Esiqeshini sami sokugcina, sisungule iqiniso eliyisisekelo: ukuze abasebenzisi basebenzise futhi bathembele ku-AI, kufanele bayithembe. Sikhulume ngokuthembana njengokwakhiwa okunezici eziningi, okwakhelwe emibonweni ye-AI, Ikhono, Ukulunga, Ubuqotho, kanye Nokubikezela. Kodwa kwenzekani lapho i-AI, ngokuhlakanipha kwayo okuthulile, kwe-algorithmic, yenza isinqumo esishiya umsebenzisi edidekile, ekhungathekile, noma elimele? Isicelo semali mboleko siyanqatshelwa, ingoma eyintandokazi ivele ingekho ohlwini lwadlalwayo, futhi ukuqalisa kabusha okufanelekayo kunqatshwa ngaphambi kokuba umuntu akubone. Kulezi zikhathi, ikhono nokubikezela kuyashabalala, futhi umusa uzwa izwe elikude. Ingxoxo yethu manje kumele isuke kokuthi kungani ukwethembana iye endleleni yokubeka izinto obala. Umkhakha we-Explainable AI (XAI), ogxile ekuthuthukiseni izindlela zokwenza imiphumela ye-AI iqondakale kubantu, uye wavela ukuze kubhekwane nalokhu, kodwa kuvame ukuhlelwa njengenselele yobuchwepheshe yososayensi bedatha. Ngiphikisa ukuthi kuyinselelo ebalulekile yokuklama kwemikhiqizo ethembele ku-AI. Kungumsebenzi wethu njengezingcweti ze-UX ukuvala igebe phakathi kokuthatha izinqumo nge-algorithmic nokuqonda komuntu. Lesi sihloko sinikeza isiqondiso esisebenzayo, esingasetshenzwa sendlela yokucwaninga nokuklama ukuze kuchazeke. Sizodlulela ngale kwe-buzzwords nasezithombeni, sihumushe imiqondo eyinkimbinkimbi ye-XAI ibe amaphethini edizayini aphathekayo ongaqala ukuwasebenzisa namuhla. I-De-mystifying XAI: I-Core Concepts Yabasebenzi be-UX I-XAI imayelana nokuphendula umbuzo womsebenzisi: "Kungani?" Kungani ngiboniswe lesi sikhangiso? Kungani le movie inconywa kimi? Kungani isicelo sami sinqatshelwe? Kucabange njenge-AI ekhombisa umsebenzi wayo enkingeni yezibalo. Ngaphandle kwakho, unempendulo nje, futhi uyaphoqeleka ukuba uyithathe ngokukholwa. Ekuboniseni izinyathelo, wakha ukuqonda nokwethemba. Futhi uvumela umsebenzi wakho ukuthi uhlolwe kabili futhi uqinisekiswe yibo kanye abantu obathintayo. Ukubaluleka Kwesici Kanye Nezingqinamba Kunamasu amaningana esingawasebenzisa ukuze sicacise noma sichaze ukuthi kwenzekani nge-AI. Nakuba izindlela zisukela ekunikezeni ukuqonda okuphelele kwesihlahla sesinqumo kuye ekukhiqizeni izifinyezo zolimi lwemvelo zokukhiphayo, izinhlobo ezimbili zolwazi olusebenzayo nezinomthelela ochwepheshe be-UX abangazethula kokuhlangenwe nakho ukubaluleka kwesici (Umfanekiso 1) kanye namaqiniso angokoqobo. Lokhu kuvame ukuqonde kakhulu kubasebenzisi ukuthi bakuqonde futhi okusebenziseka kakhulu ukuze abaklami bakusebenzise.
Ukubaluleka Kwesici Le ndlela yokuchaza iyaphendula, "Iziphi izici ezibaluleke kakhulu ezacatshangelwa yi-AI?" Imayelana nokuhlonza okuguquguqukayo okuphezulu okungu-2-3 okube nomthelela omkhulu kumphumela. Yisihloko, hhayi indaba yonke. Isibonelo: Cabanga nge-AI ebikezela ukuthi ikhasimende lizoshintsha yini (khansela isevisi yalo). Ukubaluleka kwesici kungase kuveze ukuthi “inani lezingcingo zosekelo enyangeni edlule” kanye “nokwenyuka kwentengo kwakamuva” kwakuyizici ezimbili ezibaluleke kakhulu ekunqumeni ukuthi ikhasimende lingase liphazamiseke yini.
Ama-counterfactals Le ndlela enamandla iyaphendula, “Yini engizodinga ukuyishintsha ukuze ngithole umphumela ohlukile?” Lokhu kubalulekile ngoba kunikeza abasebenzisi umuzwa wokusebenzela. Iguqula u-“cha” okhungathekisayo abe “okwamanje” okungenziwa. Isibonelo: Cabanga ngohlelo lwesicelo semalimboleko olusebenzisa i-AI. Umsebenzisi wenqatshelwa imalimboleko. Esikhundleni sokubona nje "Isicelo Senqatshiwe," incazelo engelona iqiniso nayo ingabelana ngokuthi, "Uma amaphuzu akho esikweletu abe ngamaphuzu angama-50 ngaphezulu, noma uma isilinganiso sakho sesikweletu somholo siphansi ngo-10%, imali oyibolekiwe ngabe ivunyelwe." Lokhu kunikeza u-Sarah izinyathelo ezicacile, ezibambekayo angazithatha ukuze athole imali mboleko esikhathini esizayo.
Ukusebenzisa Idatha Yemodeli Ukuze Uthuthukise Incazelo Nakuba imininingwane yobuchwepheshe ivamise ukusingathwa ososayensi bedatha, kuyasiza ukuthi ochwepheshe be-UX bazi ukuthi amathuluzi afana ne-LIME (Izincazelo Zendawo Enokuchazwa Kwemodeli-engaziwa) achaza izibikezelo ngazinye ngokulinganisa imodeli endaweni, kanye ne-SHAP (Izincazelo ze-SHApley Additive) esebenzisa indlela yethiyori yegeyimu ukuchaza umphumela wanoma yimuphi umshini osetshenziswa kumamodeli ayinkimbinkimbi “. Le mitapo yolwazi isiza ukubhidliza isinqumo se-AI sokukhombisa ukuthi yikuphi okokufaka okube nomthelela omkhulu kumphumela othile. Uma kwenziwa kahle, idatha engaphansi kwesinqumo sethuluzi le-AI ingasetshenziswa ukuxoxa indaba enamandla. Ake sidabule ukubaluleka kwesici namaqiniso angelona iqiniso futhi sibonise ukuthi isayensi yedatha ngemuva kwesinqumo ingasetshenziswa kanjani ukuze kuthuthukiswe umuzwa womsebenzisi. Manje akemboza ukubaluleka kwesici ngosizo Lwedatha Yezincazelo Zasendaweni (isb., LIME): Le ndlela iphendula, “Kungani i-AI yenze lesi sincomo kimina, khona manje?” Esikhundleni sencazelo evamile yokuthi imodeli isebenza kanjani, inikeza isizathu esigxilile sesenzakalo esisodwa, esiqondile. Kungokomuntu siqu nangokomongo. Isibonelo: Cabanga ngohlelo lokuncoma umculo olunamandla e-AI njengeSpotify. Incazelo yendawo ingaphendula, "Kungani uhlelo luncome le ngoma ka-Adele kuwena njengamanje?" Incazelo ingase ibe ukuthi: “Ngenxa yokuthi muva nje ulalele amanye ama-ballads athinta imizwa nezingoma zabaculi besifazane.”
Okokugcina, ake simboze ukufakwa Kwezincazelo Ezisekelwe Kuvelu (isb. I-Shapley Additive Explanations (SHAP) encazelweni yesinqumo: Lena inguqulo ehlukene kakhudlwana yokubaluleka kwesici ephendula ngokuthi, “Into ngayinye isiqhube kanjani isinqumo ngandlela thize?” Isiza ukubona ngeso lengqondo lokho obekubalulekile, nokuthi ingabe ithonya layo belilihle noma libi. Isibonelo: Ake ucabange ukuthi ibhange lisebenzisa imodeli ye-AI ukuze linqume ukuthi liyasivuma yini isicelo semalimboleko.
Ukubaluleka Kwesici: Okuphumayo kwemodeli kungase kubonise ukuthi amaphuzu esikweletu somfakisicelo, imali engenayo, kanye nesilinganiso sesikweletu somholo kwakuyizinto ezibaluleke kakhulu esinqumweni sakhe. Lokhu kuphendula okwakubalulekile. Ukubaluleka Kwesici Ngezincazelo Ezisekelwe Kuvelu (SHAP): Amanani e-SHAP azothatha ukubaluleka kwesici ngokuqhubekayo ngokusekelwe ezintweni zemodeli.
Ngemali mboleko egunyaziwe, i-SHAP ingase ibonise ukuthi amaphuzu aphezulu esikweletu aphushele kakhulu isinqumo ekuvunyweni (ithonya elihle), kuyilapho isilinganiso esingaphezudlwana kwesilinganiso sesikweletu semali engenayo sisikhipha kancane (ithonya elibi), kodwa akwanele ukuphika isikweletu. Ngemali mboleko enqatshiwe, i-SHAP ingase iveze ukuthi imali engenayo ephansi kanye nenani eliphezulu lemibuzo yekhredithi yakamuva kusiphushele ngokuqinile isinqumo ekunqatshelweni, ngisho noma amaphuzu esikweletu ayehloniphekile.
Lokhu kusiza isiphathimandla semalimboleko sichazele umfakisicelo ngale kwalokho obekucatshangwa, ukuthi isici ngasinye sibe negalelo kanjani esinqumweni sokugcina esithi “yebo” noma “cha”. Kubalulekile ukuqaphela ukuthi amandla okunikeza izincazelo ezinhle ngokuvamile aqala ngaphambi kwesikhathi emjikelezweni wokuthuthukiswa. Ososayensi bedatha nonjiniyela badlala indima ebalulekile ngokuhlela ngamabomu amamodeli namapayipi edatha ngezindlela ezisekela ukuchazeka ngokwemvelo, kunokuzama ukukubopha njengokucatshangelwa kwamuva. Amathimba ocwaningo nemiklamo angakhuthaza lokhu ngokuqala izingxoxo zangaphambi kwesikhathi nososayensi bedatha nonjiniyela mayelana nezidingo zabasebenzisi zokuqonda, anikele ekuthuthukisweni kwamamethrikhi achazekayo, kanye nezincazelo ze-prototyping ngokuhlanganyela ukuze kuqinisekiswe ukuthi kokubili anembile futhi asebenziseka kalula. I-XAI kanye ne-Ethical AI: Ukucazulula Ukuchema Nokuzibophezela Ngaphandle kokwakha ukwethembana, i-XAI idlala indima ebalulekile ekubhekaneni nemithelela yezimiso zokuziphatha ejulile ye-AI*, ikakhulukazi mayelana nokuchema kwe-algorithmic. Amasu okuchaza, njengokuhlaziya amanani e-SHAP, angaveza uma izinqumo zemodeli zithonywa ngokungafanele izibaluli ezizwelayo njengobuhlanga, ubulili, noma isimo senhlalo-mnotho, ngisho noma lezi zici zingazange zisetshenziswe ngokusobala njengokufakwayo okuqondile. Isibonelo, uma imodeli yokugunyaza imali mboleko inikezela ngokuqhubekayo amanani e-SHAP angemahle kubafakizicelo abavela endaweni ethile, ikhombisa ukuchema okungaba khona okudinga ukuphenywa, ukunika amandla amaqembu ukuthi aveze futhi anciphise leyo miphumela engalungile. Amandla e-XAI futhi eza namandla "okugeza okuchazayo." Njengoba nje "ukugeza okuluhlaza" kudukisa abathengi mayelana nezinqubo zemvelo, ukugeza okuchazekayo kungenzeka lapho izincazelo zenzelwe ukufihla, kunokuba zikhanyise, ukuziphatha okuyinkinga kwe-algorithmic noma ukuchema okungokwemvelo. Lokhu kungase kubonakale njengezincazelo ezilula ngokweqile ezishiya izici ezibalulekile ezinethonya, noma izincazelo ezakha imiphumela ehlelekile ukuze ibonakale ingathathi hlangothi noma ilungile kunalokho eziyikho ngempela. Igcizelela isibopho sokuziphatha sabasebenzi be-UX ukuklama izincazelo ezisobala ngempela neziqinisekayo. Ochwepheshe be-UX, ngokubambisana nososayensi bedatha kanye nezazi zezimiso zokuziphatha, banesibopho esibalulekile ekukhulumeni ukuthi kungani kwenziwa isinqumo, kanye nemikhawulo kanye nokuchema okungaba khona kwemodeli ye-AI eyisisekelo. Lokhu kuhlanganisa ukusetha okulindelwe ngabasebenzisi okungokoqobo mayelana nokunemba kwe-AI, ukukhomba lapho imodeli ingase ingathembeki khona kancane, nokuhlinzeka ngamashaneli acacile okuthola usizo noma impendulo lapho abasebenzisi bebona imiphumela engalungile noma engalungile. Ukubhekana nalokhu ngokuqhubekayoUbukhulu bokuziphatha buzosivumela ukuthi sakhe amasistimu e-AI anobulungiswa ngempela futhi athembekile. Ukusuka Ezindleleni Ukuya Kuma-Mockups: Amaphethini Wokuklama we-XAI asebenzayo Ukwazi imiqondo kuyinto eyodwa; ukuwaklama kungenye. Nansi indlela esingazihumusha ngayo lezi zindlela ze-XAI zibe amaphethini edizayini enembile. Iphethini 1: Isitatimende "Sokuba" (Ngokubaluleka Kwesici) Lena iphethini elula futhi evame ukusebenza kahle kakhulu. Isitatimende esiqondile, solimi olulula esiveza isizathu esiyinhloko sesenzo se-AI.
I-Heuristic: Yiba ngqo futhi ube mfushane. Hola ngesizathu esisodwa esinomthelela omkhulu. Gwema i-jargon ngazo zonke izindleko.
Isibonelo: Cabanga ngesevisi yokusakaza umculo. Esikhundleni sokuvele wethule uhlu lwadlalwayo oluthi "Zitholele Maviki Onke", ungeza umugqa omncane we-microcopy.Isincomo Sengoma: "I-Velvet Morning"Ngoba ulalela i-"The Fuzz" namanye ama-psychedelic rock.
Iphethini yesi-2: I-"What-If" Interactive (ye-Counterfactuals) Ama-counterfactals ngokwemvelo amayelana nokunikezwa amandla. Indlela engcono kakhulu yokuwamela ukunikeza abasebenzisi amathuluzi asebenzayo ukuze bahlole amathuba ngokwabo. Lokhu kulungile kwezezimali, ezempilo, noma ezinye izinhlelo zokusebenza ezigxile kumgomo.
I-Heuristic: Yenza izincazelo zihlanganyele futhi zinikeze amandla. Vumela abasebenzisi babone imbangela kanye nomphumela wezinqumo zabo.
Isibonelo: Isixhumi esibonakalayo sesicelo semalimboleko. Ngemva kokuphika, esikhundleni sokuvala isiphetho, umsebenzisi uthola ithuluzi lokunquma ukuthi izimo ezihlukahlukene (ukuthi-uma) zingadlala kanjani (Bheka Umfanekiso 1).
Iphethini 3: I-reel egqamisayo (Ukuze uthole Izincazelo Zasendaweni) Uma i-AI yenza isenzo kokuqukethwe komsebenzisi (njengokufingqa idokhumenti noma ukukhomba ubuso ezithombeni), incazelo kufanele ixhunywe ngokubonakalayo kumthombo.
I-Heuristic: Sebenzisa izinkomba ezibukwayo njengokugqamisa, uhlaka, noma izichasiselo ukuze uxhume incazelo ngokuqondile ku-interface element eyichazayo.
Isibonelo: Ithuluzi le-AI elifingqa ama-athikili amade.Iphuzu Elifingqiwe Elikhiqizwa yi-AI:Ucwaningo lokuqala lubonise igebe lemakethe lemikhiqizo esimeme.Umthombo Kudokhumenti:“...Ukuhlaziywa kwethu kwe-Q2 yezitayela zemakethe kubonise ngokuphelele ukuthi akekho esincintisana naye omkhulu obesebenzela umthengi ngempumelelo eco-conscious, okuveza igebe elibalulekile lemakethe lemikhiqizo esimeme.
Iphethini yesi-4: I-Push-and-Pull Visual (Ukuze uthole Izincazelo Ezisekelwe Kunani) Ukuze uthole izinqumo eziyinkimbinkimbi, abasebenzisi bangadinga ukuqonda ukusebenzisana kwezinto. Ukubukwa kwedatha okulula kungenza lokhu kucace ngaphandle kokuxaka.
I-Heuristic: Sebenzisa ukubonwa kwedatha okulula, okunemibala enemibala (njengamabha amashadi) ukuze ubonise izici ezibe nomthelela omuhle noma omubi esinqumweni.
Isibonelo: I-AI ihlolela iphrofayili yekhandidethi ukuthola umsebenzi.Kungani leli khandidethi lifana no-75%:Izinto ezikhuphula amaphuzu:5+ Iminyaka engu-5+ Isipiliyoni Sokucwaninga se-UXUchwepheshe ku-PythonFactors ephusha umphumela:Akunalwazi nge-B2B SaaS
Ukufunda nokusebenzisa lawa maphethini wokuklama ku-UX yomkhiqizo wakho we-AI kuzosiza ukwandisa ukuchazeleka. Ungasebenzisa futhi amasu engeziwe engingabazi ngokujulile lapha. Lokhu kuhlanganisa okulandelayo:
Izincazelo zolimi lwemvelo: Ukuhumusha okuphumayo kobuchwepheshe be-AI kube ulimi olulula, oluyingxoxo lwabantu abangaluqonda kalula. Izincazelo zomongo: Ukunikeza isizathu sokuphuma kwe-AI ngesikhathi esithile nasendaweni, kuhambisana kakhulu nomsebenzi womsebenzisi. Ukubuka okufanelekile: Kusetshenziswa amashadi, amagrafu, noma amamephu okushisa ukuze kumelele ngokubonakalayo inqubo yokwenza izinqumo ye-AI, okwenza idatha eyinkimbinkimbi ibe enembile futhi kube lula kubasebenzisi ukuyiqonda.
Inothi Lokuphela Kwangaphambili: Ukuhumusha le miphumela echazekayo ibe kokuhlangenwe nakho komsebenzisi okungenazihibe nakho kuveza isethi yakhona yokucatshangelwa kobuchwepheshe. Onjiniyela bangaphambili bavame ukubhekana nedizayini ye-API ukuze bathole ngempumelelo idatha yencazelo, kanye nemithelela yokusebenza (njengokukhiqiza izincazelo zesikhathi sangempela sakho konke ukusebenzisana komsebenzisi) idinga ukuhlela ngokucophelela ukugwema ukubambezeleka. Ezinye Izibonelo Zomhlaba Wangempela I-UPS Capital's DeliveryDefense I-UPS isebenzisa i-AI ukuze yabele “umphumela wokuzethemba wokulethwa” kumakheli ukuze ibikezele ukuthi kungenzeka yini ukuthi iphakheji lebiwe. Isofthiwe yabo ye-DeliveryDefense ihlaziya idatha yomlando ngendawo, imvamisa yokulahlekelwa, nezinye izici. Uma ikheli linemiphumela ephansi, isistimu ingakwazi ukuqondisa kabusha iphakheji ngokuqhubekayo ku-UPS Access Point evikelekile, inikeze incazelo yesinqumo (isb., "Iphakheji libuyiselwe endaweni evikelekile ngenxa yomlando wokweba"). Lolu hlelo lubonisa ukuthi i-XAI ingasetshenziswa kanjani ekunciphiseni ubungozi kanye nokwakha ukwethenjwa kwamakhasimende ngokusebenzisaobala. Izimoto Ezizimele Lezi zimoto zesikhathi esizayo zizodinga ukusebenzisa i-XAI ngokuphumelelayo ukuze zisize izimoto zazo zenze izinqumo eziphephile, nezichazekayo. Uma imoto ezishayelayo ibhuleka kungazelelwe, isistimu inganikeza incazelo yesikhathi sangempela ngesenzo sayo, isibonelo, ngokukhomba umuntu ohamba ngezinyawo ongena emgwaqeni. Lokhu akubalulekile nje kuphela ekunethezekeni nasekuthembekeni kwabagibeli kodwa kuyisidingo sokulawula ukufakazela ukuphepha nokuziphendulela kohlelo lwe-AI. I-IBM Watson Health (nezinselelo zayo) Yize kuvame ukukhonjwa njengesibonelo esijwayelekile se-AI ekunakekelweni kwezempilo, futhi kuyisibonelo esibalulekile sokubaluleka kwe-XAI. Ukwehluleka kwephrojekthi yayo ye-Watson ye-Oncology kugqamisa lokho okungahamba kahle lapho izincazelo zingacacile, noma lapho idatha eyisisekelo ichemile noma ingenziwanga endaweni. Izincomo zesistimu kwesinye isikhathi bezingahambisani nezinqubo zomtholampilo zendawo ngoba bezisekelwe kuzinkombandlela ze-U.S. Lokhu kusebenza njengendaba eyisixwayiso ngesidingo sokuchazwa okuqinile, okuqaphela umongo. Iqhaza Lomcwaningi we-UX: Ukukhomba Nokuqinisekisa Izincazelo Izixazululo zethu zedizayini zisebenza kuphela uma ziphendula imibuzo yomsebenzisi efanele ngesikhathi esifanele. Incazelo ephendula umbuzo umsebenzisi angenawo umsindo nje. Kulapho ucwaningo lwe-UX luba yisicubu esibalulekile esixhumanisayo esu le-XAI, siqinisekisa ukuthi sichaza ukuthi lokho kubaluleke ngani futhi kanjani kubasebenzisi bethu. Indima yomcwaningi ikabili: okokuqala, ukwazisa isu ngokukhomba lapho kudingeka khona izincazelo, futhi okwesibili, ukuqinisekisa imiklamo eletha lezo zincazelo. Ukwazisa Isu le-XAI (Okufanele Ukuchaze) Ngaphambi kokuthi siklame incazelo eyodwa, kufanele siqonde imodeli yengqondo yomsebenzisi yohlelo lwe-AI. Bakholelwa ukuthi yenzani? Aphi amagebe phakathi kokuqonda kwabo kanye neqiniso lesistimu? Lona umsebenzi oyisisekelo womcwaningi we-UX. Izingxoxo Zemodeli Yengqondo: Ukuvula Imibono Yabasebenzisi Yezinhlelo ze-AI Ngezingxoxo ezijulile, ezihleleke kancane, abasebenzi be-UX bangathola imininingwane ebalulekile yokuthi abasebenzisi babona futhi baqonde kanjani izinhlelo ze-AI. Lezi zikhathi zenzelwe ukukhuthaza abasebenzisi ukuthi badwebe ngokoqobo noma bachaze "imodeli yengqondo" yabo yangaphakathi yokuthi bakholelwa ukuthi i-AI isebenza kanjani. Lokhu kuvame ukubandakanya ukubuza imibuzo evulekile eyenza abasebenzisi bachaze ukucabanga kwesistimu, okokufaka kwayo, kanye nemiphumela yayo, kanye nobudlelwano phakathi kwalezi zinto. Lezi zingxoxo zinamandla ngoba zivame ukuveza imibono eyiphutha ejulile kanye nokuqagela abasebenzisi abanakho mayelana ne-AI. Isibonelo, umsebenzisi osebenzisana nenjini yokuncoma angase agomele ngokuzethemba ukuthi isistimu isekelwe kuphela emlandweni wabo wokubuka odlule. Bangase bangaboni ukuthi i-algorithm iphinda ihlanganise inqwaba yezinye izici, njengesikhathi sosuku abaphequlula ngaso, izinto ezithrendayo manje kuyo yonke ingxenyekazi, noma imikhuba yokubuka yabasebenzisi abafanayo. Ukwembula leli gebe phakathi kwemodeli yengqondo yomsebenzisi kanye nomqondo wangempela oyisisekelo we-AI kubaluleke kakhulu. Lisitshela ngokunembile ukuthi yiluphi ulwazi oluthile esiludingayo ukuze sixhumane nabasebenzisi ukuze sibasize bakhe imodeli yengqondo enembe kakhudlwana neqinile yesistimu. Lokhu-ke kuyisinyathelo esibalulekile ekukhuthazeni ukwethembana. Uma abasebenzisi beqonda, ngisho nasezingeni eliphezulu, ukuthi i-AI ifinyelela kanjani eziphethweni noma ezincomweni zayo, maningi amathuba okuthi bathembele ekuphumeni kwayo futhi bathembele ekusebenzeni kwayo. Imephu Yohambo Lwe-AI: Ukugxumela Okujulile Ekuthembekeni Kwabasebenzisi Nokuchazwa Ngokwenza imephu ngokucophelela uhambo lomsebenzisi ngesici esinamandla e-AI, sithola imininingwane ebalulekile ezikhathini ezinembile lapho kuvela khona ukudideka, ukukhungatheka, noma ukungathembani okujulile. Lokhu kwembula izikhathi ezibucayi lapho imodeli yengqondo yomsebenzisi yokuthi i-AI isebenza kanjani ingqubuzana nokuziphatha kwayo kwangempela. Cabangela isevisi yokusakaza umculo: Ingabe ukwethenjwa komsebenzisi kuyehla lapho isincomo sohlu lwadlalwayo sizwa “ngokungahleliwe,” sintula noma yikuphi ukuxhumana okubonakalayo emikhubeni yakhe yangaphambili yokulalela noma izinto azincamelayo? Lokhu kungenzeki okucatshangelwayo kuyinselelo eqondile ekulindeleke komsebenzisi kokulungiswa okuhlakaniphile kanye nokwephulwa kwesithembiso esisobala sokuthi i-AI iyakuqonda ukunambitheka kwabo. Ngokufanayo, ohlelweni lokusebenza lokuphatha izithombe, ingabe abasebenzisi bahlangabezana nokukhungatheka okukhulu lapho isici sokumaka isithombe se-AI sihlala sidukisa ilungu lomndeni eliyigugu? Leli phutha lingaphezu kwephutha lobuchwepheshe; ishaya enhliziyweni yokunemba, ukwenza kube ngokwakho, ngishoukuxhumana ngokomzwelo. La maphuzu ezinhlungu ayizimpawu ezicacile ezibonisa ngokunembile lapho kudingeka khona incazelo ebekwe kahle, ecacile, nemfishane. Izincazelo ezinjalo zisebenza njengezindlela zokulungisa ezibalulekile, ezilungisa ukwephulwa kokwethenjwa okuthi, uma kuyekwa kungalungiswa, kuholele ekushiyweni komsebenzisi. Amandla emephu yohambo lwe-AI asemandleni ayo okusihambisa ngaphezu kokuchaza umphumela wokugcina wesistimu ye-AI. Ngenkathi ukuqonda lokho okukhiqizwe yi-AI kubalulekile, ngokuvamile akwanele. Kunalokho, le nqubo isiphoqa ukuthi sigxile ekuchazeni inqubo ngezikhathi ezibucayi. Lokhu kusho ukukhuluma:
Kungani kukhiqizwe okuphumayo okuthile: Ingabe kwakungenxa yedatha ethile yokokufaka? Imodeli ethile yezakhiwo? Yiziphi izici ezibe nomthelela esinqumweni se-AI: Ingabe izici ezithile bezinesisindo esikhulu? I-AI yafinyelela kanjani esiphethweni sayo: Singanikeza incazelo eyenziwe lula, efana nayo yokusebenza kwayo kwangaphakathi? Imiphi imibono eyenziwa yi-AI: Ingabe kube khona ukuqonda okucacile kwenhloso yomsebenzisi noma idatha okudingeka ivezwe? Imiphi imikhawulo ye-AI: Ukuxhumana ngokusobala lokho i-AI engakwazi ukukwenza, noma lapho ukunemba kwayo kungase kuntengantenge, kwakha okulindelwe okungokoqobo.
Imephu yohambo lwe-AI iguqula umqondo ongabonakali we-XAI uwenze uhlaka olusebenzayo, olusebenzayo lwabasebenzi be-UX. Kusenza sikwazi ukudlulela ngale kwezingxoxo zethiyori zokuchazeka futhi esikhundleni salokho sikhombe izikhathi eziqondile lapho ukwethenjwa komsebenzisi kusengozini, sinikeza imininingwane edingekayo yokwakha okuhlangenwe nakho kwe-AI okunamandla, okusobala, okuqondakalayo, nokwethembeka. Ekugcineni, ucwaningo yindlela esidalula ngayo okungaziwa. Ithimba lakho kungenzeka liphikisana ngokuthi lichaza ukuthi kungani isikweletu sinqatshelwe, kodwa ucwaningo lungase luveze ukuthi abasebenzisi bakhathazeke kakhulu ngokuqonda ukuthi idatha yabo yasetshenziswa kanjani kwasekuqaleni. Ngaphandle kocwaningo, simane siqagela ukuthi abasebenzisi bethu bazibuza ini. Ukusebenzisana Ngomklamo (Indlela Yokuchaza I-AI Yakho) Uma ucwaningo seluhlonze ukuthi yini okufanele ichazwe, iluphu yokuhlanganyela enomklamo iyaqala. Abaklami bangenza isibonelo samaphethini esixoxe ngawo ngaphambili—isitatimende esithi “Ngoba”, izilayidi ezisebenzisanayo—futhi abacwaningi bangabeka leyo miklamo phambi kwabasebenzisi ukuze babone ukuthi bayabambelela yini. Ukusebenziseka Okuhlosiwe Nokuhlolwa Kokuqonda: Singaklama izifundo zocwaningo ezihlola ngokuqondile izingxenye ze-XAI. Asivele sibuze, "Ingabe lokhu kulula ukukusebenzisa?" Siyabuza, “Ngemva kokubona lokhu, ungangitshela ngamazwi akho ukuthi kungani isistimu income lo mkhiqizo?” noma “Ngibonise ukuthi ungenzani ukuze ubone ukuthi ungathola yini umphumela ohlukile.” Umgomo lapha uwukulinganisa ukuqonda nokusebenza, ngokuhambisana nokusebenziseka. Ukulinganisa Ukuzithemba Kwako: Singasebenzisa izinhlolovo ezilula nezikali zokulinganisa ngaphambi nangemva kokuboniswa kwencazelo. Isibonelo, singabuza umsebenzisi esikalini samaphoyinti angu-5, "Usethemba kangakanani lesi sincomo?" ngaphambi kokuba babone isitatimende esithi “Ngoba,” bese ubabuza futhi ngemva kwalokho. Lokhu kunikeza idatha yobuningi yokuthi ingabe izincazelo zethu empeleni zihambisa inaliti ekuthembekeni. Le nqubo idala iluphu enamandla, ephindaphindayo. Imiphumela yocwaningo yazisa ukwakheka kokuqala. Lowo mklamo ube usuhlolwa, futhi okutholakele okusha kubuyiselwa ethimbeni lokuklama ukuze kuthuthukiswe. Mhlawumbe isitatimende esithi “Ngoba” besiyi-jargony kakhulu, noma isilayidi esithi “What-If” besidida kakhulu kunokunikeza amandla. Ngalokhu kuqinisekisa okuhlanganyelwe, siqinisekisa ukuthi izincazelo zokugcina zinembile ngokobuchwepheshe, ziyaqondakala ngempela, ziwusizo, futhi zakha ukwethenjwa kubantu abasebenzisa umkhiqizo. I-Goldilocks Zone Of Explanation Isixwayiso esibucayi: kungenzeka ukuchaza ngokweqile. Njengasenganekwaneni, lapho u-Goldilocks efuna khona iphalishi 'eyayilungile', umgomo wencazelo enhle ukunikeza inani elifanele lemininingwane-hhayi kakhulu futhi hhayi encane kakhulu. Ukuhlasela umsebenzisi ngakho konke okuguquguqukayo kumodeli kuzoholela ekugcwalisweni kwengqondo futhi kunganciphisa ukwethembana. Umgomo awukona ukwenza umsebenzisi abe usosayensi wedatha. Isixazululo esisodwa ukudalula okuqhubekayo.
Qala ngokulula. Hola ngesitatimende esifushane esithi “Ngoba”. Kubasebenzisi abaningi, lokhu kuzokwanela. Nikeza indlela eya emininingwaneni. Nikeza isixhumanisi esicacile, esinokungqubuzana okuphansi njengokuthi “Funda Kabanzi” noma “Bona ukuthi lokhu kunqunywe kanjani.” Veza ubunkimbinkimbi. Ngemuva kwaleso sixhumanisi, unganikela ngezilayidi ezisebenzisanayo, okubonwayo, noma uhlu olunemininingwane eminingi yezinto ezinomthelela.
Le ndlela enezingqimba ihlonipha ukunaka komsebenzisi kanye nobuchwepheshe, ihlinzeka ngenani elifaneleyolwazi ngezidingo zabo. Ake sicabange usebenzisa idivayisi yasekhaya ehlakaniphile encoma ukushisisa okufanele okusekelwe ezintweni ezihlukahlukene. Qala ngokulula: “Okwamanje ikhaya lakho lishisisa kufika kumadigri angu-72, okuyizinga lokushisa elilungile lokonga amandla nokunethezeka.” Nikeza indlela eya emininingwaneni: Ngaphansi kwalokho, isixhumanisi esincane noma inkinobho: "Kungani u-72 degrees elungile?" Veza ubunkimbinkimbi: Ukuchofoza leso sixhumanisi kungavula isikrini esisha esibonisa:
Izilayidi ezisebenzisanayo zethempelesha yangaphandle, umswakama, neleveli yakho yokunethezeka oyikhethayo, ebonisa ukuthi lokhu kulungisa kanjani izinga lokushisa elinconyiwe. Ukuboniswa kokusetshenziswa kwamandla kumazinga okushisa ahlukene. Uhlu lwezinto ezinomthelela njengokuthi “Isikhathi sosuku,” “Izinga lokushisa langaphandle lamanje,” “Ukusetshenziswa kwamandla okungokomlando,” kanye “Nezinzwa zokuhlala.”
Kuyasebenza ukuhlanganisa izindlela eziningi ze-XAI kanye nale phethini Yencazelo Yendawo Ye-Goldilocks, ekhuthaza ukudalulwa okuqhubekayo, ikhuthaza lokhu ngokungagunci. Ungase uqale ngesitatimende esilula esithi “Ngoba” (Iphethini 1) ukuze usiqonde ngokushesha, bese unikeza isixhumanisi esithi “Funda Kabanzi” esidalula “Ingabe-Ukube” Okusebenzisanayo (Iphethini 2) noma “Okubonakalayo Push-and-Pull” (Iphethini 4) ukuze uthole ukuhlola okujulile. Ngokwesibonelo, isistimu yesicelo semalimboleko ingase iqale isho isizathu esiyinhloko sokwenqaba (ukubaluleka kwesici), bese ivumela umsebenzisi ukuthi asebenzisane nethuluzi elithi “What-If” ukuze abone ukuthi izinguquko emalini yakhe engenayo noma esikweletini zingawushintsha kanjani umphumela (okunye), futhi ekugcineni, linikeze ngeshadi elinemininingwane elithi “Push-and-Pull” (incazelo esekelwe enanini) ukuze abonise izici ezinhle nezingezinhle zazo zonke iminikelo. Le ndlela enezingqimba ivumela abasebenzisi ukuthi bafinyelele izinga lemininingwane abayidingayo, uma beyidinga, ivimbele ukugcwala kwengqondo ngenkathi isahlinzeka ngokusobala okuphelele. Ukunquma ukuthi yimaphi amathuluzi nezindlela ze-XAI okufanele zisetshenziswe ngokuyinhloko kuwumsebenzi wocwaningo olunzulu lwe-UX. Izingxoxo zemodeli yengqondo kanye nokwenza imephu kohambo lwe-AI kubalulekile ukuze ukhombe izidingo zabasebenzisi kanye nezindawo zobuhlungu ezihlobene nokuqonda kwe-AI nokwethenjwa. Izingxoxo zemodeli yengqondo zisiza ukuthola imibono eyiphutha mayelana nendlela i-AI esebenza ngayo, okubonisa izindawo lapho kudingeka khona izincazelo ezibalulekile (njengokubaluleka kwesici noma izincazelo zendawo). Imephu yohambo lwe-AI, ngakolunye uhlangothi, ikhomba izikhathi ezibucayi zokudideka noma zokungathembeki ekusebenzelaneni komsebenzisi ne-AI, isayinda lapho izincazelo ezicacile noma ezisebenzayo (njengezinto ezingelona iqiniso noma izincazelo ezisekelwe enalulweni) kungaba yinzuzo kakhulu ukwakha kabusha ukwethembana nokuhlinzeka nge-ejensi.
Ekugcineni, indlela engcono kakhulu yokukhetha indlela yokwenza iwukuba uvumele ucwaningo lomsebenzisi luqondise izinqumo zakho, uqinisekise ukuthi izincazelo oziklamayo zibhekana ngqo nemibuzo yangempela yabasebenzisi nezinto ezikukhathazayo, kunokumane unikeze imininingwane yobuchwepheshe ngenxa yabo. I-XAI yama-ejenti okucabanga ajulile Ezinye zezinhlelo ezintsha ze-AI, ezaziwa ngokuthi ama-ejenti okucabanga ajulile, zikhiqiza “uchungechunge lokucabanga” olucacile kuwo wonke umsebenzi oyinkimbinkimbi. Abagcini nje ngokucaphuna imithombo; bakhombisa indlela enengqondo, yesinyathelo nesinyathelo abayithathile ukuze bafinyelele esiphethweni. Yize lokhu kuveza ngale kunikeza umongo obalulekile, umdlalo ngamunye ohlanganisa izigaba ezimbalwa ungazizwa ukhungathekile kumsebenzisi ozama ukuqedela umsebenzi. Izimiso ze-XAI, ikakhulukazi i-Goldilocks Zone of Explanation, zisebenza ngqo lapha. Singahlela uhambo, sisebenzisa ukudalula okuqhubekayo ukuze sibonise kuphela isiphetho sokugcina kanye nesinyathelo esibaluleke kakhulu enqubweni yokucabanga kuqala. Abasebenzisi bangangena ukuze babone ukucabanga okugcwele, okunemininingwane, okunezinyathelo eziningi lapho bedinga ukuhlola kabili ingqondo noma ukuthola iqiniso elithile. Le ndlela ihlonipha ukunaka komsebenzisi ngenkathi igcina ukubonakala okugcwele kwe-ejenti. Izinyathelo Ezilandelayo: Ukunikeza Amandla Uhambo Lwakho lwe-XAI Ukuchazwa kuyinsika eyisisekelo yokwakha imikhiqizo ye-AI ethembekile nephumelelayo. Kubasebenzi abathuthukile abafuna ukuqhuba lolu shintsho enhlanganweni yabo, uhambo ludlulela ngale kwamaphethini okuklama ukuya ekukhulumeni nasekufundeni okuqhubekayo. Ukuze ujulise ukuqonda kwakho nokusebenzisa okungokoqobo, cabanga ukuhlola izinsiza ezifana nekhithi yamathuluzi ye-AI Explainability 360 (AIX360) evela ku-IBM Research noma i-What-If Tool ye-Google, enikeza izindlela ezisebenzisanayo zokuhlola imodeli yokuziphatha nezincazelo. Ukuxhumana nemiphakathi efana ne-Responsible AI Forum noma amaqembu athile ocwaningo agxile ku-AI egxile kubantu kunganikeza imininingwane ebalulekile kanye namathuba okusebenzisana. Okokugcina, yiba ummeli we-XAI enhlanganweni yakho.Ukuchazwa kohlaka njengotshalomali olunamasu. Cabangela isiphakamiso esifushane sobuholi bakho noma amaqembu ahlukahlukene: "Ngokutshala imali ku-XAI, sizodlulela ngalé kokwakha ukwethembana; sizosheshisa ukutholwa kwabasebenzisi, sinciphise izindleko zosekelo ngokunika abasebenzisi amandla okuqonda, futhi sinciphise ubungozi obubalulekile bezimiso zokuziphatha nokulawula ngokudalula ukuchema okungaba khona. Lokhu ukuklama okuhle nebhizinisi elihlakaniphile."
Izwi lakho, elisekelwe ekuqondeni okungokoqobo, libalulekile ekukhipheni i-AI ebhokisini elimnyama futhi libe nobudlelwano bokusebenzisana nabasebenzisi.
