Kwiqhekeza lam lokugqibela, siseke inyani esisiseko: ukuze abasebenzisi bamkele kwaye bathembele kwi-AI, kufuneka bayithembe. Siye sathetha ngokuthembana ukuba lulwakhiwo olunamacandelo amaninzi, olwakhelwe kwiimbono ze-AI yoBunako, ububele, iMfezeko, kunye nokuqikelela. Kodwa kwenzeka ntoni xa i-AI, ekuthuleni kwayo, ubulumko be-algorithmic, yenza isigqibo esishiya umsebenzisi edidekile, edanile, okanye ebuhlungu? Isicelo semali yokubambisa siyakhatywa, ingoma ethandwayo ayibikho ngequbuliso kuluhlu lokudlalwayo, kwaye inkqubo efanelekileyo iyaliwa phambi kokuba umntu ayibone. Ngeli xesha, amandla kunye nokuqikelelwa konakele, kwaye ububele buvakala kwihlabathi elikude. Incoko yethu ngoku kufuneka isuke kwisizathu sokuthembana ukuya kwindlela yokungafihli. Intsimi ye-AI ecacileyo (i-XAI), egxile ekuphuhliseni iindlela zokwenza iziphumo ze-AI ziqondeke ebantwini, ziye zavela ukulungisa oku, kodwa zihlala zicwangciswa njengomngeni wobugcisa ococekileyo kwizazinzulu zedatha. Ndiphikisa ukuba lucelomngeni olubalulekileyo loyilo lweemveliso ezixhomekeke kwi-AI. Ngumsebenzi wethu njengeengcali ze-UX ukuvala umsantsa phakathi kokwenza izigqibo nge-algorithmic kunye nokuqonda komntu. Eli nqaku libonelela ngesikhokelo esisebenzayo, esisebenzayo malunga nendlela yophando kunye noyilo lokucaciswa. Siza kudlulela ngaphaya kwe-buzzwords nakwi-mockups, siguqulela iikhonsepthi ze-XAI ezintsonkothileyo kwiipateni zoyilo olubambekayo onokuqala ukuzisebenzisa namhlanje. I-De-mystifying XAI: IiNgcaciso eziPhambili kwii-UX Practitioners I-XAI imalunga nokuphendula umbuzo womsebenzisi: "Kutheni?" Kutheni ndiboniswe le ntengiso? Kutheni le bhanyabhanya icetyiswa kum? Kutheni isicelo sam saliwe? Yicinge njenge-AI ebonisa umsebenzi wayo kwingxaki yezibalo. Ngaphandle kwayo, unempendulo nje, kwaye unyanzelekile ukuba uyithathe ngokholo. Ekuboniseni amanyathelo, wakha ukuqonda kunye nokuthembela. Ukwavumela ukuba umsebenzi wakho ujongwe kabini kwaye ungqinwe kwabona bantu ubachaphazelayo. Ukubaluleka kophawu kunye nezinto ezichaseneyo Kukho inani leendlela esinokuzisebenzisa ukucacisa okanye ukucacisa okwenzekayo nge-AI. Ngelixa iindlela zivela ekuboneleleni yonke ingqiqo yomthi wesigqibo ekuveliseni isishwankathelo solwimi lwendalo lwesiphumo, ezimbini zezona ntlobo zisebenzayo kunye nefuthe lolwazi lwe-UX iingcali ezinokuzisa kumava zibaluleke kakhulu (Umfanekiso 1) kunye nezinto ezichaseneyo. Ezi zihlala zezona zithe ngqo kubasebenzisi ukuba baziqonde kwaye ezona zinto zinokwenzeka ukuba abayili baziphumeze.
Ukubaluleka Kophawu Le ndlela yokucacisa iyaphendula, "Zeziphi ezona zinto zibalulekileyo ezijongwe yi-AI?" Kumalunga nokuchonga iinguqu eziphezulu ze-2-3 ezazinempembelelo enkulu kwisiphumo. Ingumxholo, hayi ibali lonke. Umzekelo: Yiba nomfanekiso-ngqondweni we-AI eqikelela ukuba ngaba umthengi uya kugungqa (arhoxise inkonzo yakhe). Ukubaluleka kophawu kunokutyhila ukuba "inani leefowuni zenkxaso kwinyanga edlulileyo" kunye "nokunyuka kwamaxabiso kutshanje" yayiyeyona miba mibini ibalulekileyo ekumiseleni ukuba umthengi angangena.
Izinto ezingeyonyani Le ndlela inamandla iyaphendula, "Yintoni endifuna ukuyitshintsha ukuze ndifumane iziphumo ezahlukeneyo?" Oku kubalulekile kuba kunika abasebenzisi imvakalelo yokusebenzela. Iguqula u-"hayi" okhathazayo abe yinto enokwenzeka "okwangoku." Umzekelo: Yiba nomfanekiso wenkqubo yesicelo semali-mboleko esebenzisa i-AI. Umsebenzisi uvinjwa imali-mboleko. Endaweni yokubona nje "iSicelo Saliwe," inkcazo echaseneyo nayo iya kwabelana, "Ukuba amanqaku akho etyala ayengama-50 amanqaku aphezulu, okanye ukuba umlinganiselo wetyala lakho kwingeniso ubuyi-10% ephantsi, imali mboleko yakho iya kuvunywa." Oku kunika uSarah amanyathelo acacileyo, anokuthi awathathe ukuze afumane imali-mboleko kwixesha elizayo.
Ukusebenzisa iDatha yeModeli ukunyusa iNkcazo Nangona iinkcukacha zobugcisa zihlala ziphathwa zizazinzulu zedatha, kuluncedo kwiingcali ze-UX ukwazi ukuba izixhobo ezifana ne-LIME (Iinkcazo zeNdawo ezitolika iModeli-agnostic) echaza uqikelelo lomntu ngamnye ngokuthelekisa imodeli yendawo, kunye ne-SHAP (i-SHApley Additive exPlanations) esebenzisa indlela yethiyori yomdlalo ukucacisa isiphumo sawo nawuphi na umatshini osetyenziswa kwimodeli entsonkothileyo. La mathala eencwadi anceda ngokusisiseko ukwaphula isigqibo se-AI sokubonisa ukuba ngawaphi amagalelo abenempembelelo enkulu kwisiphumo esinikiweyo. Xa yenziwe ngokufanelekileyo, idatha ephantsi kwesigqibo sesixhobo se-AI ingasetyenziselwa ukuxelela ibali elinamandla. Masihambe ngokubaluleka kweempawu kunye nezinto ezichaseneyo kwaye sibonise ukuba isayensi yedatha emva kwesigqibo ingasetyenziswa njani ukuphucula amava omsebenzisi. Ngoku makhegubungela ukubaluleka kophawu ngoncedo lweeNkcazo zeNdawo (umzekelo, i-LIME) idatha: Le ndlela iyaphendula, "Kutheni i-AI indenzele le ngcebiso, ngoku?" Endaweni yenkcazo jikelele yendlela imodeli esebenza ngayo, inika isizathu esigxininisekileyo somzekelo omnye, okhethekileyo. Yeyomntu kunye nomxholo. Umzekelo: Khawube nomfanekiso we-AI-powered inkqubo yokuncoma umculo njengeSpotify. Ingcaciso yasekhaya iya kuphendula, "Kutheni le nkqubo income le ngoma ka-Adele kuwe ngoku?" Ingcaciso isenokuba yile: “Kungenxa yokuba kutshanje uye waphulaphula ezinye iingoma neengoma eziculwa ngamabhinqa.”
Okokugqibela, makhe sigubungele ukubandakanywa kweeNkcazo eziSekwe kwiXabiso (umzekelo. Idatha yeShapley Additive Explanations (SHAP) kwingcaciso yesigqibo: Olu luguqulelo olunentsingiselo engakumbi yokubaluleka kwento ephendulayo, “Into nganye isityhale njani isigqibo ngenye indlela okanye enye?” Inceda ukuba nomfanekiso-ngqondweni weyona nto ibalulekileyo, nokuba impembelelo yayo ibintle okanye imbi. Umzekelo: Khawucinge ukuba ibhanki isebenzisa imodeli ye-AI ukwenza isigqibo sokuba iyasivuma isicelo semali-mboleko.
Ukubaluleka koMsebenzi: Imveliso yemodeli ingabonisa ukuba amanqaku amatyala omfaki-sicelo, umvuzo, kunye nomlinganiselo wetyala-kwingeniso yayizezona zinto zibalulekileyo kwisigqibo sayo. Oku kuphendula into ebalulekileyo. Ukubaluleka kophawu ngeeNkcazelo eziSekwe kwiXabiso (SHAP): Amaxabiso e-SHAP angathabatha ukubaluleka kweempawu ezibhekiselele kwimiba yomfuziselo.
Kwimali-mboleko evunyiweyo, i-SHAP ingabonisa ukuba amanqaku aphezulu etyala atyhalele kakhulu kwisigqibo ekuvunyweni (impembelelo entle), ngelixa umlinganiselo ongaphezulu kuno-avareji wetyala ukuya kwingeniso ulitsalile kancinane (impembelelo embi), kodwa akwanelanga ukuyikhanyela imali-mboleko. Kwimali-mboleko ekhatyiweyo, i-SHAP inokuveza ukuba umvuzo ophantsi kunye nenani eliphezulu lemibuzo yekhredithi yamva nje isityhalele ngamandla isigqibo sokukhanyela, nokuba inqaku letyala belindilisekile.
Oku kunceda igosa lemali-mboleko licacisele umenzi-sicelo ngaphaya koko bekuqwalaselwe, ukuba into nganye ibenegalelo njani kwisigqibo sokugqibela sika “ewe” okanye “hayi”. Kubalulekile ukuqaphela ukuba ukukwazi ukunika iinkcazo ezilungileyo kuhlala kuqala kwangethuba kumjikelo wophuhliso. Izazinzulu zedatha kunye neenjineli zidlala indima ebaluleke kakhulu ngokwenza ngabom iimodeli kunye nemibhobho yedatha ngeendlela ezixhasa ukucaciswa ngokwemvelo, kunokuzama ukuyivala njengengcinga yamva. Amaqela ophando kunye noyilo anokukhuthaza oku ngokuqalisa iincoko zakwangoko kunye noosonzululwazi bedatha kunye neenjineli malunga neemfuno zabasebenzisi zokuqonda, ukufaka isandla kuphuhliso lweemetriki zokucacisa, kunye neenkcazo zeprototyping ngentsebenziswano ukuqinisekisa ukuba zombini zichanekile kwaye zisebenziseka lula. I-XAI kunye ne-Ethics AI: Ukukhupha i-Bias kunye noxanduva Ngaphaya kokwakha ukuthembana, i-XAI idlala indima ebalulekileyo ekujonganeni neziphumo ezibi zeenqobo ezisesikweni ze-AI*, ingakumbi ngokuthambekela kwi-algorithmic. Ubuchule bokucacisa, obufana nokuhlalutya amaxabiso e-SHAP, kunokutyhila ukuba izigqibo zemodeli azichatshazelwanga ngokulungeleleneyo ziimpawu ezinovakalelo ezifana nobuhlanga, isini, okanye imeko yentlalo-ntle yezoqoqosho, nokuba ezi zinto azizange zisetyenziswe ngokucacileyo njengegalelo elithe ngqo. Umzekelo, ukuba imodeli yolwamkelo lwemali-mboleko ihlala inika amaxabiso angalunganga e-SHAP kubafaki-zicelo abasuka kwindawo ethile, oko kubonisa ukuthambekela okunokwenzeka okudinga uphando, ukuxhobisa amaqela ukuba avele kwaye athobe ezo ziphumo zingalunganga. Amandla e-XAI akwaza nokubanakho "ukuhlamba okucacileyo." Kanye njengokuba "i-greenwashing" ilahlekisa abathengi malunga nezenzo zokusingqongileyo, ukuhlamba okucacileyo kunokwenzeka xa iinkcazo zenzelwe ukufihla, kunokuba zikhanyise, ukuziphatha okuyingxaki kwe-algorithmic okanye ukuchasana okungokwemvelo. Oku kunokubonakala njengeengcaciso ezilula kakhulu ezishiya izinto ezibalulekileyo ezinempembelelo, okanye iingcaciso ezibonisa ukuba iziphumo zibonakala zingathathi hlangothi okanye zilungile kunokuba zinjalo ngokwenene. Igxininisa uxanduva lokuziphatha lwabasebenzi be-UX ukuyila iinkcazo ezicacileyo ngokwenene nezinokuqinisekiswa. Iingcali ze-UX, ngentsebenziswano nezazinzulu zedatha kunye neengcali zokuziphatha, zibambe uxanduva olubalulekileyo ekwaziseni isizathu sokuba kuthathwe isigqibo, kunye nemida kunye nokuthambekela okunokwenzeka kwemodeli ye-AI esisiseko. Oku kubandakanya ukuseta ulindelo oluyinyani lomsebenzisi malunga nokuchaneka kwe-AI, ukuchonga apho imodeli inokuthi ingathembeki kangako, kunye nokubonelela ngamajelo acacileyo oncedo okanye impendulo xa abasebenzisi bebona iziphumo ezingalunganga okanye ezingalunganga. Ukujongana neziimilinganiselo yokuziphatha iya kusivumela ukuba sakhe iinkqubo ze-AI ezinobulungisa kunye nezithembekileyo ngokwenene. Ukusuka kwiindlela ukuya kwiiMockups: Iipateni zoYilo zeXAI ezisebenzayo Ukwazi imiba yinto enye; ukuyila yenye. Nantsi indlela esinokuguqulela ngayo ezi ndlela ze-XAI zibe ziipateni zoyilo olucacileyo. Ipateni yoku-1: Inkcazo ethi "Ngenxa" (Yokubaluleka Kophawu) Le yeyona pateni ilula kwaye ihlala isebenzayo. Yingxelo ethe ngqo, yolwimi olucacileyo ebonisa isizathu esiphambili sesenzo se-AI.
I-Heuristic: Thetha kwaye ushwankathele. Khokela ngesona sizathu sinempembelelo. Yiphephe ijargon ngazo zonke iindleko.
Umzekelo: Yiba nomfanekiso wenkonzo yokusasaza umculo. Endaweni yokubonisa nje uluhlu lokudlalayo "Fumana ngeVeki", wongeza umgca omncinci we-microcopy.Isincomo seNgoma: "I-Velvet Morning"Kuba umamela "i-Fuzz" kunye nenye i-psychedelic rock.
Ipateni yesi-2: "Kuthekani-Ukuba" Unxibelelwano (yeeNtengiso) Ii-counterfactals ngokwendalo zimalunga nokuxhotyiswa. Eyona ndlela ingcono yokubamela kukunika abasebenzisi izixhobo ezisebenzayo zokujonga okunokwenzeka ngokwabo. Oku kugqibelele kwezemali, impilo, okanye ezinye izicelo ezijolise kwiinjongo.
I-Heuristic: Yenza iinkcazo zibe nokusebenzisana kunye nokuxhobisa. Vumela abasebenzisi babone unobangela kunye nesiphumo sokhetho lwabo.
Umzekelo: Ujongano lwesicelo semali-mboleko. Emva kokukhanyela, endaweni yesiphelo esifileyo, umsebenzisi ufumana isixhobo sokumisela ukuba iimeko ezahlukeneyo (yintoni-ukuba) zinokudlala (Jonga uMfanekiso 1).
Ipateni yesi-3: I-Highlight Reel (kwiNkcazelo yasekuhlaleni) Xa i-AI yenza isenzo kumxholo womsebenzisi (njengokushwankathela uxwebhu okanye ukuchonga ubuso kwiifoto), inkcazo kufuneka idibaniswe ngokubonakalayo kumthombo.
I-Heuristic: Sebenzisa imiqondiso yokubonwayo njengokuqaqambisa, amagqabantshintshi, okanye amagqabantshintshi ukudibanisa ingcaciso ngokuthe ngqo kwi-interface element eyichazayo.
Umzekelo: Isixhobo se-AI esishwankathela amanqaku amade.Inqaku elisishwankathelo le-AI-Generated:Uphando lokuqala lubonise umsantsa kwimarike yeemveliso ezizinzileyo.Umthombo kuXwebhu:“...Uhlalutyo lwethu lwe-Q2 yeentsingiselo zemarike lubonise ngokugqibeleleyo ukuba akukho sikhuphisana siphambili ebesisebenzela umthengi ngokufanelekileyo, eveza umsantsa obalulekileyo wemarike.” kwiimveliso ezizinzileyo.
Ipatheni yesi-4: Ukutyhala-kunye-Tsalo okuBonakalayo (kwiiNgcaciso eziSekwe kwiXabiso) Ukufumana izigqibo ezinzima ngakumbi, abasebenzisi banokufuna ukuqonda ukudibana kwezinto. Ukubonwa kwedatha elula kunokukwenza oku kucace ngaphandle kobunzima.
I-Heuristic: Sebenzisa imbonakalo yedatha elula, enombala (njengeetshathi zebha) ukubonisa izinto eziphembelele ngokufanelekileyo nangokungalunganga isigqibo.
Umzekelo: I-AI ivavanya iprofayile yomgqatswa womsebenzi.Kutheni lo mviwa engumdlalo we-75%:Iizinto ezinyusa amanqaku:5+ Iminyaka ye-UX yamava oPhandoUnolwazi kwiPythonFactors etyhala amanqaku ezantsi:Akukho mava ngeB2B SaaS
Ukufunda kunye nokusebenzisa ezi patheni zoyilo kwi-UX yemveliso yakho ye-AI kuya kunceda ukwandisa ukucaciswa. Unokusebenzisa iindlela ezongezelelweyo endingazibandakanyi nzulu apha. Oku kuquka oku kulandelayo:
Iingcaciso zolwimi lwendalo: Ukuguqulela imveliso ye-AI yobugcisa kulwimi olulula, olunencoko lwabantu abangengabo abachwephesha abanokuqonda ngokulula. Iinkcazo zomxholo: Ukubonelela ngengqiqo yesiphumo se-AI ngexesha elithile kunye nendawo, kubaluleke kakhulu kumsebenzi womsebenzisi. Ukubonwa okufanelekileyo: Ukusebenzisa iitshathi, iigrafu, okanye iimephu zokufudumala ukumela ngokubonakalayo inkqubo yokwenza izigqibo ze-AI, ukwenza idatha enzima ibe lula kwaye kulula ukuba abasebenzisi bayibambe.
Inqaku leSiphelo sangaphambili: Ukuguqulelwa kwezi ziphumo zokuchazwa kumava omsebenzisi angenamthungo kukwabonelela ngeseti yayo yoqwalaselo lobugcisa. Abaphuhlisi be-front-end bavame ukulwa noyilo lwe-API ukuze bafumane ngokufanelekileyo idatha yenkcazo, kunye neempembelelo zokusebenza (njengexesha langempela lemveliso yeenkcazo zentsebenziswano nganye yomsebenzisi) zidinga ukucwangciswa okucokisekileyo ukuphepha ukulinda. Eminye imizekelo yelizwe lokwenyani I-UPS Capital's DeliveryDefense I-UPS isebenzisa i-AI ukwabela "inqaku lonikezelo lokuzithemba" kwiidilesi ukuqikelela ukuba kunokwenzeka ukuba ipakeji ibiwe. Isoftware yabo yeDeliveryDefense ihlalutya idatha yembali kwindawo, ilahleko frequency, kunye nezinye izinto. Ukuba idilesi inamanqaku aphantsi, inkqubo inokubuyisela ngokukhawuleza ipakethe kwindawo ekhuselekileyo ye-UPS Access Point, inika inkcazo yesigqibo (umzekelo, "Ipakethi ibuyiselwe kwindawo ekhuselekileyo ngenxa yembali yobusela"). Le nkqubo ibonisa indlela i-XAI enokuthi isetyenziswe ngayo ukunciphisa umngcipheko kunye nokwakha ukuthembana kwabathengiukungafihli. Izithuthi ezizimeleyo Ezi zithuthi zexesha elizayo kuya kufuneka zisebenzise i-XAI ngokusebenzayo ukunceda izithuthi zabo ukuba zenze izigqibo ezikhuselekileyo nezicacileyo. Xa i-self-driving car brakes ngokukhawuleza, inkqubo inokubonelela ngenkcazo yexesha langempela ngesenzo sayo, umzekelo, ngokuchonga umhambi ngeenyawo ongena endleleni. Oku akubalulekanga nje kuphela ekuthuthuzelekeni komkhweli kunye nokuthembela kodwa yimfuneko yokulawula ukubonisa ukhuseleko kunye noxanduva lwenkqubo ye-AI. I-IBM Watson Health (kunye nemingeni yayo) Ngelixa uhlala ecatshulwa njengomzekelo jikelele we-AI kukhathalelo lwempilo, ikwaluphononongo olubalulekileyo lokubaluleka kwe-XAI. Ukungaphumeleli kweprojekthi yayo ye-Watson ye-Oncology igxininisa into enokuthi ihambe kakubi xa iinkcazo zingacaci, okanye xa idatha engaphantsi ithatha icala okanye ingekho kwindawo. Iingcebiso zenkqubo ngamanye amaxesha zazingahambelani nezenzo zeklinikhi zendawo kuba zazisekelwe kwizikhokelo ze-US-centric. Oku kusebenza njengentsomi yesilumkiso malunga nesidingo sokuqina, ukuqonda umxholo. Indima yoPhando we-UX: Ukukhomba kunye nokuqinisekisa iiNkcazo Izisombululo zethu zoyilo zisebenza kuphela ukuba zijongana nemibuzo yabasebenzisi abafanelekileyo ngexesha elifanelekileyo. Ingcaciso ephendula umbuzo umsebenzisi angenawo yingxolo nje. Kulapho uphando lwe-UX luba sisihlunu esibalulekileyo esinxibelelanayo kwisicwangciso se-XAI, siqinisekisa ukuba sicacisa ukuba yintoni kwaye ibaluleke kangakanani kubasebenzisi bethu. Indima yomphandi iphindwe kabini: okokuqala, ukwazisa isicwangciso ngokuchonga apho iinkcazo zifuneka khona, kwaye okwesibini, ukuqinisekiswa koyilo oluhambisa ezo ngcaciso. Ukwazisa iQhinga le-XAI (Yintoni emayichaze) Ngaphambi kokuba siyile inkcazo enye, kufuneka siqonde imodeli yengqondo yomsebenzisi yenkqubo ye-AI. Bakholelwa ukuba yenza ntoni? Ziphi izikhewu phakathi kokuqonda kwabo kunye nenyani yenkqubo? Lo ngumsebenzi osisiseko womphandi we-UX. Udliwano-ndlebe lweModeli yeNgqondo: Ukukhupha iiMvo zabasebenzisi kwiiNkqubo ze-AI Ngodliwano-ndlebe olunzulu, olulungelelanisiweyo, abaqeqeshi be-UX banokufumana ulwazi oluxabisekileyo malunga nendlela abasebenzisi ababona ngayo kwaye baqonde iinkqubo ze-AI. Ezi seshoni zenzelwe ukukhuthaza abasebenzisi ukuba bazobe ngokoqobo okanye bachaze "imodeli yengqondo" yabo yangaphakathi yendlela abakholelwa ngayo ukuba i-AI isebenza. Oku kuhlala kubandakanya ukubuza imibuzo evulelekileyo eyenza ukuba abasebenzisi bachaze ingqiqo yenkqubo, igalelo layo, kunye neziphumo zayo, kunye nobudlelwane phakathi kwezi zinto. Olu dliwano-ndlebe lunamandla kuba luhlala lutyhila iingcinga ezingezizo ezinzulu kunye neengcinga abanazo abasebenzisi malunga ne-AI. Umzekelo, umsebenzisi osebenzisana ne-injini yengcebiso angatsho ngokuzithemba ukuba inkqubo isekelwe kuphela kwimbali yabo yokujonga yangaphambili. Basenokungaqondi ukuba i-algorithm ikwabandakanya inkitha yezinye izinto, ezifana nexesha losuku abakhangela ngalo, izinto eziqhubekayo ngoku kwiqonga, okanye imikhwa yokujonga yabasebenzisi abafanayo. Ukutyhila lo msantsa phakathi kwemodeli yengqondo yomsebenzisi kunye neyona nto ingundoqo ye-AI logic ibaluleke kakhulu. Isixelela ngokuchanekileyo ukuba loluphi ulwazi esiludingayo ukunxibelelana nabasebenzisi ukubanceda bakhe imodeli yengqondo echaneke ngakumbi neyomeleleyo yenkqubo. Oku ke, linyathelo elibalulekileyo lokukhuthaza ukuthembana. Xa abasebenzisi beqonda, nakwinqanaba eliphezulu, indlela i-AI efikelela ngayo kwizigqibo okanye iingcebiso zayo, banokuthembela ngakumbi kwiziphumo zayo kwaye bathembele ekusebenzeni kwayo. Imephu yohambo lwe-AI: Ukuntywila ngokunzulu kwiTrasti yoMsebenzisi kunye nokucaciswa Ngokwenza imaphu yobuchule bohambo lomsebenzisi ngenqaku elinamandla e-AI, sifumana ulwazi oluxabiseke kakhulu kumaxesha achanekileyo apho kuvela khona ukubhideka, ukuphazamiseka, okanye ukungathembani okunzulu. Oku kuveza iindawo ezibalulekileyo apho imodeli yengqondo yomsebenzisi yendlela i-AI esebenza ngayo ingqubana nokuziphatha kwayo kwangempela. Qwalasela inkonzo yokusasaza umculo: Ngaba ukuthembela komsebenzisi kuyehla xa isincomo soluhlu lokudlala siziva “ngokungaqhelekanga,” siswele naluphi na unxibelelwano olubonakalayo kwimikhwa yabo yokuphulaphula yangaphambili okanye izinto abazikhethayo? Oku kubonwa ngokungaqhelekanga ngumceli mngeni othe ngqo kulindelo lomsebenzisi lokucutshungulwa okukrelekrele kunye nokwaphulwa kwesithembiso esicacileyo sokuba i-AI iyayiqonda incasa yabo. Ngokufanayo, kwisicelo solawulo lweefoto, ngaba abasebenzisi bafumana unxunguphalo olubalulekileyo xa i-AI yokuthega ifoto ye-AI isoloko ingabonakalisi ilungu losapho elixabisekileyo? Le mpazamo ingaphezulu kwe-glitch yobugcisa; ibetha entliziyweni yokuchaneka, ukwenziwa komntu, kunye nokubaunxibelelwano lweemvakalelo. Ezi ngongoma zentlungu ziyimpawu ezicacileyo ezibonisa ngokuchanekileyo apho inkcazo ebekwe kakuhle, ecacileyo kunye necacileyo iyimfuneko. Ezo ngcaciso zisebenza njengezona ndlela zibalulekileyo zokulungisa, ukulungisa ukophulwa kwentembeko, okuthi, ukuba kuyekwa kungalungiswa, kunokukhokelela ekulahlweni komsebenzisi. Amandla emephu yohambo lwe-AI alele ekukwazini ukusihambisa ngaphaya kokuchaza isiphumo sokugqibela senkqubo ye-AI. Ngelixa ukuqonda ukuba i-AI ivelise ntoni ibalulekile, ihlala ingonelanga. Endaweni yoko, le nkqubo isinyanzela ukuba sigxile ekuchazeni inkqubo ngamaxesha anzima. Oku kuthetha ukujongana:
Kutheni imveliso ethile yenziwe: Ngaba kwakungenxa yedatha yegalelo elithile? Imodeli ethile yoyilo? Zeziphi izinto eziphembelele isigqibo se-AI: Ngaba izinto ezithile bezinobunzima obungaphezulu? Indlela i-AI efikelele ngayo kwisigqibo sayo: Ngaba sinokunika inkcazo elula, efana neyokusebenza kwayo kwangaphakathi? Zeziphi iingcamango ezenziwa yi-AI: Ngaba kukho ukuqonda okucacileyo kwenjongo yomsebenzisi okanye idatha efuna ukuvela? Yiyiphi imida ye-AI: Ukunxibelelana ngokucacileyo oko i-AI ayinakuyenza, okanye apho ukuchaneka kwayo kunokugungqa, kwakha ulindelo oluyinyani.
Imephu yohambo ye-AI iguqula ingqikelelo engabonakaliyo ye-XAI ibe yinkqubo-sikhokelo esebenzayo, enokusetyenzwa kubasebenzi be-UX. Isenza ukuba sihambe ngaphaya kweengxoxo zethiyori zokuchazwa kwaye endaweni yoko sichonge amaxesha achanekileyo apho ukuthenjwa komsebenzisi kusemngciphekweni, ukubonelela ngeembono eziyimfuneko ukwakha amava e-AI anamandla, acacileyo, aqondakalayo, kwaye athembekileyo. Ekugqibeleni, uphando yindlela esityhila ngayo izinto ezingaziwayo. Iqela lakho lisenokuba lixoxa ngendlela yokuchaza ukuba kutheni imali-mboleko yaliwe, kodwa uphando lunokutyhila ukuba abasebenzisi bazixhalabele kakhulu ukuqonda ukuba idatha yabo yasetyenziswa njani kwasekuqaleni. Ngaphandle kophando, siqikelela nje ukuba abasebenzisi bethu bazibuza ntoni. Ukusebenzisana kuYilo (Indlela yokuchaza i-AI yakho) Nje ukuba uphando luchonge into emayichaze, i-loop yentsebenziswano kunye noyilo iyaqala. Abaqulunqi banokubonisa iipateni ebesixoxile ngazo ngaphambili-ingxelo ethi "Ngenxa", izilayidi ezisebenzisanayo-kwaye abaphandi banokubeka ezo ziyilo phambi kwabasebenzisi ukubona ukuba babambelele na. UkuSetyenziswa okuJolisiweyo kunye noVavanyo lokuQonda: Sinokuyila izifundo zophando ezivavanya ngokukodwa amacandelo e-XAI. Asivele sibuze, "Ngaba kulula ukuyisebenzisa le?" Siyabuza, "Emva kokubona oku, ungandixelela ngawakho amazwi ukuba kutheni le nkqubo incoma le mveliso?" okanye “Ndibonise into obuya kuyenza ukuze ubone ukuba ungafumana iziphumo ezahlukileyo.” Injongo apha kukulinganisa ukuqonda kunye nokusebenza, kunye nokusebenziseka. Ukulinganisa Ukuthembela Kwako: Sinokusebenzisa uphando olulula kunye nezikali zokukala phambi nasemva kokuba ingcaciso ibonisiwe. Umzekelo, sinokubuza umsebenzisi kwisikali se-5-point, "Uyithemba kangakanani le ngcebiso?" phambi kokuba babone ingxelo ethi “Kuba”, uze ubabuze kwakhona emva koko. Oku kubonelela ngedatha yobuninzi malunga nokuba iinkcazo zethu zihambisa inaliti kwi-trust. Le nkqubo idala i-loop enamandla, ephindaphindayo. Iziphumo zophando zazisa uyilo lokuqala. Olo yilo luye luvavanywe, kwaye izinto ezintsha ezifunyenweyo zibuyiselwa kwiqela loyilo ukuze licokiseke. Mhlawumbi ingxelo ethi “Ngenxa yokuba” ibiyijargony kakhulu, okanye isilayida esithi “Yintoni-Ukuba” besibhidekisa ngaphezu kokuxhobisa. Ngolu kuqinisekiswa kwentsebenziswano, siqinisekisa ukuba iinkcazo zokugqibela zichanekile ngokobugcisa, ziyaqondakala, ziluncedo, kwaye ziyakha ukuthemba abantu abasebenzisa imveliso. Indawo yeGoldilocks yeNkcazo Igama elibalulekileyo lokulumkisa: kunokwenzeka ukucacisa ngokugqithisileyo. Njengoko kwi-fairy tale, apho i-Goldilocks ifuna i-porridge 'eyayilungile', injongo yenkcazo efanelekileyo kukubonelela ngeenkcukacha ezifanelekileyo-kungekhona kakhulu kwaye kuncinci. Ukubhobhoza umsebenzisi ngayo yonke into eguquguqukayo kwimodeli kuya kukhokelela kumthwalo onzima wokuqonda kwaye kunokuncipha ukuthembana. Injongo ayikokwenza umsebenzisi abe yinzululwazi yedatha. Esinye isisombululo kukubhengeza okunenkqubela.
Qala ngezinto ezilula. Khokela ngenkcazo emfutshane ethi “Kuba”. Kubasebenzisi abaninzi, oku kuya kwanela. Nikeza indlela eya kwiinkcukacha. Nikeza ikhonkco elicacileyo, elinokruthakruthwano oluphantsi njengo "Funda Ngakumbi" okanye "Jonga ukuba oku kuye kwamiselwa njani." Bonisa ukuntsokotha. Emva kwelo khonkco, unokubonelela ngezilayidi ezisebenzisanayo, imiboniso, okanye uluhlu oluneenkcukacha ezininzi zezinto ezinegalelo.
Le ndlela yokwenziwa komsebenzi iyayihlonipha ingqalelo yabasebenzisi kunye nobuchule, ibonelela ngomlinganiselo ochanekileyoyolwazi kwiimfuno zabo. Makhe sicinge ukuba usebenzisa isixhobo sasekhaya esilumkileyo esicebisa ukufudumeza okufanelekileyo okusekwe kwizinto ezahlukeneyo. Qala ngento elula: "Ikhaya lakho ngoku lifudunyezwe ukuya kutsho kuma-72 degrees, elona qondo lobushushu lifanelekileyo lokonga amandla kunye nokuthuthuzela." Nikeza indlela eya kwiinkcukacha: Ezantsi koko, ikhonkco elincinane okanye iqhosha: "Kutheni i-72 degrees ilungile?" Bonisa ubunzima: Ukucofa elo khonkco kunokuvula isikrini esitsha esibonisa:
Izilayidi ezisebenzayo zobushushu obungaphandle, ukufuma, kunye nenqanaba lokuthuthuzela olithandayo, ebonisa ukuba ezi zilungisa njani na ubushushu obucetyiswayo. Umbono wokusetyenziswa kwamandla kumaqondo obushushu ahlukeneyo. Uluhlu lwezinto ezinegalelo ezifana “nexesha lemini,” “Iqondo lobushushu langaphandle langoku,” “Ukusetyenziswa kwamandla ngokwembali,” kunye “Nezivamvo zokuhlala.”
Kuyasebenza ukudibanisa iindlela ezininzi ze-XAI kwaye le ndawo yeGoldilocks yepatheni yeNkcazo, ekhuthaza ukubhengezwa okuqhubelekayo, ikhuthaza ngokungagungqiyo oku. Unokuqala ngenkcazo elula ethi "Ngenxa yokuba" (iPattern 1) yokuqonda ngokukhawuleza, kwaye unikeze ikhonkco elithi "Funda ngakumbi" elibonisa i-"What-If" Interactive (Pattern 2) okanye i-"Push-and-Pull Visual" (iPattern 4) yokuhlola nzulu. Umzekelo, inkqubo yesicelo semali-mboleko inokuqala ngokuchaza esona sizathu siphambili sokukhanyela (ukubaluleka kophawu), emva koko ivumele umsebenzisi ukuba anxibelelane nesixhobo esithi "Yintoni-Ukuba" ukubona ukuba utshintsho kwingeniso okanye kwityala labo lingasitshintsha njani isiphumo (izinto ezingeyonyani), kwaye ekugqibeleni, inike itshati eneenkcukacha "Tyhila-kunye-Tsalo" (ingcaciso esekwe kwixabiso) ukubonisa yonke imiba emihle kunye nengalunganga. Le ndlela yolwaleko ivumela abasebenzisi ukuba bafikelele kwinqanaba leenkcukacha abazidingayo, xa bezifuna, ukuthintela ukuxinana kwengqondo ngelixa besabonelela ngokucace okubanzi. Ukumisela ukuba zeziphi izixhobo ze-XAI kunye neendlela zokusetyenziswa ngumsebenzi wophando olucokisekileyo lwe-UX. Udliwano-ndlebe lwemodeli yengqondo kunye nemephu yohambo lwe-AI zibalulekile ekuchongeni iimfuno zabasebenzisi kunye neentlungu ezinxulumene nokuqonda kunye nokuthembela kwe-AI. Udliwano-ndlebe lwemodeli yengqondo lunceda ukufumanisa ukungaqondi kakuhle komsebenzisi malunga nendlela i-AI esebenza ngayo, ebonisa iindawo apho iinkcazo ezisisiseko (ezifana nokubaluleka kophawu okanye iingcaciso zendawo) zifuneka. Imephu yohambo lwe-AI, ngakolunye uhlangothi, ichaza amaxesha abalulekileyo okudideka okanye ukungathembeki ekusebenzisaneni komsebenzisi kunye ne-AI, ukubonisa apho iinkcazo ezingaphezulu okanye ezidibeneyo (njengezinto ezichaseneyo okanye iinkcazo ezisekelwe kwixabiso) kuya kuba luncedo kakhulu ukuvuselela ukuthembela kunye nokubonelela nge-arhente.
Ekugqibeleni, eyona ndlela ingcono yokukhetha ubuchule kukuvumela uphando lomsebenzisi lukhokele izigqibo zakho, uqinisekise ukuba iinkcazelo oziyileyo zijongana ngqo nemibuzo yokwenyani yabasebenzisi kunye neenkxalabo, kunokuba banike iinkcukacha zobugcisa ngenxa yabo. I-XAI yee-Arhente eziNxibeleleyo Ezinye zezona nkqubo ze-AI zintsha, ezaziwa ngokuba zii-arhente zokuqiqa ezinzulu, zivelisa “ikhonkco lokucinga” elicacileyo kuwo wonke umsebenzi onzima. Abachazi nje imithombo yolwazi; babonisa indlela esengqiqweni, inyathelo nenyathelo abalithathileyo ukuze bafikelele kwisigqibo. Ngelixa oku kungafihlwa kubonelela ngomxholo obalulekileyo, umdlalo-ngomdlalo ogubungela imihlathi emininzi unokuziva usoyisa kumsebenzisi ngokuzama ukugqibezela umsebenzi. Imigaqo ye-XAI, ngakumbi iNdawo yeNkcazo yeGoldilocks, isebenza ngqo apha. Sinokululungisa uhambo, sisebenzisa ukubhengeza okunenkqubela ukubonisa kuphela isiphelo sokugqibela kunye nelona nyathelo libalulekileyo kwinkqubo yokucinga kuqala. Abasebenzisi banokukhetha ukungena ukuze babone ingqikelelo epheleleyo, eneenkcukacha, enamanyathelo amaninzi xa befuna ukuphinda bajonge ingqiqo okanye bafumane inyani ethile. Le ndlela ihlonipha ingqalelo yabasebenzisi ngelixa igcina ukungafihli ngokupheleleyo kwe-arhente. Amanyathelo Alandelayo: Ukuxhobisa Uhambo lwakho lwe-XAI Ukuchazwa yintsika esisiseko yokwakha iimveliso ze-AI ezithembekileyo nezisebenzayo. Kubasebenzi abaphambili abafuna ukuqhuba olu tshintsho kwintlangano yabo, uhambo ludlulela ngaphaya kweepatheni zoyilo ukuya kubhengezo kunye nokufunda okuqhubekayo. Ukuzinzisa ukuqonda kwakho kunye nesicelo esisebenzayo, qwalasela ukuhlola izixhobo ezifana ne-AI Explainability 360 (AIX360) toolkit evela kuPhando lwe-IBM okanye i-Google's What-If Tool, enikezela ngeendlela ezisebenzisanayo zokuphonononga imodeli yokuziphatha kunye neenkcazo. Ukusebenzisana noluntu olunje ngeForam ye-AI enoxanduva okanye amaqela ophando athile agxile kwi-AI egxile emntwini anokubonelela ngolwazi oluxabisekileyo kunye namathuba entsebenziswano. Okokugqibela, yiba ngummeli we-XAI kumbutho wakho.Ukucaciswa kwesakhelo njengotyalo-mali olucwangcisiweyo. Qwalasela isandi esifutshane kubunkokeli bakho okanye amaqela asebenza ngokudibeneyo: "Ngotyalo-mali kwi-XAI, siya kugqitha ngaphaya kokwakha ukuthembana; siya kukhawulezisa ukwamkelwa kwabasebenzisi, sinciphise iindleko zenkxaso ngokuxhobisa abasebenzisi ukuqonda, kwaye sithobe imingcipheko ebalulekileyo yeenqobo ezisesikweni kunye nolawulo ngokubhenca ucalucalulo olunokubakho. Olu luyilo oluhle kunye neshishini elilumkileyo."
Ilizwi lakho, elisekelwe ekuqondeni okusebenzayo, libalulekile ekukhupheni i-AI ngaphandle kwebhokisi emnyama kunye nentsebenziswano yentsebenziswano nabasebenzisi.
