linepornmovies

EDITORLETTERArtificialIntelligence,DeepfakesandaFutureofEctypesLucianoFloridi1,2Publishedonline:1August2018#SpringerNatureB.
V.
2018Theartworldisfullofreproductions.
Someareplainreplicas,forexampletheMonaLisa.
Othersarefakesorforgeries,liketheBVermeers^paintedbyHanvanMeegerenthatsoldfor$60million(KreugerandvanMeegeren2010).
Thedistinctionbetweenareplicaandafakeisbasedontheconceptofauthenticity.
Isthisartefactwhatitclaimstobe1Theanswerseemssimplebut,inreality,thingsarecomplicated.
Today,thepaintingsoftheforgerJohnMyattaresofamousthattheyarevaluedatupto$40,000each,asBgenuinefakes^(Furlong1986).
Theyarenotwhattheysaytheyare,buttheyareauthenticallypaintedbyhimandnotbyanotherforger.
Andtheyarebeautiful.
Abitasifoneweretoutterabeautifullie,notanyordinarylie.
AndanartistlikeMagritteseemstohavepaintednotonlyfalsePicassosandRenoirsduringtheNazioccupationofBelgium(Marin1983),butalsofakedhisownwork,sotospeak,inthefamouscaseofthetwocopiesofthepaintingBTheFlavourofTears^(1948),bothbyMagritte,butoneofwhichhepassedoffasfalse—partlyasasurrealistactandpartlytomakemoney.
Inthismess,andasifthingswerenotconfusingenough,digitaltechnologiesfurtherreshufflewhatispossibleandourunderstandingofit.
Thankstodigitaltechnologies,todayitismucheasiertoestablishtheauthenticityofawork.
Therearedatabaseswhereyoucancheckauthors'signatures,andmillionsofimagesthatcanbeviewedwithafewclicks.
Sellingafakeismoredifficult.
Figure1showsareproductionoftheBLodgeonLakeComo^byCarlFrederikPederAagaard(1833–1895),aDanishlandscapepainteranddecorativeartist.
Itwasonsalein2016oneBay.
Thepaintingisverypopularontheweb,andthereareplentyofgoodreplicas.
Nothingwrongwiththem.
However,ifyoucheckFig.
1carefully,youwillnoticethatthisissoldasanunsignedBoriginal^,whichismisleadingtosaytheleast.
Boththequalityofthepaintingandthepricearesuspicious,andaGoogleimagesearchquicklyPhilos.
Technol.
(2018)31:317–321https://doi.
org/10.
1007/s13347-018-0325-31Ihavediscussedthenatureofquestionsandepistemicrelevancein(Floridi2008).
*LucianoFloridiluciano.
floridi@oii.
ox.
ac.
uk1OxfordInternetInstitute,UniversityofOxford,1StGiles,OxfordOX13JS,UK2TheAlanTuringInstitute,96EustonRoad,LondonNW12DB,UKrevealsthatthisisamerereplica.
Atthetimeofwriting,thepaintingwasnolongeravailableandthesellerdidnotseemtobeactiveoneBayanymore.
Ofcourse,fakesarenotalwaysreproductions;theycanalsobeBnewworks^byafamousartist,likePollockorVanGogh.
Inthiscase,sophisticatedscientifictechniquestoestablishauthenticityincludetestsrunusingAI.
Aresearchpaper,publishedlastNovemberbyAhmedElgammal,YanKangandMilkoDenLeeuw(Elgammaletal.
2017)proposedBacomputationalapproachforanalysisofstrokesinlinedrawingsbyartists^,basedonneuralnetworks.
Thetrainingcollectionconsistedofadatasetof300digitiseddrawingswithover80,000strokes,byPabloPicasso,HenryMatisseandEgonSchiele,andafewworksbyotherartists.
Bysegmentingindividualstrokes,thesystemlearnedtoquantifythecharacteristicsofindividualstrokesindrawings,thusidentifyingtheuniquepropertiesforeachartist.
ThesoftwaremanagedtoclassifyBindividualstrokeswithaccuracy70%-90%,andaggregateoverdrawingswithaccu-racyabove80%,whilebeingrobusttobedeceivedbyfakes(withaccuracy100%fordetectingfakesinmostsettings)^.
Itturnsoutthatthewayinwhichindividualsdrawlinesisasuniqueastheirfingerprintsortheirgait,andAIcanhelponetodiscoverit,asifitwereamicroscope.
ButAIisnotjustforidentifyingfakes.
LetusstayintheNetherlands,averyinterestingproject2byMicrosoft,incollaborationwiththeRembrandtHouseMuseum,hasledtothecreationofaportraitofagentleman,whichbothisandisnotaRembrandt(seeFig.
2).
Figure1Afake,theoriginalisBLodgeonLakeComo^byCarlFrederikPederAagaard(1833–1895)2Seehttps://news.
microsoft.
com/europe/features/next-rembrandt/318L.
FloridiAnalysingtheknownworksofRembrandt,analgorithmidentifiedthemostcom-monsubject(aportraitofaCaucasianman,30–40yearsold),themostcommontraits(facialhair,facingtotheright,wearingahat,acollaranddarkclothing,etc.
),themostsuitablestyletoreproducethesecharacterisingproperties,thebrushstrokes,inshort,alltheinformationneededtoproduceanewpaintingbyRembrandt.
Havingcreatedit,itwasreproducedusinga3Dprinter,toensurethatthedepthandlayeringofthecolourwouldbeascloseaspossibletoRembrandt'sstyleandwayofpainting.
Theresultisamasterpiece.
ARembrandtthatRembrandtneverpainted,butwhichchallengesourconceptsofBauthenticity^andBoriginality^,giventhepainting'sstronglinkwithRembrandthimself.
Idonotknowthevalueofthepainting.
MybetisthatitwouldbequiteexpensiveifitwereauctionedasreliablyauthenticatedasthatuniqueMicrosoft'sRembrandt.
WedonothaveawordtodefineanartefactsuchasMicrosoft'sRembrandt.
Soletmesuggestectype.
ThewordcomesfromGreekandithasasubtlemeaningthatisquiteusefulhere:anectypeisacopy,yetnotanycopy,butratheracopythathasaspecialrelationwithitssource(theoriginofitscreation),thearchetype.
Inparticular,anectypeistheimpressionleftbyaseal.
Itisnottherealthing,butitisclearlylinkedinasignificant,authenticwaywiththerealthingitself.
LockeusedBectypes^torefertoideasorimpressionsthatcorrespond,althoughsomewhatinadequately,tosomeexter-nalrealities(thearchetypes)towhichtheyrefer(Locke2008).
Digitaltechnologiesareabletoseparatethearchetypalsource—whatwasinthemindoftheartist,forexample—fromtheprocess(style,method,procedure)thatleadsfromthesourcetotheartefact(Floridi2017).
Oncethislinkissevered,onecanhaveectypesthatareBauthentic^instyleandcontent,butnotBoriginal^,intermsofarchetypalsource,likeMicrosoft'sRembrandt.
ButonecanalsohaveectypesthatareBoriginal^intermsofarchetypalsource(theydocomefromwheretheypurporttocome)yetnotBauthentic^intermsofproduction,performance,ormethod(theyarenottheonesusedbythesourcetodelivertheartefact).
Inotherwords,ectypescanbeauthenticbutunoriginalartefacts,likeMicrosoft'sRembrandt,orinauthenticbutoriginalartefacts.
AgreatexampleofaninauthenticoriginalectypewasprovidedinMarchbyanaudiorecordingofJohnF.
Kennedy'slastspeech.
Despitebeinganordinaryspeechfromadecades-oldcampaigntrail,itsuddenlymadeheadlinenews.
BecauseitwastheDallasTradeMartFig.
2TheRembrandtthatisnotaRembrandt.
MicrosoftProjectwiththeRembrandtHouseMuseumArtificialIntelligence,DeepfakesandaFutureofEctypes319speechof22November1963,thetextthatJFKwouldhaveread,hadhenotbeenassassinatedmeremomentsbefore,onhiswaytodeliverit.
Thetextisoriginal:itcomesfromthesource.
Butthevoicethatrecitesisinauthentic,becauseitwassynthesisedbysoftwarethatanalysed831recordingsofKennedy'sspeechesandinterviews,inordertoBlearn^howtospeaklikehim.
ThesoftwarefinallygavevoicetoJFK'slastspeech55yearslate.
SohereisaKennedywhoisandisnotaKennedy,similarandyetdifferentfromtheRembrandtthatisandisnotaRembrandt.
Theyarebothectypes(seeTable1).
Wesawthattheproductionofectypesdoesnotstopattheworkofart,butinvolvesanyartefact,fromtextstophotos,fromaudiorecordingstovideos.
Itiswellknownthatthehistoryofmanuscripts,printing,photography,cinemaandtelevisionispavedwithfakes.
Expectmoreectypestoo.
Inparticular,artistslovetobreakboundariesanditiseasytoimaginethat,likeMagrittefakinghisownpainting,theywillstartproducingtheirownectypes.
ImagineapainterusingthesoftwaredevelopedbyMicrosofttoproduceherownnewworks.
Itwouldstillbeanectype,andthiswouldexplainwhy(withqualifications)theprocesswouldcapturesomeauthenticity.
Thereproductionoftheworkofartbymechanicalmeanswillhaveacquiredanewmeaning(Benjamin2008).
Withectypes,weusuallyknowwherethingsstand.
Butsomeonecouldcheat.
LastMay,GooglepresentedGoogleDuplex,aversionofitsAIassistantthatsimulatesbeinghumantohelpuserswithsimpleinteractivetasks,likebookingarestauranttable.
Thecompanywasquicktostatethatitwillnotintentionallymisleadanyone,andthatitwillmakesurealwaystoclarifywhenauserisinteractingwithanartificialagent.
Butsomeoneelsecouldusethesetechnologiesforcriminalorevilpurposes.
ThisiswhathappenswithDeepfake,asetoftechniquesusedtosynthesisenewvisualproducts,forexamplebyreplacingfacesintheoriginals.
ThetypicalcasesinvolvepornmoviesinwhichthefacesoffamousactresseslikeGalGadotorScarlettJohansson(thisisregularlyaboutwomen'sfaces)areusedtoreplacetheoriginalfaces.
Inthiscasetoo,largedatabasesareneededtoinstructthesoftware(whichisavailableforfree,andthereisalsoanapp),soifyouarenotapublicfiguretherisksarelower.
Deepfakealsoconcernspoliticians,likePresidentObama,forexample.
WhatisthefutureaheadofusDigitaltechnologiesseemtoundermineourconfidenceintheoriginal,genuine,authenticnatureofwhatweseeandhear.
Butwhatthedigitalbreaksitcanalsorepair,notunliketheendlessstrugglebetweensoftwarevirusandantivirus.
Inourcase,inadditiontoeducatingpeople,acquiringnewsensitivitiesandhavingtherightlegalframework,thereareatleastacoupleofinterestingdigitalstrategies.
Forartefactsthatarealreadyavailable,itiseasytoimagineTable1Archetype,fakeandectypesOriginalsourceAuthenticproductionLeonardo'sMonaLisaYesYesHanvanMeegeren'sforgedVermeersNoNoMicrosoft'sRembrandtNo(Qualified)YesJFK'sTradeMartspeechYesNo320L.
FloridiAIsystemsthatgiveusahand.
ItwouldbeinterestingtoanalyseMicrosoft'sRembrandtandKennedy'sspeechwithanartificialsystemtoseewhetheritdiscoveredthemtobeectypes.
ResearchisalreadyavailableonmethodstoexposeDeepfakevideosgeneratedwithneuralnetworks(Lietal.
2018).
Inshort,letusrememberthesoftwaredevelopedtoanalysedrawings:thereareplentyofsophisticatedtoolsfordetectionofimageforgery.
Andmorearelikelytobedevelopedasthedemandforthemincreases.
Next,asregardsnewartefacts,becauseoriginalityandauthenticityarealsoamatterofprovablehistoricalcontinuityfromthesourcetotheproductthroughtheprocessofproduction,themuch-vauntedblockchain,orasimilarsolution,couldmakeabigdifference.
Blockchainislikearegisterthatstorestransactionsinanaccruable,safe,transparentandtraceableway.
Asasecureanddistributedregisteroftransactions,blockchainisbeingexploredasameansofreliablycertifyingtheoriginsandhistoryofparticularproducts:whetherintermsofsecuringfoodsupplychains,orinrecordingthemanylinkedactsofcreationandownershipthatdefinetheprovenanceofanartwork.
Inthefuture,wemayadoptthesamesolutionwhereverthereisaneedtoensure(orestablish)theoriginalityandauthenticityofsomeartefact,beitawrittendocument,aphoto,avideoorapainting.
Andofcourse,afutureartistmaywanttoensure,throughablockchain,thatherworkofartasanectypeisreallywhatitsaysitis.
Atthatpointweshallhavetravelledfullcircle,forweshallhaveBgenuineectypes^,liketheMicrosoft'sRembrandt,orKennedy'sspeech.
ReferencesBenjamin,W.
(2008).
Theworkofartintheageofmechanicalreproduction.
London:Penguin.
Elgammal,A.
,Kang,Y.
,&DenLeeuw,M.
(2017).
Picasso,matisse,orafakeAutomatedanalysisofdrawingsatthestrokelevelforattributionandauthentication.
arXivpreprintarXiv:1711.
03536.
Floridi,L.
(2008).
Understandingepistemicrelevance.
Erkenntnis,69(1),69–92.
Floridi,L.
(2017).
Digital'scleavingpoweranditsconsequences.
Philosophy&Technology,30(2),123–129.
Furlong,M.
(1986).
Genuinefake:abiographyofAlanWatts.
Portsmouth:Heinemann.
Kreuger,F.
H.
,&vanMeegeren,H.
(2010).
HanvanMeegerenrevisited:hisart&listofworks.
Delft:F.
H.
Kreuger.
Li,Y,Chang,M.
-C.
,Farid,H.
,&Lyu,S.
(2018).
InIctuOculi:exposingAIgeneratedfakefacevideosbydetectingeyeblinking.
arXivpreprintarXiv:1806.
02877.
Locke,J.
(2008).
Anessayconcerninghumanunderstanding.
NewYorkOxford:OxfordUniversityPress.
Marin,M.
(1983).
Leradeaudelamémoire:souvenirsdéterminés.
Paris:Pré-aux-Clercs.
ArtificialIntelligence,DeepfakesandaFutureofEctypes321