Isikhathi sokufunda esilinganiselwe: 9 imizuzu
Ekuqaleni kuka-2023, i-OpenAI's ChatGPT chatbot edumile neyimpikiswano yaba nenkinga enkulu yezomthetho: ukuvinjelwa okusebenzayo e-Italy. I-Italian Data Protection Authority (GPDP) isole i-OpenAI ngokwephula imithetho ye-EU yokuvikela idatha, futhi inkampani ivumile ukukhawulela ukufinyelela kwenkonzo e-Italy njengoba izama ukuxazulula lolu daba. Ngomhla zingama-28 kuMbasa, i-ChatGPT yabuyela ezweni, i-OpenAI ikhuluma kancane ngezinto ezikhathazayo ze-GPDP ngaphandle kokwenza izinguquko ezinkulu enkonzweni yayo - ukunqoba okusobala.
I-GPDP yaqinisekisa "ukwamukela" izinguquko ezenziwe yi-ChatGPT. Kodwa-ke, izindaba zezomthetho zenkampani - kanye nalezo zezinkampani ezakha ama-chatbots afanayo - cishe zisaqala. Abalawuli emazweni amaningana baphenya ukuthi la mathuluzi e-AI aqoqa futhi akhiqize kanjani ulwazi, icaphuna izinhlobonhlobo zokukhathazeka okuvela ezinkampanini eziqoqa idatha yokuqeqeshwa engenalayisense ukuya ethambekeni lama-chatbot ukusabalalisa ulwazi oluyi-disinformation.
E-EU baphoqelela i-General Data Protection Regulation (GDPR), enye yezinhlaka zezomthetho zobumfihlo eziqine kakhulu emhlabeni, imithelela yayo okungenzeka izwakale nangaphandle kwe-Europe. Ngaleso sikhathi, abenzi bomthetho base-Europe basebenzela umthetho ozobhekana ngqo nobuhlakani bokwenziwa, okungenzeka ungenise inkathi entsha yokulawula amasistimu afana ne-ChatGPT.
I-ChatGPT ingesinye sezibonelo ezidume kakhulu ze-AI ekhiqizayo, igama lesambulela elihlanganisa amathuluzi akhiqiza umbhalo, izithombe, ividiyo nomsindo ngokusekelwe ezicelweni zabasebenzisi. Le nkonzo kuthiwa isiphenduke enye yazo izicelo zabathengi ezikhula ngokushesha emlandweni ngemuva kokufinyelela abasebenzisi abasebenzayo abayizigidi eziyi-100 ezinyangeni ezimbili nje ngemuva kokwethulwa ngoNovemba 2022 (i-OpenAI ayikaze iqinisekise lezi zibalo).
Abantu bayisebenzisela ukuhumusha umbhalo ezilimini ezahlukene, bhala izindatshana zasenyuvesi futhi ukhiqize ikhodi. Kodwa abagxeki, okuhlanganisa nabalawuli, bagqamise okukhiphayo okungathembeki kwe-ChatGPT, izinkinga ze-copyright ezididayo, nezinqubo zokuvikela idatha ezimnyama.
I-Italy yaba yizwe lokuqala ukuthutha. NgoMashi 31, wagqamisa izindlela ezine akholelwa ukuthi i-OpenAI yephula i-GDPR:
Alikho elinye izwe elike lathatha isinyathelo esinjalo. Kepha kusukela ngoMashi, okungenani amazwe amathathu e-EU - Germania , EFrance e Spain – sebeqalise olwabo uphenyo nge-ChatGPT.
Ngaleso sikhathi, ngakolunye uhlangothi lwe-Atlantic, Canada ihlola ukukhathazeka kobumfihlo ngaphansi koMthetho Wokuvikelwa Kolwazi Lomuntu Siqu kanye ne-Electronic Documents Act, noma i-PIPEDA. I-European Data Protection Board (EDPB) isisungule ngisho eyodwa iqembu elisebenzayo elizinikele ukusiza ukuxhumanisa uphenyo. Futhi uma lezi zikhungo zicela izinguquko ku-OpenAI, zingathinta indlela isevisi esebenza ngayo kubasebenzisi emhlabeni jikelele.
Ukukhathazeka kwabalawuli kungahlukaniswa kabanzi ngezigaba ezimbili:
I-ChatGPT isebenzisa i-OpenAI's GPT-3.5 kanye nezinhlobo zezilimi ezinkulu ze-GPT-4 (LLMs), eziqeqeshwe ngenani elikhulu lombhalo okhiqizwe umuntu. I-OpenAI iyaqaphela ukuthi yimuphi umbhalo wokuqeqesha ewusebenzisayo, kodwa ithi idonsela "emithonjeni yedatha etholakala esidlangalaleni, edaliwe, nenelayisense, engafaka phakathi ulwazi lomuntu siqu olutholakala esidlangalaleni."
Lokhu kungase kudale izinkinga ezinkulu ngaphansi kwe-GDPR. Umthetho washaywa ngo-2018 futhi uhlanganisa zonke izinsizakalo eziqoqa noma ezicubungula idatha yezakhamizi ze-EU, kungakhathaliseki ukuthi inhlangano enesibopho isekelwe kuphi. Imithetho ye-GDPR idinga ukuthi izinkampani zibe nemvume ecacile ngaphambi kokuqoqa idatha yomuntu siqu, zibe nezizathu ezingokomthetho zokuthi kungani iqoqwa, futhi zibe sobala mayelana nendlela esetshenziswa futhi igcinwe ngayo.
Abalawuli baseYurophu bathi imfihlo yedatha yokuqeqeshwa kwe-OpenAI isho ukuthi ayikho indlela yokuqinisekisa ukuthi ulwazi lomuntu siqu olufakiwe lwalunikezwe imvume yomsebenzisi ekuqaleni, futhi i-GPDP yaphikisa ngokuqondile ukuthi i-OpenAI "yayingenasisekelo esingokomthetho" sokuyiqoqa kwasekuqaleni. Kuze kube manje i-OpenAI nabanye bahambe ngokucutshungulwa okuncane, kodwa lesi sitatimende sengeza uphawu lombuzo omkhulu emizamweni yesikhathi esizayo yokukhuhla idatha.
Bese kuba khona" ilungelo lokukhohlwa ” ye-GDPR, evumela abasebenzisi ukuthi bacele izinkampani ukuthi zilungise imininingwane yazo yomuntu siqu noma zilususe ngokuphelele. Vula i-AI ibuyekeze inqubomgomo yayo yobumfihlo ngaphambilini ukwenza lula izicelo ezinjalo, kodwa yebo kunjalo xoxa kungakhathaliseki ukuthi ngokobuchwepheshe kuyenzeka yini ukuzilawula, uma kubhekwa ukuthi kungaba yinkimbinkimbi kangakanani ukuhlukana idatha ethile uma sezifakwe kulezi zibonelo zolimi olukhulu.
I-OpenAI iphinde iqoqe ulwazi ngokuqondile kubasebenzisi. Njenganoma iyiphi inkundla ye-inthanethi, iqoqa a isethi ejwayelekile yedatha yomsebenzisi (isb. igama, imininingwane yokuxhumana, imininingwane yekhadi, njll.). Kepha okubaluleke kakhulu, ifaka ukuxhumana abasebenzisi abanakho ne-ChatGPT. Njengoba kushiwo ku-FAQ , le datha ingabuyekezwa ngabasebenzi be-OpenAI futhi isetshenziselwa ukuqeqesha izinguqulo zesikhathi esizayo zemodeli yayo. Uma kubhekwa imibuzo ejulile abantu abayibuza ku-ChatGPT, besebenzisa i-bot njengomelaphi noma udokotela, lokhu kusho ukuthi inkampani iqoqa zonke izinhlobo zedatha ebucayi.
Okungenani enye yale datha kungenzeka iqoqwe kusuka ezinganeni, njengoba inqubomgomo ye-OpenAI ithi "ayiqoqi ngamabomu ulwazi lomuntu siqu ezinganeni ezingaphansi kweminyaka engu-13," akukho ukulawula okuqinile kweminyaka. Lokhu akudlali kahle ngemithetho ye-EU, evimbela ukuqoqwa kwedatha kubantu abangaphansi kweminyaka engu-13 futhi (kwamanye amazwe) idinga imvume yabazali ezinganeni ezingaphansi kweminyaka engu-16. Ngasohlangothini lokuphumayo, i-GPDP ithe ukuntuleka kwezihlungi zeminyaka ye-ChatGPT kuveza izingane a "izimpendulo ezinganele neze uma ziqhathaniswa nezinga lokuthuthuka nokuzazi kwazo".
Futhi ukuthambekela kwe-ChatGPT kokuthi nikeza ulwazi olungamanga kungaba inkinga. Imithethonqubo ye-GDPR ibeka ukuthi yonke idatha yomuntu siqu kufanele inembile, into i-GPDP eyiqokomisa esimemezelweni sayo. Kuya ngokuthi iza kanjani definite, ingapela inkinga kubakhiqizi abaningi bemibhalo ye-AI, abathambekele kuyo ” ama-hallucinations “: Itemu elihle lomkhakha lezimpendulo ezingalungile noma ezingabalulekile embuzweni. Lokhu sekuvele kubone imiphumela yomhlaba wangempela kwenye indawo, njengoba imeya yesifunda sase-Australia ibonile wasabisa ngokumangalela i-OpenAI ngokwehlisa isithunzi ngemuva kokuthi i-ChatGPT iqambe amanga ukuthi udonse isigwebo sasejele ngenkohlakalo.
Ukuduma kwe-ChatGPT kanye nokubusa kwamanje kwemakethe ye-AI kuyenza ibe ithagethi ekhangayo, kodwa asikho isizathu sokuthi kungani izimbangi zayo kanye nabanikelayo, njenge-Google ene-Bard noma i-Microsoft ene-Azure AI esekelwe ku-OpenAI, kungadingeki babhekane nokubhekwa. Ngaphambi kwe-ChatGPT, i-Italy yavala inkundla yezingxoxo Replika ukuqoqwa kolwazi lwezingane futhi kuze kube manje kusanqatshelwe.
Nakuba i-GDPR iyisethi yemithetho enamandla, ayizange idalelwe ukubhekana nezinkinga ezithile ze-AI. Imithetho lokho , nokho, kungenzeka ukuthi zisemkhathizwe.
Ngo-2021, i-EU yethula uhlaka lwayo lokuqala lwe-I-Artificial Intelligence Act (AIA) , umthetho ozosebenzisana ne-GDPR. Umthetho ulawula amathuluzi e-AI ngokusekelwe engcupheni ecatshangelwayo, kusukela “kokuncane” (izinto ezifana nezihlungi zogaxekile) kuye “phezulu” (amathuluzi e-AI okuqinisa umthetho noma ezemfundo) noma “okungamukeleki ” ngakho-ke kwenqatshelwe (njengesistimu yezikweletu zomphakathi). Ngemva kokuqhuma kwamamodeli ezilimi ezinkulu njenge-ChatGPT ngonyaka odlule, abenzi bomthetho manje sebegijima ukuze bengeze imithetho “yamamodeli ayinhloko” kanye “namasistimu e-General Purpose Artificial Intelligence (GPAI)” - amagama amabili esikali sokwenziwa sezinhlelo zezobunhloli okuhlanganisa i-LLM - futhi okungenzeka hlukanisa njenge izinsiza ezinobungozi obukhulu.
Abenzi bezomthetho be-EU bafinyelele isivumelwano esimisiwe ngoMthetho we-AI ngo-Ephreli 27. Ikhomishini izovotela lolu hlaka ngoMeyi 11, kanti isiphakamiso sokugcina silindeleke maphakathi noJuni. Ngakho-ke, uMkhandlu WaseYurophu, iPhalamende kanye neKhomishini kuyodingeka xazulula noma yiziphi izingxabano ezisele ngaphambi kokusebenzisa umthetho. Uma konke kuhamba kahle, kungase kuthathwe engxenyeni yesibili ka-2024, ngemuva kancane kokuhlosiwe isikhulu okhethweni lwase-Europe lwangoMeyi 2024.
I-OpenAI isenemigomo okufanele ifinyelelwe. Kukhona kuze kube uSepthemba 30 ukuze udale umkhawulo weminyaka oqinile wokugcina abaneminyaka engaphansi kwe-13 ngaphandle futhi kudinga imvume yabazali kwentsha engaphansi kweminyaka yobudala. Uma ihluleka, ingase ivinjwe futhi. Kodwa inikeze isibonelo salokho iYurophu ekuthatha njengokuziphatha okwamukelekayo enkampanini ye-AI, okungenani kuze kuphasiswe imithetho emisha.
Ercole Palmeri
I-Coveware ye-Veeam izoqhubeka nokuhlinzeka ngezinsizakalo zokuphendula izigameko zokuntshontshwa kwe-inthanethi. I-Coveware izohlinzeka ngama-forensics kanye nekhono lokulungisa…
Ukulungiswa okuqagelayo kuguqula umkhakha kawoyela negesi, ngendlela emisha nesebenzayo yokuphatha izitshalo.…
I-CMA yase-UK ikhiphe isexwayiso mayelana nokuziphatha kwe-Big Tech emakethe yezobunhloli bokwenziwa. Lapho…
Isinqumo esithi "Case Green", esakhiwe yi-European Union ukuze kuthuthukiswe ukusebenza kahle kwamandla ezakhiwo, siphothule inqubo yaso yomthetho ngokuthi...
