I-Privacy Loop: ubuhlakani bokwenziwa ku-labyrinth yobumfihlo kanye ne-copyright

Lesi yisihloko sokuqala kwezimbili lapho ngikhuluma khona ngobudlelwano obubucayi phakathi kobumfihlo kanye ne-copyright ngakolunye uhlangothi, kanye ne-Artificial Intelligence ngakolunye.

Ubudlelwano obuyinkinga lapho ukuziphendukela kwezobuchwepheshe kubonakala kushesha kangangokuthi kwenza noma yikuphi ukulungiswa kokulawula kuphelelwe yisikhathi kusukela ekusetshenzisweni kwakho kokuqala.

Ukubhekana nezindaba ezibuhlungu ezihilela amalungelo abantu nedatha yomuntu siqu kudinga ukunakwa, ikhono kanye nengxoxo ebalulekile phakathi kongqondongqondo nochwepheshe besikhathi sethu. Sithola ukuthi asisheshi ngokwanele ukulungisa imithetho yenhlalo nezinselele ukuqanjwa kabusha kwezobuchwepheshe okusibekela kona. Ubuchwepheshe obusafufusa buzithola busebenza endaweni evulekile, lapho kungekho ngokuphelele imithethonqubo ebeka umkhawulo ekusetshenzisweni kwayo, okukhululekile ukudala umonakalo ngakho-ke ukwenza kanjalo ngaphandle kokujeziswa ngokuphelele.

Kungenzeka yini ukucabanga ukulawula okubuyisela emuva uchungechunge lwentuthuko yezobuchwepheshe ocwaningweni lwesayensi kanye nezinjongo zakho zamasu?

Ingabe kunengqondo ukubusa ukuvela kwezinhlobo zethu zezilwane kuyilapho silondoloza inhlonipho eqinile yenkululeko yomuntu ngamunye?

Ubumfihlo?

“Lapho uzama ukucasha, kulapho udonsa khona ukunaka. Kungani kubaluleke kangaka ukuthi kungabikho muntu owaziyo ngawe?" - evela kwifilimu ethi "Anon" ebhalwe futhi yaqondiswa ngu-Andrew Niccol - 2018

Ku-movie "Anon” ka-2018, umphakathi wesikhathi esizayo uyindawo emnyama, ngaphansi kokulawulwa okuqondile kwesistimu yekhompiyutha enkulu ebizwa ngokuthi i-Ether, ekwazi ukuqapha wonke amagumbi esizwe ngokuyibheka ngamehlo abantu abafanayo abagcwele kuyo. Wonke umuntu ungumbonisi esikhundleni sika-Ether futhi umthwalo wabo wokuqala, yiqiniso, ukuzibheka bona kanye nokuziphatha kwabo.

U-Ether ungumngane ongcono kakhulu webutho lamaphoyisa: nge-Ether, ama-ejenti angakwazi ukulandelela isipiliyoni sanoma yimuphi umuntu ngokukukhumbula ngamehlo abo futhi axazulule noma yiluphi uhlobo lobugebengu.

Iphoyisa u-Sal liyazibuza ukuthi kungani kufanele ulwe ukuze uvikele ubumfihlo bakho: kusizani uma ungenaso isizathu sokucasha? Phela, enkathini lapho ubuchwepheshe esakhayo ukuze kwandiswe ukuphepha kwemizi yethu kanye nemigwaqo yethu budinga ukuqoshwa, ukugadwa kanye nokuqinisekiswa kolwazi olunjalo ukuze kuzuze abantu ngokwabo abacela ukuvikelwa, singalindela kanjani ukuqinisekisa ubumfihlo babo?

Ukukhombisa ukuthi kuyingozi kangakanani ukufinyelela ezimpilweni zabanye, i-hacker izolawula u-Ether futhi iphupho elibi lizokwehlela empilweni yezigidi zabantu: usongo lokubuka njengezibukeli ezingenakuzisiza izithombe zabaningi kakhulu. izikhathi ezihlukumezayo zokuphila kwabo, zisakazwa ngqo ku-retina yabo.

I-Loop

Le amanethiwekhi e-neural okwenziwa okuyisisekelo sokusebenza kobuhlakani bokwenziwa besimanje, izungeza izici ezintathu eziyinhloko: ulwazi oluyisisekelo olubizwa ngenye indlela. Corpus, a i-algorithm ukuze kuhlanganiswe ulwazi kanye nokukodwa inkumbulo ukuze babambe ngekhanda.

I-algorithm ayikhawulelwe ekulayishweni okungavumelekile kolwazi kumemori, iyayiskena ukuze ifune izinto ezizihlobanisa zodwa. Ingxube yedatha nobudlelwano izodluliselwa kumemori ezokwakha okuthi a modello.

Ngaphakathi kwemodeli, idatha nobudlelwano abuhlukaniseki ngokuphelele, yingakho ukwakha kabusha ikhophasi yolwazi lwangempela lokuqeqeshwa kusuka kunethiwekhi ye-neural eqeqeshiwe cishe akunakwenzeka.

Lokhu kuyiqiniso ikakhulukazi uma izidumbu ziqukethe inani elikhulu ledatha. Lena indaba yezinhlelo ezinkulu zezilimi ezaziwa ngokuthi Large Language Models (i-LLM ngamafuphi) kufaka phakathi i-ChatGpt edume kabi. Bakweleta ukuphumelela kwabo ngenani elikhulu lolwazi olusetshenziswa ekuqeqesheni: okwamanje ukuqeqeshwa okuhle kudinga okungenani ama-terabyte ambalwa edatha futhi kunikezwe ukuthi i-terabyte eyodwa ihambisana nezinhlamvu eziyizigidi eziyizinkulungwane ezingu-90, cishe amakhasi ombhalo ayizigidi ezingu-75, kulula ukuqonda ukuthi kukhona ulwazi oluningi oludingekayo.

Kodwa uma amamodeli engakwazi ukwakhiwa kabusha, kungani kufanele sizibuze inkinga yokuphulwa kobumfihlo?

Ukubusa kwedatha

"Noma ngubani osile angacela ukukhululwa emisebenzini yendiza, kodwa noma ngubani ocela ukukhululwa emisebenzini yendiza akahlanyi." - kususelwa kunoveli ethi "Catch 22" kaJoseph Heller.

Innovation newsletter

Ungaphuthelwa yizindaba ezibaluleke kakhulu zokuqamba. Bhalisa ukuze uthole nge-imeyili.

Ukuqoqwa kwedatha enosayizi ovumela ukwakhiwa kwamaphrojekthi afana ne-ChatGpt noma amanye afana nawo namuhla kuyilungelo lezinkampani ezinkulu zamazwe ngamazwe ezithi, ngemisebenzi yazo yedijithali, ezikwazile ukuthola izandla zazo endaweni yokugcina enkulu yolwazi. emhlabeni: Iwebhu.

I-Google kanye ne-Microsoft, osekuyiminyaka bephethe izinjini zokusesha eziskena iWebhu futhi zikhiphe imininingwane eminingi, zingamakhandidethi okuqala ekwakhiweni kwe-LLM, okuwukuphela kwamamodeli we-AI akwazi ukugaya inqwaba yolwazi njengaleyo echazwe ngenhla.

Kunzima ukukholelwa ukuthi i-Google noma i-Microsoft ingakwazi ukufihla imininingwane yomuntu siqu kudatha yabo ngaphambi kokuyisebenzisa njengekhorasi ekuqeqesheni inethiwekhi ye-neural. Ukungadaluli ulwazi esimweni sezinhlelo zezilimi kuhumusha ekuhlonzweni kwedatha yomuntu siqu ngaphakathi kwekhorasi kanye nokushintshwa kwayo ngedatha engelona iqiniso. Ake sicabange ikhophasi enosayizi wamaterabhayithi ambalwa esifuna ukuqeqesha ngawo imodeli futhi ake sizame ukucabanga ukuthi ungakanani umsebenzi ongadingeka ukuze wenze idatha equkethwe ingaziwa ingaziwa: cishe ngeke kwenzeke. Kodwa uma besifuna ukuthembela ku-algorithm ukuze sikwenze ngokuzenzakalelayo, okuwukuphela kwesistimu ekwazi ukwenza lo msebenzi kungaba enye imodeli enkulu ngokulinganayo neyinkimbinkimbi.

Sibhekene nenkinga yakudala ye-Catch-22: “ukuze siqeqeshe i-LLM ngedatha engaziwa sidinga i-LLM ekwazi ukuyenza ingaziwa, kodwa uma sine-LLM ekwazi ukwenza idatha ingaziwa, ukuqeqeshwa kwayo akwenziwanga ngedatha engaziwa. .”

I-GDPR ayisebenzi

I-GDPR ebeka (cishe) emhlabeni jikelele imithetho yokuhlonipha ubumfihlo babantu, uma kucatshangelwa lezi zihloko isivele iyizindaba ezindala futhi ukuvikelwa kwedatha yomuntu siqu ehilelekile kusethi yokuqeqeshwa akucatshangelwa.

Ku-GDPR, ukucubungula idatha yomuntu siqu ngenjongo yokufunda ukuhlobana okuvamile nokuxhumana kulawulwa kancane kuphela I-Athikili 22 ethi: “Isihloko sedatha sinelungelo lokungathwesi isinqumo esisekelwe kuphela ekucutshungulweni okuzenzakalelayo, okuhlanganisa nokwenza iphrofayela, okuyinto ikhiqiza imiphumela yezomthetho kuye noma emthinta ngendlela efanayo futhi ebalulekile".

Lesi sihloko sethula ukuvinjelwa kokuthi abalawuli bedatha basebenzise idatha yomuntu siqu yesihloko njengengxenye yenqubo yokuthatha izinqumo ezenzakalelayo enemiphumela eqondile yezomthetho esihlokweni. Kodwa amanethiwekhi e-neural, afana kalula nezinqubo zokwenza izinqumo ezizenzakalelayo, uma eseqeqeshiwe athola ikhono lokwenza izinqumo ezizenzakalelayo ezingaba nomthelela ezimpilweni zabantu. Kodwa lezi zinqumo azihlali "zinengqondo". Ngesikhathi sokuqeqeshwa, empeleni, inethiwekhi ye-neural ngayinye ifunda ukuhlobanisa ulwazi nenye, imvamisa iluhlobanisa nolunye ngendlela engaqondile ngokuphelele. Futhi ukungabi bikho kwe-"logic" akuwenzi umsebenzi ube lula kusishayamthetho esifuna ukuphakamisa isihlangu sokuvikela ubumfihlo babantu.

Uma umuntu ephinde ekhetha ukusebenzisa inqubomgomo ekhawulela ngokwedlulele, ngokwesibonelo ukuvimbela ukusetshenziswa kwanoma iyiphi idatha ebucayi ngaphandle uma kugunyazwe ngokucacile umnikazi, ukusetshenziswa okusemthethweni kwamanethiwekhi e-neural ngeke kwenzeke. Futhi ukuyeka ubuchwepheshe benethiwekhi ye-neural kungaba ukulahlekelwa okukhulu, cabanga nje ngamamodeli okuhlaziya aqeqeshwe ngedatha yomtholampilo yezifundo zenani labantu eliye lathinteka ngokwengxenye yisifo esithile. Lawa mamodeli asiza ukuthuthukisa izinqubomgomo zokuvimbela ngokuhlonza ukuhlobana phakathi kwezinto ezikhona kudatha nesifo ngokwaso, ukuhlobana okungalindelekile lokho emehlweni odokotela kungase kubonakale kungenangqondo ngokuphelele.

Ukuphatha izidingo

Ukuveza inkinga yokuhlonipha ubumfihlo babantu ngemuva kokugunyaza ngokungakhethi ukuqoqwa kwayo iminyaka kuwubuzenzisi ukusho okuncane. I-GDPR ngokwayo ngobunkimbinkimbi bayo inesibopho sokukhohlisa okuningi okuvumela ukuthola ukugunyazwa ukucubungula idatha yomuntu siqu ngokusebenzisa ukungaqondakali kwezigaba kanye nobunzima bokuqonda.

Impela sidinga ukwenziwa lula komthetho ovumela ukusebenza kwawo kanye nemfundo yangempela ekusetshenzisweni kolwazi lomuntu siqu.

Isiphakamiso sami asikona ukuvumela izinkampani ukuthi zazi idatha yomuntu siqu yabasebenzisi ababhalisela amasevisi abo, ngisho noma beyizinsizakalo ezikhokhelwayo. Ukusetshenziswa kwedatha yomuntu siqu mbumbulu ngabantu abazimele kufanele kwenzeke ngokuzenzakalelayo lapho besebenzisa amasistimu aku-inthanethi. Ukusetshenziswa kwedatha yangempela kufanele kuvalelwe inqubo yokuthenga iyodwa, kuqinisekiswe ukuthi ihlale ihluke ngokuphelele kusizindalwazi sesevisi.

Ukwazi okuthandwayo nokuthandwayo kwesihloko ngaphandle kokuvumela igama noma ubuso ukuthi buhlotshaniswe nale phrofayela kuzosebenza njengendlela yokungaziwa okwenziwa phezulu nomfula okungavumela ngokuzenzakalelayo ukuqoqwa kwedatha nokusetshenziswa kwayo ngaphakathi kwezinhlelo ezizenzakalelayo ezifana nobuhlakani bokwenziwa.

Isigaba se- Gianfranco Fedele