I-Deepfakes yokuzijabulisa: Lapho ama-deepfakes eba ukuzijabulisa

• About the Author:
• Igama lomlobi

  I-Quantumrun Foresight
• December 7, 2023

Isifinyezo sokuqonda

Ubuchwepheshe be-Deepfake, ukusebenzisa i-AI ne-ML, buguqula ukudalwa kokuqukethwe kuzo zonke izimboni ezihlukahlukene. Ivumela ukuguqulwa okulula kwezithombe namavidiyo, okudumile ezinkundleni zokuxhumana ngezici zokushintsha ubuso. Kwezokungcebeleka, ama-deepfakes athuthukisa ikhwalithi yevidiyo futhi enze lula ukukopishwa kwezilimi eziningi, ukuthuthukisa ukuzizwisa kokubukwa kwamazwe ngamazwe. Kufinyeleleka ngezinkundla ezisebenziseka kalula, ama-deepfake asetshenziselwa izithuthukisi zefilimu, ukudala ama-avatar afana nempilo endaweni ye-VR/AR, ukulingisa okufundisayo kwemicimbi yomlando, nokukhangisa komuntu siqu. Ziphinde zisize ekuqeqeshweni kwezokwelapha ngokulingisa okungokoqobo futhi zinike amandla amabhrendi emfashini ukuze abonise amamodeli ahlukahlukene abonakalayo, anikeze izixazululo ezingabizi nezihlanganisayo ekudaleni okuqukethwe.

Ama-Deepfakes womongo omuhle wokudala okuqukethwe

Ubuchwepheshe be-Deepfake buvame ukufakwa ezinhlelweni ezidumile ze-smartphone nedeskithophu ezivumela abasebenzisi ukuthi baguqule isimo sobuso babantu ezithombeni nakumavidiyo. Ngokufanelekile, lobu buchwepheshe bufinyeleleka kalula ngezixhumi ezibonakalayo nokucubungula ngaphandle kwedivayisi. Isibonelo, ukusetshenziswa okwandile kwama-deepfake ezinkundleni zokuxhumana kwaholwa yisihlungi sokushintsha ubuso esidumile lapho abantu beshintshana ubuso bomunye nomunye kumadivayisi abo eselula. 

Ama-Deepfakes enziwa kusetshenziswa i-Generative Adversarial Network (GAN), indlela lapho izinhlelo ezimbili zekhompyutha zilwa zodwa ukuze kukhiqizwe imiphumela engcono kakhulu. Olunye uhlelo lwenza ividiyo, kanti olunye luzama ukubona amaphutha. Umphumela uba ividiyo ehlanganisiwe engokoqobo ngokuphawulekayo. 

Kusukela ngo-2020, ubuchwepheshe be-deepfake bufinyeleleka kakhulu emphakathini. Abantu abasadingi amakhono wobunjiniyela bekhompyutha ukuze benze i-deepfake; kungenziwa ngemizuzwana. Kunamakhosombe amaningana e-GitHub ahlobene ne-deepfake lapho abantu benikela ngolwazi lwabo nokudala. Ngaphandle kwalokho, kunemiphakathi yendalo ejulile engama-20 namabhodi ezingxoxo abonakalayo (2020). Eminye yale miphakathi inababhalisile nabahlanganyeli ababalelwa ku-100,000. 

Umthelela ophazamisayo

Ubuchwepheshe be-Deepfake buzuza ngokushesha embonini yezokuzijabulisa ukuthuthukisa ikhwalithi yevidiyo ekhona. Ngenxa yokuthi ama-deepfake angakwazi ukuphindaphinda ukunyakaza kwezindebe zomuntu kanye nesimo sobuso ukuze afane nalokhu akushoyo, angasiza ekuthuthukisweni kwefilimu. Ubuchwepheshe bungathuthukisa amafilimu anombala omnyama nomhlophe, buthuthukise ikhwalithi yamavidiyo angafundi noma anesabelomali esiphansi, futhi budale okuhlangenwe nakho okungokoqobo okwengeziwe kwababukeli bamazwe ngamazwe. Isibonelo, i-deepfakes ingase ikhiqize umsindo obizwa ngezilimi eziningi ngezilimi eziningi ngokuqasha abalingisi bezwi bendawo. Ukwengeza, ama-deepfakes angasiza ekukhiqizeni izwi lomlingisi olahlekelwe ikhono lakhe lokukhuluma ngenxa yokugula noma ukulimala. Ama-Deepfakes nawo anenzuzo ukuwasebenzisa uma kunezinkinga ekurekhodweni komsindo ngesikhathi sokukhiqizwa kwefilimu. 

Ubuchwepheshe be-Deepfake buzuza ukuduma phakathi kwabadali bokuqukethwe abasebenzisa izinhlelo zokusebenza ezishintshana ubuso njenge-Ukraine-based Reface. Inkampani, i-Reface, inentshisekelo yokwandisa ubuchwepheshe bayo ukuze ifake ukushintshwa komzimba ogcwele. Abathuthukisi be-Reface bathi ngokuvumela lobu buchwepheshe ukuthi bufinyelelwe abantu abaningi, wonke umuntu angathola ukuphila okuhlukile kwevidiyo elingisiwe ngesikhathi. 

Kodwa-ke, ukukhathazeka ngokuziphatha kuphakanyiswa inani elandayo lamavidiyo ama-deepfakes ezinkundleni zokuxhumana. Okokuqala ukusetshenziswa kobuchwepheshe be-deepfake embonini yezocansi, lapho abantu belayisha izithombe zabesifazane abagqokile ohlelweni lokusebenza olujulile futhi "bahlubule" izingubo zabo. Kuphinde kusetshenziswe amavidiyo ashintshiwe emikhankasweni eminingi enemininingwane engamanga, ikakhulukazi ngesikhathi sokhetho lukazwelonke. Ngenxa yalokho, i-Google ne-Apple bavimbele isofthiwe ye-deepfake edala okuqukethwe okunonya ezitolo zabo zezinhlelo zokusebenza.

Imithelela yokusebenzisa i-deepfakes ekudaleni okuqukethwe

Imithelela ebanzi ye-deepfakes ekudaleni okuqukethwe ingase ihlanganise: 

• Ukwehliswa kwezindleko zemiphumela ekhethekile kwabadali bokuqukethwe abaqopha izigcawu ezibandakanya abantu abanephrofayili ephezulu, abalingisi abagugayo, abalingisi abangena esikhundleni abangatholakali ukushuthwa kabusha, noma okubonisa indawo ekude noma eyingozi. 
• Ukuvumelanisa ngokweqiniso ukunyakaza kwezindebe zabadlali ngomsindo obizwa ngezilimi ezahlukene, okuthuthukisa umuzwa wokubuka wababukeli bamazwe ngamazwe.
• Dala ama-avatar edijithali afana nempilo nezinhlamvu ngaphakathi kwezindawo ze-VR ne-AR, uthuthukise umuzwa ogxilile wabasebenzisi.
• Ukudala kabusha izibalo zomlando noma izehlakalo ngezinjongo zokufundisa, okuvumela abafundi ukuthi bezwe izinkulumo zomlando noma izehlakalo ngokucace kakhudlwana.
• Amabhrendi adala ukukhangisa komuntu siqu kakhudlwana, okufana nokufaka umkhulumeli wosaziwayo odumile ezimakethe zesifunda ezihlukahlukene ngokushintsha ukubukeka noma ulimi lwabo kuyilapho egcina ubuqiniso.
• Amabhrendi emfashini abonisa izingubo zokugqoka nezinsiza ngokudala amamodeli abonakalayo ahlukahlukene akhuthaza ukumelwa okubandakanyayo ngaphandle kwezinselele zokuhlela zokuthwebula izithombe okuvamile.
• Izikhungo zokuqeqesha zezokwelapha zidala ukulingisa kwesiguli okungokoqobo kokuqeqeshwa kwezokwelapha, ukusiza odokotela ukuthi bafunde ukuxilonga nokwelapha izimo ezihlukahlukene endaweni elawulwayo, ebonakalayo.

Imibuzo ongaphawula ngayo

• Abantu bangazivikela kanjani olwazini olungamanga olujulile?
• Yiziphi ezinye izinzuzo ezingaba khona noma izingozi zobuchwepheshe be-deepfake?