Izingozi zokuphepha ezipheqululini ezinama-ejenti e-AI

Informatec Digital » Izinsiza » Izingozi zokuphepha ezipheqululini ezinama-ejenti e-AI

Igagasi leziphequluli ezine-ejenti ze-AI ezihlanganisiwe selifikile kakade: i-Atlas ye-OpenAI, i-Perplexity's Comet, i-Brave with Leo, i-Chrome with Gemini, i-Edge with Copilot, i-Firefox efaka izici ezihlakaniphile… nokuningi okuzayo. Lesi sizukulwane esisha sesofthiwe sithembisa ukuthi sizoyeka nje "ukuphequlula" bese siqala ukutshela umshini lokho esifuna ukwenze: sifundele, sigcwalise amafomu, siphathe ukuthenga, noma sizenzele imisebenzi eyinkimbinkimbi kuwebhu.

Kodwa-ke, lokhu kulula kunohlangothi olumnyama: isiphequluli ngokwaso siba yi-ejenti ezimele enokufinyelela ezimvumeni zakho , amaseshini avulekile, amafayela, futhi, ezimweni eziningi, idatha yakho yomuntu siqu noma yenkampani. Uma kukhona okungahambi kahle—kungaba ngenxa yamaphutha okuklama, ubuthakathaka bokuphepha, noma ukuhlaselwa okuhlelwe kahle—umthelela ungaba mkhulu kakhulu kunowesiphequluli sendabuko noma i-chatbot elula esekelwe efwini.

Iyini isiphequluli esisebenzisa i-AI futhi kungani sishintsha umdlalo?

Isiphequluli esisebenza nge-AI empeleni siyisiphequluli "esivamile" esinemodeli yolimi ehlanganiswe ngokujulile ebona amakhasi owavakashelayo, eqonda okuqukethwe kwawo, futhi egunyaziwe ukwenza izenzo njengokungathi ngumsebenzisi. Asifinyezi nje iwebhusayithi: singachofoza izinkinobho, sigcwalise amafomu, silande amafayela, siphathe amathebhu, noma sixhumane nezinsizakalo lapho usuvele ungene khona ngemvume.

Lokhu kwenza umehluko omkhulu uma kuqhathaniswa nokusebenzisa i-chatbot yakudala kwenye ithebhu, lapho ukopisha futhi unamathisela khona okuqukethwe: manje i-ejenti "ingaphakathi" kwesiphequluli , inokufinyelela okuqondile kumongo wakho wokuphequlula, amaseshini akho avulekile, kanye namafayela akho endawo noma ezinye izinsiza zesistimu.

Lo mklamo ukhanga kakhulu izinkampani ezinkulu zobuchwepheshe. I-Perplexity iqalise isiphequluli sayo se-Comet futhi yaze yacabanga nokuthenga i-Chrome , kuyilapho i-Google ne-Microsoft befaka amamodeli abo e-Gemini ne-Copilot ngqo ku-Chrome naku-Edge. I-OpenAI, yona, iveze i-ChatGPT Atlas njengesiphequluli sayo esisekelwe ku-Chromium, esinesendlalelo se-AI esisebenza ngezinqubo ezihlukene zokunciphisa izingozi.

Isisusa sebhizinisi sicacile: izigidi zabasebenzisi, ukusetshenziswa njalo, kanye nedatha eningi . Isiphequluli esinjengalesi sikhiqiza ukwethembeka okukhulu kwabasebenzisi ( kunzima ukushintsha iziphequluli, yingakho kuvamile ukusebenzisa iziphequluli eziningi ) futhi, ngaphezu kwalokho, sinikeza i-telemetry kuyo yonke ithrafikhi yewebhu kanye nendlela abasebenzisi abaxhumana ngayo nenethiwekhi, into ewusizo kakhulu kumamodeli okuqeqesha, ukuhlola izici ezintsha, kanye nokunciphisa izindleko zengqalasizinda ngokuhambisa omunye womsebenzi kumadivayisi abasebenzisi.

I-OpenAI Atlas: Izici Eziyinhloko kanye Nemingcele Yokuphepha Yokuqala

Ngo-Okthoba 21, 2025, i-OpenAI ikhiphe inguqulo yokuqala yomphakathi ye -ChatGPT Atlas ye-macOS , isiphequluli sayo esisebenza nge-AI, ngokusho kwenkampani, esizonwetshwa siye ku-Windows, Android, kanye ne-iOS. Umqondo uwukuthi ukuzulazula "kube" yi-ChatGPT: umsebenzisi akasadingeki ukukopisha nokunamathisela umbhalo ukuze imodeli iwucubungule, kodwa kunalokho ucela ngqo imisebenzi ekhasini abalibukayo.

Ukuze kuncishiswe izingozi, i-OpenAI isebenzise imikhawulo eminingana yobuchwepheshe eyenzelwe ngqo i-ejenti yesiphequluli . I-Atlas isekelwe ku-Chromium, kodwa i-AI isebenza ngezinqubo ezihlukene, ezifakwe kulokho abakubiza ngokuthi i-OpenAI Web Layer (OWL). Ngokusho kwemibhalo yokuqala, i-ejenti:

• Awukwazi ukusebenzisa ikhodi noma ukufaka izandiso. ngokwakho
• Ayinawo amandla okwenza lokho landa amafayela ngokuzimela noma finyelela kwezinye izinhlelo zokusebenza zasendaweni.
• Akaboni futhi akasebenzisi amaphasiwedi agciniwe noma idatha yokuqedela ngokuzenzakalela isiphequluli.
• Awukwazi ukufinyelela Ulwazi lwedivayisi olungaphandle kobubanzi balo ukuzulazula.

Ngokuphathelene nobumfihlo, i-OpenAI ithembisa indlela "yobumfihlo ngokuzenzakalelayo": ukusebenzisana ne-Atlas akusetshenziselwa ukuqeqesha amamodeli ngaphandle kokuthi kugunyazwe ngokusobala ngumsebenzisi. Ngaphezu kwalokho, inkumbulo yesiphequluli—ikhono lokukhumbula amakhasi nezimo—ikhutshazwa ngokuzenzakalelayo futhi kumele inikwe amandla ngesandla.

Umsebenzisi angasusa nomlando nezinkumbulo , achaze ukuthi yimaphi amawebhusayithi i-ChatGPT engasebenzisana nawo, noma avimbele ngokuphelele i-AI ngokuchofoza okukodwa. Kumasayithi abalulekile njengamabhange noma izinsizakalo ezibucayi, i-Atlas idinga ukuqinisekiswa ngesandla ngaphambi kokuba i-ejenti ithathe isinyathelo , ukuze kuvinjelwe ukuthengiselana okungafuneki noma ukudluliselwa kwezimali okuzenzakalelayo.

Lezi zivikelo ziyisinyathelo esiqonde ngqo, kodwa aziyisusi inkinga eyisisekelo: isiphequluli esisebenza nge-AI sinamandla amakhulu kunesiphequluli sakudala , futhi ubuthakathaka obubodwa, ukungalungiselelwa kahle, noma isinqumo sokuklama kwanele ukusiguqula sibe yi-vector yokuhlasela enenzuzo enkulu.

Izingozi ezithile zokuphepha zeziphequluli ezinama-ejenti e-AI

Ukuhlanganiswa kwe-AI kudala indawo entsha yokuhlasela. Asikhulumi nje ngamaphutha esiphequluli esidala, kodwa futhi namaphutha ekusebenzisaneni phakathi kwemodeli yolimi, okuqukethwe kwewebhu, kanye nezenzo ezenzakalelayo . Lezi yizingozi ezibaluleke kakhulu ezitholakele kuze kube manje.

Umjovo osheshayo kanye nemiyalelo efihliwe

Lokhu okubizwa ngokuthi i-prompt injection kuhlanganisa ukufihla imiyalelo enonya ngaphakathi kwekhasi lewebhu uqobo, ezingcezwini zombhalo ezibonakala zingenangozi, ku-HTML efihliwe, kumazwana eforamu, kumathegi okumaka ngisho nasezithombeni ezicutshungulwa nge-OCR.

Uma umsebenzisi ecela umenzeli ukuthi afingqe, ahlaziye, noma asebenzisane nalelo khasi, imodeli ingachaza leyo miyalelo njengemiyalo yomsebenzisi esemthethweni futhi iyisebenzise ngawo wonke amalungelo eseshini yayo: ukuzulazula kuma-URL amasha, ukuchofoza izinkinobho, ukukopisha idatha yangasese, ukugcwalisa amafomu, noma ukuthumela ulwazi olubucayi.

Isibonelo esisebenzayo sibhalwe yi-Brave with Perplexity Comet: umsebenzisi uvula intambo ye-Reddit efihla, ngaphakathi kombhalo "we-spoiler", isethi yemiyalo enonya . Uma umsebenzisi echofoza inkinobho ethi "ikhasi lokuqalisa kabusha" nge-ejenti evuliwe, isiphequluli:

• Finyelela izilungiselelo ze-akhawunti ye-Perplexity bese uthola i-imeyili yomsebenzisi.
• Lingisa ukungena ngemvume kwe- dala ikhodi ye-OTP Ukuqinisekiswa
• Vula i-Gmail (uma usuvele ungene ngemvume), thola umlayezo onekhodi bese uwufunda.
• Thumela ku-Reddit thread uqobo i-imeyili yomsebenzisi kanye ne-OTP, bekushiya emehlweni omhlaseli.

Konke lokhu kwenzeka kuphela ngombhalo, ngaphandle kwesidingo sezikripthi ezinonya noma ukuxhaphaza okuphansi , futhi ngokuchofoza okukodwa okuvela kumsebenzisi. Okuqukethwe kwekhasi "okuhlelela" umenzeli ukuthi asebenzise kabi izimvume zakhe.

Umjovo webhodi lokunamathisela kanye ne-OCR

Enye i-vector ethola ukuthandwa ukukhohliswa kwebhodi lokunamathisela . I-malware yendabuko isivele iyisebenzisa ukushintsha amakheli e-cryptocurrency noma ukuvimba iziqinisekiso; esimweni sesiphequluli esisebenza nge-AI, i-ejenti ingafunda lokho okukopishayo futhi yenze ngokufanele, noma inamathisele okuqukethwe okuguquliwe ngaphandle kokuqaphela.

Okuyinkimbinkimbi nakakhulu ukusetshenziswa kwemiyalelo efihliwe ezithombeni ezicutshungulwa yi-OCR . Isiphequluli singafunda umbhalo ofakwe kuzithombe noma izithombe-skrini, futhi uma umhlaseli efihla imiyalo ngamaphikseli angabonwa yiso lomuntu kodwa hhayi uhlelo lokubona, i-ejenti ingayisebenzisa njengokungathi iyingxenye esemthethweni yekhasi.

Kuzo zombili izimo, inkinga iyafana: imodeli ayihlukanisi kahle phakathi kokuqukethwe "okuchazayo" nokuqukethwe "okufundisayo" , futhi ivame ukulalela noma yini ezwakala sengathi i-oda, ikakhulukazi uma ifomethwe ngendlela ekholisayo.

Ukudalulwa nokusetshenziswa kabi kwedatha yomuntu siqu neyimfihlo

Iziphequluli ezisebenzisa i-AI zivame ukufaka izici zokugcwalisa ngokuzenzakalela noma zokuphatha amafomu ezidlula ukuqedela ngokuzenzakalela okuvamile. Uma ikhasi elinonya lilingisa ifomu elisemthethweni (isibonelo, elivela ebhange, kunethiwekhi yezenhlalo, noma kumhlinzeki we-SaaS) futhi i-ejenti ikholelwa ukuthi kufanele "ikusindise isikhathi," ingafaka idatha efana negama lakho eligcwele, ikheli, inombolo yocingo, noma ngisho nezihlonzi zenkampani yangaphakathi ngaphandle kokucela ukuqinisekiswa ngokusobala.

Ngaphezu kwalokho, uma isiphequluli sinemvume yokubuka wonke amathebhu avuliwe noma akamuva , ingozi iyaphindaphindeka: umhlaseli angasebenzisa ngesikhathi esisodwa ulwazi oluvela ku-imeyili yakho, ikhalenda, oxhumana nabo, i-CRM yenkampani, noma uhlelo lokuphatha iphrojekthi ukuze aqedele umsebenzi, ngaleyo ndlela andise ububanzi bokwephulwa kwedatha. Nakuba lokhu kungaba wusizo ngezinjongo ezisemthethweni, ezandleni zomhlaseli kuba umthombo wedatha ebucayi.

Ezindaweni zebhizinisi, inkinga iba yimbi kakhulu. Izandiso zesiphequluli ezisebenzisa i-AI, ezihlanganiswe emisebenzini yokusebenza, zivame ukucela izimvume ezibanzi kakhulu : ukufunda nokushintsha okuqukethwe kwawo wonke amawebhusayithi, ukufinyelela amakhukhi nezikhathi, ukusebenzisana nama-API, njalo njalo. Le nhlanganisela ilungele ukweba impahla yobuhlakani, imibiko yezezimali, imiklamo eqhubekayo, ikhodi yomthombo, noma izingxoxo eziyimfihlo.

Izikhathi eziqhubekayo kanye nokuntshontshwa kwama-akhawunti

Isiphequluli sesimanje sigcina amaseshini amaningi aqinisekisiwe evulekile : i-imeyili, isitoreji samafu, amanethiwekhi omphakathi, ibhange eliku-inthanethi, amathuluzi omsebenzi, amaphaneli okuphatha… Uma i-ejenti inamandla okuxhumana nazo zonke lezi zinsizakalo futhi, ngaphezu kwalokho, iseshini ayiphelelwa yisikhathi noma isetshenziswa kabusha unomphela, umhlaseli angasebenzisa noma yikuphi ukujova okuphumelelayo ukwenza izenzo zochungechunge ngaphandle kokudinga ukuyekethisa umsebenzisi kabusha.

Lokhu kuvula umnyango wokuhlaselwa kweseshini kanye nokunyakaza okuseceleni : uma i-ejenti isikhohlisiwe, ingalanda noma isuse amafayela, ishintshe izilungiselelo, ingeze okhiye be-SSH ezindaweni zokugcina, iguqule amarekhodi ku-CRM, noma ivume ama-oda okuthenga, konke kusukela kumongo wezinsizakalo ezahlukene lapho umsebenzisi esevele engene khona ngemvume.

Ubugebengu bokweba imininingwane eyimfihlo, ulwazi olungesilo iqiniso, kanye nokuntuleka kokucaca kwezenzo

Iziphequluli ezisebenzisa i-AI nazo zisengozini enkulu yemikhankaso yobugebengu bokweba imininingwane eyimfihlo "engabonakali" . I-ejenti ye-AI ngokwayo ingasetshenziswa ukuvakashela amakhasi okukhwabanisa, ukugcwalisa amafomu okuqinisekisa, noma ukuqinisekisa ukuthengiselana ngaphandle kokuthi umsebenzisi abone isikhombikubona sangempela, kuphela isifinyezo "esinobungane" esikhiqizwe yimodeli.

Ngaphezu kwalokho, izinhlelo eziguqula okuqukethwe ngaphambi kokukubonisa kumsebenzisi zenza kube nzima ukuthola amasayithi angamanga: I-AI ingathambisa izimpawu zesixwayiso, ihlele kabusha izinto, futhi ikhiqize nezincazelo ezinengqondo ezenza kube nzima ukuhlukanisa iwebhusayithi esemthethweni nekhophi eyenzelwe ukweba iziqinisekiso.

Enye ingozi engabambeki kangako kodwa engathi sína ngokulinganayo ukuthambekela kolwazi olungelona iqiniso nokuqukethwe okushintshiwe . Imodeli engahlelwanga kahle, noma umuntu oqeqeshwe ngedatha enobuthi, angenza imithombo ebandlululayo ibe phambili, ashiye izixwayiso, noma akhiqize izimpendulo eziqinisa izindaba ezingamanga. Uma isiphequluli sisetshenziswa kakhulu ekuqoqweni kwezindaba, umthelela emibonweni yeqiniso noma ezinqumweni ezibucayi (isib., ukutshalwa kwezimali, impilo, ezombusazwe) ungaba mkhulu kakhulu.

Kuzo zonke lezi zimo, kukhona inkinga evamile: ukungabonakali kwezenzo zomenzeli . Ukusebenzisana okuningi kwenzeka "ngaphansi kwe-hood," ngaphandle kokuba umsebenzisi abone ukuthi yimaphi amathebhu avuliwe, ukuthi iyiphi idatha ekopishiwe, noma ukuthi yiziphi izinkinobho ezicindezelwe. Lokhu kwenza ukuhlolwa, ukwabela umthwalo wemfanelo osemthethweni (ingabe kwakungumsebenzisi noma i-AI?), kanye nokuthola ukuziphatha okungavamile kusenesikhathi kube nzima.

I-Omnibox, imiyalo efihliwe, kanye nokusetshenziswa kwe-zero-click

I-omnibox, noma ikheli elilodwa kanye nebha yokusesha, isici seziphequluli zanamuhla, senezela ungqimba lwayo lwengozi. Esimweni esilawulwa yi-AI, umyalo ononya ungafihlwa njenge-URL engenacala noma umbuzo wokusesha . Uma imodeli ihlaziya lokho okuthayiphayo futhi izama "ukukusiza" ngokwenza izenzo ngqo, ingagcina ikuqondise kumasayithi alawulwa abahlaseli noma isebenzisa imisebenzi engafuneki.

Okwamanje, ukuqhuma kobuthakathaka be -zero-click ezindaweni zemvelo ze-AI kubonisa ukuthi ngezinye izikhathi ukusebenzisana okuqondile akudingekile: ​​ukuthola nje i-imeyili, isimemo sekhalenda, noma umlayezo epulatifomu ehlanganiswe ne-ejenti kwanele ukuqala i-logic enonya. Izibonelo ezifana ne-EchoLeak ku-Microsoft 365 Copilot zibonisa ukuthi, ngokusebenzisa ukulayishwa kwezithombe okuzenzakalelayo, ukuhumusha kwe-Markdown, kanye nama-proxies angaphakathi , umhlaseli angakhulisa amalungelo futhi akhiphe idatha ngaphandle kokungenelela komsebenzisi.

Lolu hlobo lokuhlasela alugcini nje emahhovisi amakhulu: noma yisiphi isiphequluli esine-ejenti ye-AI evezwe ku-webmail, imiyalezo, noma amaphaneli okuphatha singaba yisisulu uma imodeli ixhumana ngokuzenzakalelayo nokuqukethwe okutholakele.

Izandiso ezisebenzisa i-AI, uchungechunge lokuhlinzeka, kanye ne-Shadow AI

Ngaphandle kweziphequluli "ezisemthethweni" ezine-AI eyakhelwe ngaphakathi, kukhona ukwanda kwezandiso zesiphequluli ezisebenza ngamamodeli olimi athembisa ukukhiqiza okusheshayo: abasizi bokubhala, abahloli bemizwa, ukuqedela okuzenzakalelayo okuthuthukisiwe, izifinyezo zemihlangano, njalo njalo. Lezi zandiso zivame ukucela izimvume eziphazamisayo kakhulu futhi, ezimweni eziningi, zithumela idatha kumasevisi angaphandle angacacile.

Lesi simo sidala ukuxubana okukhulu kobuthakathaka bochungechunge lokuhlinzeka : imitapo yolwazi yangaphandle, ama-API angakahlolwa, amaseva okucabanga atholakala kwamanye amazwe, izibuyekezo ezizenzakalelayo ezingalawulwa... Isandiso esisemthethweni ekuqaleni singathengwa ngumuntu ononya futhi sibuyekezwe buthule ukuze kuqoqwe idatha ebucayi noma kutshalwe i-malware ezindaweni zezinkampani.

Konke lokhu kwenziwa ngcono yi -Shadow AI , okungukuthi, ukusetshenziswa kwala mathuluzi ngaphandle kwemvume noma ukubonakala komnyango we-IT. Izisebenzi ezinezinhloso ezinhle ezifaka izandiso ze-AI ukuze zisebenze ngokushesha zingase zithumele ulwazi oluvikelwe yi-GDPR, i-HIPAA, noma i-PCI DSS ngokungazi kubahlinzeki bezinkampani zangaphandle abangahlangabezani nezindinganiso zomthetho noma zokuphepha ezidingekayo.

Umthelela kubumfihlo, ukuthobela imithetho kanye nobugebengu be-inthanethi

Umphumela oqondile walesi simo esisha ukwanda okukhulu kwezingozi zobumfihlo bomuntu siqu kanye nobumfihlo bebhizinisi . Isiphequluli se-AI esiphethwe kabi noma isandiso singathwebula izincwadi, izihloko zesayensi ezisekelwe kokubhaliselwe, imibiko yezezimali ebhaliwe, izinhlelo zamasu, ikhodi yobunikazi, noma idatha yamakhasimende ngaphandle kokuthi othile aqaphele.

Ekhaya, sesivele sibone izinkinga ezithile: I-ChatGPT ibonise ngisho nezingcezu zezingxoxo zabanye abasebenzisi ngenxa yephutha lobuchwepheshe, futhi isici sokwabelana ngengxoxo sakha ama-URL angafakwa ohlwini yizinjini zokusesha, okushiya izinkulungwane zezingxoxo zangasese zibonakala kunoma ubani. Kulula ukucabanga ukuthi yini engenzeka uma umsizi enokufinyelela okugcwele kuthrafikhi yewebhu namafayela endawo.

Ezweni lezinkampani, ukudalulwa kwempahla yobuhlakani noma idatha yokulawula kungaholela ezinhlawulweni zezigidi zamaRandi, ukulahlekelwa yinzuzo yokuncintisana, kanye nomonakalo wedumela . Uma isandiso sokubhalwa komhlangano sigcina izingxoxo namakhasimende abucayi kumaseva ezinkampani zangaphandle, noma uma isiphequluli se-AI sisebenzisa ulwazi lwezezimali oluyimfihlo ukuqeqesha amamodeli, inhlangano ingase yephule imithetho ye-GDPR, CCPA, HIPAA, noma eminye imikhakha.

Ngesikhathi esifanayo, izigebengu ze-inthanethi zisebenzisa wona kanye amakhono afunwa amabhizinisi. I-ransomware esizwa yi-AI kanye ne-malware ye-polymorphic sekuvele kubonwe endle: imindeni efana ne-PromptLock ikhiqiza izikripthi ngokushesha ukuze idlule ezinhlelweni ze-Windows, macOS, noma ze-Linux, igweme ukutholwa, futhi ibhale ngemfihlo idatha. Lokho okubizwa ngokuthi “ama-LLM amnyama”—amamodeli avulekile aguqulwe ngezinhloso zobugebengu—avumela ukuzenzekela kokuhlola, ukwebiwa kweziqinisekiso, kanye nemikhankaso yonke yokungena.

Ucwaningo lwakamuva lubonise ukuthi ama-LLM afakwe emisebenzini yezinkampani angaphoqwa ukuthi afake i-malware, akhiphe idatha, noma aguqule imiphumela ngokusebenzisa izixwayiso eziklanywe kahle noma idatha yokuqeqesha enobuthi. Njengoba iziphequluli ezisebenzisa i-AI ziba yi-interface eyinhloko yewebhu, lolu songo luzosabalala cishe kuyo yonke imisebenzi eku-inthanethi.

Izindlela zokunciphisa ziqalisiwe kanye nezincomo zokunciphisa izingozi

Abathengisi besiphequluli se-AI sebeqala ukusabela. Ngaphezu kwemikhawulo esivele ixoxwe ngayo ku-Atlas, i-Brave iphakamise izindlela ezithile ezinhle zokunciphisa ukuhlaselwa okusheshayo kokujova kanye nokusetshenziswa kabi kwama-agent:

• Hlukanisa ngokucacile imiyalo yomsebenzisi kumbhalo wewebhuukugwema ukuxuba kokubili esiteshini esifanayo semiyalelo nemodeli.
• Ukudinga Ukuqinisekiswa okucacile kwanoma yisiphi isenzo esibucayi (ukuthenga, izinguquko zokucushwa, ukudluliselwa kwemali, ukufinyelela kuma-imeyili).
• Hlukanisa imisebenzi ethuthukisiwe ye-ejenti yokuzulazula eyisisekelongakho ukuvakashela ikhasi nje akusho ukuthi umuntu uyakwazi ukwenza izinto ngokuzenzakalelayo.
• Zama konke okuqukethwe kwewebhu njengokungathembekile ngokuzenzakalelayo, edinga izimpawu ezicacile zenhloso yomsebenzisi ngaphambi kokwenza izenzo.

Okwamanje, ezinye izinkampani, njenge-OpenAI, zinikeza ukubonakala okuhlanganisiwe kanye nokulawula imemori : ukuvimba i-AI ezizindeni ezithile, ukukhubaza ukusetshenziswa kwedatha yokuqeqesha ngokuzenzakalelayo, okuvumela umsebenzisi ukuthi asuse izingxoxo nezinkumbulo noma nini lapho efuna, noma okudinga ukuvunyelwa ngesandla kumasayithi abhekwa njengabalulekile.

Ngokombono wenkampani yomsebenzisi, kunezindlela eziningi zokuhlanzeka eziyisisekelo okufanele ziqaliswe ngokushesha okukhulu:

• Inqubomgomo ecacile yokusebenzisa i-AI: yiziphi iziphequluli nezandiso ezivunyelwe, iyiphi idatha ezingayicubungula futhi ngaphansi kwaziphi izimo.
• Ukuqeqeshwa okuqhubekayo kwabasebenzi, kuchaza Izingozi zokufakwa ngokushesha, ubugebengu bokweba imininingwane ebucayi obuqondiswe ku-AI, kanye nokwephulwa kwedatha.
• Ukusetshenziswa izindawo ezingokoqobo noma ezihlukanisiwe (VDI, amabhokisi esihlabathi) emisebenzini enobungozi obukhulu, ukuze kube lula ukuyilawula.
• Ukuqaliswa kwe izixazululo zokuphepha ezigxile kusiphequluli eziqapha izandiso, ukuziphatha okungajwayelekile, kanye nokukhishwa kwedatha.

“Isiphequluli se-AI esifanele,” ngokusho kochwepheshe, kufanele sivumele abasebenzisi ukuthi banike amandla noma bakhubaze ukucubungula okuhlakaniphile ngesisekelo sesayithi ngalinye ngokuchofoza okukodwa, bahlukanise umongo wemodeli kuzo zonke izizinda, bahlale beqinisekisa ukungena kwedatha ebucayi, banciphise ukufinyelela kwamafayela ngesistimu yokusebenza, futhi bavumele abasebenzisi ukuthi bakhethe amamodeli endawo ngaphandle kokuthumela idatha efwini. Akukho mikhiqizo yamanje ehlangabezana nazo zonke lezi zidingo, okwenza kube kubalulekile ukuzinezela ngezendlalelo zokuphepha zangaphandle.

Isimo esivezwa yiziphequluli ezinama-ejenti e-AI sinamandla njengoba sibucayi: uma siphathwa kahle, singavumela ukwenziwa kwe-automation okuthuthukisiwe kube yintando yeningi empilweni yansuku zonke; uma singanakwa, singaba ithuluzi elikhethwa ngabahlaseli . Isihluthulelo sizoba ukuthi izinyathelo zobuchwepheshe zishintsha kanjani (ukuhlukaniswa, ukulawulwa komsebenzisi, ukutholwa komjovo) nokuthi izinhlangano nabantu ngabanye batshala malini ekuqondeni kahle ukuthi lo "msizi" ohlala ngaphakathi kwesiphequluli sakhe wenzani.