- AmaNethiwekhi Aphikisayo Akhiqizayo (ama-GAN) avumela ukudalwa kobunikazi bokwenziwa obungokoqobo kakhulu obugwema i-biometrics yendabuko.
- Ukuzivikela okuqine kakhulu kusekuhleleni izimpawu kanye nokutholwa okunezingqimba ekubanjweni, ekuthuthweni nasekuqhathanisweni.
- Izitifiketi ezizimele njenge-iBeta Level 3 kanye nemithethonqubo efana ne-EU AI Act zibalulekile ekuqinisekiseni ukuphepha.
Sesingene esigabeni lapho ukwebiwa kobunikazi kuthathe igxathu elikhulu phambili. Asisakhulumi ngamakhophi alula noma amamaski enjoloba, kodwa ngemidiya yokwenziwa eyenziwe nge-AI ekwazi ukukhohlisa ngisho nezinhlelo eziyinkimbinkimbi kakhulu. Abahlaseli manje basebenzisa ama-deepfake kanye nokuhlaselwa okuhlanganisiwe okwenza izindlela zokuphepha zeminyaka embalwa edlule zibukeke njengokudlala kwengane.
Ubungozi obukhulu bezinkampani eziningi ukukholelwa ukuthi isithiyo esisodwa sanele. Kodwa-ke, idatha isitshela ukuthi imizamo yokukhwabanisa ethuthukisiwe engaphezu kuka-70% iyekwa kuphela uma sekukhona izendlalelo eziningi zokuqinisekisa . Akukhona nje ukuthi usongo lungokoqobo yini—okuyinto engokoqobo kakhulu—kodwa mayelana nendlela yokwakha udonga lokuzivikela olusebenza kahle ngempela ngaphandle kokukhungathekisa abasebenzisi abasemthethweni.
Inkinga ngokuthembela kuphela ekuphileleni impilo
Isikhathi eside, ukuhlolwa kokuphila okusebenzayo—okukucela ukuthi ucwayize noma uphendule ikhanda lakho—kwakuyinkanyezi. Kodwa-ke, ama-deepfake athuthukile futhi manje angalingisa ukubonakaliswa kobuso futhi asabele ngesikhathi sangempela emiyalweni yesistimu. Inkinga ukuthi la mathuluzi abeka phambili ukuphendula, kodwa hhayi ngempela ubuqiniso beqiniso bevidiyo.
Ngaphezu kwalokho, ukucela ikhasimende ukuthi lenze i-acrobatics phambi kwekhamera kudala ukungezwani okukhulu. Kwezinye izimo, ukushiywa komsebenzisi kufinyelela ku-40%, kuyilapho ukushintshela ekutholeni ukuqina okungapheli (i-selfie elula) kukhulisa izinga lokuqedela libe ngaphezu kuka-95%. Icebo liwukuba nomsebenzi wokuphepha ngemuva ngaphandle kokuphazamisa umsebenzisi.
Ingabe i-Deepfakes ibaluleke kakhulu futhi yenzani?
Empeleni, lawa amafayela omsindo, ezithombe, noma evidiyo asetshenziswa kusetshenziswa ukufunda okujulile ukuze abonakale eyiqiniso. Ubuchwepheshe obuyinhloko lapha yi-Generative Adversarial Networks (GANs), lapho amanethiwekhi amabili e-neural encintisana khona: elinye lidala inkohliso kanti elinye lizama ukulithola, okuphoqa ijeneretha ukuthi ilungise amanga kuze kube yilapho cishe ingasahlukaniseki eqinisweni.
Kunezinhlobo ezimbili eziyinhloko zamagatsha: i-Deepfaces , ethatha indawo yobuso noma idale abantu abasha ngokuphelele, kanye ne -Deepvoices , eguqula izwi lomuntu. Lokhu kokugcina kuyingozi kakhulu endaweni yezinkampani, lapho kube khona amacala abaphathi abadlulisela izinkulungwane zamaRandi ngemuva kokukholelwa ukuthi bakhuluma ne-CEO yabo.
Isu lokuthola elihlukaniswe ngezingqimba
Uma usebenzisa i-deepfake detector kuphela, unesikhala sokuphepha. I-algorithm ingase ibe yinhle kakhulu ekutholeni ukuthungwa kwesikhumba okwenziwe, kodwa ingayiboni i- template injection attack . Ngakho-ke, ikhambi ukusebenzisa ukwakheka kwezigaba ezintathu:
- Ukutholwa ekubanjweni: Ihlaziya ukuthi ngabe umuntu wangempela ukhona yini futhi ithole izinto zobuciko ze-AI ngesikhathi sokuqoshwa.
- Ukutholwa kwethrafikhi: Iqinisekisa ukuthi ividiyo ayibanjwanga noma ayiguqulwanga phakathi kwedivayisi yeselula yomsebenzisi kanye neseva, ivimbela ukuthi ingabi khona faka okuqukethwe okwenziwe ku-API.
- Ukutholwa uma kuqhathaniswa: Qondanisa ubuwena obuveziwe namaphethini okukhwabanisa aziwayo bese uhlaziya ukuthi ukuziphatha komsebenzisi kuyasolisa yini.
Le ndlela ivumela umkhohlisi okwazi ukugwema ubufakazi bokuphila ukuba awele ogibeni lobuqotho besiteshi noma ekuhlaziyweni kokuziphatha. Kuyinethiwekhi lapho i-mesh ngayinye isebenza khona umsebenzi ohlukile.
Ubunikazi Bokwenziwa kanye Nomkhakha Wezezimali
Ezweni le-fintech kanye nebhange, ingozi inkulu kakhulu. Izigebengu azisantshontshi nje kuphela ama-identity; zakha ama-identity okwenziwa . Lokhu kuhilela ukuxuba idatha yangempela, evuvukile (njenge-ID evumelekile) nolwazi oluqanjiwe kanye nobuso obukhiqizwe yi-AI. Umphumela uba iphrofayili ebonakala isemthethweni futhi ingavula ama-akhawunti noma ifake isicelo semali mboleko ngaphandle kokuphakamisa izinsolo.
Ukuze kulwiwe nalokhu, uhlelo lushintsha lusuka ekuqinisekisweni okuhlukile luye ezimpawini ezihleliwe . Akusanele ukuthi ubuso bufane nedokhumenti; manje indawo, izigxivizo zeminwe zedivayisi, kanye nesivinini sokuthayipha kuyahlaziywa. Uma i-biometrics iphelele kodwa idivayisi iyi- emulator esolwayo , uhlelo luqala isexwayiso.
Amazinga, Imithethonqubo kanye Nezitifiketi
Akuwona wonke amathuluzi adalwe ngokulinganayo. Emkhakheni wobungcweti, isitifiketi se -iBeta Level 3 siyindinganiso yegolide, njengoba siqinisekisa ukumelana nokuhlaselwa ngomjovo ezindaweni ezilawulwayo. Ngaphezu kwalokho, uMthetho we-EU AI uzoqala ukufuna ukuthi konke okuqukethwe okukhiqizwe yi-AI kubhalwe ngokucacile , ngaphansi kwesijeziso sezinhlawulo ezinkulu.
Kubalulekile ukuthi izinkampani zifune izixazululo ezihambisana ne -ISO/IEC 30107 futhi zivumele ukucutshungulwa kwedatha kudivayisi uqobo. Lokhu akugcini nje ngokuthuthukisa ukuphepha kodwa futhi kuqinisekisa ukulandelwa kwe-GDPR ngokuvimbela idatha eluhlaza ye-biometric ukuthi ingahambi kunethiwekhi, ngaleyo ndlela kuncishiswe indawo yokuhlasela.
Amathiphu awusizo okuthola ukuphazamiseka
Nakuba i-AI isiya ngokuya iba yinkimbinkimbi, kusenezinkomba ezingasisindisa. Naka ukucwayiza okungaguquki , njengoba ama-algorithms elwela ukulingisa amaphethini emvelo okucwayiza kwabantu. Kubalulekile futhi ukubheka imiphetho yobuso kanye nengaphakathi lomlomo ; amazinyo nolimi ngokuvamile kuyizindawo lapho i-AI yenza khona amaphutha abonakalayo.
Enye imininingwane okufanele uyicabangele ukuvumelanisa izindebe nomsindo. Ngokuvamile, umsindo awuhambisani kahle nokunyakaza kwezindebe. Uma unemibuzo, kungcono ukunciphisa ijubane levidiyo ukuze ubheke ukugxuma okungazelelwe esithombeni noma izinguquko ezingavamile ngemuva ezingase zembule ukuguqulwa.
Ukuphepha kwedijithali namuhla kudinga ukuqapha okuqhubekayo kanye nokwamukelwa kwezinhlelo ezinganciki esicini esisodwa. Ukuhlanganiswa kwe-biometrics ethuthukisiwe, ukuhlaziywa kokuziphatha, kanye nokuthobela imithetho eqinile kuyindlela kuphela yokuvimbela ukuthuthuka kwezindlela zokwenziwa nokuvikela ubuqotho bokuthengiselana ezweni lapho ukubona kungasakholwa khona.
