Indlela yokuthola ama-Deepfakes: umhlahlandlela ophelele wokuzibona nokuzivikela kumavidiyo nomsindo mbumbulu

Eminyakeni yamuva nje sibone ukuthi ubuhlakani bokwenziwa buholele kanjani ekudalweni kokuqukethwe okungokoqobo kakhulu, okwaziwa kangcono ngokuthi deepfakesLawa mavidiyo, izithombe, kanye nokuqoshwa komsindo okushintshiwe kuyinselele yangempela endleleni esibona ngayo ubuqiniso bolwazi ku-inthanethi. Kuya ngokuya kuba nzima ukuhlukanisa iqiniso namanga, futhi ikhono lokwenza ubuso, amazwi, noma ngisho nezimo eziphelele liveze izinhlobo ezintsha zolwazi olungelona iqiniso, ukukhwabanisa kwedijithali, kanye nokwebiwa kobunikazi.

Ukwanda kwalobu buchwepheshe, okubangelwa ukutholakala kwamathuluzi ayinkimbinkimbi ngokwengeziwe, kuphoqa wonke umuntu ukuba aqaphele kakhulu ngaphambi kokwamukela noma yikuphi okuqukethwe okulalelwayo noma okulalelwayo njengokwangempela. Funda ukuthola ama-deepfake Kubalulekile ukuzivikela thina mathupha nangokomsebenzi, kanye nokuvimbela ukusabalala kwezindaba ezingamanga, ukukhwabanisa kanye nokukhwabanisa kwabezindaba.

Ayini ama-deepfakes futhi asebenza kanjani?

Leli gama faka Kuvela ekuhlanganisweni "kokufunda okujulile" kanye "nokungamanga," futhi kubhekisela kokuqukethwe okukhohlisiwe kusetshenziswa amasu okufunda okujulile okuthuthukisiwe noma ubuhlakani bokwenziwa. Empeleni, Lezi yizithombe, amavidiyo, noma imisindo elawulwa yi-algorithms ekwazi ukuphinda, ngobuqiniso obukhulu, ubuso babantu, amazwi, kanye nezinkulumo.Icebo lisekusetshenzisweni kwamanethiwekhi e-neural — njenge-GANs (Generative Adversarial Networks) — kanye nekhono lalezi zinhlelo lokuhlaziya izinkulungwane zamaphuzu edatha bese kukhiqiza izithombe ezintsha ezingokoqobo ezilingisa ngokuphelele umuntu wangempela.

Nakuba ekuqaleni kwakusetshenziswa imiphumela ekhethekile kumafilimu, namuhla kunezinhlelo zokusebenza ezifinyeleleka emphakathini ezivumela noma yimuphi umsebenzisi ukuthi adlale ngalobu buchwepheshe namavidiyo, izithombe nomsindo mbumbulu ngezinga elikhulu, futhi lokhu kwandisa ingozi yokukhwabanisa, imikhankaso yolwazi olungesilo iqiniso kanye nokwebiwa kobunikazi emhlabeni jikelele.

Umthelela kanye nezingozi zama-deepfake

Ngokusho kwedatha yakamuva, i- Inani lama-deepfake ku-inthanethi landa kakhulu minyaka yonkeIsibonelo, i-World Economic Forum ilinganisele ukukhula kwaminyaka yonke okungu-900% kulolu hlobo lokuqukethwe ngo-2023. Ingozi yama-deepfake ingokoqobo kangangokuthi, ngokusho kwemibiko ekhethekile, ukuhlaselwa okuhilela ukuzenza umuntu ongeyena kanye nokuphathwa kabi kwamavidiyo kukhule ngo-450% eLatin America kuphela ngo-2023. Ukusetshenziswa kwawo sekuvele kubangele izinkinga endaweni ye- ubuntatheli, ezombusazwe, ezezimali, ezokuzijabulisa, kanye nokuphepha komphakathi.

Izingozi ezinkulu zifaka:

• Ukwehliswa isithunzi noma ukulimala kwedumela kubantu ngabanye kanye nezinkampani ngokusebenzisa amasethingi athembekile.
• Ukusetshenziswa kwamavidiyo mbumbulu ukuqinisekiswa kwe-biometric kwe-pass futhi uzenze abanye.
• Ukugcwaliseka kwe ukukhwabanisa kumapulatifomu edijithali kanye nokuphanga.
• Ukukhula ukusabalala kwezindaba ezingamanga kanye nokuphathwa kabi kwezenhlalo nezepolitikiokungaphazamisa izimakethe kanye nobudlelwano bamazwe ngamazwe.
• Ubungozi be ukweba noma ukudlulisa imali ukusebenzisa umsindo namavidiyo mbumbulu, njengoba kwenzekile ngama-fraud fraud kanye nokukhwabanisa ku-inthanethi.

Ukuhlonza i-deepfake kudinga amakhono ahlanganisa ukuhlaziywa kobuchwepheshe kanye nokucabanga okubucayi, njengoba ubuhlakani balezi zinto ezidaliwe buqhubeka buthuthuka.

Izimpawu namaqhinga okuthola ama-deepfake

Intuthuko kwezobuchwepheshe obujulile yenza la mavidiyo akholiseke kakhulu, kodwa ngisho nalawo angcono kakhulu asabonisa amaphutha athile. amaphutha noma izinto ezingavamile ezicashile lokho kungasisiza ukuthi sithole inkohliso. Ngezansi, sikutshela izimpawu ezibalulekile okufanele uziqaphele:

1. Amehlo angavamile nokucwayiza

Elinye lamaphutha avame kakhulu ukuthi Inani lokucwayiza noma isigqi sazo akuyona into yemvelo.Ezinye izigcawu zibonakala ziphoqelelwe noma zibonisa ukucwayiza okuncane kakhulu, into abantu abavame ukuyenza njalo. Le mininingwane, nakuba icwengisisiwe, isalokhu iyinkomba ebalulekile.

2. Ukunyakaza kobuso nomzimba okungajwayelekile

Bheka ukungafani phakathi kokunyakaza kobuso nomzimba, noma ukubonakaliswa okuqinile, okungavamile okungahambisani nephimbo lezwiUma ubuso buhamba ngendlela engahambisani nomzimba, kungaba yinkohliso ejulile. Kungase kube khona futhi amaphutha okunyakaza kwezindebe Ngokuphathelene nomsindo.

3. Ukukhanya okungaguquguquki kanye nezithunzi

Ama-Deepfake avame ukubonakala izithunzi ezingamanga, ezifiphele, noma ezingewona amaqinisoUma ukukhanya kuwela ebusweni ngendlela engavamile noma kungathinti isizinda nesikhumba ngokulinganayo, cabanga ngokungabaza. Umbala wobuso ungase uhluke entanyeni noma emzimbeni wonke.

4. Izici zobuso ezifiphele, izinwele, nezindlebe

La indawo yezinwele kanye nokwakheka kobuso Ngokuvamile kubonisa ukungapheleli: imiphetho engajwayelekile, ukufiphala, noma izindawo ezifiphele. Izinwele kunzima kakhulu ukuzicubungula, njengoba kunjalo ngezakhi ezifana namacici, izibuko, noma indawo ehlangana nesilevu nentamo.

5. Izinkinga emlonyeni nasemazinyweni

Uma umuntu ekhuluma, angakwazi Kungase kuvele amazinyo afiphele noma angewona angempela, noma umlomo onesimo esingajwayelekile noma ongavumelani nhlobo nomsindo.Ubuchwepheshe busalokhu buhluleka ukumelela ulimi, amazinyo kanye nomlomo ngendlela engokoqobo.

6. Ukususa ukuvumelanisa umsindo nezwi

El umsindo oshintshiweIkakhulukazi kuma-deepfake ezwini, kungazwakala sengathi kuyirobhothi, ngokumiswa okwenziwayo noma ithoni engeyona eyemvelo. Ukuxhumana phakathi kokunyakaza kwezindebe nokukhuluma kungahluleka, okwenza kube lula ukuyibona.

7. Isikhumba esibushelelezi kakhulu noma esinokuphazamiseka

Eminye imibukiso ye-deepfake isikhumba esingenazo izimbobo, imibimbi, ama-moles angajwayelekile, noma ukusonteka lapho uhambaImibala engavamile noma ukukhanya okukhulu emehlweni nakho kungabonisa ukukhohliswa kwe-AI.

8. Ubude kanye nomongo wamavidiyo

Ama-former ayinkimbinkimbi kakhulu avame ukudinga amahora amaningi okusebenza, ngakho-ke, Ngokuvamile zinesikhathi esilinganiselwe.Uma ividiyo iyinde, kulapho kungenzeka khona ukuthi kube namaphutha amaningi aqongelelekayo.

9. Ukuvumelana nolwazi lomphakathi

Uma ividiyo ephikisanayo yomuntu odumile emphakathini Usola ukuthi kubonakala kuphela epulatifomu engathembekile, noma kungekho omunye umthombo wezindaba othola lolo lwazi. Uqhathanisa okuqukethwe kwalo neminye imithombo ukuze abone ukuthi kuyahambisana yini nomlando womuntu, umlando, noma izimo zakhe.

10. Imizwa ephoqelelwe kanye nokukhanga imizwa

Okuningi kokuqukethwe okuqanjwe kakhulu kuhlose ukukhiqiza ukusabela okukhulu ngokomzwelo (Intukuthelo, ukwesaba, ukumangala) ukukhuthaza ukusabalala kwemizwa. Uma ubona ukuthi umyalezo ubonakala uklanyelwe ukulawula imizwa yakho, qaphela: ulwazi olungelona iqiniso lusakazeka kangcono lapho ludala ukuxhumana okukhulu ngokomzwelo.

Amathuluzi obuchwepheshe okuhlonza ama-deepfake

Kuyiqiniso ukuthi kunezinkomba esingazibona ngamehlo enyama, kodwa ukuze kusetshenziswe ngendlela ebucayi neyobungcweti, kuyalulekwa ukuthi kusetshenziswe amathuluzi akhethekileLokhu kusebenzisa kokubili ubuhlakani bokwenziwa kanye nokuhlaziywa kwedatha, futhi kuhlanganiswe ngezigaba eziningana:

1. Amathuluzi asekelwe ku-AI okuthola ama-deepfake

Kunezixazululo ezisebenzisa ukufunda komshini okuthuthukisiwe ukuqhathanisa amaphethini obuso, ukunyakaza, noma ukungahambisani okungezwakali noma okungabonakali emehlweni omuntu. Nazi ezinye izibonelo ezihamba phambili kanye nemisebenzi yazo eyinhloko:

• SentinelIsevisi esekelwe efwini evumela ukutholwa kwesikhathi sangempela ngokuhlaziya izimpawu zobuso, ukucwayiza, kanye nokuvumelana kwesikhathi. Ingahlaziya amavidiyo, izithombe, nomsindo, ikhiqize umbiko obonakalayo wezindawo ezithintekile.
• UkuzwelaIpulatifomu esetshenziswa kabanzi yokuqapha nokuvikela ubuwena ezinhlanganweni, kohulumeni, nasemabhizinisini. Ihlaziya ulwazi oluningi olubonakalayo futhi ithole ukushintshana kobuso, umsindo oshintshiwe, kanye nobufakazi bokuphila. Iwusizo kakhulu emikhakheni enengozi enkulu yokwebiwa kobunikazi.
• Ukuphila kwe-OzNjengoba iqinisekiswe ngokwezinga le-ISO 30107, isebenza njengesivikelo ekuhlaselweni kobuqili nge-SDK okulula ukuyihlanganisa nezinhlelo zokusebenza zeselula nezewebhu.
• WeVerifyIthuluzi elifanele labo abaqinisekisa okuqukethwe okusabalele ezinkundleni zokuxhumana, njengezintatheli, izishoshovu, noma abaqondisi. Lisebenzisa i-blockchain ukugcina isizindalwazi somphakathi sezinto mbumbulu ezaziwayo.
• I-HyperVergeIsixazululo esithuthukisiwe esihlanganisa ukuqinisekiswa kwe-biometric, ukuqashelwa kobuso, kanye nokuhlaziywa kokulwa nokukhwabanisa emikhakheni eminingi yamazwe ngamazwe. Sithole ukuqashelwa kanye nezitifiketi zomhlaba wonke.
• I-FakeCatcher ye-IntelI-real-time detector enesilinganiso sokunemba esingu-96%. Ihlaziya ividiyo ifuna umkhondo omncane wokugeleza kwegazi kumaphikseli, okungenakwenzeka ukuwulingisa kahle nge-AI yamanje.
• Isiqinisekisi se-AI sevidiyo ye-Microsoft: Ithuluzi lamahhala elinikeza amaphuzu okuzethemba abonisa amathuba okuguqulwa kwamavidiyo noma izithombe, ukuthola izinguquko ezimpunga ezingatholakali kubantu.
• I-Deepware: Ikuvumela ukuthi uhlaziye amavidiyo alayishiwe ukuze uhlole ukuthi asetshenziswe yini ngamasu e-AI, ngendlela elula nefinyelelekayo kubo bonke abasebenzisi.
• Ukungafani kwe-Phoneme-VisemeKugxilwe ekutholeni ukungafani phakathi kwemisindo ezwakalayo (ama-phoneme) nalokho okubonwayo ekunyakazeni kwezindebe (ama-viseme). Uma kukhona ukungahambisani, mhlawumbe kuyinkohliso ejulile.
• I-DuckDuckGooseUkugxila ekuvikeleni amabhizinisi nabasebenzisi ezinsongweni ze-inthanethi, idumela, kanye nezezezimali. Kunikeza izincazelo ezicacile zokukhwabanisa futhi kubaluleke kakhulu ezindaweni zezinkampani.

2. Amathuluzi okuhlaziya i-Metadata

Hlaziya imethadatha yefayela (usuku lokudalwa, usuku lokuhlela, indawo, njll.) kungasiza ekutholeni izinkomba mayelana nobuqiniso bayo. Amathuluzi anjenge MediaInfo o I-ExifTool Zikuvumela ukuthi uhlole le mininingwane efihliwe kumafayela e-multimedia.

3. Amathuluzi athola ukungalingani kokuqukethwe

Ezinye izixazululo ezifana I-FaceForensics noma amasistimu we I-Audio Fingerprinting Balandelela amaphethini angajwayelekile ekunyakazeni kobuso, izwi, noma izingxenye zevidiyo ngokwayo, befuna ukungalingani okungenakwenzeka kwezibalo emikhiqizweni yangempela.

Amathiphu awusizo kanye nemikhuba emihle yokuzivikela kuma-deepfake

Ngaphezu kwamathuluzi obuchwepheshe, ukulwa nama-deepfakes nakho kudinga ukwahlulela komuntu kanye nokuqonda okuvamile. Nazi izincomo ezibalulekile:

• Njalo qinisekisa umsuka kwesithombe, ividiyo noma umsindo, ikakhulukazi uma kuvela ezinkundleni zokuxhumana noma emilayezweni esheshayo.
• Ukungathembi okuqukethwe okuklanyelwe ukuvusa imizwa enamandla njengentukuthelo, ukwesaba, noma ukumangala. Uma imizwa ikhuphukile, kulapho kungenzeka khona ukuthi kube ukuzama ukukhohlisa.
• Yenza eyodwa ukusesha isithombe okuhlanekezelwe kusetshenziswa izinjini zokusesha ezikhethekile ukuhlola ukuthi isithombe sivela kwezinye izimo noma nezinye izinsuku.
• Ungabelani ngamavidiyo noma izithombe uma ungakakwazi ukuqinisekisa ubuqiniso bawoikakhulukazi uma zithinta idumela lomuntu, inhlangano, noma zingase zibe nemiphumela emphakathini.
• Hlala usesikhathini izitayela ezintsha namasu obuhlakani bokwenziwa kanye namathuluzi okutholaUbuchwepheshe buthuthuka ngokushesha, futhi umzabalazo uhlala ukhona.

Imibono yesikhathi esizayo kanye nokulwa nama-deepfake

Ibhizinisi le-deepfake ngeke lilahlekelwe amandla: kulindeleke ukuthi libe yimakethe yezigidigidi zamaRandi, enokuhlaselwa okuyinkimbinkimbi ngokwengeziwe. Abathuthukisi bezobuchwepheshe abasemthethweni, izinkampani, kanye nohulumeni bathatha isinyathelo, bebambisana ukuqinisa ukuphepha kwedijithali kanye nokuziphatha kwe-AI.

Kunemiklamo ethembisayo kanye nezinsizakusebenza zamahhala. Kulabo abafisa ukuhlaziya ngokuphelele ubuqiniso bokuqukethwe kwe-multimedia. Nakuba kungekho sixazululo esiphelele, ukuhlanganiswa kokuhlaziywa kwabantu namathuluzi obuchwepheshe kuzoba yisihluthulelo sokubhekana nale nselele.

Ukwanda kobuchwepheshe be-deepfakes kudala inselele eqhubekayo ekwethembekeni nasekuthembekeni kolwazi oluku-inthanethi. Ukuzithola kudinga ukucabanga okubucayi, ukusetshenziswa kobuchwepheshe bokuthola, kanye nokuzibophezela ekungahlanganyelini ochungechungeni lolwazi olungalungile. Lapho siqaphela kakhulu lezi zingozi, mancane amathuba okuthi siwele ekukhohlisweni kwedijithali futhi ukuzivikela kwethu kuzosebenza kahle kakhulu ezinsongweni esezikhona kakade futhi ngokungangabazeki ezizoba zinkulu nakakhulu esikhathini esizayo esiseduze.

Indlela yokuthola amavidiyo ajulile kwi-inthanethi: izindlela, izimpawu namathuluzi