Malunga ne-Artificial Intelligence Bias

tl; dr:

• Ukufunda koomatshini kukhangela iipateni kwidatha. Kodwa ubukrelekrele bokwenziwa bunokuba "nomkhethe" - oko kukuthi, ukufumana iipateni ezingalunganga. Umzekelo, inkqubo yokukhangela umhlaza wolusu esekwe kwifoto inokunika ingqwalasela eyodwa kwimifanekiso ethathwe kwiofisi kagqirha. Ukufunda ngomatshini akukwazi ukuqonda: i-algorithms yayo ichonga kuphela iipateni kumanani, kwaye ukuba idatha ayimeleli, kuya kuba njalo umphumo wokuqhubekeka kwayo. Kwaye ukubamba iibhugi ezinjalo kunokuba nzima ngenxa yobuchwephesha bokufunda koomatshini.
• Eyona ngxaki icacileyo neyoyikisayo kukwahluka kwabantu. Kukho izizathu ezininzi zokuba kutheni idatha malunga nabantu inokulahlekelwa yinjongo nakwinqanaba lokuqokelela. Kodwa ungacingi ukuba le ngxaki ichaphazela abantu kuphela: ubunzima obufanayo buvela xa uzama ukufumanisa umkhukula kwindawo yokugcina impahla okanye i-turbine yegesi engaphumeleli. Ezinye iisistim zinokuthambekela kumbala wolusu, ezinye ziya kuthambekela kwicala leenzwa ze-Siemens.
• Ezo ngxaki aziqali kukufunda ngoomatshini, kwaye azikho kwaphela kuzo. Iingcinga ezingalunganga zenziwa kuyo nayiphi na isakhiwo esiyinkimbinkimbi, kwaye ukuqonda ukuba kutheni isigqibo esithile senziwe sihlala sinzima. Kufuneka sikulwe oku ngendlela ebanzi: ukwenza izixhobo kunye neenkqubo zokuqinisekisa - kwaye sifundise abasebenzisi ukuze bangazilandeli ngokumfamekileyo iingcebiso ze-AI. Ukufunda ngoomatshini kwenza ezinye izinto ngcono kakhulu kunathi - kodwa izinja, umzekelo, zisebenza ngakumbi kunabantu ekuchongeni iziyobisi, engesosizathu sokuzisebenzisa njengamangqina kunye nokwenza izigwebo ezisekelwe kubungqina bazo. Kwaye izinja, ngendlela, zihlakaniphe kakhulu kunayo nayiphi na inkqubo yokufunda koomatshini.

Ukufunda ngoomatshini yenye yezona ndlela zibalulekileyo zetekhnoloji namhlanje. Le yenye yeendlela eziphambili apho itekhnoloji iya kutshintsha ihlabathi elisingqongileyo kule minyaka ilishumi izayo. Eminye imiba yolu tshintsho ibangela inkxalabo. Ngokomzekelo, impembelelo enokubakho yokufunda koomatshini kwimarike yabasebenzi, okanye ukusetyenziswa kwayo ngeenjongo ezingekho mthethweni (umzekelo, ngoorhulumente abagunyazisiweyo). Kukho enye ingxaki esi sithuba sijongana nayo: bias intelligence bias.

Eli asilobali lilula.

I-AI kaGoogle inokufumana iikati. Ezi ndaba zango-2012 zaziyinto ekhethekileyo ngelo xesha.

Yintoni "i-AI Bias"?

"Idatha eluhlaza" zombini i-oxymoron kunye nombono ombi; idatha kufuneka ilungiswe kakuhle kwaye ngononophelo. —UGeoffrey Bocker

Kwindawo ethile ngaphambi kwe-2013, ukuze wenze inkqubo ethi, ithi, ibone iikati kwiifoto, kufuneka uchaze amanyathelo anengqiqo. Indlela yokufumana iikona emfanekisweni, qaphela amehlo, uhlalutye ukuthungwa koboya, ukubala i-paws, njalo njalo. Emva koko hlanganisa onke amacandelo kwaye ufumanise ukuba ayisebenzi ngokwenene. Ngokufana nehashe lomatshini - ngokwethiyori inokwenziwa, kodwa ekusebenzeni kunzima kakhulu ukuyichaza. Isiphelo ngamakhulu (okanye amawaka) emithetho ebhalwe ngesandla. Kwaye akukho modeli isebenzayo.

Ngokufika kokufundwa komatshini, sayeka ukusebenzisa imithetho "ye-manual" yokuqaphela into ethile. Endaweni yoko, sithatha iisampuli eziliwaka "le", X, iisampuli eziliwaka "ezinye", i-Y, kwaye ikhompyutha yakha imodeli ngokusekelwe kuhlalutyo lwabo lwamanani. Emva koko sinika le modeli idatha yesampulu kwaye imisela ngokuchanekileyo ukuba iyalingana na enye yeeseti. Ukufunda ngoomatshini kuvelisa imodeli esuka kwidatha kunokuba ibhalwe ngumntu. Iziphumo ziyamangalisa, ngakumbi kwintsimi yomfanekiso kunye nokuqatshelwa kwepateni, kwaye yiyo loo nto yonke imboni yezobuchwephesha ngoku ihambela ekufundeni komatshini (ML).

Kodwa akukho lula ngolo hlobo. Kwilizwe lokwenyani, amawaka akho emizekelo ka X okanye Y ikwaqulathe A, B, J, L, O, R, kunye no L. Ezi zinokungasasazwa ngokulinganayo, kwaye ezinye zinokwenzeka rhoqo kangangokuba inkqubo iya kuhlawula ngaphezulu ingqalelo kuzo kunokuba kwizinto ezinomdla kuwe.

Kuthetha ukuthini oku xa kusenziwa? Umzekelo wam endiwuthandayo kuxa iinkqubo zokuqaphela umfanekiso jonga induli enengca kwaye uthi, "iigusha". Kucacile ukuba kutheni: uninzi lwemizekelo yeefoto "zezimvu" zithathwa emadlelweni apho zihlala khona, kwaye kule mifanekiso ingca ithatha indawo eninzi kuneengcongconi ezimhlophe, kwaye yingca ebonwa njengeyona nto ibalulekileyo. .

Kukho imizekelo enzulu ngakumbi. Enye yakutshanje iprojekthi ukufumanisa umhlaza wolusu kwiifoto. Kwavela ukuba i-dermatologists ihlala ifota irula kunye nokubonakaliswa komhlaza wesikhumba ukurekhoda ubungakanani bokubunjwa. Akukho balawuli kumzekelo weefoto zesikhumba esinempilo. Kwinkqubo ye-AI, abalawuli abanjalo (ngokuchanekileyo, iipikseli esizichaza ngokuthi "umlawuli") ziye zaba ngomnye umahluko phakathi kweeseti zemizekelo, kwaye ngamanye amaxesha kubaluleke kakhulu kunokugqithisa okuncinci kwesikhumba. Ke inkqubo edalwe ukuchonga umhlaza wolusu ngamanye amaxesha waqaphela abalawuli endaweni.

Inqaku eliphambili apha kukuba inkqubo ayinakho ukuqonda kwesemantic malunga nokuba ijonge ntoni. Sijonga isethi yeepikseli kwaye sibone kuzo igusha, ulusu okanye iirula, kodwa inkqubo ngumgca manani kuphela. Ayiboni sithuba samacala amathathu, ayiboni izinto, ukumila, okanye iigusha. Ubona nje iipateni kwidatha.

Ubunzima bokufumanisa iingxaki ezinjalo kukuba inethiwekhi ye-neural (imodeli eveliswe yinkqubo yokufunda ngomatshini wakho) iqulathe amawaka amakhulu amawaka eendawo. Akukho ndlela ilula yokujonga kwimodeli kwaye ubone ukuba yenza njani isigqibo. Ukuba nendlela enjalo kuya kuthetha ukuba inkqubo ilula ngokwaneleyo ukuchaza yonke imithetho ngesandla, ngaphandle kokusebenzisa umatshini wokufunda. Abantu banenkxalabo yokuba ukufunda ngomatshini kuye kwaba yinto yebhokisi emnyama. (Ndiza kuchaza kamva ukuba kutheni lo mzekeliso usemkhulu kakhulu.)

Oku, ngokuqhelekileyo, yingxaki yecala kwingqondo yokwenziwa okanye ukufunda ngomatshini: inkqubo yokufumana iipatheni kwidatha inokufumana iipatheni ezingalunganga, kwaye awuyi kuqaphela. Olu luphawu olusisiseko lwetekhnoloji, kwaye icacile kuye wonke umntu osebenza nayo kwizifundiswa nakwiinkampani ezinkulu zetekhnoloji. Kodwa iziphumo zayo zintsonkothile, kwaye kunjalo nezisombululo zethu ezinokubakho kwezo ziphumo.

Makhe sithethe ngeziphumo kuqala.

I-AI inokuthi, ngokungafihlisiyo kuthi, yenze ukhetho ngokuthanda iindidi ezithile zabantu, ngokusekwe kwinani elikhulu lemiqondiso engabonakaliyo.

I-AI ye-Bias Scenarios

Ngokucacileyo kwaye iyoyikisa, le ngxaki inokuzibonakalisa xa ifika kwiintlobo ezahlukeneyo zabantu. Kutshanje kwavakala amarheukuba iAmazon yazama ukwakha inkqubo yokufunda ngomatshini ukulungiselela ukuhlolwa kokuqala kwabagqatswa bemisebenzi. Ekubeni kukho amadoda amaninzi phakathi kwabasebenzi base-Amazon, imizekelo "yokuqeshwa ngempumelelo" iphinda iphinde ibe ngamadoda, kwaye kwakukho amadoda amaninzi ekukhetheni kwakhona okucetyisiweyo yinkqubo. I-Amazon yaqaphela oku kwaye ayizange ikhulule inkqubo kwimveliso.

Eyona nto ibalulekileyo kulo mzekelo kukuba le nkqubo yayinamarhe okuba ixhasa abafaki-zicelo abangamadoda, nangona isini sasingakhange sichazwe kwi-resume. Inkqubo yabona ezinye iipatheni kwimizekelo "yengqesho elungileyo": umzekelo, abafazi banokusebenzisa amagama akhethekileyo ukuchaza okufezekisiweyo, okanye ukuzonwabisa okukhethekileyo. Kakade ke, inkqubo yayingazi ukuba yintoni na "ihoki", okanye "abantu" ngubani, okanye yintoni "impumelelo" - yenza nje uhlalutyo lwamanani esicatshulwa. Kodwa iipateni azibonileyo zisenokungaqatshelwa ngabantu, kwaye ezinye zazo (umzekelo, into yokuba abantu besini esahlukileyo bachaza impumelelo ngokwahlukileyo) mhlawumbi bekuya kuba nzima kuthi ukuzibona nokuba sizijongile.

Ngaphezulu - kubi. Inkqubo yokufunda ngomatshini elunge kakhulu ekufumaneni umhlaza eluswini oluluthuthu isenokungasebenzi kakuhle eluswini olumnyama, okanye ngokuphambanayo. Akunjalo ngenxa yecala, kodwa ngenxa yokuba mhlawumbi ufuna ukwakha imodeli eyahlukileyo yombala wesikhumba, ukhetha iimpawu ezahlukeneyo. Iinkqubo zokufunda ngoomatshini azitshintshi nakwindawo emxinwa njengokuqonda umfanekiso. Kufuneka udibanise inkqubo, ngamanye amaxesha nje ngolingo kunye nempazamo, ukufumana isibambo esihle kwiifitsha kwidatha onomdla kuyo de ube ufezekise ukuchaneka oyifunayo. Kodwa into ongayiqaphelayo kukuba inkqubo ichanekile i-98% yexesha kunye neqela elinye, kwaye kuphela i-91% (nokuba ichaneke ngakumbi kunohlalutyo lwabantu) kunye nomnye.

Ukuza kuthi ga ngoku ndisebenzise ubukhulu becala imizekelo enxulumene nabantu neempawu zabo. Ingxoxo ejikeleze le ngxaki ikakhulu igxile kwesi sihloko. Kodwa kubalulekile ukuqonda ukuba umkhethe ebantwini yinxalenye yengxaki kuphela. Siza kusebenzisa ukufunda koomatshini kwizinto ezininzi, kwaye impazamo yesampulu iya kuhambelana nazo zonke. Ngakolunye uhlangothi, ukuba usebenza nabantu, i-bias kwidatha ayinakunxulumana nabo.

Ukuyiqonda le nto, makhe sibuyele kumzekelo womhlaza wolusu kwaye siqwalasele izinto ezintathu ezinokwenzeka zokungaphumeleli kwenkqubo.

1. Ukusasazwa okungafaniyo kwabantu: inani elingalinganiyo leefoto zolusu olwahlukileyo, olukhokelela kwizinto ezingeyonyani okanye ezingezizo ezingalunganga ngenxa yebala.
2. Idatha apho le nkqubo iqeqeshwa khona iqulethe into eyenzeka rhoqo kwaye isasazwe ngokungafaniyo nabantu kwaye ayinalo ixabiso lokuxilonga: irula kwiifoto zomhlaza wolusu okanye ingca kwiifoto zeegusha. Kule meko, umphumo uya kuhluka ukuba inkqubo ifumana iipikseli kumfanekiso wento ebonwa liso lomntu "njengomlawuli".
3. Idatha iqulethe uphawu lomntu wesithathu umntu angenakulubona nokuba uyayikhangela.

Ingaba ithetha ntoni? Siyazi i-priori ukuba idatha inokumela amaqela ahlukeneyo abantu ngokwahlukileyo, kwaye ubuncinci sinokucwangcisa ukukhangela izinto ezinjalo. Ngamanye amazwi, kukho izizathu ezininzi zentlalo zokucinga ukuba idatha malunga namaqela abantu sele iqulethe i-bias. Ukuba sijonga ifoto kunye nomlawuli, siya kubona lo mlawuli - sivele singayinaki ngaphambili, sisazi ukuba akunandaba, kwaye silibale ukuba inkqubo ayazi nto.

Kodwa kuthekani ukuba zonke iifoto zakho zesikhumba esingenampilo zithathwe kwiofisi phantsi kokukhanya kwe-incandescent, kwaye ulusu lwakho olunempilo luthathwe phantsi kokukhanya kwe-fluorescent? Kuthekani ukuba, emva kokuba ugqibile ukudubula ulusu olusempilweni, ngaphambi kokudubula ulusu olungenampilo, uhlaziye inkqubo yokusebenza kwifowuni yakho, kwaye i-Apple okanye iGoogle yatshintsha kancinci i-algorithm yokunciphisa ingxolo? Umntu akanakuqaphela oku, kungakhathaliseki ukuba ukhangeleka kangakanani na iimpawu ezinjalo. Kodwa inkqubo yokusetyenziswa komatshini iya kuyibona kwangoko kwaye isebenzise oku. Akazi nto.

Ukuza kuthi ga ngoku sithethe ngolungelelwaniso olukhohlisayo, kodwa kusenokwenzeka ukuba idatha ichanekile kwaye iziphumo zichanekile, kodwa awufuni ukuzisebenzisa ngenxa yezizathu zokuziphatha, ezisemthethweni, okanye zolawulo. Ngokomzekelo, eminye imimandla ayibavumeli abafazi ukuba bafumane isaphulelo kwiinshorensi yabo, nangona abafazi besenokuba ngabaqhubi abakhuselekileyo. Sinokucinga ngokulula inkqubo ethi, xa sihlalutya idatha yembali, iya kunika umngcipheko ophantsi kumagama abasetyhini. Kulungile, masisuse amagama kukhetho. Kodwa khumbula umzekelo we-Amazon: inkqubo inokugqiba isini ngokusekwe kwezinye izinto (nangona ingayazi ukuba yintoni na isini, okanye nokuba yintoni imoto), kwaye awuyi kuqaphela oku de umlawuli ahlalutye kwakhona amaxabiso kuwe. uya kuhlawuliswa.

Ekugqibeleni, kudla ngokucingelwa ukuba siya kusebenzisa kuphela iinkqubo ezinjalo kwiiprojekthi ezibandakanya abantu kunye nobudlelwane boluntu. Oku akulunganga. Ukuba wenza iiturbines zerhasi, uya kufuna ukusebenzisa umatshini wokufunda kwitelemetry esasazwa ngamashumi okanye amakhulu ezivamvo kwimveliso yakho (umsindo, ividiyo, ubushushu, kunye nazo naziphi na ezinye izibonisi zoluvo zivelisa idatha enokuhlengahlengiswa ngokulula ukwenza umatshini. imodeli yokufunda). Ngokucingayo, unokuthi, "Nantsi idatha evela kwiwaka leeinjini zomoya ezisileleyo ngaphambi kokuba zisilele, kwaye nantsi idatha evela kwiwaka leeinjini zomoya ezingakhange zisilele. Yakha imodeli ukuchaza ukuba yintoni umahluko phakathi kwabo. " Ewe, ngoku khawucinge ukuba i-Siemens sensors ifakwe kwi-75% yee-turbines ezimbi, kwaye kuphela i-12% yezinto ezilungileyo (akukho nxu lumano kunye nokungaphumeleli). Inkqubo iya kwakha imodeli yokufumana ii-turbines ezineenzwa ze-Siemens. Yhu!

Umfanekiso-Moritz Hardt, UC Berkeley

Ukulawula i-AI Bias

Yintoni esinokuyenza ngayo? Unokujongana nalo mba ngee-engile ezintathu:

1. Ubukhali bemethodological ekuqokeleleni nasekulawuleni iinkcukacha zoqeqesho lwenkqubo.
2. Izixhobo zobugcisa zokuhlalutya kunye nokuxilonga imodeli yokuziphatha.
3. Qeqesha, fundisa, kwaye ulumke xa usebenzisa ukufunda koomatshini kwiimveliso.

Kukho ihlaya kwincwadi kaMolière ethi "The Bourgeois in the Nobility": enye indoda yaxelelwa ukuba uncwadi lwahlulahlulwe lube yiprose kunye neembongo, kwaye wayevuyiswa kukufumanisa ukuba wayethetha ngeprose ubomi bakhe bonke, engazi. Mhlawumbi le yindlela abavakalelwa ngayo abamanani namhlanje: bengayiqondi loo nto, baye banikela imisebenzi yabo kubukrelekrele bokwenziwa kunye nempazamo yesampulu. Ukukhangela impazamo yesampulu kunye nokuba nexhala malunga nayo ayisiyongxaki entsha, kufuneka nje sisondele kwisisombululo sayo. Njengoko kukhankanyiwe ngasentla, kwezinye iimeko kulula ngokwenene ukwenza oku ngokufunda iingxaki ezinxulumene data yabantu. Thina a priori sicingela ukuba sinokuba nocalucalulo malunga namaqela ahlukeneyo abantu, kodwa kunzima kuthi nokuba sicinge ucalucalulo malunga ne-Siemens sensors.

Yintoni entsha ngayo yonke le nto, ngokuqinisekileyo, kukuba abantu abasazenzi uhlalutyo lweenkcukacha-manani ngokuthe ngqo. Yenziwa ngoomatshini abakha iimodeli ezinkulu, ezinzima ekunzima ukuziqonda. Umba wokwenziwa kwezinto elubala ngomnye wemiba ephambili yengxaki yokuthatha icala. Sinoloyiko lokuba le nkqubo ayikhethi cala nje, kodwa akukho ndlela yokubona i-bias yayo, kwaye ukufunda koomatshini kwahlukile kwezinye iindlela zokuzisebenzela, ezimele ukuba zibandakanya amanyathelo acacileyo asengqiqweni anokuhlolwa.

Kukho iingxaki ezimbini apha. Sisenokukwazi ukwenza uphicotho oluthile lweenkqubo zokufunda ngoomatshini. Kwaye ukuphicotha nayiphi na enye inkqubo ngokwenene akukho lula.

Okokuqala, enye yezalathiso zophando lwangoku kwinkalo yokufunda koomatshini kukukhangela iindlela zokuchonga ukusebenza okubalulekileyo kweenkqubo zokufunda koomatshini. Oko kuthethiweyo, ukufunda ngoomatshini (kwimeko ekuyo ngoku) licandelo lezenzululwazi elitsha ngokupheleleyo elitshintsha ngokukhawuleza, ngoko ungacingi ukuba izinto ezingenakwenzeka namhlanje azinakuba yinto yokwenene. Iprojekthi I-OpenAI - umzekelo onomdla wale nto.

Okwesibini, uluvo lokuba umntu unokuvavanya kwaye aqonde inkqubo yokwenza izigqibo kwiinkqubo esele zikhona okanye imibutho ilungile ngokwethiyori, kodwa ke ngoko xa isenziwa. Ukuqonda indlela izigqibo ezenziwa ngayo kwintlangano enkulu akulula. Nokuba kukho inkqubo yokwenziwa kwezigqibo esesikweni, ayibonisi ukuba abantu banxibelelana njani na, kwaye bona ngokwabo kaninzi abanayo indlela esengqiqweni, ecwangcisiweyo yokwenza izigqibo zabo. Njengoko watshoyo umlingane wam Vijay Pande, abantu nabo ziibhokisi ezimnyama.

Thatha iwaka labantu kwiinkampani ezininzi ezidibanayo kunye namaziko, kwaye ingxaki iba nzima ngakumbi. Siyazi emva kwento yokuba iSpace Shuttle yayimiselwe ukwaphuka xa kubuya imbuyekezo, kwaye abantu abakwiNASA babenolwazi olwabanika isizathu sokucinga ukuba kukho into embi enokwenzeka, kodwa inkqubo ngokubanzi Bendingayazi le nto. I-NASA ide yadlula kuphicotho olufanayo emva kokuba iphulukene nenqanawa yayo yangaphambili, ukanti yaphulukana nenye ngesizathu esifanayo. Kulula ukuphikisa ngelithi imibutho nabantu balandela imithetho ecacileyo nesengqiqweni enokuvavanywa, iqondwe, ize itshintshwe—kodwa amava angqina okwahlukileyo koko. Oku "Ukukhohlisa kweGosplan».

Ndihlala ndithelekisa ukufundwa komatshini kumaziko edatha, ngakumbi ezobudlelwane-iteknoloji entsha eyisiseko etshintshe izakhono zesayensi yekhompyuter kunye nehlabathi elijikelezileyo, eliye laba yinxalenye yayo yonke into, esiyisebenzisa rhoqo ngaphandle kokuyiqonda. Iidatabase nazo zineengxaki, kwaye zinesimo esifanayo: inkqubo inokwakhiwa kwiingcamango ezimbi okanye idatha embi, kodwa kuya kuba nzima ukuqaphela, kwaye abantu abasebenzisa inkqubo baya kwenza oko ibaxelela ngaphandle kokubuza imibuzo. Zininzi iziqhulo ezindala malunga nabantu berhafu abakhe bapela igama lakho ngokungalunganga, kwaye ukubaqinisekisa ukuba balungise impazamo kunzima kakhulu kunokutshintsha igama lakho. Kukho iindlela ezininzi zokucinga ngale nto, kodwa akucaci ukuba yeyiphi engcono: njengengxaki yobugcisa kwi-SQL, okanye njenge-bug ekukhululweni kwe-Oracle, okanye njengokusilela kwamaziko e-bureaucratic? Kunzima kangakanani ukufumana ibug kwinkqubo ekhokelele kwinkqubo engenalo uphawu lolungiso lokuchwetheza? Ngaba oku kwakunokuqondwa ngaphambi kokuba abantu baqalise ukukhalaza?

Le ngxaki ibonakaliswa ngakumbi ngamabali xa abaqhubi beqhubela emilanjeni ngenxa yedatha yakudala kwi-navigator. Kulungile, iimephu kufuneka zihlaziywe rhoqo. Kodwa ingaba iTomTom inetyala elingakanani ngenqwelo-mafutha yakho eqhutyelwe elwandle?

Isizathu sokuba ndithethe oku kukuba ewe, i-bias yokufunda ngomatshini iya kudala iingxaki. Kodwa ezi ngxaki ziya kufana nezo siye sajamelana nazo kwixesha elidlulileyo, kwaye zinokuqatshelwa kwaye zisonjululwe (okanye zingasonjululwa) malunga nendlela ebesikwazi ngayo ngaphambili. Ke ngoko, imeko apho i-AI bias ibangela ingozi ayinakwenzeka ukuba yenzeke kubaphandi abaphezulu abasebenza kumbutho omkhulu. Inokwenzeka ukuba, enye ikontraka yetekhnoloji engabalulekanga okanye umthengisi wesoftware uya kubhala into emadolweni, esebenzisa amacandelo omthombo ovulekileyo, amathala eencwadi kunye nezixhobo abangaziqondiyo. Kwaye umxhasi ongenathamsanqa uya kuthenga ibinzana elithi "ubukrelekrele bokufakelwa" kwinkcazo yemveliso kwaye, ngaphandle kokubuza nayiphi na imibuzo, ayisasaze kubasebenzi bakhe abahlawula kancinci, ebayalela ukuba benze oko i-AI ikutshoyo. Yile nto kanye eyenzekileyo ngoovimba beenkcukacha. Oku akuyongxaki yobukrelekrele bokwenziwa, okanye ingxaki yesoftware. Le yinkalo yomntu.

isiphelo

Ukufunda ngoomatshini kunokwenza nantoni na onokuyifundisa inja - kodwa awunakuze uqiniseke ukuba yintoni kanye oyifundise inja.

Ndihlala ndivakalelwa kukuba igama elithi "ubukrelekrele bokwenziwa" lingena kuphela kwindlela yeencoko ezinje. Eli gama linika umbono ongeyonyani wokuba siwudalile - obu bulumko. Ukuba sisendleleni eya HAL9000 okanye Skynet - into eneneni uyayiqonda. Kodwa hayi. Aba ngoomatshini nje, kwaye kuchaneke ngakumbi ukubathelekisa, masithi, umatshini wokuhlamba. Uhlamba iimpahla ngcono kakhulu kunomntu, kodwa ukuba ufaka izitya kuye endaweni yokuhlamba, yena ... uya kuzihlamba. Izitya ziya kuhlambuluka. Kodwa oku akuyi kuba yile nto ubuyilindele, kwaye oku akuyi kwenzeka ngenxa yokuba inkqubo inomkhethe malunga nezitya. Umatshini wokuhlamba akazi ukuba zeziphi izitya okanye iimpahla zithini - ngumzekelo nje wokuzenzekelayo, ngokwengqiqo akukho nto yahlukileyo kwindlela iinkqubo ezazizenzekela ngayo ngaphambili.

Nokuba sithetha ngeemoto, iinqwelomoya, okanye oovimba beenkcukacha, ezi nkqubo ziya kuba namandla kakhulu kwaye zilinganiselwe kakhulu. Baya kuxhomekeka ngokupheleleyo kwindlela abantu abasebenzisa ngayo ezi nkqubo, nokuba iinjongo zabo zilungile okanye zimbi, kwaye bayiqonda kangakanani indlela abasebenza ngayo.

Ngoko ke, ukuthi “ubukrelekrele bokwenziwa yimathematika, ngoko ke ayinakuba nomkhethe” bubuxoki ngokupheleleyo. Kodwa kububuxoki ngokufanayo ukuthi ukufunda ngoomatshini “kuxhomekeke kwindalo.” Ukufunda ngoomatshini kufumanisa iipateni kwidatha, kwaye zeziphi iipateni ezizifumanayo zixhomekeke kwidatha, kwaye idatha ixhomekeke kuthi. Kanye njengokuba sisenza nabo. Ukufunda ngoomatshini kwenza ezinye izinto ngcono kakhulu kunathi - kodwa izinja, umzekelo, zisebenza ngakumbi kunabantu ekuchongeni iziyobisi, engesosizathu sokuzisebenzisa njengamangqina kunye nokwenza izigwebo ezisekelwe kubungqina bazo. Kwaye izinja, ngendlela, zihlakaniphe kakhulu kunayo nayiphi na inkqubo yokufunda koomatshini.

Ukuguqulelwa: UDiana Letskaya.
Ukuhlela: Aleksey Ivanov.
Uluntu: @PonchikNews.

umthombo: www.habr.com