Kuyini i-opacity noma ibhokisi elimnyama ku-AI futhi kungani kubalulekile?

La lokho okubizwa ngokuthi “ibhokisi elimnyama” lobuhlakani bokwenziwa Sekungenye yezihloko eziphikisana kakhulu njalo uma sikhuluma ngazo ama-algorithms enza izinqumo kithi. Sithembele ezinhlelweni ezincoma imithi, izimali mboleko zesibonelelo, noma ama-resume okuhlunga… kodwa ngokuvamile Asazi ukuthi kungani befinyelela kulezo zinqumongisho noma zithinta ngqo amalungelo ethu.

Lokhu kuntuleka kokubonakala akuyona nje inkinga yobuchwepheshe: Inemiphumela yokuziphatha, yezomthetho, yezenhlalo kanye neyebhizinisi.Yingakho kunezinkulumo eziningi kangaka mayelana nokungacaci kwe-algorithmic, ukuchazeka (i-XAI), kanye nemithetho emisha efana ne-European AI Act, ehlose ngqo ukuletha ukuhleleka kule ndawo. Ake sibheke lokhu ngokuthula kodwa ngokuningiliziwe. Kuyini ngempela i-opacity noma "ibhokisi elimnyama" ku-AI?Kungani kuvela, yiziphi izingozi ezikubangelayo, nokuthi kwenziwa kanjani imizamo yokuvula lelo bhokisi ngaphandle kokulahlekelwa izinzuzo zobuchwepheshe.

Kusho ukuthini ukuthi “ibhokisi elimnyama” kanye ne-opacity ku-AI?

Ngokwesimo sobuhlakani bokwenziwa, “Ibhokisi elimnyama” uhlelo olunezinqubo zangaphakathi ezingaqondakali kahleSiyazi ukuthi idatha ingena kuphi nokuthi umphumela uphuma kanjani, kodwa "indlela" ephakathi ayiqondakali noma ayifinyeleleki kubantu, ngisho nakubathuthukisi abaningi.

Lesi simo sihlotshaniswa kakhulu ne- amamodeli okufunda komshini ayinkimbinkimbi, njengamanethiwekhi ajulile e-neuralezisebenza nezinkulungwane noma izigidi zamapharamitha asatshalaliswa ezingqimbeni eziningi. Ngokungafani ne-algorithm yakudala esekelwe emithethweni esobala, lapha imodeli ifunda kokuhlangenwe nakho, ilungisa izisindo zangaphakathi ukuze Akekho ongakwazi ukulandelela ngesandla ukuthi iyiphi inhlanganisela eqondile yama-neurons eholele empendulweni ethile.

Ukungacaci kahle kungavela ngezindlela ezimbili ezahlukene kodwa ezihambisanayo: ngakolunye uhlangothi, ngoba Inkampani inquma ukungayivezi ikhodi noma imininingwane yomodeli. (ukuvikela impahla yabo yobuhlakani noma isu lokuhweba kuphela); ngakolunye uhlangothi, ngoba Ubunzima bezibalo kanye nezibalo obukhona ngaphakathi kwenza ukuhumusha komuntu okunembile kube nzima kakhulu.noma ikhodi ingumthombo ovulekile.

Kulesi simo sesibili, sivame ukukhuluma ngokuthi “amabhokisi amnyama e-organic"Ngisho nabadali bohlelo abakwazi ukuchaza ngokunembile ukuthi yimaphi amaphethini angaphakathi i-AI ewafundile noma ukuthi iwahlanganisa kanjani esinqumweni ngasinye. Ngamamodeli okufunda okujulile, lokhu kuyinto evamile, hhayi okuhlukile."

Uma sibhekene nalezi zinhlelo, singabona ngokucacile kuphela izingcingo Izendlalelo ezibonakalayo: isendlalelo sokufaka kanye nesendlalelo sokukhiphaSibona idatha efakiwe (izithombe, umbhalo, iziguquguquko zezinombolo) kanye nezibikezelo noma ukuhlukaniswa okuvelayo (okuvunyelwe/okunqatshiwe, ukuxilongwa, isincomo, njll.). Kodwa kwenzekani ku-multiple izendlalelo eziphakathi ezifihliwe Ngokwezinga elikhulu, kusalokhu kungaphezu kwamandla ethu okuqonda.

Indlela amamodeli e-black box asebenza ngayo: amanethiwekhi e-neural kanye nokufunda okujulile

Ukuze uqonde ukuthi lokhu kungabonakali kuvelaphi, kuyasiza ukubuyekeza, noma ngabe kukhulunywa kakhulu, Amamodeli okufunda okujulile ahlelwe kanjani?Esikhundleni sefomula eyodwa elula, lezi zinhlelo zakhiwe ngamanethiwekhi e-neural anezendlalelo eziningi (ngezinye izikhathi amakhulu) kanye nenani elikhulu lama-neurons kungqimba ngayinye.

I-neuron ngayinye ngokuyisisekelo ibhlogo elincane lekhodi elithola okokufaka, lisebenzise ukuguqulwa kwezibalo, futhi likhiqize umphumelaInqubo yokufunda ihilela ukulungisa, ngezigidi zezibonelo, izisindo kanye nemingcele yazo zonke lezi zinzwa ukuze uhlelo lunciphise amaphutha okubikezela. Inkinga iwukuthi, uma seziqeqeshwe, umphumela uba inethiwekhi enkulu yemingcele Akuhambelani nemibono yabantu ecacile nehlukene.

Lolu hlobo lwenethiwekhi lungangena inani elikhulu ledatha eluhlaza (izithombe, umsindo, umbhalo wamahhala, idatha yenzwa) bese bebona amaphethini obunzima obukhulu: ubudlelwano obungewona oqondile, inhlanganisela yezici ezicashile kakhulu, ubudlelwano obuphikisana nomuzwa wethu. Ngenxa yalokhu, bayakwazi ukuhumusha izilimi, ukukhiqiza izithombe, ukubhala imibhalo ehambisanayo, noma ukuhlaziya ngokunembile ama-X-ray okufana nokwabachwepheshe.

Kodwa lawo mandla afika ngentengo: izethulo zangaphakathi abazidalayo (isibonelo, odumile ukushumeka kwe-vector) yizakhiwo zezinombolo ezinobukhulu obuphezulu ezithi Azivumelani ngqo nezigaba ezilula zabantuSingabona ukuthi ama-vector athile ahlanganisa izincazelo ezifanayo noma ukuthi ama-neurons athile asabela emaphethini athile, kodwa imephu ephelele cishe ayinakulawuleka.

Ngisho noma imodeli ivulekile futhi singabona yonke imigqa yokuhlela, Lokho akusho ukuthi singachaza yonke imininingwane ngokuningiliziwe.Kungenzeka ukulandelela ukuthi idatha igeleza kanjani phakathi kwezingqimba kanye nokuthi yimiphi imisebenzi esetshenziswayo, kodwa akunakwenzeka ukuqonda ukuthi kungani inhlanganisela ethile yezigidi zamapharamitha iholela "ekuvunyelweni" komuntu oyedwa kanye "nokunqatshwa" komunye.

Ngokufingqa, Ibhokisi elimnyama alibangelwa nje kuphela yimfihlo yenkampaniFuthi kungumphumela wokukhetha izakhiwo eziyinkimbinkimbi kakhulu ezithuthukisa ukunemba, kodwa ezilahla ukuhunyushwa.

Ukungacaci, ubandlululo, kanye nokubandlulula: lapho ibhokisi elimnyama libangela ingozi

Ukuntuleka kokubonakala obala akuyona nje inkinga engokwemfundiso. Ukungacaci kwe-algorithm kungaholela ezinqumweni ezingafanele, ezibandlululayo, noma ezingalungile ngokuphelele.ngaphandle kokuba nendlela ecacile yokuthola inkinga noma ukuyilungisa ngesikhathi.

Isibonelo esivame ukucashunwa yiphrojekthi Ubulili ShadesnguJoy Buolamwini noTimnit Gebru, abahlaziye izinhlelo ezahlukene zokuqaphela ubuso kwezentengiselwano. Ucwaningo lukhombisile ukuthi Amazinga amaphutha ayephezulu kakhulu lapho kutholakala abesifazane abanesikhumba esimnyama. ukuthi lapho kuhlonzwa amadoda anesikhumba esikhanyayo: kwezinye izimo, amaphutha angaphezu kuka-34% uma kuqhathaniswa nangaphansi kuka-1% eqenjini eliphathwe kahle kakhulu.

Ngokusekelwe emiphumeleni iyonke, lezi zinhlelo zibonakala zisebenza kahle. Kodwa-ke, ngokusekelwe emiphumeleni iyonke, lezi zinhlelo zibonakala zisebenza kahle. hlukanisa amaphutha ngobulili kanye nesikhumba Ukungalingani okukhathazayo kakhulu kwavela. Leso ngesinye sezingibe zebhokisi elimnyama: Amaphutha amakhulu angafihlwa ngokwesilinganiso futhi ungaqashelwa uma kungekho muntu ohlolisisa imiphumela ngokucophelela.

Lolu hlobo lokubandlulula ngokuvamile aluhlosiwe. I-AI ifunda kudatha esiyinikezayo, futhi uma leyo datha ibonisa ukungalingani komlando noma imele amaqembu athile kancane, Le modeli iphinda futhi ikhulise lokhu kungabi nabulungisa ngaphandle kokuthi othile "ayiyale" ngokusobala ukuba yenze kanjalo.Futhi ngenxa yokuthi akucaci, ukuthola ukuthi yiziphi izinto eziguquguqukayo noma izinhlanganisela ezidala ubandlululo kuba umsebenzi oyinkimbinkimbi kakhulu.

Ukungacaci kwenza kube nzima futhi ukuhlonza amaphutha noma ubuthakathaka obuhlelekileUma singazi ukuthi imodeli "icabanga kanjani," kunzima kakhulu ukubikezela ukuthi yiziphi izinhlobo zokufaka ezingase ziholele "ekuboneni izinto ezingekho" (ukukhiqiza izimpendulo ezingamanga kodwa ezikholisayo) noma ziwele ezingibeni zokuphikisana ezenzelwe ukuyilawula.

Konke lokhu kunomphumela owodwa ocacile: Ukwethembana phakathi kwabasebenzisi, amakhasimende, kanye neziphathimandla kuyaphela.Uma umuntu ebhekana nesinqumo esibi esisekelwe ku-AI futhi kungekho muntu ongachaza ngokucacile ukuthi yiziphi izici ezicatshangelwe, kuvamile ukuba kuvele ukungabaza mayelana nokulunga kanye nokuba semthethweni kohlelo.

Umthelela wokuziphatha, wezomthetho kanye nowesibopho

Ngokombono wezomthetho, ibhokisi elimnyama lidala inkinga enkulu: Kuyinkimbinkimbi ukufakazela ukuxhumana okuyimbangela phakathi kohlelo lwe-AI kanye nomonakalo owenzekileUkuze kuqinisekiswe icala lomphakathi, inhlanganisela yomonakalo, ukuziphatha okunecala noma okunephutha, kanye nesixhumanisi sembangela ngokuvamile kuyadingeka. Uma isinqumo sisekelwe kumodeli engacacile, lesi sici sesithathu siba esingazinzile.

Ezweni le-analog, ukuxoxa ngokuxoshwa, ukwenqatshwa kwesikweletu, noma isihlungi sokufinyelela kwenziwa ngokubukeza imibhalo, izindlela zokuhlola, ofakazi kanye nezisusa ezicacileNgamamodeli e-AI, izendlalelo zokuphetha okunzima ukuzakha kabusha zihlanganiswa phakathi kwedatha yokufaka kanye nesinqumo sokugcina, esivame ukuphathwa uchungechunge lwabalingisi (umhlinzeki wemodeli, umhlanganisi, inkampani yomsebenzisi, abantu besithathu abahlinzeka ngedatha) lokho ukunciphisa ukuthi ubani olawula ini.

Ngaphezu kwalokho, kunesizathu esicacile sokugcina i-till ivaliwe: Umsebenzisi angacasha ngemuva kwemfihlo yezohwebo noma ubunzima bobuchwepheshe ukugwema ukudalula ulwazi olufanele enkantolo. Uma isisulu singakwazi ukufinyelela amarekhodi, imibhalo yobuchwepheshe, noma imikhondo yezinqumo, ukufakazela ukuthi umonakalo ubangelwa uhlelo lwe-AI kuba cishe akunakwenzeka.

Impendulo evela kumenzi wezomthetho waseYurophu inamandla: uma kungenzeki ukuchaza ngokugcwele, Umthwalo wonke wokuhlolwa awukwazi ukubekwa engxenyeni ebuthakathaka kakhulu.Ngakho-ke, sibona imithetho emisha edinga ukugcinwa kwamalogi, imibhalo yokuthi uhlelo lusebenza kanjani, ukugunyazwa kokuhlolwa kwamabhuku, kanye, ezingeni lenqubo, vula umnyango wokwethulwa kobufakazi kanye nokucatshangelwa okuvuna iqembu elilimele lapho umqhubi engabambisani.

Ake sicabangele, isibonelo, inkampani esebenzisa amathuluzi e-AI kwezabasebenzi ukuze ama-resume esikrini, amaphuzu okusebenza, noma ukuncoma ukukhushulwaNgokomthetho, isinqumo sokugcina silele kumuntu, kodwa empeleni, sincike kakhulu emibikweni ekhiqizwe yi-AI. Uma isisebenzi senqatshwa noma sixoshwa futhi singanikezwa ukufinyelela ku- Yimuphi idatha esetshenzisiwe, isisindo sayo sasingakanani, yimaphi amaphethini atholakele Noma yimiphi imibhalo yobuchwepheshe esekela uhlelo; ibhokisi elimnyama alinqumi nje kuphela: liyanquma futhi kuvimbela ngempumelelo ukuphonsela inselelo isinqumo.

I-AI echazayo kanye nokuhunyushwa kwayo: ukuzama ukuvula ibhokisi

Ukuze kuncishiswe lezi zinkinga, insimu ye I-AI Echazayo noma i-XAI (Ubuhlakani Bokwenziwa Obuchazayo)Umgomo awukona nje "ukuhumusha" umugqa nomugqa lokho okwenziwa yi-algorithm, kodwa ukunikeza izincazelo eziwusizo, eziqondakalayo, nezisebenzisekayo zokuthi kungani imodeli yenze isinqumo esithile.

Kunezindlela ezimbili eziyinhloko. Ngakolunye uhlangothi, kukhona amamodeli angahunyushwa ngokwemvelo noma amabhokisi amhlopheAma-algorithm alula njengokubuyela emuva okuqondile, izihlahla zesinqumo esingajulile, noma imithetho enengqondo ikhombisa ngokusobala ukuthi yiziphi iziguquguquko ezifakiwe, yimiphi imithetho esetshenziswayo, nokuthi umphumela ufinyelelwa kanjani. Lezi zinhlobo zamamodeli zenza kube lula ukuhlola nokulandelela, yize ngezinye izikhathi zidela ukunemba okuthile.

Ngakolunye uhlangothi, sinefayela le- amamodeli ayinkimbinkimbi (ibhokisi elimnyama) lapho kusetshenziswa khona amasu okuchaza angemuvaYilapho amathuluzi anjenge-LIME, i-SHAP, amamephu e-salience noma i-Grad-CAM eqala khona ukusebenza, okusivumela ukuthi silinganisele ukuthi yiziphi izici ezibe nesisindo esikhulu esibikezelweni esithile, noma ukubona ngeso lengqondo ukuthi yiziphi izindawo zesithombe ezibe nomthelela omkhulu ekuxilongweni.

Isibonelo, ezindaweni zezokwelapha, amasu ohlobo lwe-SHAP asetshenziswe ukwenza hlaziya amamodeli okucabanga okuxilonga nokuthola ukuthi, kwezinye izimo, uhlelo belunaka kakhulu izimpawu noma izichasiselo ku-X-ray kunokuba lubheke amaphethini omtholampilo afanele. Ukuthola lokhu kuphambuka kuvumela ukulungiswa kumodeli kanye nokunciphisa izingozi.

Ngaphezu kwalokho, ukuchazeka kunesici esibalulekile somuntu: Incazelo ayisizi ngalutho uma umuntu oyitholayo engayiqondi.Udokotela akanazo izidingo ezifanayo nonjiniyela wedatha, ijaji alinazo izidingo ezifanayo nesiguli noma ikhasimende lasebhange. Yingakho sisebenza ngendlela ehlanganisa imikhakha eminingi, sihlanganisa ubuchwepheshe kanye ne-cognitive psychology kanye nomklamo we-interface ukuze sivumelanise incazelo nephrofayili yomuntu oyitholayo.

Ibhokisi elimnyama vs ibhokisi elimhlophe vs i-AI echazekayo: zihluke kanjani?

“Ibhokisi elimhlophe”, “ibhokisi elimnyama” kanye “ne-AI echazwe kahle” zivame ukusetshenziswa ngokushintshana, kodwa azifani ncamashiKubalulekile ukucacisa amagama ngoba lokhu kudideka kudala ukungaqondani okukhulu.

Un imodeli yebhokisi elimhlophe ingabe nguyena Ukusebenza kwangaphakathi kusobala futhi kuyaqondakalaKulula ukubona ukuthi yiziphi iziguquguquko ezihilelekile, ukuthi zihlangana kanjani, yimiphi imithetho esebenzayo, nokuthi okokufaka kuba kanjani umphumela. Izibonelo ezijwayelekile yilezi: ukuhlehla okuqondile okuchazwe kahle noma izihlahla zesinqumo ezilulaLawa mamodeli ahunyushwa ngokwawo: isakhiwo sawo sesivele sisebenza njengencazelo.

Un imodeli yebhokisi elimnyamaNgakolunye uhlangothi, yileyo enengqondo yayo yangaphakathi engenakulandelwa kalula. Lokhu kuzofaka phakathi amanethiwekhi ajulile e-neural, amahlathi angahleliwe ayinkimbinkimbi kakhulu, ukukhulisa uhlobo lwe-XGBoost futhi, ngokuvamile, noma yiluphi uhlelo olunezendlalelo eziningi zemingcele okunzima ukuzihumusha zibe imithetho yabantu ecacile.

La I-AI echazekayo (XAI) Isambulela esibanzi esihlanganisa kokubili amamodeli ebhokisi elimhlophe kanye amasu asetshenziswa emabhokisini amnyama ukukhiqiza izincazelo zangemuva kwe-hocImodeli eyinkimbinkimbi kakhulu ingabhekwa njenge "echazekayo" uma ihambisana namathuluzi avumela, isibonelo, ukuhlukanisa ukubaluleka kwezinto eziguquguqukayo, ukubona amaphuzu abalulekile, noma ukudala izibonelo eziphikisanayo ("ukube umholo wakho ubungu-X kanye nobudala bakho u-Y, umphumela ubuzoshintsha").

Empeleni, izinhlangano eziningi zihlanganisa zombili izindlela: Basebenzisa amamodeli alula lapho ukucaca kudlula ukunemba (amacala alawulwa kakhulu) futhi basebenzise amamodeli anamandla kakhulu ahambisana ne-XAI lapho kudingeka bandise amandla okubikezela, kodwa ngaphandle kokuyeka ngokuphelele ukuhumusha.

Umthetho waseYurophu: Umthetho we-AI, i-GDPR kanye nesibopho somkhiqizo

I-European Union inqume ukubhekana nokungacaci kwe-algorithm kusuka ezinhlangothini eziningana. Ngakolunye uhlangothi, Umthethonqubo Wokuvikelwa Kwemininingwane Jikelele (RGPD) Sekuvele kubeka izibopho ezithile lapho kwenziwa izinqumo ezizenzakalelayo ngokusekelwe kudatha yomuntu siqu, okudinga ukuthi ulwazi "olunenjongo" mayelana nomqondo osetshenzisiwe lunikezwe ngendlela eqondakalayo eqenjini elithintekayo.

Okungeziwe kulokhu yi- Umthetho we-AI noma uMthethonqubo Wobuhlakani Bokwenziwa WaseYurophu, esisebenza kusukela ngo-Agasti 2024, esisungula uhlaka oluthile lokuthuthukiswa nokusetshenziswa kwezinhlelo ze-AI e-EU. Umthethonqubo ihlukanisa izinhlelo ngamazinga engozi, evimbela ngokuqondile labo “abanengozi engamukeleki” (njengokuthola amaphuzu emphakathini ngendlela yokubonga okukhulu emphakathini noma amasu athile okuphatha ukuziphatha ngokweqile).

Izinhlelo ingozi enkulu (isibonelo, ezinye izindlela zokwelapha ezempilo, ezezimali, izinsiza zabantu, imfundo noma ezokuphepha) zingaphansi kwezibopho eziqinile: kumele zibe nazo Imibhalo yobuchwepheshe enemininingwane, amarekhodi azenzakalelayo (ukungena ngemvume) avumela ukulandelelwa, ulwazi olucacile noluqondakalayo kubasebenzisi kanye nezindlela zokuqapha ezisebenzayo zabantu.

Ngaphezu kwalokho, uMthetho we-AI ubeka izibopho zokungafihli lutho Ezimweni ezifana nokusetshenziswa kwama-chatbot noma abakhiqizi bokuqukethwe, kuyadingeka ukuxwayisa abasebenzisi uma bexhumana ne-AI, futhi kwezinye izimo, ukumaka okuqukethwe okukhiqizwe ngokuzenzakalelayo. Eziningi zalezi zibopho zizosetshenziswa kancane kancane eminyakeni ezayo, kuqala ngamacala anomthelela omkhulu.

Okwamanje, okusha Isiqondiso (EU) 2024/2853 mayelana nesibopho semikhiqizo enephutha Ibuyekeza uhlaka lwesibopho somphakathi ukuze luvumelane nendawo lapho imikhiqizo ingaba yisofthiwe futhi lapho ukwehluleka kungavela khona ngenxa yemisebenzi yedijithali. I-Directive iyaqaphela ngokucacile ukuthi ubunzima bezobuchwepheshe kanye nesayensi bezinhlelo ze-AI futhi kwenza abahluleli bakwazi ukufuna ukwethula ubufakazi obufanele, okuhlanganisa nobufakazi bedijithali, ngendlela efinyelelekayo neqondakalayo.

Uma umqhubi ehluleka ukubambisana noma ephula izibopho zokuphepha, okulandelayo kungasebenza: ukucabanga kokungabi naphutha kanye nokuba imbangelaNgamanye amazwi, uma umuntu olimele enikeza ubufakazi obufanele futhi ummangalelwa enganikezi amarekhodi noma imibhalo eceliwe yinkantolo, umthetho uvala ukungalingani kobufakazi ngokubeka izikali eceleni ukuze kuzuze isisulu.

Lonke leli phakethe lokulawula lithumela umyalezo ocacile: Noma ngubani oletha ubunzima be-algorithm emakethe kumele athathe umthwalo wemfanelo wokukwenza kuhlolweIbhokisi elimnyama aliseyona inzuzo yokuzivikela futhi liba yingozi yokuthobela imithetho kanye nedumela.

Ukucaca, amamodeli avulekile kanye nezinselele ezisalindile

Enye indlela yokunciphisa ukungabonakali kahle ukutshala imali ku- amamodeli omthombo ovulekile kanye nemikhuba ephelele yokubhala amadokhumentiIzinhlelo ezivulekile zivumela abacwaningi, abalawuli, kanye nomphakathi wezobuchwepheshe ukuthi bahlole ikhodi, baphinde bahlole izivivinyo, futhi bathole ukucwasa noma ubuthakathaka obungaba khona.

Noma kunjalo, ngisho noma sinomthombo ovulekile, sisenenkinga eyisisekelo: ukuhunyushwa kwemingcele kanye nokumelwa kwangaphakathiUkucaca kokufinyelela akusho ukuthi ngokuzenzakalela ukucaca kokuqonda. Yingakho kugcizelelwa kakhulu ekuhlanganiseni ukuvuleleka namasu okufinyelela okuvulekile kanye nezinqubo zokuphatha ezicacile kanye nokuhlola.

Iziphathimandla nochwepheshe bagcizelela ukubaluleka ukukhuthaza isiko lokungafihli lutho kanye nokuziphendulelaGcina amarekhodi okuqeqeshwa nokusetshenziswa anemininingwane, ubhale phansi izinguquko zemodeli, uchaze izinqubo zokuqapha abantu, kanye nezixhumi zokuklama ezichaza amakhono esistimu, ukulinganiselwa, kanye nezingozi kumsebenzisi.

Umsebenzi uyenziwa amasu amasha okuhumusha, njenge-sparse autoencoders kanye nezinye izindlela ezifuna ukukhipha izinto ezifihliwe "ezihlanzekile" nezifundeka kalula kumamodeli ayinkimbinkimbi kakhulu. Umqondo uwukusondela kancane kancane ohlotsheni "lwebhokisi lengilazi," lapho ubunzima bangaphakathi buhlala khona, kodwa ngezingqimba eziqinile zencazelo.

Nokho, ochwepheshe bayavuma ukuthi Ngeke senze wonke amamodeli abe sobala ngokuphelele.Inselele yangempela ukulinganisela ukunemba, ukusebenza kahle, kanye nokuchazeka, ukugxila ekwenzeni lezo zinhlelo eziqondakalayo ikakhulukazi ezenza izinqumo ezinomthelela omkhulu emalungelweni ayisisekelo.

Ekugcineni, ukusebenza ne-AI namuhla kudinga ukucabanga ukuthi Ubudlelwano kufanele bube nokubambisana, hhayi ukungaboni.Imishini inikeza amandla okusebenzisa amakhompyutha kanye namakhono okuthola amaphethini, kodwa abantu kumele baqhubeke nokubeka izindinganiso zokuziphatha, baqinisekise imiphumela ebalulekile, futhi bafune izincazelo ezinengqondo lapho okuthile kungasebenzi kahle.

Kulesi simo, lokho okubizwa ngokuthi “i-opacity” noma umphumela webhokisi elimnyama we-AI akuyona nje inkinga yobuchwepheshe kodwa iphuzu eliyinhloko lokungqubuzana phakathi kokusungula izinto ezintsha kanye nomthethonqubo kanye nokwethembana komphakathiNjengoba imithetho yaseYurophu, amasu e-AI kanye nemikhuba yokubusa okuhle iqhubekela phambili, ibhokisi elimnyama liyayeka ukuba yimfihlakalo engafinyeleleki futhi liqala ukubonwa njengohlelo, nakuba luyinkimbinkimbi, olungakhanyiswa futhi okufanele lukhanyiselwe ngokwanele ukuze izakhamuzi, izinkampani kanye nezinkantolo zikwazi ukwethemba izinqumo zalo.