ACCA+為值得信賴的人工智能奠定基礎(chǔ)

BUILDINGTHEFOUNDATIONSFOR

TRUSTEDARTIFICIALINTELLIGENCE

ResponsetotheUKgovernmentwhitepaper,Apro-innovationapproachtoAIregulation

AboutACCA

ACCA(theAssociationofCharteredCertifiedAccountants)istheglobalprofessionalbodyforprofessionalaccountants.

We’reathrivingglobalcommunityof241,000membersand542,000futuremembersbasedin178countriesandregions,whoworkacrossawiderangeofsectorsandindustries.Weupholdthehighestprofessionalandethicalvalues.

Weoffereveryoneeverywheretheopportunitytoexperiencearewardingcareerinaccountancy,financeandmanagement.Ourqualificationsandlearningopportunitiesdevelopstrategicbusinessleaders,forward-thinkingprofessionalswiththefinancial,businessanddigitalexpertiseessentialforthecreationofsustainableorganisationsandflourishingsocieties.

Since1904,beingaforceforpublicgoodhasbeenembeddedin

ourpurpose

.InDecember2020,wemadecommitmentstothe

UNSustainableDevelopment

Goals

whichwearemeasuringandwillreportoninourannualintegratedreport.

Webelievethataccountancyisacornerstoneprofessionofsocietyandisvitalinhelpingeconomies,organisationsandindividualstogrowandprosper.Itdoesthisbycreatingrobusttrustedfinancialandbusinessmanagement,combatingcorruption,ensuringorganisationsaremanagedethically,drivingsustainability,andprovidingrewardingcareeropportunities.

Throughourcutting-edgeresearch,weleadtheprofessionbyansweringtoday’squestionsandpreparingforthefuture.We’reanot-for-profitorganisation.

Findoutmoreat

AboutEY

EYexiststobuildabetterworkingworld,helpingtocreatelong-termvalueforclients,peopleandsocietyandbuildtrustinthecapitalmarkets.

Enabledbydataandtechnology,diverseEYteamsinover150countriesprovidetrustthroughassuranceandhelpclientsgrow,transformandoperate.

Workingacrossassurance,consulting,law,strategy,taxandtransactions,EYteamsaskbetterquestionstofindnewanswersforthecomplexissuesfacingourworldtoday.

EYreferstotheglobalorganization,andmayrefertooneormore,ofthememberfirmsofErnst&YoungGlobalLimited,eachofwhichisaseparatelegalentity.Ernst&YoungGlobalLimited,aUKcompanylimitedbyguarantee,doesnotprovideservicestoclients.InformationabouthowEYcollectsandusespersonaldataandadescriptionoftherightsindividualshaveunderdataprotectionlegislationareavailablevia/privacy.EYmemberfirmsdonotpracticelawwhereprohibitedbylocallaws.

Formoreinformationaboutourorganization,pleasevisit

ThisnewsreleasehasbeenissuedbyEYGMLimited,amemberoftheglobalEYorganizationthatalsodoesnotprovideanyservicestoclients.

?2023AssociationofCharteredCertifiedAccountantsandEYGlobalServicesLimited.

July2023

TheviewsreflectedinthisarticlearetheviewsoftheauthoranddonotnecessarilyreflecttheviewsoftheglobalEYorganizationoritsmemberfirms.

Thispublicationcontainsinformationinsummaryformandisthereforeintendedforgeneralguidanceonly.Itisnotintendedtobeasubstitutefordetailedresearchortheexerciseofprofessionaljudgment.MemberfirmsoftheglobalEYorganizationcannotacceptresponsibilityforlosstoanypersonrelyingonthisarticle.

BUILDINGTHEFOUNDATIONSFOR

TRUSTEDARTIFICIALINTELLIGENCE

RESPONSETOTHEUKGOVERNMENTWHITEPAPER,

APRO-INNOVATIONAPPROACHTOAIREGULATION

InMarch2023,theUKgovernment’sDepartmentforScience,

Technology&Innovationpublishedawhitepaperentitled

APro-innovationApproachtoAIRegulation.Thewhitepaper

setsoutthegovernment’sstrategyforsupportinginnovation,

whileprovidingaframeworktoensurethattherisksassociated

withAIareidentified,addressedandmitigated.

Thispaper,fromACCAandEY,isaresponsetothatwhite

paper.Ouraimistoprovidepolicymakerswithaconstructive

critiqueofthewhitepaper,informedbyourpositionswithin

theaccountancyandbusinessecosystem,andtheaccessto

feedbackandinputswehaveacrossourstakeholderbases.

Forfurtherinformation:

NarayananVaidyanathan,HeadofPolicyDevelopment,ACCA

Dr.AnsgarKoene,GlobalAIEthicsandRegulatoryLeader,EY

4

Forewords

HelenBrandOBE

Chiefexecutive,ACCA

NotadaygoesbywithoutAIbeinginthenews,anditfeaturesstronglyinthepublicconsciousnessduetoitspotentialtointroducebothnewopportunitiesandpreviouslyunseenrisks.ThereleaseoftheUKgovernment’sAIwhitepaperhasprovidedatimelyopportunitytotakestock.

Asthispapernotes,we’resupportiveoftheoverallpro-innovationapproach,andtheimportanceofmulti-stakeholderfeedback,suchascapturingthevoiceofSMEsandtheUKregionsoutsideLondon.Wealsosupportthedesiretobuildbridgesinternationally,whichalignsstronglywithACCA’sownglobalapproach.WestronglyendorsetheexplicitreferencetotheroleofauditandassuranceandrelatedtoolsasakeypartofthesolutiontoensureanAIecosystemthatisunderpinnedbyethicalpractices.

Aflexibleapproachtomanagethefastpaceofnewdevelopmentcanhelp–thecurrentwaveofgenerativeAI,forexample,hasexplodedquiterecently.However,further,moreprescriptiveguidancemaybeneededtoestablishclarityonwhereaccountabilitieslieinsomeinstances.

Overall,weseethisasthestartofamulti-yearprocessandlookforwardtocollaboratingwithandsupportingUKgovernmentinthisimportantendeavourinanywaywecan.

ProfessorMichaelMainelli

FCCA,Chairman,Z/YenGroup,SeniorIndependentDirector,UnitedKingdom,AccreditationService,SeniorAldermanBelowtheAldermanicChair,CityofLondonCorporation,andHeadoftheEthicalAIInitiative

Thegovernment’sAIwhitepaperlaysoutaproportionateand‘pro-innovation’approachtoAIregulation.ABritishstrengthinregulatoryframeworkshasbeenabiastowardstheuseofindustryorcommunity‘standardsmarkets’,whereindustrystandardsandassuranceprocessesprovide‘a(chǎn)ccreditedconformityassessment’.Standardsmarketsbasedonproportionateandpro-innovationbusinessecosystemsrangingfromsafegasboilersuppliestofireprevention,aircraftcertification,orshippingsafety.

Standardsmarketsshouldprovideaccreditationandcertificationsuchthatpeoplecanplaceappropriatereliabilityontheassessedproductsandservices.Thispapernotestheopportunitytousesuchstandards,particularlyexistingISOstandards,forinspectionandtestingtoprovideappropriateassuranceonAIproductsandservices.Thiscombinationofminimalregulationwithmaximaluseofstandardsmarketscanprovideamiddlepathbetweentwoextremes:acompletelyunregulatedapproachleavingatangleofconfusiontothelegalsystemandanoverlyregulatedapproachwherenothingcanmoveuntilpermissionisgranted.

ThoseofuswithdecadesofexperienceinthefieldsofAIhavelongrecognisedtheneedforstandards.TheseISOstandardshavenotjust‘poppedoutofthewoodwork’duetorecentmediaattention;rathertheyhavebeentheresultofyearsofconcertedworktoreachinternationalagreements.ThispapercontendsthatitisnowappropriatetopushfirmlytowardstheiruseinaccreditedUKconformityassessment.

ShawnMaher

EYGlobalViceChair–

Publicpolicy

DrivenbyacceleratingdevelopmentandadoptionofAI-enabledsolutions,policymakersacrosstheglobeareconfrontingthechallengesofformulatingaregulatoryapproachtothisrapidlyadvancingtechnologythatwillmitigatethepotentialharmsofAIwhilesimultaneouslyenablingitssocialandeconomicbenefits.

TheAIgovernancelandscapeisevolvingrapidlyandislikelytocontinuetobeinfluxfor severalyearstocome.Withinthiscontext,theUKgovernment’sAIwhitepaperhasproposeda‘pro-innovation’approachthatfocusesontheroleofexistingregulatorybodiesindevelopinganAIregulatoryframeworkunderpinnedbyprinciplesofresponsibledevelopmentanduse.

RegulatoryapproachestoAI–iftheyaretobeeffective–mustoperateinaglobalmarketandbecomparableacrossjurisdictionsinordertoensurerulesthatmitigateriskandallowforresponsibleinnovationanduse.TheUK’salignmentofitsAIregulatoryprincipleswiththoseoftheOECDandG20isanencouragingsignthatitiscommittedtobealeaderineffortstodevelopsensiblerulesforAI’sprogress.Wesupportthisroleandurgeothernationstocontinuetoworktogethertodeveloprulesthatareeffectiveonaglobalaswellaslocalbasis.

5

Contents

1.Executivesummary6

2.Recommendationsforpolicymakers8

3.Introduction9

4.ResponsetoUKgovernmentwhitepaper

13

AIregulatorydevelopmentsintheUK

13

WhatdoestheUKgovernmentwhitepaperpropose?

13

Generalreactionstothewhitepaper

14

Specificreactionstothewhitepaper

15

ToolsfortrustworthyAIandethicaluse

16

Existingcodesandstandardsrelevanttoprofessionalaccountants

17

5.Conclusion

20

Acknowledgements

21

PastpapersfromACCA

22

PastpapersfromEY

23

References

24

6

BUILDINGTHEFOUNDATIONSFORTRUSTEDARTIFICIALINTELLIGENCE|EXECUTIVESUMMARY

Executivesummary

1.TheUKgovernmentwhitepaper,APro-innovationApproachtoAIRegulation,hasbeenlongawaitedanditspublicationiswelcome.Itadoptsaflexible,common-senseapproachtotheregulationofartificialintelligence(AI)andalignswithotherprinciplessuchasthevalues-basedAIprinciplespropoundedbytheOrganisationforEconomicCo-operationandDevelopment(OECD).

2.GiventhepaceofchangeandnewscenariosthrownupbyAI,thewhitepaper’sapproachappearswellsuitedto‘unknown-unknown’AIrisks.TheapproachoflearningthroughwatchingandwaitingrecognisesthatAIischangingfastandthatsomerisksmaybebetterassessedanddefinedasthetechnologyevolves.AtACCAandEY,wearesupportiveofregulatingoutcomesratherthanthetechnologyitself.Thisalignswiththeprinciples-basedapproachtoregulationfortheaccountancyprofessionintheUK.1

3.Weagreewiththewhitepaper’sdesiretobuildontheUK’sthrivingAIecosystemandwelcometheUKgovernment’scommitmenttointernationalcooperation

–somethingwe’dbekeentosupportwherepossible.

4.Wearehighlysupportiveoftherolethatthewhitepaperdescribesforeffectiveoversighttools,andthatofauditandassurance,forsupportingthelong-termsustainabledevelopmentoftheAIecosystem.Webelievethattheaccountancyprofessionhasakeyroletoplayindrivingtrustin,andtheethicaldeploymentof,AI.

5.Wewelcomethewhitepaper’sspecificemphasisonconsideringtheroleofAIacrossallregionsandnationsoftheUK.Nearlytwo-thirdsoftheUK’sAIindustryecosystemisbasedinLondon(BusinessWire2021).WhilethisconcentrationishelpingtomakeLondontheAIcapitalofEuropeandmustbesupported,ACCAandEYoperateacrosstheregionsandnationsoftheUK,andweseethepotentialforAItodriveeconomicopportunityinsupportoflevelling-upacrosstheUK.

6.Effectivelyengagingwithexistingcodesandstandardsspecifictoindividualsectorsanddomains

–includingtheirevolutionandfuturedirection–isfundamentaltotheapproachproposedbythewhite

paper.Accordingly,thepaperexploresafewoftheseexistingcodesandstandardsinthecontextofAI-relevantconsiderations.

7.Thereare,however,otheraspectsworthnotingtoensurethattheambitionsofthewhitepaperarerealisedasintended.Somekeyconsiderationsfromourperspectiveareasfollows:

a.Thegovernmentwillnotputitscross-cuttingprinciplesonastatutoryfootinginitiallybuthasleftthedooropenforplacingastatutorydutyonregulatorslater.SotherecouldbeaconsiderabledelayindecidingbetweenAI-specificmandatoryrequirementslinkedtoimplementingthewhitepaperprinciples,ononehand,andrelyingonself-regulationorsuggestedguidanceontheother.Organisationsvalueregulatorycertainty.Along,uncertainwaitingperiodrisksdelaystoupskillingandpreparationforfuturecompliance,particularlyamongsmall-andmedium-sizedentities(SMEs),whichwillbelesslikelytocommitcostoreffortwithoutcertainty.

b.Issuesofaccountability,ethicsandtrustrequirecarefulconsideration,particularlyintheabsenceofadesignatedentityresponsibleforAIoversight.Thecurrentapproachmaywellbeworkable,butstakeholderswouldbenefitfromadditionalsupporttohelpthemnavigatethedifferentbodiesrelevanttotheirAIcomplianceobligations.SomeareaspertinenttoAI(suchasdataprotection)haveextensivepre-existingregulationandguidance,whileothers(suchasexplicabilityofAIsystems)arelessmature.Also,considerationisneededforfoundationalmodelssuchaslargelanguagemodels(LLMs)2thatdonotneatlyfallintotheremitofanyoneexistingregulator.

c.Theconceptofcentralsupportfunctionscouldbeeffectiveandinnovative,butthewhitepaperwouldbenefitfromfurtherdetailonhowthesewillbeimplementedinpractice:howtheywillbe

resourced,themechanismthroughwhichtheywillgatherinputsfrommarketparticipants,andhowtheywillbecoordinatedacrosssectorregulators.

1Calltoaction#10inACCA’sreportonRiskCulture<

/gb/en/professional-insights/risk/risk-culture.html

>,notesthat“Furthercooperationbetweenourprofessionsandregulatorswillbebeneficialforall”,whichisbasedonthe‘Frameworkformeaningfulengagement’<

/

publications/framework-meaningful-engagement-human-rights-impact-assessments-ai

>.

2Amachinelearningmodelthatusesnaturallanguageprocessing(NLP).

BUILDINGTHEFOUNDATIONSFORTRUSTEDARTIFICIALINTELLIGENCE|EXECUTIVESUMMARY

d.Itwouldbegoodtounderstandifthereisanythinkingwithingovernmentontheneedformoreprescriptiverequirementsthatbuildonthewhitepaperprinciples.Someregimesare,ofcourse,takingamoreprescriptiveapproach.Thisdoesn’tnecessarilymakethatthepreferredchoice,buttherisktoconsideristhatUKorganisationsoperatingacrossboundarieswilldefaulttothemostprescriptiveregimeacrosstheirregionsofactivity–foroperationaleaseandtoreduceriskofnon-compliance.TheconsiderationhereisthatthismightinadvertentlymakeUKorganisationsrule-takersofotherregimes.Thereisabalancetoconsiderinrelationtomaintainingapro-innovationapproach–forexample,thismaymeanmoreprescriptiverequirementsfortheriskiestusecasesratherthanacrosstheboard.

e.Asthewhitepapertakesacontext-specificapproachtoregulation,existingsectorregulatorswillneedtoprovidetheirownguidanceontheuseofAI.Whileitisgoodtolearnfromthosewithgreaterexpertise,itwouldbehelpfultohaveanexplanationofhowtheproposedapproachwouldsupportless-maturesectorsastheyupskillonAI.

f.Thereisahugedifferencebetweenthecapabilities,resourcesandrequirementsofthelargetechnologycompaniesandthoseofSMEs.Thewhitepaper’sapproachofcarryingonwithinexistingregulatoryambits,needscaretoavoid

–despitebestintentions–entrenchingpowerimbalances.Thoseholdingcustomerchannels,accesstotrainingdataandresearchbudgets

maytendincreasinglyandirreversiblytopull

awayfromthepack.Considerationisneededto

avoidcreatingmonopoliesandstiflingthevery

innovationthatissought.Itisalsoimportanttobe

sensitivetoblindspotsininnovation,becausetruly

disruptive–asopposedtoincremental–change

oftencomesfromSMEsandAIinnovationisripe

forexactlythistypeoftransformativechange.

g.Leadersacrossbusiness,fromboardleveldownwards,areattheearlystagesofclarifyingwhereaccountability,contestabilityandredresswouldresideinrelationtoAIsystems.Itwouldbehelpful,possiblyviathecentralsupportfunctions,toprovideaforumforsharingexamplesofhowthiscanbedone.Suchaforumcouldhelpdefinetheroleandlegalliabilityofnotjustofboardmembers,andC-suite,butalsolendersandinvestors.

h.Itwouldbehelpfulforfurtherpolicyworktosetoutthegovernment’sviewoftheenvironmentalimpactofAI,forexample,itsmeasurement,monitoring,ormitigationofharm.ThisisparticularlyneededasfoundationalAIandgenerativemodels

driveever-highercomplexity,dataandenergy

consumption.Therearealsosocialconsiderations,

particularlyinrelationtotheoften-mentionedarea

ofjobsdisplacement,whichareimportantforthe

responsibledeploymentofAI.

8.Overall,wewelcomethiswhitepaperaspartof

adialoguethroughwhichtheUKAIregulatory

regimecanbecontinuallyrefinedandpositionedas

progressive,innovativeandresponsible.

ATACCAANDEY,WEARESUPPORTIVEOFREGULATINGOUTCOMESRATHERTHANTHETECHNOLOGYITSELF.THISALIGNSWITHTHEPRINCIPLES-BASEDAPPROACHTOREGULATIONFORTHEACCOUNTANCYPROFESSIONINTHEUK.

7

BUILDINGTHEFOUNDATIONSFORTRUSTEDARTIFICIALINTELLIGENCE|RECOMMENDATIONSFORPOLICYMAKERS

Recommendationsforpolicymakers

9.Actfastonthedetail.AIisrapidlydeveloping;itisthesubjectofintensemediainterestandisprominentinthepublicconsciousness.ConversationsonAIregulationareconstantlyevolving,withregimesaroundtheworldworkingfasttostaketheirplaceinthedebate.Whetheroneisregulatingtoomuchortoolittle–prioritisingconsumerprotectionandunrestrictedinnovation–isanimportantissuebutdelayingisjustasdamagingasgettingthiswrong.Businessdislikesregulatoryuncertaintyanditiscriticalthatthedetailsarefleshedoutassoonaspossibletooperationalisethewhitepaper’sapproach.

10.Aligninternationally,coordinatedomestically.ACCA’sandEY’sstakeholdersoperateinternationallyand,whiletheyacknowledgegeo-politicalfactors,theywouldwelcomeasmuchharmonisationaspossibleacrossregimestosupportinteroperability.Thisreducescosts,complexityandrisk.Atthesametime,itisvitaltoavoidthefatalflawofallowing

majorissuestofallintothegapsbetweentheaccountabilitiesofdifferentsectorregulators,ortheirduplicationofthesamepreparatorywork.

11.Seekmulti-stakeholderfeedbackandinvolvement.

AIusescitizens’data,directlyorindirectly,andits

outputsaffectthesecitizens’lives.Itisimperative

thatitsdevelopmentaltrajectoryisnotinformedby

anarrowinnercoterieofinfluencers.Thoseacross

theregionsoftheUK,acrossthefullsocio-economic

spectrum,acrosssectors/industriesandinSMEsas

wellaslargerorganisations,needtobeinvolved.

12.ChampionanecosystemfortrustworthyAI.Itisnecessarytocreate/reinforcepathwaysforsharingknowledge,toolsandexperience.Theaccountancyprofessionhasskillsforassessingrisksandcontrols,frequentlyfacessituationsrequiringethicaljudgementandseekstoupholdthepublicinterest.AnenvironmentthatleveragesandproactivelyinvolvesthematureUKaccountancysector(egeducationandawarenessaspectsofthecentralsupportfunctions,partneringtosharekeymessagesinternationally)wouldbolsteritsglobalreputationasatrustedhomeforresponsibleAI.

8

9

BUILDINGTHEFOUNDATIONSFORTRUSTEDARTIFICIALINTELLIGENCE|INTRODUCTION

Introduction

13.Overthecomingyears,AIwillcompletelytransformawiderangeofsectors,includingagriculture,communication,education,healthcare,manufacturingandtransportation.Asaresult,itissettoboosttheglobaleconomy,witheconomistspredictingthatAIcouldeventuallyboostannualgrossdomesticproduct(GDP)by7%(BriggsandKodnani2023).

14.AItechnologiescouldalsobringenormoussocialandenvironmentalbenefits,suchashelpingustocombatchildabuseandcybercrimeandprovidingcriticalinsightsthatcanbeusedtomitigateclimatechange.YetwhileAIcanpotentiallybeatremendousforceforgood,italsopresentssomesignificantthreatstosociety.

15.Ifusedinappropriately,AIcouldexacerbateexistinginequalities,damagepublictrustthroughthespreadofmisinformation,andbeharmfultothephysicalandmentalhealthofindividuals.AItechnologiesthreatenjobsandlivelihoodsandareatriskofbeingexploitedbycyberattackers.AndwhileAIcanbeapowerfultoolinthefightagainstclimatechange,itsuseisitselfaconsiderablesourceofcarbonemissions.

16.Despiteitsfar-reachingimplicationsforsociety,AIispoorlyunderstoodbymostanditsuseremainslargelyunregulated.SuchisthelevelofconcernaboutthepotentiallydetrimentalimpactofAIonsocietythatover1,000technologyleadersandresearchershavecalledforapauseonAIdevelopment,citing‘profoundriskstosocietyandhumanity’(OpenDataScience2023).

17.AwareofboththeopportunitiesandchallengesthatAIpresents,numerousjurisdictionsaroundtheworldareexploringthepossibilityofregulatingAI.OneofthesejurisdictionsistheUK,whichispositioningitselftobecomeagloballeaderintheregulationandgovernanceofAItechnologies.

18.TheUK’sambitionsintheregulatoryspacearewell-founded,giventhatitisalreadytheAIcapitalofEuropeandarecognisedleaderinthedevelopmentandapplicationofAItechnologies.ThecountryranksthirdontheGlobalAIIndexproducedbyTortoiseMedia,behindtheUSandChina,measuredonitslevelofinvestment,innovation,andimplementationofAI(TortoiseMedia2021).Itisalsohometoaroundone-thirdofEurope’sAIstart-ups(Kelnar2019).

19.InMarch2023,theUKgovernment’sDepartmentforScience,InnovationandTechnology(DSIT)publishedawhitepaperentitledAPro-innovationApproachtoAIRegulation.Thewhitepapersetsoutthegovernment’sstrategyforsupportinginnovation,whileprovidingaframeworktoensurethattherisksassociatedwithAIareidentified,addressedandmitigated.

20.Thispaper,fromACCAandEY,isaresponsetothatwhitepaper.Ouraimistoprovidepolicymakerswithaconstructivecritiqueofthewhitepaper,informedbyourpositionswithintheaccountancyandbusinessecosystem,andtheaccesstofeedbackandinputswehaveacrossourstakeholderbases.

WhatisAI?

22.AIcanbebroadlycategorisedintonarrowAI(whereamodelistrainedtoperformasingletask,suchasrecognisingfacesortranslatingtextintoanotherlanguage)andgeneralAI(whereamodelistrainedtolearn,behaveandperformactionsinthesamewaythathumanswoulddo).Inbothcases,themodelsaretrainedusingdata.

21.Thereisnosingle,commonlyagreeddefinition

ofAI.Nevertheless,AIisoftendescribedasthe

simulationofhumanintelligencebycomputer

systems,wherebyAIalgorithms,iesetsof

instructions,areprogrammedtomakedecisions

andsolveproblemsusinglargevolumesof

structured(egfiguresinrowsandcolumns)and

unstructured(egtextual,vocal,image)data.

BUILDINGTHEFOUNDATIONSFORTRUSTEDARTIFICIALINTELLIGENCE|INTRODUCTION

Foundationmodels

23.So-called‘foundationmodels’areanemergingtypeofgeneral-purposeAI.Thesepowerfulmodelsaretrainedonbroaddatasets,oftentakendirectlyfromtheinternetratherthanusingcuratedtrainingdatasets,andwithminimalhumansupervision.Theyusemachinelearningtoapplywhattheyhavelearnedaboutonesituationtoanother.

24.LargeLanguageModels(LLMs)areaspecifictypeoffoundationmodelthataretrainedonlargeamountsoftextdata.LLMsarepoweringgenerativeAImodels,whichincludeChatGPTandGoogle’sBard.

25.The‘generativeAI’thatisenabledbyfoundationmodelscanproducenoveloutputsderivedfromthesimpleinputsthatitreceives.Aswellasgeneratingvastswathesoftextanddata,generativeAIcancreatecompleximagery,videosandmusic.Itcan

generatethiscontentfromashortprompt,ratherthanhavingtobetrainedspecificallytoperformthetaskinquestion.Thisisdifferentfrommachinelearningtechniques,suchasclassification,whichsimplymapnewinformationontoexistingcategoriesdependingonhowtheyweretrained.

26.FoundationmodelsareatthecentreofmuchofthecurrentpublicdebateonAIbecausetheyaredevelopingatarapidrate,outpacingtheabilityofbothgovernmentsandregulatorstokeepupwiththem.Astheycanbeusedformanydifferentpurposes,foundationmodelscouldhaveapotentiallytransformativeeffectonbusinessecosystems.Butthecomplexitiesassociatedwithfoundationmodelsmakesthemverychallengingtoregulate.

BenefitsofAI

27.AItechnologies–andfoundationmodels,inparticular–havethepotentialtobehugelybeneficialtoboththeeconomyandsociety.AstheUKprimeministerhassaid:‘HarnessingthepotentialofAIprovidesenormousopportunitiestogrowoureconomy,createbetter-paidjobs,andbuildabetterfuturethroughadvancesinhealthcareandsecurity’(UKGovernment2023).

28.Fromaneconomicperspective,AItechnologiescanboostefficiencyandproductivity(apriorityfortheUK),reduceoperationalcosts,andenableconsumerstoenjoyawiderc