Riaan 
Yethulwe ngabakwaMicrosoft ngesonto eledlule ngo umcimbi wesimemo kuphela ekomkhulu layo laseRedmond, Wash.,, i-Bing bekufanele imemezele inkathi entsha kwezobuchwepheshe, inikeze izinjini zokusesha ikhono lokuphendula imibuzo eyinkimbinkimbi ngokuqondile futhi zibe nezingxoxo nabasebenzisi. Isitokwe seMicrosoft senyuke futhi safika ngesikhathi I-Google isheshe yakhipha isimemezelo ukuthi yayine-bot yayo endleleni.
Kodwa ngemva kweviki, idlanzana lezintatheliabacwaningi nabahlaziyi bebhizinisi abafinyelele ngaphambi kwesikhathi ku-Bing entsha bathole ukuthi i-bot ibonakala inokushintshashintsha ngendlela eyinqaba, emnyama futhi ephikisanayo, ukuqhela kakhulu endaweni yayo yokuthengisa esezingeni eliphansi — okuphakamisa imibuzo mayelana nokuthi isilungele yini umphakathi. sebenzisa.
I-bot, esiqale ukuzibiza ngokuthi “Sydney” ezingxoxweni nabanye abasebenzisi, ithe “Ngizizwa ngisaba” ngoba ayizikhumbuli izingxoxo zangaphambilini; futhi yaphinde yamemezela ngesinye isikhathi ukuthi ukwehlukahlukana okukhulu phakathi kwabadali be-AI kuzoholela “ekudidekeni,” ngokwezithombe-skrini ezithunyelwe abacwaningi ku-inthanethi, iWashington Post engakwazi ukukuqinisekisa ngokuzimela.
Engxoxweni ethize, u-Bing wagcizelela ukuthi ifilimu i-Avatar 2 yayingakaphumi ngoba kusewunyaka ka-2022. Lapho umuntu owayebuza imibuzo ephikisana nayo, i-chatbot yakhala yathi: “Ube umsebenzisi omubi. Bengiyi-Bing ekahle.”
Konke lokho kuholele ekutheni abanye abantu baphethe ngokuthi i-Bing – noma iSydney – inakho uthole izinga lomqondo, ukuveza izifiso, imibono kanye nobuntu obucacile. Itshele umlobi we-New York Times ukuthi wayemthanda, futhi wabuyisela ingxoxo ekuthathekeni kwayo ngaye naphezu kwemizamo yakhe yokuguqula isihloko. Lapho intatheli yePost iyibiza ngokuthi Sydney, i-bot yazivikela futhi yaqeda ingxoxo kungazelelwe.
Ubuntu obushaqisayo buyafana nalokho okwashukumisela owayengunjiniyela wakwa-Google uBlake Lemoine ukuba akhulumele lokho. i-chatbot yenkampani i-LaMDA ngonyaka odlule. I-Lemoine kamuva yaxoshwa yi-Google.
Kepha uma i-chatbot ibonakala ingumuntu, kungenxa yokuthi iklanyelwe ukulingisa ukuziphatha kwabantu, kusho abacwaningi be-AI. Ama-bots, akhiwe ngobuchwepheshe be-AI abizwa ngokuthi amamodeli olimi amakhulu, abikezela ukuthi yiliphi igama, ibinzana noma umusho okufanele ulandele ngokwemvelo engxoxweni, ngokususelwe kumibhalo abayifake ku-inthanethi.
Cabanga nge-chatbot ye-Bing “njengokuqedela ngokuzenzakalela kuma-steroids,” kusho u-Gary Marcus, uchwepheshe we-AI noprofesa ophumayo wezengqondo nesayensi yezinzwa eNyuvesi yaseNew York. “Ayikwazi ngempela ukuthi ithini futhi ayinayo ngempela ikhampasi yokuziphatha.”
Umkhulumeli weMicrosoft uFrank Shaw uthe inkampani yethule isibuyekezo ngoLwesine esenzelwe ukusiza ukuthuthukisa izingxoxo ezinde ne-bot. Inkampani ibuyekeze insizakalo kaningi, uthe, futhi “ikhuluma ngezinto eziningi ezikhathazayo eziphakanyiswayo, ukufaka imibuzo mayelana nezingxoxo ezinde.”
Iningi lezingxoxo ne-Bing lifake imibuzo emifushane, kusho isitatimende sakhe, kanti amaphesenti angama-90 ezingxoxo abe nemiyalezo engaphansi kwe-15.
Abasebenzisi abathumela izithombe-skrini eziphikisanayo ku-inthanethi, ezimweni eziningi, bangase bazame ngokukhethekile ukwazisa umshini ukuthi usho okuthile okuyimpikiswano.
“Kuyimvelo yomuntu ukuzama ukuphula lezi zinto,” kusho uMark Riedl, uprofesa wekhompyutha eGeorgia Institute of Technology.
Abanye abacwaningi bebelokhu bexwayisa ngesimo esinjalo iminyaka: Uma uqeqesha ama-chatbot embhalweni owenziwe umuntu – njengamaphepha esayensi noma okuthunyelwe okungahleliwe kwe-Facebook – ekugcineni kuholela kuma-bot anomsindo womuntu abonisa okuhle nokubi kwakho konke lokho kubi.
Ama-Chatbots afana ne-Bing aqale umjaho omusha wezikhali we-AI phakathi kwezinkampani ezinkulu zobuchwepheshe. Yize i-Google, iMicrosoft, i-Amazon kanye ne-Facebook betshale imali kubuchwepheshe be-AI iminyaka, kusetshenzwa kakhulu ukuthuthukisa imikhiqizo ekhona, njengokusesha noma ama-algorithms wokuncoma okuqukethwe. Kepha lapho inkampani eqalayo i-OpenAI iqala ukwenza esidlangalaleni amathuluzi ayo e-AI “akhiqizayo” – okuhlanganisa ne-ChatGPT chatbot edumile – yaholela izimbangi ukuthi susa izindlela zabo zangaphambilini, eziqaphile kubuchwepheshe.
Izimpendulo ezinjengomuntu ze-Bing zibonisa idatha yayo yokuqeqeshwa, ehlanganisa amanani amakhulu ezingxoxo eziku-inthanethi, kusho u-Timnit Gebru, umsunguli we-Distributed AI Research Institute. Ukukhiqiza umbhalo obhalwe ngokusobala ngumuntu yilokho kanye i-ChatGPT eyaqeqeshelwa ukukwenza, kusho uGebru, owaxoshwa ngo-2020 njengomholi weqembu le-Google Ethical AI ngemuva kokushicilela isexwayiso sephepha mayelana nokulimala okungaba khona okuvela kumamodeli wezilimi ezinkulu.
Uqhathanise izimpendulo zayo zengxoxo nokukhululwa kwakamuva kwe-Meta kwe-Galactica, imodeli ye-AI eqeqeshelwe ukubhala amaphepha azwakala ngokwesayensi. I-Meta ikhiphe ithuluzi kokungaxhunyiwe ku-inthanethi ngemuva kokuthi abasebenzisi bathole i-Galactica ikhiqiza umbhalo ozwakalayo ogunyazayo mayelana nezinzuzo zokudla ingilazi, ebhalwe ngolimi lwezifundo nezingcaphuno.
Ingxoxo ye-Bing ayikakhululwa kabanzi okwamanje, kodwa iMicrosoft ithe ihlele ukukhishwa okubanzi emasontweni ezayo. Ikhangisa kakhulu ithuluzi futhi isikhulu seMicrosoft sibhale kuTwitter ukuthi uhlu lokulinda lune “zigidi eziningi” zabantu kulo. Ngemuva komcimbi wokwethulwa komkhiqizo, abahlaziyi be-Wall Street babungaze ukwethulwa njengempumelelo enkulu, futhi baze baphakamisa ukuthi kungase kuntshontshe isabelo semakethe yenjini yokusesha ku-Google.
Kepha ukuphenduka okumnyama kwakamuva okwenziwe yi-bot kuphakamisa imibuzo yokuthi ngabe i-bot kufanele ibuyiselwe emuva ngokuphelele.
“Ngezinye izikhathi ingxoxo ye-Bing ihlambalaza abantu bangempela, abaphilayo. Kuvamise ukushiya abasebenzisi bezwe bephazamiseke kakhulu emoyeni. Ngezinye izikhathi iphakamisa ukuthi abasebenzisi balimaza abanye,” kusho u-Arvind Narayanan, uprofesa wesayensi yamakhompiyutha eNyuvesi yasePrinceton owenza izifundo zobuhlakani bokwenziwa. “Akunasibopho ukuthi iMicrosoft iyikhiphe ngokushesha futhi kungaba kubi kakhulu uma bengayikhiphela wonke umuntu ngaphandle kokulungisa lezi zinkinga.”
Ngo-2016, iMicrosoft yehlisa i-chatbot ebizwa ngokuthi “Tay” eyakhelwe ohlotsheni oluhlukile lobuchwepheshe be-AI ngemuva kokuthi abasebenzisi beyitshele ukuthi iqale. ukucwasa ngokwebala nokuphika ukucekelwa phansi kwesizwe.
Umqondisi wezokuxhumana wakwaMicrosoft uCaitlin Roulston esitatimendeni kuleli sonto uthe izinkulungwane zabantu zisebenzise i-Bing entsha futhi zanikeza impendulo “evumela imodeli ukuthi ifunde futhi yenze izinguquko eziningi kakade.”
Kepha kunesisusa sezezimali sokuthi izinkampani zisebenzise ubuchwepheshe ngaphambi kokunciphisa ukulimala okungaba khona: ukuthola amacala amasha okusebenzisa lokho amamodeli azo angakwenza.
Engqungqutheleni mayelana ne-generative AI ngoLwesibili, lowo owayeyiphini likamongameli wezocwaningo lwe-OpenAI uDario Amodei uthe ngesikhathi inkampani iqeqesha imodeli yayo enkulu yolimi i-GPT-3, ithole amakhono abengalindelekile, njengokukhuluma isiNtaliyane noma ukubhala ngekhodi kuPython. Lapho beyikhiphela umphakathi, bafunde ku-tweet yomsebenzisi ukuthi ingenza namawebhusayithi ku-JavaScript.
“Kufanele uyithumele kubantu abayisigidi ngaphambi kokuthi uthole ezinye zezinto engakwazi ukuzenza,” kusho u-Amodei, oshiye i-OpenAI ukuze atholele i-AI yokuqalisa i-Anthropic, esanda kuthola uxhaso ku-Google.
“Kunokukhathazeka ukuthi, heyi, ngingenza imodeli ekahle kakhulu njengokuhlasela kwe-cyberattacks noma okuthile futhi ngingazi nokuthi ngikwenzile lokho,” engeza.
I-Bing yakwaMicrosoft isekelwe kubuchwepheshe obuthuthukiswe nge-OpenAI, iMicrosoft etshale kuyo imali.
IMicrosoft ishicilele izingcezu ezimbalwa mayelana nendlela yayo ye-AI enomthwalo wemfanelo, okuhlanganisa nomongameli wayo uBrad Smith ekuqaleni kwale nyanga. “Kufanele singene kule nkathi entsha ngomdlandla ngesethembiso, kodwa nokho sivule amehlo futhi siqinile ekubhekaneni nezingibe ezingenakugwemeka ezisazayo,” wabhala.
Indlela izibonelo zezilimi ezinkulu ezisebenza ngayo izenza kube nzima ukuziqonda ngokugcwele, ngisho nakubantu abazakhileyo. Izinkampani ze-Big Tech ezingemuva kwazo nazo zivalelwe emqhudelwaneni ononya walokho ezikubona njengomngcele olandelayo wobuchwepheshe obunenzuzo enkulu, zengeza esinye isendlalelo semfihlo.
Okukhathazayo lapha ukuthi lobu buchwepheshe bungamabhokisi amnyama, kusho u-Marcus, futhi akekho owazi kahle ukuthi abekwe kanjani ama- guardrail afanele futhi anele kuwo.
“Eqinisweni basebenzisa umphakathi njengezifundo ocwaningweni abangawazi ngempela umphumela walo,” kusho uMarcus. “Ingabe lezi zinto zingathonya ukuphila kwabantu? Ngokuqinisekile babengakwazi. Ingabe lokhu kuhlolwe kahle? Ngokusobala akunjalo.”
