofeducation.

Wideavailabilityofcheapprocessingpowerandvastamountsofdatainrecentyearshaveenabledimpressivebreakthroughsinmachinelearningandcreatedextraordinarycommercialandresearchinterestinartificialneuralnetworks,i.e.computationalmodelsbasedonthestructureandfunctionsofbiologicalneuralnetworks.NeuralAI,andmachinelearningmethodsassociatedwithit,arenowusedforreal-timelanguageprocessingandtranslation,imageanalysis,driverlesscarsandautonomousvehicles,automatedcustomerservice,frauddetection,processcontrol,syntheticart,servicerobots,andinmanyotherapplications.Althoughsomeofthisexcitementmaybebasedonunrealisticexpectationsandlimitedknowledgeofthecomplexitiesoftheunderpinningtechnologies,itisreasonabletoexpectthattherecentadvancesinAIandmachinelearningwillhaveprofoundimpactsonfuturelabourmarkets,competencerequirements,aswellasinlearningandteachingpractices.Aseducationalsystemstendtoadapttotherequirementsoftheindustrialage,AIcouldmakesomefunctionsofeducationobsoleteandemphasizeothers.Itmayalsoenablenewwaysofteachingandlearning.

2CommunicationfromtheCommissiontotheEuropeanParliament,theCouncil,theEuropeanEconomicandSocialCommitteeandtheCommitteeoftheRegionsontheDigitalEducationActionPlan(COM(2018)237final).

3CommunicationfromtheCommissiontotheEuropeanParliament,theCouncil,theEuropeanEconomicandSocialCommitteeandtheCommitteeoftheRegionsCommissionWorkProgramme2018-Anagendaforamoreunited,strongerandmoredemocraticEurope(COM(2017)650final).

IntheEuropeanframeworkprogrammesforresearchandtechnologicaldevelopment,AItechnologieshavebeenstudiedandappliedineducationalcontextsinmanyprojectsfocusingontechnology-enabledlearning.TheseprojectshaveusedtechnologiesthathavedeeptieswithAIresearch,includingnaturallanguageprocessing,patternrecognition,intelligenttutoring,probabilisticAIplanning,intelligentagents,AIgameengines,andadaptiveusermodelsinpersonalizedlearningenvironments(PLE).Theimpactofthesetechnologiesinpracticaleducationalsettingshasbeenrelativelymodestuntilrecently.Technicaldevelopmentsovertherecentyears,however,suggestthatthesituationmaybechangingrapidly.

Themainintentofthepresentreportistohelpeducatorsandpolicymakerstomakesenseofthesepotentiallyveryimportanttechnicaldevelopments.TounderstandtheimpactofAI,weneedtounderstandwhatAIisandwhatitcando.Inthecurrent“AIavalanche”thisisnotalwayseasy.DeepexpertiseinAItechnologyisscarce,andmanyeducatorsandpolicymakersnowstruggletogetuptodatewithbasicknowledgeinthisarea.Inthemidstofself-drivingcars,speakingrobots,andthefloodof“AImiracles”,itmaybeeasytothinkthatAIisrapidlybecomingsuperintelligent,andgainallthegoodandevilpowersawardedtoitinpopularculture.This,ofcourse,isnotthecase.ThecurrentAIsystemsareseverelylimited,andtherearetechnical,social,scientific,andconceptuallimitstowhattheycando.Perhapssurprisingly,well-establishedresearchonhumanlearningprovidesimportanttoolsandconceptsthathelpusunderstandthestate-of-the-artandfutureofAI.ManycurrentAIsystemsuserathersimplifiedmodelsoflearningandbiologicalintelligence,andlearningtheoriesthushelpusgainbetterunderstandingofthecapabilitiesofcurrentAIsystems.

TherewillbegreateconomicincentivestouseAItoaddressproblemsthatarecurrentlyperceivedasimportantbyeducationaldecision-andpolicy-makers.Thiscreatespolicychallenges.Foreducationaltechnologyvendorsitiseasytosellproductsthatsolveexistingproblems,butitisverydifficulttosellproductsthatrequirechangesininstitutions,organizationsandcurrentpractices.Toavoidhard-wiringthepast,itwouldbeimportanttoputAIinthecontextofthefutureoflearning.PolicymaybeneededtoorientdevelopmentinAItowardssociallyusefuldirectionsthataddressthechallenges,opportunities,andneedsofthefuture.AsAIscalesup,itcaneffectivelyroutinizeoldinstitutionalstructuresandpracticesthatmaynotberelevantforthefuture.Future-orientedwork,therefore,isneededtounderstandthepotentialimpactofAItechnologies.Howthispotentialisrealizeddependsonhowweunderstandlearning,teachingandeducationintheemergingknowledgesocietyandhowweimplementthisunderstandinginpractice.Future-orientedpolicyexperimentation,assuggestedbytheDigitalEducationActionPlan,may,therefore,beaneffectivewaytoaddressthischallenge..

RecentAIbreakthroughsarebasedonsupervisedmachinelearning.Acriticalsuccessfactorofthesesystemsistheavailabilityofhugeamountsofpre-categorizedtrainingdata.Incontrasttologic-andknowledge-basedapproachestoAI,wethereforecharacterizetheseas“data-based”AIsystemsinthisreport.Manyofthese“deep-learning”neuralAIsystemsmaywellbecharacterizedas“datavores.”Atpresent,themostimportanttechnicalbottleneckofAI,therefore,istheavailabilityofdata.Thisisaqualitativelynewdevelopmentinthehistoryofcomputingandinformationprocessing.Withoutaccesstovasttrainingdatasets,itisverydifficulttodevelopsuccessfulAIsystems.Inthisreport,weputforwardanargumentthatEUpoliciescouldcreatedataplatformsthatcouldredefinethecompetitivelandscapeforlearning-andeducation-orientedAIsystems.

AsthesesupervisedAIlearningalgorithmsarebasedonhistoricaldata,theycanonlyseetheworldasarepetitionofthepast.Thishasdeepethicalimplications.When,forexample,studentsandtheirachievementsareassessedusingsuchAIsystems,theassessmentisnecessarilybasedoncriteriathatreflectculturalbiasesandhistoricallysalientmeasuresofsuccess.Supervisedlearningalgorithmscreateunavoidablebiases,andthesearecurrentlyextensivelydebated.Fromamorefundamentalethicalpointofview,however,theexpressionofhumanagencyrequirescapabilitytomakeauthenticchoicesthatdonotonlyrepeatthepast.AlthoughtherearealreadyAIsystemsthatdealwithcreativeactivities,AIsystemswillhavegreatdifficultiesindealingwithpeoplewhoarecreative,innovative,andnotonlyaveragerepresentationsofvastcollectionsofhistoricalexamples.

ItisoftenassumedthatAIsystemsenablenewlevelsofpersonalisationanddiversityforinformationsystems;muchofthis,however,resultsfromfine-grainedcategorizationthatputsusersintopre-definedclasses.Althoughthesesystemsmaybeabletoefficientlysimulatepersonalisation,theydonotnecessarilysupportdeeperlevelsofdiversity.AtpresentwecansaythattheuseAIsystemsineducationalsettingswillshapethedevelopmentofhumancognitionandself-efficacy,butwedon’tknowhow.Itisthereforeimportanttocontinuouslyevaluate,forexample,howtheuseofAIineducationalcontextsconstrainsandenableshumanpossibilitiesforresponsibleandethicalaction.AIsystemscanbeexcellentpredictivemachines,butthisstrengthmaybeanimportantweaknessindomainswherelearninganddevelopmentareimportant.AcontributionofthisreportistoshowthatdifferenttypesofAIandmachinelearningsystemsoperateondifferentlayersofhumanbehaviour4.Mostimportantly,thelevelofmeaningfulactivity—whichinsocio-culturaltheoriesoflearningunderpinsadvancedformsofhumanintelligenceandlearning—remainsbeyondthecurrentstateoftheAIart.

OneofthemostsuccessfulapplicationareasinAIhasbeenvideoprocessing.Therewillbestrongeconomicinterestsinusingvideo-connectedAIsystemsinclassroomsandtocomplementthecollecteddatawithdatafromsocialmediaandInternetofthings(IoT)platforms.Asitbecomestechnicallypossibletomonitorstudentemotionsandattentioninrealtimeandusesuchdatatohelpteachersandstudents,AIprivacyandsecuritybecomeimportanttopicsalsoineducation.Similarly,AIsystemsarewellsuitedforcollectinginformalevidenceofskills,experience,andcompetencefromopendatasources,includingsocialmedia,learnerportfolios,andopenbadges.Thiscreatesbothethicalandregulatorychallenges.

Severalhigh-profileeconometricstudiesonthefutureofworkhaveshownthatmanyoccupationscanbeautomatedwithcurrentAItechnologies.Thesestudieshavereliedontask-andskill-biasedmodelsoftechnicalchange.Inthisreport,wearguethatadata-biasedmodelismoreappropriateforcurrentAIsystems.Wealsoexploreasimilarmethodologytoseehowthefutureoftheteachingprofessionmightlooklike.Theresultssuggestthatmanycurrentlydefinedhigh-priorityteachertasksmightbeautomated.However,thisisbasedontheassumptionthattheroleofteachersisrathermechanicalandpurelyinstructionalwithsummativeassessmentplayingacentralrole,reflectingdeepbeliefsaboutthefunctionsofeducationandthesocialinstitutionsaroundit.Ineducationalsystemsthatemphasizedevelopmentand,forexample,socialcompetences,formativeassessmentmightbehigheronthelist.Asaresult,thereisariskthatAImightbeusedtoscaleupbadpedagogicalpractices.IfAIisthenewelectricity,itwillhaveabroadimpactinsociety,economy,andeducation,butitneedstobetreatedwithcare.

4Readersmayalsobeinterestedin“HUMAINT”,aninterdisciplinaryJRCprojectaimingtounderstandtheimpactofmachineintelligenceonhumanbehaviour,withafocusoncognitiveandsocio-emotionalcapabilitiesanddecisionmaking(seehttps://ec.europa.eu/jrc/communities/community/humaint).

Introduction

Allhumanactionsarebasedonanticipatedfutures.Wecannotknowthefuturebecauseitdoesnotexistyet,butwecanuseourcurrentknowledgetoimaginefuturesandmakethemhappen.Thebetterweunderstandthepresentandthehistorythathascreatedit,thebetterwecanunderstandthepossibilitiesofthefuture.Toappreciatetheopportunitiesandchallengesthatartificialintelligence(AI)creates,weneedbothgoodunderstandingofwhatAIistodayandwhatthefuturemaybringwhenAIiswidelyusedinthesociety.AIcanenablenewwaysoflearning,teachingandeducation,anditmayalsochangethesocietyinwaysthatposenewchallengesforeducationalinstitutions.Itmayamplifyskilldifferencesandpolarizejobs,oritmayequalizeopportunitiesforlearning.TheuseofAIineducationmaygenerateinsightsonhowlearninghappens,anditcanchangethewaylearningisassessed.Itmayre-organizeclassroomsormakethemobsolete,itcanincreasetheefficiencyofteaching,oritmayforcestudentstoadapttotherequirementsoftechnology,deprivinghumansfromthepowersofagencyandpossibilitiesforresponsibleaction.Allthisispossible.NowisagoodtimetostartthinkingaboutwhatAIcouldmeanforlearning,teaching,andeducation.Thereisalotofhype,andthetopicisnotaneasyone.Itis,however,bothimportant,interesting,andworththeeffort.

Since2013,whenFreyandOsborne5estimatedthatalmosthalfofU.S.jobswereatahighriskofbecomingautomated,AIhasbeenontopofpolicymakers’agendas.Manystudieshavereplicatedandrefinedthisstudy,andthegeneralconsensusnowisthatAIwillgeneratemajortransformationsinthelabourmarket.6Manyskillsthatwereimportantinthepastarebecomingautomated,andmanyjobsandoccupationswillbecomeobsoleteortransformedwhenAIwillbeincreasinglyused.Atthesametime,therehasbeenatremendousdemandforpeoplewithskillsinAIdevelopment,leadingtosevenfiguresalariesandsign-upfees.ChinahasannouncedthatitaimstobecometheworldleaderinAIandgrowa150billionAIecosystemby2030.TheU.S.DepartmentofDefenseinvestedabout2.5billionUSDinAIin2017,andthetotalprivateinvestmentintheU.S.isnowprobablyover20billionUSDperyear.In2017,therewereabout1200AIstart-upsinEurope,7andtheEuropeanCommissionaimstoincreasethetotalpublicandprivateinvestmentinAIintheEUtobeatleast20billioneurosbytheendof2020.8

Inlimitedtasks,AIalreadyexceedshumancapabilities.Lastyear,withjustaboutonemonthofsystemdevelopment,researchersatStanfordwereabletouseAItodiagnose

14typesofmedicalconditionsusingfrontal-viewX-rayimages,exceedingthehumandiagnosticaccuracyforpneumonia.9In2017,givennodomainknowledgeexceptthegamerules,anartificialneuralnetworksystem,AlphaZero,achievedwithin24hoursasuperhumanlevelofplayinthegamesofchess,shogi,andGo.10InMay2018,GoogleCEOSundarPichaicausedafirestormwhenhedemonstratedinhiskeynoteanAIsystem,Duplex,thatcanautonomouslyscheduleappointmentsonthephone,foolingpeopletothinktheyarediscussingwithanotherhuman.Inthemidstofself-drivingcars,speakingrobots,andthefloodofAImiracles,itmaybeeasytothinkthatAIisrapidlybecomingsuperintelligent,andgainallthegoodandevilpowersawardedtoitinpopularculture.This,ofcourse,isnotthecase.ThecurrentAIsystemsareseverelylimited,andtherearetechnical,social,scientific,andconceptuallimitstowhattheycando.Asone

5FreyandOsborne(2013,2017).

6E.g.,EuropeanPoliticalStrategyCentre(EPSC2018),UnitedStatesGovernmentAccountabilityOffice(GAO2018),FinnishSteeringGroupofArtificialIntelligenceProgramme(2017),andUKHouseofLords(2018).

7 DatafromtheU.K.HouseofLordsSelectCommitteeonArtificialIntelligencereport(HouseofLords2018,48).

8 ArtificialIntelligenceforEurope(EC2018b).

9 Rajpurkaretal.(2017).

10 Silveretal.(2017).

recentauthornoted,AImayberidingaone-trickponyasalmostallAIadvancesreportedinthemediaarebasedonideasthataremorethanthreedecadesold.11AparticularchallengeofthecurrentlydominantlearningmodelsusedinAIisthattheycanonlyseetheworldasarepetitionofthepast.Theavailablecategoriesandsuccesscriteriathatareusedfortheirtrainingaresuppliedbyhumans.Personalandculturalbiases,thus,areaninherentelementinAIsystems.Athree-levelmodelofhumanactionpresentedinthenextsectionsuggeststhatnormsandvaluesareoftentacitandexpressedthroughunarticulatedemotionalreactions.Perhapssurprisingly,therecentsuccessesinAIalsorepresenttheoldestapproachtoAIandonewherealmostalltheintelligencecomesfromhumans.

InsteadofabeginningofanAIrevolution,wecouldbeattheendofone.This,ofcourse,dependsonwhatwemeanbyrevolution.ElectricitydidnotrevolutionizetheworldwhenVoltafoundawaytostoreitin1800orwhenEdisonGeneralElectricCompanywasincorporatedin1889.Thetransformativeimpactofgeneralpurposetechnologiesbecomesvisibleonlygradually,whensocietiesandeconomiesreinventthemselvesasusersofnewtechnologies.Technologicalchangerequiresculturalchangethatisreflectedinlifestyles,norms,policies,socialinstitutions,skills,andeducation.Becauseofthis,AI—nowoftencalledthe"newelectricity"—mayrevolutionizemanyareasoflifewhenitistakenintouseevenifitkeepsondrivingits"one-trick"ponyfortheforeseeablefuture.Manyinterestingthingswillhappenwhenalreadyexistingtechnologieswillbeadopted,adapted,andappliedforlearning,teaching,andeducation.Forexample,AImayenablebothnewlearningandteachingpractices,anditmaygenerateanewsocial,cultural,andeconomiccontextforeducation.

BelowweasksimplequestionsthatillustratetherelevanceofAIforeducationalpoliciesandpractices.Whichvocationsandoccupationswillbecomeobsoleteinthenearfuture?Whatarethe21stCenturyskillsinaworldwhereAIiswidelyused?HowshouldAIbeincorporatedintheK-12curriculum?HowwillAIchangeteaching?Shouldreal-timemonitoringofstudentemotionsbeallowedinclassrooms?CanAIfairlyassessstudents?DoweneedfewerclassroomsbecauseofAI?DoesAIreducetheimpactofdyslexia,dyscalculia,orotherlearningdifficulties?Thesequestionsaresimpletoask,andrelevantforunderstandingthefutureoflearning,teaching,andeducation.Theanswers,ofcourse,aremorecomplex.

Themainaimofthisreportistoputtheseandothersimilarquestionsinacontextwheretheycanbemeaningfullyaddressed.Wedonotaimtoprovidefinalanswers;instead,wehopetoprovidebackgroundthatwillfacilitatediscussionontheseandotherimportantquestionsthatneedtobeaskedasAIbecomesincreasinglyvisibleinthesocietyandeconomyaroundus.Todothis,wehavetofirstopenthe"blackbox"ofAIandpeekinside.ThereareseveralthingsAIcandowell,andmanythingsitcannotdo.AtpresentthereisanavalancheofreportsandnewspaperarticlesonAI,anditisnotalwayseasytodistinguishimportantmessagesfromnoise.Itis,however,importanttounderstandsomekeycharacteristicsofcurrentAItobeabletoimaginerealisticfutures.Inthenextsections,weputAIinthecontextoflearning,teaching,andeducation,andthenfocusonthespecificformofAI,adaptiveartificialneuralnetworks,thathavegeneratedtherecentinterestinAI.

11 Somers(2017