解读欧盟人工智能法案（英）-毕马威.docx

umf¾rJrfWDecodingtheEUAlActUnderstandingtheAlAct,simpactandhowyoucanrespondKPMG.MaketheDifference.KPMGInternationalContents07ExecutivesummaryExaminingtheAlAcfsimpactandscopeUnravellingtheAlAcfskeycomponentsNextstepsIntroductionArtificial Intelligence (AI) is offering new benefits to society and businesses, aiming to transform the workplace and major industries along the way.Simply put, the race is on to embrace the remarkable and evolving power of Al and automation.AsKPMG'sGlobalTeChReDOrt2023reveals,mostglobalexecutives(62percent)reportanincreaseinperformanceorprofitabilityfromdigitaltransformationinitiativesrelatedtoAlandmachinelearningoverthepast24months.And68percentsaythesetechnologieswillplaya*vital,roleinhelpingthemachievetheirbusinessobjectivesoverthenextthreeyears,while57percentbelieveAlandmachinelearningwillbe'important*inmeetingshort-termobjectives.ButastheworldwideAlproliferationinbusinessandoreverydaylivesunfolds,thereisacriticalneedforguardrailsandlegislationtodealwithsignificantnewrisksregardingtheappropriateandethicaluse,developmentanddistributionofAl.AccordingtoTnJStinartificialintellience,aglobalsurveyconductedbyKPMGAustraliaandtheUniversityofQueensland,threeinfivepeoplearewaryabouttrustingAlsystems,and71percentexpectAltoberegulated.Morerecently,CEOsfromglobaltechgiantscalledforgreaterAlregulationatameetingonCapitolHilltoprotectpeoplefromtheworsteffectsofAl.Inresponse,theEuropeanUnion(EU)hasreachedaground-breakingprovisionalagreementonacomprehensiveArtificialIntelligenceAct(AlAct)thattakesarisk-basedapproachtoprotectingfundamentalrights,democracy,theruleoflawandenvironmentalsustainability.iThoughit'sexpectedtobecomelawin2024,withcomplianceexpectedby2025,thislegislationthefirstofitskindisanticipatedtoemergeasthede-factonewglobalstandardforAlregulation.WiththeintroductionoftheAlAct,theEUaimstostrikeabalancebetweenfosteringAladoptionandensuringindividuals'righttoresponsible,ethicalandtrustworthyuseofAl.Inthispaper,weexplorewhattheAlActmaymeantoyourorganizationandexaminethestructureoftheAlAct,theobligationsitimposes,thetimelinesforcomplianceandtheactionplanthatorganizationsshouldconsider.OrganizationalleadershipshoulddriveinitiativesinlinewiththeAlAct,companybrand,valuesandrisktolerancetopromoteresponsibleuseofAl.Thiscanhelppromoteethicaldevelopment,regulatorycompliance,riskmitigationandstakeholdertrust.，DavidRowlandsGlobalAlLeaderKPMGInternationalAlshouldbedevelopedandusedwithafocusonsafetyandethics,turningtechnologicaladvancementintoapositiveforceforsociety.TheEUAlActwillhelpfosterinnovationwhileprotectingend-users.，1.aurentGobbiGlobalTrustedAlLeaderKPMGInternationaliEuropeanParliament.(December9,2023).ArtificialIntelligenceAct:dealoncomprehensiverulesfortrustworthyAlPressrelease.ExecutivesummaryTheAlActaimstoregulatetheethicaluseofAlAlholdsimmensepromisetoexpandthehorizonofwhatisachievableandtoimpacttheworldforourbenefit-butmanagingAsrisksandpotentialknownandunknownnegativeconsequenceswillbecritical.TheAlActissettobefinalizedin2024andaimstoensurethatAlsystemsaresafe,respectfundamentalrights,fosterAlinvestment,improvegovernance,andencourageaharmonizedsingleEUmarketforAl.MostAlsystemsneedtocomplywiththeAlActbythefirsthalfof2026TheAlAct'sdefinitionofAlisanticipatedtobebroadandincludevarioustechnologiesandsystems.Asaresult,organizationsarelikelytobesignificantlyimpactedbytheAlAct.Mostoftheobligationsareexpectedtotakeeffectinearly2026.However,prohibitedAlsystemswillhavetobephasedoutsixmonthsaftertheAlActcomesintoforce.Therulesforgoverninggeneral-purposeAlareexpectedtoapplyinearly2025.2Providersandusersofhigh-riskAlsystemsfacestringentobligationsTheAlActappliesarisk-basedapproach,dividingAlsystemsintodifferentrisklevels:unacceptable,high,limitedandminimalrisk.3High-riskAlsystemsarepermittedbutsubjecttothemoststringentobligations.Theseobligationswillaffectnotonlyusersbutalsoso-called'providers'ofAlsystems.Theterm,provider,intheAlActcoversdevelopingbodiesofAlsystems,includingorganizationsthatdevelopAlsystemsforstrictlyinternaluse.Itisimportanttoknowthatanorganizationcanbebothauserandaprovider.Providerswilllikelyneedtoensurecompliancewithstrictstandardsconcerningriskmanagement,dataquality,transparency,humanoversight,androbustness.UsersareresponsibleforoperatingtheseAlsystemswithintheAlAct'slegalboundariesandaccordingtotheprovider'sspecificinstructions.Thisincludesobligationsontheintendedpurposeandusecases,datahandling,humanoversightandmonitoring.GuardrailsforgeneralAlsystemsNewprovisionshavebeenaddedtoaddresstherecentadvancementsingeneral-purposeAl(GPAI)models,includinglargegenerativeAlmodels.”ThesemodelscanbeusedforavarietyoftasksandcanbeintegratedintoalargenumberofAlsystems,includinghigh-risksystems,andareincreasinglybecomingthebasisformanyAlsystemsintheEU.ToaccountforthewiderangeoftasksAlsystemscanaccomplishandtherapidexpans