- Ama-Deepfake okuqukethwe okulalelwayo nokubukwayo okukhiqizwe yi-AI okusongela ubumfihlo, idumela, kanye nokwethenjwa komphakathi.
- Umthelela wayo usukela ekuhlukunyezweni kanye nezithombe zobulili ezingcolile ezingavunyelwene kuya ekukhwabaniseni izinkampani kanye nokuphathwa kabi kwezepolitiki.
- Uhlaka lwezomthetho lwaseSpain naseYurophu seluvele lunikeza amathuluzi okushushisa ukusetshenziswa okuyingozi, yize ubuchwepheshe buthuthuka ngokushesha okukhulu.
- Ukuzivikela kudinga ukuhlanganisa ukutholwa kobuchwepheshe, izinqubo zokuhlela, kanye nokuqeqeshwa ekucabangeni okubucayi kanye nokuphepha kwe-inthanethi.
Ama-Deepfakes asuke ekubeni yilukuluku le-inthanethi aya kwenye yezinselelo ezinkulu kakhulu zesikhathi sedijithali. Amavidiyo, umsindo, nezithombe ezisetshenziswe kusetshenziswa ubuhlakani bokwenziwa kungasenza sibone othile ekhuluma noma enza into engakaze yenzeke, ngeqiniso elingenza ngisho namehlo ochwepheshe aphuce. Futhi lokhu akugcini nje ngokuthinta osaziwayo noma osopolitiki: noma ubani okhona ezinkundleni zokuxhumana angaba yisisulu.
Lesi simo sihlanganisa izithako ezintathu eziyingozi kakhulu: ubuchwepheshe obunamandla, ukusebenziseka kalula, kanye nokufinyelela okukhulu . Umphumela uba yindlela ephelele yokunikeza ulwazi olungamanga, ukukhwabanisa ngezezimali, ukuhlukunyezwa, kanye nokuwohloka okubanzi kokuthembana kulokho esikubonayo nesikuzwayo. Ake sichaze, ngokuthula nangokuningiliziwe, ukuthi ziyini ngempela izinganekwane ezijulile, ukuthi zidalwe kanjani, yiziphi izingozi ezizibeka kubantu ngabanye, ezinkampanini, nasezikhungweni, ukuthi umthetho uthini, nokuthi yiziphi izinyathelo zangempela esingazithatha ukuze sizivikele.
Ayini ama-deepfake futhi kungani ebalulekile?
Igama elithi deepfake livela ekuhlanganisweni kwe- deep learning kanye ne-fake . Libhekisela kunoma yikuphi okuqukethwe okulalelwayo nokubonakalayo—ividiyo, isithombe, noma umsindo—okukhiqizwe noma okuguquliwe yi -algorithms yobuhlakani bokwenziwa ukuze kubonakale kuyiqiniso. Asikhulumi ngezihlungi noma ukuphinda kuthinte, kodwa ukwakhiwa kabusha okuphelele kobuso, ukushukuma komzimba, noma amazwi alingisa ngokunembile umuntu wangempela.
Ngamathuluzi atholakala kalula, ngisho namahhala, kungenzeka ukufaka ubuso bomuntu emzimbeni womunye umuntu, ukopishe izwi lakhe, noma udale kabusha izigcawu ezingakaze zenzeke . Intuthuko kumanethiwekhi ezinzwa kanye nama-GAN (amanethiwekhi aphikisanayo akhiqizayo) ikhulise kakhulu izinga lokuba ngokoqobo, okwenza kube nzima kakhulu ukuhlukanisa phakathi kokuqukethwe kwangempela nokuguquliwe.
Ama-Deepfake manje abhekwa njengokuguquka kwezindaba ezingamanga : ngenkathi umbhalo noma izithombe ngaphambili zazishintshwa, manje kukhiqizwa ividiyo engamanga noma ukuqoshwa komsindo okuthembekile okunomthelela omkhulu kakhulu ngokomzwelo nangokwengqondo. Ubuchopho bethu buvame ukuthemba lokho "obukubonayo nabakuzwayo" ngaphezu kwesihloko esilula, esikhulisa amandla abo okukholisa.
Nakuba lobu buchwepheshe bukhona kusukela ngasekupheleni kweminyaka yama-90, bavelela ngo-2017 lapho umsebenzisi we-Reddit eqala ukuthumela okuqukethwe kocansi okungekho emthethweni okubonisa ubuso babadlali abadumile. Kusukela ngaleso sikhathi, ukusetshenziswa kwabo kusakazeke kakhulu, kokubili kwezokuzijabulisa nasezimweni ezisobala ukuthi zinonya.
Indlela ama-deepfake adalwe ngayo: uhlangothi lobuchwepheshe olungenalo ulimi lobuchwepheshe
Ngemuva kwe-deepfake kunezinhlelo zokufunda komshini ezihlaziya izinkulungwane zezithombe, ozimele, noma izingcezu zomsindo . Inqubo evamile enobuso ilandela izinyathelo ezimbili eziyinhloko: ukufaka ikhodi nokukhipha ikhodi. Okokuqala, i-algorithm yokufaka ikhodi ifunda ukufana phakathi kobuso obubili bese ikhipha isethi yezici ezivamile. Ngemuva kwalokho, i-decoder yakha kabusha isithombe esiqondiwe isebenzisa izinkulumo zomunye umuntu.
Empeleni, kuqeqeshwa ama-decoder amabili ahlukene, elilodwa lobuso ngabunye. Uma idatha evela komunye umuntu ifakwa ku-decoder yomunye , sithola ividiyo lapho isihloko A sibonakala senza khona ukushukuma kanye nokunyakaza kwesihloko B. Uma kwenziwe kahle, umphumela ubukeka njengokuqoshwa kwasekuqaleni.
Enye indlela evamile ukusetshenziswa kwamanethiwekhi adala izingxabano (ama-GAN) . Lokhu kuhilela amanethiwekhi amabili: ijeneretha, edala izithombe noma amavidiyo amanga, kanye ne-discriminator, ezama ukuhlukanisa amanga kwangempela. Ancintisana izikhathi eziyizinkulungwane, ethuthuka njalo kuze kube yilapho i-discriminator ingasakwazi ukubona kalula inkohliso futhi umphumela usondela eduze kweqiniso.
Iningi lama-deepfake evidiyo adalwe ngokuphakela i-algorithm amasethi amakhulu ezithombe zomuntu olingiswayo . Uma imodeli inama-engeli amaningi, ukushukuma, kanye nezimo zokukhanya, umphumela wokugcina uzoba ngokholisayo. Ngakho-ke, ngokwesiko, izisulu zazivame ukuba ngabantu ababona kakhulu imidiya, yize namuhla isithombe esisodwa noma imizuzwana embalwa yokuqoshwa kwezwi kwanele ukukhiqiza i-clone efanelekile.
Uma kukhulunywa ngezwi, izinhlelo zokubhala ama-cloning zivumela, kusukela emizuzwaneni embalwa nje yokuqopha, ukukhiqizwa kabusha kwe-timbre, i-accent, kanye nesigqi senkulumo . Ngokuhlanganisa ukuhlanganiswa kwenkulumo nezikripthi ezenziwe nge-AI, kungenzeka ukukhiqiza umsindo lapho isisulu "sisho" noma yini umhlaseli afuna ukuyibeka emlonyeni wakhe.
Izinhlobo zama-deepfake: isithombe, ividiyo, kanye nezwi
Ngaphakathi kwesambulela sama-deepfakes, singahlukanisa izinhlobo eziningana, kuye ngohlobo lokuqukethwe okushintshiwe kanye nendlela esetshenzisiwe. Izigaba ezimbili sezithandwa kakhulu: ama-deepfaces nama-deepvoices.
Ama-Deepfaces amavidiyo lapho ubuso bomuntu bukhiqizwa khona noma bushintshwa ngokuphelele . Ngokusebenzisa izithombe eziningi, uhlelo ludala ubuso obungaguquki kangangokuthi bubonakala sengathi yizithombe zangempela, bese luzihlanganisa ngokulandelana ukuze kudalwe ividiyo ebushelelezi. Umgomo uwukuthi, lapho ubuka, akekho ongasola ukuthi ubheka inhlanganisela yokwenziwa.
Okwamanje, ama-deepvoice agxila ekuzibonakaliseni ngezwi . I-algorithm ifunda izici zomsindo womuntu (imvamisa, ithoni, ukuphumula, amagama okugcwalisa, njll.) bese ingafunda noma yimuphi umbhalo, iwudlulise njengalowo muntu. Lolu hlobo lwe-deepfake luyingozi kakhulu emikhakheni yezinkampani kanye neyezezimali.
Isibonelo esiyinhloko senzeka ngo-2019: izigebengu ze-inthanethi zalingisa izwi le-CEO ukuze zibize umphathi omkhulu enkampanini yakhe futhi zi-ode ukudluliselwa okuphuthumayo okungaphezu kwama-$250.000. Umphathi wavuma ngoba izwi lalizwakala lifana ncamashi nelomphathi wakhe, futhi umuzwa wokuphuthuma wawungakholeleki.
Ngaphandle kwalezi zigaba ezimbili, kukhona futhi ama-deepfake ahlanganisiwe ahlanganisa ividiyo ekhiqizwe yi-AI, umsindo, kanye nombhalo , ukuze konke okuqukethwe—lokho okubonayo nalokho okuzwayo—kube yinto yokwenziwa kodwa kuhambisane. Njengoba ubuchwepheshe bukhula, imigqa phakathi kwalezi zinhlobo iyafiphala.
Ukusetshenziswa okusemthethweni kanye nezinhlelo zokusebenza ezinhle
Nakuba sivame ukuhlobanisa ama-deepfakes nokukhwabanisa kanye nokuxhashazwa kwe-inthanethi, ubuchwepheshe obufanayo bunezinhlelo zokusebenza ezisemthethweni neziwusizo ngokuphelele uma busetshenziswa ngokusobala nangemvume. Embonini yamafilimu, isibonelo, busetshenziselwa ukwehlisa ubudala babalingisi, ukuphinda abalingiswa asebeshonile, noma ukukopisha ingxoxo ngaphandle kokulahlekelwa ukuvumelana kwezindebe.
Emkhakheni wezemfundo, amamodeli akhiqizayo avumela okuqukethwe okulalelwayo nokubukwayo ukuthi kulungiswe ngezilimi eziningi ngenkathi kulondolozwa ukushukuma kokuqala kwesikhulumi. Lokhu kuthuthukisa ukuqonda, kugwema umuzwa "wokuhunyushwa kwesabelomali esiphansi," futhi kwenza kube lula ukufinyeleleka emhlabeni wonke ezifundweni nasezingqungqutheleni.
Izinhlelo zokusebenza zokufinyeleleka nazo ziyahlolwa, njengokudala ama-avatar ahumusha ulimi lwezandla ngendlela eyenzelwe wena, noma amazwi okwenziwa ngokwezifiso abantu abangakwazi ukukhuluma. Ngendlela efanele yokuziphatha, ubuchwepheshe obufanayo obusikhathazayo namuhla bungavula iminyango ebalulekile yokufakwa kwedijithali.
Inselele enkulu kakhulu akuyona ithuluzi ngokwalo, kodwa kunalokho umongo, inhloso, kanye nokucaca okusetshenziswa ngakho. Yingakho imithetho yamuva yaseYurophu igcizelela isibopho sokumaka ngokucacile okuqukethwe okukhiqizwe yi-AI ukuze umphakathi ukwazi ukuhlukanisa phakathi kwezinhlobo ezahlukene.
Izingozi nezinsongo ze-deepfake emphakathini
Umthelela omubi wama-deepfakes usukela empilweni yomuntu siqu kuya ekuzinzeni kwezepolitiki, kwezomnotho, kanye nokwezenhlalo . Enye yezingozi ezisobala kakhulu ulwazi olungesilo iqiniso: amavidiyo mbumbulu abonisa abaholi bezepolitiki benza izitatimende ezivusa ulaka noma bevuma ubugebengu obungakaze bube khona.
Ngisho noma kamuva ziphikiswa, umonakalo usuvele wenziwe, ngoba umthelela ongokomzwelo wokubukwa kokuqala uvame ukwedlula noma yikuphi ukulungiswa okulandelayo . Lokhu kunegalelo ekuhlukaniseni imibono yomphakathi, kuvuselele imibono yozungu, futhi kwenze buthaka ukwethembana kwabezindaba kanye nezikhungo zentando yeningi.
Enye ingozi enkulu kakhulu ukusetshenziswa kwezithombe ezijulile zokuhlukumeza, ukuhlambalaza, kanye nokwephulwa kobumfihlo . Ukudalwa kwezithombe zobulili ezingcolile mbumbulu kusetshenziswa ubuso babesifazane—ikakhulukazi abezombusazwe, izishoshovu, kanye nezintatheli—sekuyindlela yobudlova bedijithali enemiphumela emibi edumela labo, izimpilo zabo, kanye nempilo yengqondo.
Ucwaningo lwamuva nje lubonisa ukuthi cishe abantu abangu-2,2% ababuzwe emibhalweni ethile yezemfundo baziveza njengezisulu zezithombe ezijulile, kuyilapho abangu-1,8% bevuma ukuthi badale noma basakaza lolu hlobo lokuqukethwe. Nakuba lezi zibalo zingase zibonakale ziphansi, zibonisa umkhuba okhathazayo nokhulayo.
Ngaphezu kwalokho, ama-deepfake anegalelo "enkingeni yeqiniso" : uma kukhona ubufakazi obuzwakalayo obungaqanjwa, umphakathi usengozini yokucabanga ukuthi akukho lutho oluthembekile ngokuphelele. Mhlawumbe lokhu kungumthelela oyingozi kakhulu wesikhathi eside, ngoba kuqeda ukwethembana okuyisisekelo okudingekayo ukuze kube khona ukuhlala ndawonye kwedijithali kanye nengxoxo yomphakathi enolwazi.
Umthelela ekukhwabaniseni nasekuphepheni kwe-inthanethi kwezinkampani
Ezweni lebhizinisi, ama-deepfakes aba ithuluzi lokuhlasela eliphumelela kakhulu kubagebengu be-inthanethi. Lokhu akuyona inganekwane yesayensi: sekuvele kunezimo eziningi zabaphathi abalingiswa besebenzisa ividiyo noma umsindo ukuze ba-ode ukudluliselwa, bacele idatha ebucayi, noma banikeze imiyalelo yamasu engamanga.
Esinye sezimo ezivame kakhulu sihilela [ okungacacile - mhlawumbe "ikheli le-imeyili" noma "inombolo yocingo"). Ngaphezu kwe-imeyili yokukhwabanisa yakudala, umhlaseli ufaka ividiyo ejulile noma ukuqoshwa komsindo kwe-CEO noma i-CFO, okuqinisa ukuphuthuma kwesicelo. Lapho ebona noma ezwa "umphathi" wakhe, isisulu siyehlisa ukuqapha futhi silandele umyalo.
Ama-Deepfake asetshenziswa futhi kwezobunjiniyela bezenhlalo obuthuthukisiwe . Isibonelo, ividiyo ebonakala ijwayelekile ithunyelwa enkampanini yabameli lapho umlingani efuna khona ukusayinwa okusheshayo kwezinkontileka noma ukufinyelela imibhalo ethile. Umongo unengqondo, isithombe nezwi kuyahambisana, futhi uhlelo lokuqinisekisa luthembele kuphela ekubonweni kobuso.
Ezindaweni zokusebenza ezikude ezivame ukwanda, abahlaseli bangena ngezingcingo zevidiyo ngokuzenza amalungu eqembu . Besebenzisa ama-avatar akhiqizwe yi-AI, bahlanganyela emihlanganweni, bacele ukufinyelela ezinhlelweni zangaphakathi, noma bacindezelwe ukuthola ulwazi oluyimfihlo. Ukuzenza umuntu ongeyena kungabonakala uma ukuziphatha kungabonakali kungavamile kakhulu.
Konke lokhu kubeka inselele eqondile ezinhlelweni zokuphepha ngokusekelwe ekubonweni kobuso noma ngezwi. Izifundo zakamuva zibonisa ukuthi ukulungiswa okuncane ezithunzini, ukufiphaza okuhlakaniphile, noma umsindo wesitha kunganciphisa kakhulu ukusebenza kahle kwezitholi ze-biometric, okuvumela ubuso obungamanga ukuthi budlule ukulawula kokufinyelela.
Izinselele zokuthola: kungani isofthiwe yokulwa namagciwane yodwa inganele
Ukuthola ama-deepfake akulula njengokufaka isofthiwe bese ukhohlwa ngakho. Amamodeli okuthola amanje abhekene nenkinga enkulu: ukuguquguquka kwawo kulinganiselwe . Ngokuvamile aqeqeshwa ngokusebenzisa amasu nezitayela ezithile zokuphatha, futhi lapho kuvela i-algorithm yesizukulwane esisha, alahlekelwa ukusebenza kahle.
Ngaphezu kwalokho, abahlaseli bangaletha ukuphazamiseka okuphambene okuklanyelwe ngqo ukukhohlisa abatholi . Okusho ukuthi, banezela umsindo ongabonakali esweni lomuntu, kodwa odida izinhlelo zokuhlola futhi okuvumela okuqukethwe okushintshiwe ukuthi kudlule ezihlungini njengokungathi kuyiqiniso.
Okungeziwe kulokhu inselele yokukwazi ukukhulisa. Amapulatifomu kanye nezinkundla zokuxhumana zithola amavidiyo amaningi nomsindo nsuku zonke , okwenza ukuhlaziywa okujulile kwengxenye ngayinye kube nzima kakhulu. Abaningi baphoqeleka ukusebenzisa amasampula, ukuhlaziywa okungaphezulu, noma izinhlelo ezenzakalelayo ngezinye izikhathi ezihlulekayo.
Ngakho-ke, okwamanje akukho "inhlamvu yesiliva" ngokumelene nama-deepfake. Ukuzivikela kudinga inhlanganisela yamasu : ukuhlaziywa kwe-metadata, ukuhlolwa kwezinto ezibonakalayo noma ezilalelwayo, amanethiwekhi ezinzwa aqeqeshwe ukuthola amaphethini angabonakali, futhi, okubaluleke kakhulu, ukuqeqeshwa komsebenzisi kanye nokucabanga okubucayi.
Ukuthuthuka kwe-AI kudinga ukubuyekezwa okuqhubekayo kwamamodeli okuthola . Isizukulwane ngasinye esisha samathuluzi okukhiqiza sidinga ukubuyekezwa nokuqeqeshwa kabusha kwezinhlelo zokuzivikela, okudala umdlalo ohlala njalo phakathi kwabahlaseli nabavikeli.
Uhlaka lwezomthetho: indlela iSpain ne-European Union ezisabela ngayo
Ngokombono wezomthetho, ama-deepfake awabhekwa njengobugebengu obuthile ngaso sonke isikhathi, kodwa ukusetshenziswa kwawo okulimazayo kuwela ngaphansi kwamacala nemithethonqubo eminingi yobugebengu ekhona . ESpain, i-Penal Code inikeza izinhlobo ezahlukene zokuvikela ekuziphatheni okungesona kanye nokwephulwa kobumfihlo.
Isigaba 197 kanye nezinye ezilandelanayo zivikela ilungelo lobumfihlo kanye nesithombe sakho, zijezisa ukuthola, ukusebenzisa, noma ukusatshalaliswa kwezithombe eziyimfihlo noma okuqoshiwe ngaphandle kwemvume , noma ngabe kusetshenziswe ngedijithali. Ukuziphendukela kocansi okusakazwa ngaphandle kwemvume kungabhekwa njengobugebengu obumelene nobumfihlo.
Isigaba 401 senza ukwebiwa kobunikazi kube yicala , okungasetshenziswa lapho kusetshenziswa inkohliso ejulile ukuzenza omunye umuntu, isibonelo, ukukhohlisa abantu besithathu noma ukwenza ukukhwabanisa. Izigaba 208-210 zilawula ukuhlambalaza nokunyundela, ziyasebenza uma okuqukethwe okungamanga konakalisa kakhulu udumo noma idumela lesisulu.
Izihloko 248 no-249, eziphathelene nokukhwabanisa, zimboza amacala lapho amazwi noma ubuso obushintshiwe busetshenziswa khona ukuthola inzuzo yezezimali engekho emthethweni . Kulezi zimo, ubuchwepheshe be-deepfake buyindlela esetshenziswayo yokwenza ubugebengu, yize bungabhekwa njengecala elihlukile.
Ngesikhathi esifanayo, uMthetho We-Organic 3/2018 (LOPDGDD) kanye noMthetho Wokuvikelwa Kwedatha Ojwayelekile (GDPR) bavikela idatha yomuntu siqu, okuhlanganisa idatha ye-biometric efana nobuso nezwi. Ukusebenzisa isithombe noma izwi lomuntu othile eqoqweni elijulile ngaphandle kwemvume yakhe kungaba ukwephulwa okukhulu kokuvikelwa kwedatha, okunezijeziso zokuphatha kanye nesibopho sobugebengu esingaba khona.
UMthetho We-Organic 1/1982, ophathelene nelungelo lokuhlonipha, ubumfihlo kanye nesithombe sakho, uvumela isisulu ukuthi sifake icala lombango ngokumelene nanoma ubani osakaza inkohliso ephula isithunzi noma idumela lakhe , ecela ukususwa kokuqukethwe, isinxephezelo somonakalo kanye nezinyathelo zokuqapha eziphuthumayo zokuyeka ukusakazwa kwakho.
Ezingeni laseYurophu, uMthetho we-AI (i-EU Artificial Intelligence Regulation) uzodinga ukuthi okuqukethwe okukhiqizwe yi-AI, okuhlanganisa nama-deepfake, kuhlonzwe ngokucacile kanjalo. Okwamanje, uMthetho Wezinsizakalo Zedijithali (i-DSA) uphoqa amapulatifomu nezinsizakalo zedijithali ukuthi zisabele ngokushesha kokuqukethwe okungamanga, okuhlambalazayo, noma okwephula amalungelo ayisisekelo.
Ngaphezu kwalokho, ohulumeni abathile—kuhlanganise nabaseSpain—sebeqalile ukuchaza ubugebengu obuhlobene nokukhohlisa ngokocansi kanye nokuzilungisa, baqinisa ukuvikelwa kobugebengu ezithombeni zobulili ezingcolile ezingavunyelwene ngazo ezidalwe nge-AI kanye nokuhlukunyezwa kwezingane ngokusebenzisa ukuqondiswa kwemisindo nokubukwayo.
Izingozi ezithile ezinganeni, kwabesifazane kanye namaqembu asengozini
Ama-Deepfake awathinti wonke umuntu ngokulinganayo. Idatha iphakamisa ukuthi abesifazane, izingane ezincane, kanye namaqembu athile omphakathi anengxenye enkulu yomthelela, ikakhulukazi maqondana nokuqukethwe okuyimfihlo okwabiwe ngaphandle kwemvume.
Imibiko yakamuva ikhombisa ukuthi ingxenye enkulu yamanga aqoshwe ku-inthanethi ayizithombe zobulili ezingcolile futhi ahlose abesifazane, abaningi babo abadumile emphakathini. Lokhu kuhunyushwa kube imikhankaso yokuhlukumeza, ukusongela, kanye nokuhlambalaza, kanye nengxenye enamandla yobudlova obusekelwe kubulili bedijithali.
Ezinye izifundo kanye nemibiko zilinganisela ukuthi intsha eyodwa kwabahlanu eSpain ibika ukuthi ibe nokuhlangenwe nakho okuthile okuhlobene nezenzo ezijulile ngesikhathi sobuntwana noma sobusha, kungaba yisisulu esiqondile, umbukeli, noma umhlanganyeli. Lokhu kuvezwa kusenesikhathi kwandisa ingozi yokwenza imikhuba yokuhlukunyezwa kanye nokudlwengulwa ngokocansi ibe yinto evamile.
Kuma-criminology kanye nohlelo lwezobulungiswa, ama-deepfake abangela ikhanda elengeziwe: angavezwa "njengobufakazi" ezinqubweni zomthetho , okwenza kube nzima ngabahluleli, abashushisi, kanye nochwepheshe ukunquma ubuqiniso balezi zinto.
Lesi simo siphinde siphakamise izinkinga ezinkulu endaweni yokuphepha komuntu siqu kanye nokuphepha kwe-inthanethi komphakathi, njengoba i-biometrics yokulawula ukufinyelela okusekelwe ebusweni noma ngezwi ingaba yinto engathembeki kakhulu uma izindlela eziqinile zokuthola okuqukethwe okwenziwe zingathuthukiswa ngesikhathi esifanayo.
Ngaphandle kokulimala komuntu ngamunye, ukwanda kwama-deepfake kunegalelo esimweni sokungathembani okusabalele kolwazi lwedijithali . Uma noma iyiphi ividiyo noma umsindo kungaba mbumbulu, umqondo uyaqhubeka ukuthi "akukho okuphephile," futhi ekugcineni, abantu abaningi bayayeka ukuzama ukuhlukanisa phakathi kwalokho okungokoqobo nalokho okushintshiwe.
Lokhu kukhathala kolwazi kuyingozi kakhulu ngoba kuvula indlela yemikhankaso yolwazi olungesilo iqiniso kanye nokuphathwa kabi kwabantu abaningi . Ukuhlaselwa kolwazi olungesilo iqiniso okuhlanganisa izindaba ezingamanga nezithombe ezijulile ezizwakalayo kungathinta ukhetho, amareferendum, izinqumo zokutshalwa kwezimali, kanye nobudlelwano bamazwe ngamazwe.
Imithombo yezokuxhumana isebenza njengesisheshisi sale nqubo. Okuqukethwe okungamanga kodwa okunomthelela kungasakazeka ngemizuzu embalwa , kuyilapho ukuqinisekiswa nokulungiswa kufika kamuva kakhulu futhi kungafinyeleleki kakhulu. Lokhu kubambezeleka kusobala ukuthi kusebenza kahle kulabo abasebenzisa i-AI yokukhiqiza ngezinjongo ezimbi.
Amapulatifomu amakhulu—i-Facebook, i-X (eyayikade iyi-Twitter), i-Google, njll.—amemezele izinyathelo zokunciphisa ukusabalala kwama-deepfake ayingozi , kusukela ekuvinjelweni okucacile kuya ekulebheleni nasekususeni ngokushesha izinhlelo. Kodwa-ke, ubukhulu benkinga kanye nesivinini ubuchwepheshe obushintsha ngaso kusho ukuthi impendulo ihlala ingaphelele.
Ingozi enkulu kakhulu ukuthi lezi zimo zinegalelo ekubhubhiseni ukucabanga okubucayi phakathi kwezakhamuzi . Uma “konke kungaba amanga,” kuba lula ukuhlwanyela ukungabaza, ukusabalalisa imibono eyeqisayo, kanye nokwehlisa ukwethenjwa kwabezindaba, izikhungo, ngisho nohlelo lwentando yeningi uqobo.
Amasu obuchwepheshe nawokuhlela okuzivikela
Njengoba sibhekene nenkinga eyinkimbinkimbi kangaka, impendulo kumele ibe nezilinganiso eziningi ngokulinganayo. Ezingeni lobuchwepheshe, kuthuthukiswa ama-algorithm okuthola asekelwe ku-AI afuna ukungahambisani okucashile kumavidiyo: amaphutha okuvumelanisa izindebe, ukucwayiza okungavamile, izinto zokucindezela, isikhumba esibushelelezi ngokweqile, noma imiphetho efiphele endaweni yobuso.
Enye indlela ihilela ukuvikela okuqukethwe "emthonjeni" ngokufaka ama-watermark angabonakali noma imethadatha engaqinisekiswa eqinisekisa ubuqiniso bevidiyo noma isithombe. Lawa ma-mark angasiza ukuhlukanisa izinto zokuqala kumakhophi ashintshiwe, uma nje kukhona indinganiso eyamukelekayo namathuluzi okuqinisekisa lokho.
Ezindaweni zezinkampani, ukuqinisekiswa kwezinto eziningi okuthuthukisiwe kuthola ukuthandwa okwandayo . Lokhu kusho ukungathembeli kuphela esithombeni noma ezwini lomuntu, kodwa ukulihlanganisa nezinye izici: indawo, amaphethini okusetshenziswa ajwayelekile, ukuqinisekiswa ngeziteshi ezihlukile (i-SMS, izinhlelo zokusebenza eziphephile), noma amakhodi esikhathi esisodwa.
Kunconywa futhi ukusebenzisa izinqubo ezicacile zokuthengiselana okubucayi , njengokudluliselwa kwebhange noma izinguquko zeziqinisekiso. Isibonelo, qinisekisa ukuthi akukho myalelo wezezimali obalulekile owenziwa nge-imeyili noma ngevidiyo kuphela, kodwa njalo udinga ukuqinisekiswa okwengeziwe ngesiteshi esihlukile.
Okokugcina, ukuqeqeshwa okuqhubekayo kubalulekile. Izinhlangano ezenza ukuvivinya nge-phishing kanye nokujula - ukufundisa abasebenzi bazo ukuthola izimpawu ezicashile njengezinkinga zokuvumelanisa umsindo, ukushukuma okungavamile, noma ukuziphatha okungavamile - zakha isiko lokuphepha eliqinile kakhulu.
Indlela yokuthola i-deepfake: izimpawu ezisebenzayo zokuphila kwansuku zonke
Ngokombono womsebenzisi, kunezinkomba eziningana ezingasiza ekuboneni okuqukethwe okungamanga, yize kungenakuphikwa ngaso sonke isikhathi. Isincomo sokuqala ukubheka amaphutha abonakalayo asobala : imiphetho efiphele ezungeze ubuso, ukukhanya okungaguquki uma kuqhathaniswa nesizinda, ukunyakaza okuxekethile, noma ukubukeka kobuso "okungajwayelekile".
Ukucwayiza kungenye inkomba ewusizo. Ama-deepfake amaningi asabonisa amaphethini okucwayiza okungavamile : amehlo ahlala evulekile kakhulu isikhathi eside noma acwayiza ngesigqi esingajwayelekile. Nakuba amamodeli amasha ethuthuka kule ndawo, kuseyiphuzu elibuthakathaka ezimweni eziningi.
Kubalulekile futhi ukubheka umzimba womuntu, intamo, namahlombe. Iningi lezindlela zokuphatha ligxila ebusweni , ngakho-ke ingxenye esele yomzimba ingabonisa ukukhwabanisa uma ubukhulu, ukuma, noma ukushukuma komzimba kungafani nalokho okulindelekile kulowo muntu.
Ubude bevidiyo bungaba enye inkomba. Ukudala i-deepfake ende kudinga izinsiza ezengeziwe futhi kwandisa amathuba okuba namaphutha, ngakho okuqukethwe okuningi okushintshiwe kufushane kakhulu futhi kugxile emushweni owodwa noma endaweni eyodwa enomthelela. Isiqeshana esifushane kakhulu esinomyalezo oqhumayo kufanele siphakamise ifulegi elibomvu.
Okokugcina, kubalulekile ukuhlaziya umongo kanye nomsuka wokuqoshwa . Ubani owabelane ngakho kuqala? Ingabe kuvela ezinkundleni zokuxhumana ezithembekile noma kuma-akhawunti angaziwa kuphela? Ingabe kukhona eminye imithombo ezimele eqinisekisa lokho okubonayo? Ukunciphisa ijubane lokudlala noma ukubuka ifreyimu yevidiyo ngefreyimu kungasiza ekuboneni izinguquko ezisheshayo ngemuva, ukulungiswa kobuso, noma ingaphakathi lomlomo (ulimi, amazinyo)—izindawo lapho ama-algorithm amaningi esahluleka khona.
Ukufundisa abezindaba kanye nesiko lokuqinisekisa
Akukho buchwepheshe bokuthola ulwazi obwanele ngaphandle kwesakhamuzi esinamakhono okucabanga abucayi kanye namakhono okuqinisekisa ayisisekelo . Izinhlangano ezifana ne-INCIBE kanye ne-Internet User Security Office zigcizelela isidingo sokuqaphela ulwazi esilutholayo, sihlaziye okuqukethwe ngokuningiliziwe, futhi sikuqinisekise njalo ngemithombo ethembekile.
Ukuthuthukisa umkhuba wokungabelani lutho ngokuphamazela , ikakhulukazi uma kudonsela imizwa enamandla (ukwesaba, intukuthelo, intukuthelo), kungenye yezindlela ezinhle kakhulu zokuzivikela ekusabalaleni kwezindaba ezijulile nezindaba ezingamanga. Ukuzithatha imizuzwana embalwa uzibuze, "Ubani ozuzayo ngokukholelwa kwami lokhu?" kungenza umehluko omkhulu.
Ukuqeqeshwa kokuphepha kwe-inthanethi, ucwaningo oluku-inthanethi, kanye nobuhlakani be-inthanethi akuzona nje ezochwepheshe. Ukuqonda ukuthi ukuhlasela nge-phishing, ubunjiniyela bezenhlalo, kanye nokuhlasela okungamanga kusebenza kanjani kusiza noma ubani ukunciphisa kakhulu ubuthakathaka bakhe, kokubili empilweni yakhe yomuntu siqu kanye neyomsebenzi.
Kubantwana kanye nentsha, kubaluleke kakhulu ukuhlanganisa lezi zihloko emfundweni yedijithali neyemizwa , ukuchaza izingozi zokuxhashazwa ku-inthanethi, ukuhlukumeza ngokocansi, kanye nokwabelana ngezithombe eziseduze. Ukwazi ukuthi ividiyo ingaba yiqiniso, kodwa imiphumela yayo ingaba ngokoqobo kakhulu, kwenza umehluko omkhulu.
Kumabhizinisi nezikhungo, ukutshala imali ezinhlelweni zokuqwashisa ezidlula izeluleko zokuphepha kwe-inthanethi ezivamile manje sekuyinto ebalulekile. Labo abangaqondi ama-deepfake, umthelela wawo, nokuthi bangasabela kanjani kuwo bazoba sengozini enkulu endaweni entsha yedijithali.
Zonke izinkomba zisikisela ukuthi ama-deepfake azoqhubeka nokwanda ngobuningi nangobunyoninco , athinta ukuphepha, umnotho, ezombusazwe, kanye nokuphila kwangasese kwezigidi zabantu. Ukwamukela ukuthi ayingxenye yendawo yedijithali, kudinga izinhlaka zomthetho ezicacile, ukutshala imali kubuchwepheshe obuqinile bokuthola, futhi ngaphezu kwakho konke, ukuhlakulela isiko lokuqinisekisa kanye nokwabelana ngemithwalo yemfanelo kuyizindlela ezinhle kakhulu zokuvimbela leli thuluzi elinamandla ekubeni yisikhali esiqeda ngokuphelele ukwethembana ezimpilweni zethu eziku-inthanethi.