{"id":32781,"date":"2019-10-31T21:48:54","date_gmt":"2019-10-31T18:48:54","guid":{"rendered":"https:\/\/prohoster.info\/blog\/o-predvzyatosti-iskusstvennogo-intellekta\/"},"modified":"2019-10-31T21:48:54","modified_gmt":"2019-10-31T18:48:54","slug":"o-predvzyatosti-iskusstvennogo-intellekta","status":"publish","type":"post","link":"https:\/\/prohoster.info\/ro\/blog\/news\/o-predvzyatosti-iskusstvennogo-intellekta","title":{"rendered":"Despre biasul inteligen\u021bei artificiale","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"<p><img decoding=\"async\" alt=\"Despre biasul inteligen\u021bei artificiale\" src=\"\/wp-content\/uploads\/2019\/04\/05a731677e1231acceff953db8e1e26a.jpg\" style=\"display:block;margin: 0 auto;\" \/><\/p>\n<p><\/p>\n<h2 id=\"tldr\">tl;dr:<\/h2>\n<p><\/p>\n<ul>\n<li>\u00cenv\u0103\u021barea automat\u0103 caut\u0103 tipare \u00een date. \u00cens\u0103 inteligen\u021ba artificial\u0103 poate fi \u201esupus\u0103 prejudec\u0103\u021bii\u201d - adic\u0103, poate g\u0103si tipare eronate. De exemplu, un sistem de detectare a cancerului de piele \u00een fotografii poate acorda o aten\u021bie deosebit\u0103 imaginilor realizate \u00eentr-un cabinet medical. \u00cenv\u0103\u021barea automat\u0103 nu \u0219tie <em>s\u0103 \u00een\u021beleag\u0103<\/em>: algoritmii ei doar identific\u0103 tipare \u00een numere, iar dac\u0103 datele nu sunt reprezentative, rezultatul proces\u0103rii va fi similar. Identificarea acestor erori poate fi dificil\u0103 din cauza mecanicii \u00eenv\u0103\u021b\u0103rii automate. <noindex><a rel=\"nofollow\" name=\"habracut\"><\/a><\/noindex><\/li>\n<li>Cea mai evident\u0103 \u0219i \u00eenfrico\u0219\u0103toare zon\u0103 problem\u0103 este diversitatea uman\u0103. Exist\u0103 multe motive pentru care datele despre oameni pot pierde obiectivitatea chiar din etapa de colectare. Dar nu trebuie s\u0103 credem c\u0103 aceast\u0103 problem\u0103 se refer\u0103 doar la oameni: acelea\u0219i dificult\u0103\u021bi apar \u0219i atunci c\u00e2nd \u00eencerc\u0103m s\u0103 detect\u0103m o inunda\u021bie \u00eentr-un depozit sau o turbin\u0103 de gaz defect\u0103. Unele sisteme pot avea prejudec\u0103\u021bi legate de culoarea pielii, altele vor fi p\u0103rtinitoare \u00een ceea ce prive\u0219te senzorii Siemens. <\/li>\n<li>Aceste probleme nu sunt noi pentru \u00eenv\u0103\u021barea automat\u0103 \u0219i nu \u00eei sunt caracteristice doar ei. Presupunerile eronate sunt facute \u00een orice structur\u0103 complex\u0103, iar a \u00een\u021belege de ce a fost luat\u0103 o anumit\u0103 decizie este \u00eentotdeauna dificil. Trebuie s\u0103 ne confrunt\u0103m cu acest lucru \u00een mod complex: s\u0103 cre\u0103m instrumente \u0219i procese pentru verificare - \u0219i s\u0103 educ\u0103m utilizatorii, astfel \u00eenc\u00e2t s\u0103 nu urmeze orbe\u0219te recomand\u0103rile AI. \u00cenv\u0103\u021barea automat\u0103 face unele lucruri mult mai bine dec\u00e2t noi, - dar c\u00e2inii, de exemplu, sunt mult mai eficien\u021bi dec\u00e2t oamenii \u00een detectarea substan\u021belor ilegale, ceea ce nu este un motiv pentru a-i aduce ca martori \u0219i s\u0103 pronun\u021b\u0103m sentin\u021be bazate pe declara\u021biile lor. \u0218i c\u00e2inii, de altfel, sunt mult mai inteligen\u021bi dec\u00e2t orice sistem de \u00eenv\u0103\u021bare automat\u0103. <\/li>\n<\/ul>\n<p>\u00cenv\u0103\u021barea automat\u0103 este ast\u0103zi una dintre cele mai importante tendin\u021be tehnologice fundamentale. Este una dintre metodele principale prin care tehnologia va schimba lumea \u00een urm\u0103torul deceniu. Unele aspecte ale acestor schimb\u0103ri suscit\u0103 \u00eengrijorare. De exemplu, impactul poten\u021bial al \u00eenv\u0103\u021b\u0103rii automate asupra pie\u021bei muncii sau utilizarea acesteia \u00een scopuri neetice (s\u0103 zicem, de c\u0103tre regimuri autoritare). Exist\u0103 o alt\u0103 problem\u0103, la care este dedicat acest post: <strong>prejudecata inteligen\u021bei artificiale<\/strong>. <\/p>\n<p><\/p>\n<p>Aceasta este o poveste complicat\u0103. <\/p>\n<p><\/p>\n<p><img decoding=\"async\" alt=\"Despre biasul inteligen\u021bei artificiale\" src=\"\/wp-content\/uploads\/2019\/04\/d692fbfff47614758ab1d8e1516b2742.jpg\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<em>AI-ul de la Google poate g\u0103si pisici. Aceast\u0103 \u0219tire din 2012 a fost atunci ceva special.<\/em><\/p>\n<p><\/p>\n<h2 id=\"chto-takoe-predvzyatost-ii\">Ce este \u00abpartialitatea AI\u00bb?<\/h2>\n<p><\/p>\n<blockquote><p><em>\u00abDatele brute\u00bb sunt at\u00e2t un oxymoron, c\u00e2t \u0219i o idee proast\u0103; datele trebuie preg\u0103tite bine \u0219i cu aten\u021bie.<\/em> \u2014Jeffrey Bockert <\/p><\/blockquote>\n<p>P\u00e2n\u0103 prin 2013, pentru a crea un sistem care, s\u0103 spunem, recunoa\u0219te pisicile \u00een fotografii, trebuia s\u0103 descrii pa\u0219ii logici. Cum s\u0103 g\u0103se\u0219ti unghiurile pe imagine, s\u0103 recuno\u0219ti ochii, s\u0103 analizezi texturile pentru a detecta blana, s\u0103 num\u0103ri labele \u0219i a\u0219a mai departe. Apoi trebuie s\u0103 aduni toate componentele \u2014 \u0219i s\u0103 descoperi c\u0103, de fapt, nu func\u021bioneaz\u0103. Aproape ca un cal mecanic \u2014 teoretic \u00eel po\u021bi face, dar \u00een practic\u0103 este prea complicat de descris. La final, aveai sute (sau chiar mii) de reguli scrise de m\u00e2n\u0103. \u0218i niciun model func\u021bional. <\/p>\n<p><\/p>\n<p>Odat\u0103 cu apari\u021bia \u00eenv\u0103\u021b\u0103rii automate, am \u00eencetat s\u0103 mai folosim reguli \u201emanuale\u201d pentru recunoa\u0219terea unui anumit obiect. \u00cen schimb, lu\u0103m o mie de exemple de \u00abacesta\u00bb, X, o mie de exemple de \u00abacel altceva\u00bb, Y, \u0219i for\u021b\u0103m computerul s\u0103 construiasc\u0103 un model pe baza analizei lor statistice. Apoi oferim acestui model un set de date, iar acesta determin\u0103, cu o anumit\u0103 precizie, dac\u0103 se potrive\u0219te cu unul dintre seturi. \u00cenv\u0103\u021barea automat\u0103 genereaz\u0103 un model pe baza datelor, nu cu ajutorul unei persoane care \u00eel scrie. Rezultatele sunt impresionante, mai ales \u00een domeniul recunoa\u0219terii imaginilor \u0219i a modelelor, \u0219i de aceea toat\u0103 industria tehnologic\u0103 se \u00eendreapt\u0103 acum spre \u00eenv\u0103\u021barea automat\u0103 (ML). <\/p>\n<p><\/p>\n<p>Dar nu este totul at\u00e2t de simplu. \u00cen lumea real\u0103, cele mii de exemple X sau Y con\u021bin, de asemenea, A, B, J, L, O, R \u0219i chiar L. Ele pot fi distribuite neuniform, iar unele dintre ele pot ap\u0103rea at\u00e2t de des \u00eenc\u00e2t sistemul va acorda mai mult\u0103 aten\u021bie acestora dec\u00e2t obiectelor care te intereseaz\u0103.<\/p>\n<p><\/p>\n<p>Ce \u00eenseamn\u0103 asta \u00een practic\u0103? Exemplul meu preferat este atunci c\u00e2nd sistemele de recunoa\u0219tere a imaginilor <noindex><a rel=\"nofollow\" href=\"http:\/\/aiweirdness.com\/post\/171451900302\/do-neural-nets-dream-of-electric-sheep\">privesc o colin\u0103 acoperit\u0103 cu iarb\u0103 \u0219i spun: \u00aboaie\u00bb<\/a><\/noindex>. E clar de ce: cea mai mare parte a fotografiilor-example cu \u00aboi\u00bb sunt f\u0103cute pe p\u0103\u0219uni, unde acestea tr\u0103iesc, iar \u00een aceste imagini iarba ocup\u0103 mult mai mult spa\u021biu dec\u00e2t miciile bl\u0103ni\u021be albe, \u0219i tocmai iarba este ceea ce sistemele consider\u0103 cel mai important. <\/p>\n<p><\/p>\n<p>Exist\u0103 \u0219i exemple mai serioase. Unul recent \u2014 unul <noindex><a rel=\"nofollow\" href=\"https:\/\/www.medpagetoday.com\/dermatology\/generaldermatology\/70513\">proiect<\/a><\/noindex> despre detectarea cancerului de piele \u00een fotografii. S-a dovedit c\u0103 dermatologii fotografiaz\u0103 adesea o rigl\u0103 \u00eempreun\u0103 cu manifest\u0103rile cancerului de piele pentru a \u00eenregistra dimensiunea forma\u021biunilor. \u00cen exemplele de fotografii cu pielea s\u0103n\u0103toas\u0103 nu exist\u0103 rigle. Pentru sistemul de inteligen\u021b\u0103 artificial\u0103, aceste rigle (mai precis, pixelii pe care \u00eei definim ca \u201erigl\u0103\u201d) au devenit una dintre distinc\u021biile \u00eentre seturile de exemple \u0219i uneori mai importante dec\u00e2t o mic\u0103 erup\u021bie cutanat\u0103. Astfel, sistemul creat pentru a recunoa\u0219te cancerul de piele a recunoscut uneori rigle \u00een loc de acesta. <\/p>\n<p><\/p>\n<p>Punctul cheie aici este c\u0103 sistemul nu are o \u00een\u021belegere semantic\u0103 a ceea ce se uit\u0103. Noi ne uit\u0103m la un set de pixeli \u0219i vedem \u00een ei o oaie, piele sau rigle, iar sistemul vede doar un \u0219ir numeric. Nu vede spa\u021biul tridimensional, nu vede obiecte sau texturi, nu vede nici oile. Vede doar tipare \u00een date. <\/p>\n<p><\/p>\n<p>Dificultatea diagnostic\u0103rii acestor probleme este c\u0103 re\u021beaua neuronal\u0103 (modelul generat de sistemul dvs. de \u00eenv\u0103\u021bare automat\u0103) const\u0103 din zeci de mii de noduri. Nu exist\u0103 o modalitate simpl\u0103 de a privi \u00een model \u0219i a vedea cum ia deciziile. A avea o astfel de modalitate ar \u00eensemna c\u0103 procesul este suficient de simplu pentru a descrie toate regulile manual, f\u0103r\u0103 a folosi \u00eenv\u0103\u021barea automat\u0103. Oamenii se \u00eengrijoreaz\u0103 c\u0103 \u00eenv\u0103\u021barea automat\u0103 a devenit un fel de \u201ecutie neagr\u0103\u201d. (Voi explica mai t\u00e2rziu de ce aceast\u0103 compara\u021bie este totu\u0219i exagerat\u0103.) <\/p>\n<p><\/p>\n<p>Aceasta, \u00een linii mari, este problema prejudec\u0103\u021bii inteligen\u021bei artificiale sau \u00eenv\u0103\u021b\u0103rii automate: un sistem pentru g\u0103sirea de tipare \u00een date poate g\u0103si tipare gre\u0219ite, iar dumneavoastr\u0103 s-ar putea s\u0103 nu observa\u021bi acest lucru. Aceasta este o caracteristic\u0103 fundamental\u0103 a tehnologiei \u0219i este evident\u0103 pentru oricine lucreaz\u0103 cu aceasta \u00een cercurile \u0219tiin\u021bifice \u0219i \u00een mari companii tehnologice. \u00cens\u0103 consecin\u021bele sale sunt complexe, iar solu\u021biile noastre posibile la aceste consecin\u021be sunt, de asemenea. <\/p>\n<p><\/p>\n<p>S\u0103 discut\u0103m mai \u00eent\u00e2i despre consecin\u021be. <\/p>\n<p><\/p>\n<p><img decoding=\"async\" alt=\"Despre biasul inteligen\u021bei artificiale\" src=\"\/wp-content\/uploads\/2019\/04\/43748412bd82ae53a9c2e47afec9c655.jpg\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<em>AI poate face alegeri \u00een mod implicit pentru noi \u00een favoarea anumitor categorii de oameni, baz\u00e2ndu-se pe o mul\u021bime de semnale subtile.<\/em><\/p>\n<p><\/p>\n<h2 id=\"scenarii-predvzyatosti-ii\">Scenarii de prejudecat\u0103 AI<\/h2>\n<p><\/p>\n<p>Cel mai evident \u0219i \u00eenfrico\u0219\u0103tor este c\u0103 aceast\u0103 problem\u0103 ar putea ap\u0103rea c\u00e2nd vine vorba de diversitatea uman\u0103. Recent, <noindex><a rel=\"nofollow\" href=\"https:\/\/www.reuters.com\/article\/us-amazon-com-jobs-automation-insight\/amazon-scraps-secret-ai-recruiting-tool-that-showed-bias-against-women-idUSKCN1MK08G\">a circulat zvonul<\/a><\/noindex>, c\u0103 Amazon a \u00eencercat s\u0103 construiasc\u0103 un sistem de \u00eenv\u0103\u021bare automat\u0103 pentru a realiza o selec\u021bie ini\u021bial\u0103 a candida\u021bilor pentru locuri de munc\u0103. Deoarece printre angaja\u021bii Amazon sunt mai mul\u021bi b\u0103rba\u021bi, exemplele de \u201eangajare reu\u0219it\u0103\u201d erau, de asemenea, predominant masculine, iar \u00een selec\u021bia de CV-uri propus\u0103 de sistem, erau mai mul\u021bi b\u0103rba\u021bi. Amazon a observat acest lucru \u0219i nu a lansat sistemul \u00een produc\u021bie. <\/p>\n<p><\/p>\n<p>Cel mai important lucru \u00een acest exemplu este c\u0103 sistemul, conform zvonurilor, prefera candida\u021bii de sex masculin, de\u0219i sexul nu era specificat \u00een CV.<strong> Sistemul observa alte modele \u00een exemplele de \u201eangajare reu\u0219it\u0103\u201d: de exemplu, femeile ar putea folosi cuvinte speciale pentru a-\u0219i descrie realiz\u0103rile sau ar putea avea hobby-uri specifice. Desigur, sistemul nu \u0219tia ce \u00eenseamn\u0103 \u201ehochei\u201d, cine sunt \u201eoamenii\u201d sau ce \u00eenseamn\u0103 \u201esucces\u201d \u2013 el doar efectua o analiz\u0103 statistic\u0103 a textului.<\/strong> Dar regularit\u0103\u021bile pe care le-a observat probabil c\u0103 ar fi trecut neobservate de un om, iar unele dintre ele (de exemplu, c\u0103 oamenii de diferite sexe descriu succesul \u00een mod diferit) ne-ar fi greu s\u0103 le observ\u0103m chiar \u0219i privindu-le. <\/p>\n<p><\/p>\n<p>Mai departe \u2013 mai r\u0103u. Sistemul de \u00eenv\u0103\u021bare automat\u0103, care g\u0103se\u0219te foarte bine cancerul pe pielea deschis\u0103, poate func\u021biona mai slab pe pielea \u00eenchis\u0103, sau invers. Nu neap\u0103rat din cauza prejudec\u0103\u021bii, ci pentru c\u0103, probabil, trebuie s\u0103 construie\u0219ti un model separat pentru o alt\u0103 culoare a pielii, aleg\u00e2nd alte caracteristici.<strong> Sistemele de \u00eenv\u0103\u021bare automat\u0103 nu sunt interschimbabile, chiar \u0219i \u00eentr-un domeniu at\u00e2t de restr\u00e2ns precum recunoa\u0219terea imaginilor.<\/strong> Trebuie s\u0103 configurezi sistemul, uneori doar prin \u00eencerc\u0103ri \u0219i gre\u0219eli, pentru a observa bine caracteristicile din datele care te intereseaz\u0103, p\u00e2n\u0103 c\u00e2nd ajungi la acurate\u021bea dorit\u0103. Dar s-ar putea s\u0103 nu observi c\u0103 sistemul are o acurate\u021be de 98% c\u00e2nd lucreaz\u0103 cu un grup \u0219i doar de 91% (de\u0219i este mai precis dec\u00e2t analiza realizat\u0103 de un om) \u2013 cu altul. <\/p>\n<p><\/p>\n<p>P\u00e2n\u0103 acum am folosit \u00een principal exemple legate de oameni \u0219i caracteristicile acestora. Acesta este, de asemenea, subiectul principal al discu\u021biei din jurul acestei probleme. Dar este important s\u0103 \u00een\u021belegem c\u0103 prejudecata fa\u021b\u0103 de oameni este doar o parte a problemei. Vom utiliza \u00eenv\u0103\u021barea automat\u0103 pentru o mul\u021bime de lucruri, iar eroarea de e\u0219antionare va fi relevant\u0103 pentru toate acestea. Pe de alt\u0103 parte, dac\u0103 lucrezi cu oameni, prejudecata datelor poate s\u0103 nu fie legat\u0103 de ei. <\/p>\n<p><\/p>\n<p>Pentru a \u00een\u021belege acest lucru, s\u0103 ne \u00eentoarcem la exemplul cu cancerul de piele \u0219i s\u0103 analiz\u0103m trei scenarii ipotetice de e\u0219ec al sistemului. <\/p>\n<p><\/p>\n<ol>\n<li>Distribu\u021bia heterogen\u0103 a persoanelor: un num\u0103r dispropor\u021bionat de fotografii ale pielii \u00een tonuri diferite, ceea ce duce la rezultate fals pozitive sau fals negative legate de pigmenta\u021bie. <\/li>\n<li>Datele pe care se antreneaz\u0103 sistemul con\u021bin o caracteristic\u0103 frecvent \u00eent\u00e2lnit\u0103 \u0219i distribuit\u0103 inegal, care nu este legat\u0103 de oameni \u0219i nu are valoare diagnostic\u0103: o rigl\u0103 \u00een fotografiile cu manifest\u0103ri ale cancerului de piele sau iarba \u00een fotografiile cu oi. \u00cen acest caz, rezultatul va varia daca pe imagine sistemul g\u0103se\u0219te pixeli ai unui obiect pe care ochiul uman \u00eel va identifica ca fiind \u00abrigl\u0103\u00bb. <\/li>\n<li>Datele con\u021bin o caracteristic\u0103 extern\u0103 care nu poate fi v\u0103zut\u0103 de om, chiar dac\u0103 o caut\u0103. <\/li>\n<\/ol>\n<p><\/p>\n<p>Ce \u00eenseamn\u0103 acest lucru? \u0218tim dinainte c\u0103 datele pot reprezenta diferit grupuri diverse de oameni \u0219i, cel pu\u021bin, putem planifica c\u0103utarea unor asemenea excep\u021bii. Cu alte cuvinte, exist\u0103 o mul\u021bime de motive sociale pentru a presupune c\u0103 datele despre grupuri de oameni con\u021bin deja un anumit prejudiciu. Dac\u0103 ne uit\u0103m la o fotografie cu o rigl\u0103, o vom observa \u2014 am ignorat-o doar \u00eenainte, \u0219tiind c\u0103 nu are importan\u021b\u0103 \u0219i uit\u00e2nd c\u0103 sistemul nu \u0219tie nimic.<\/p>\n<p><\/p>\n<p>Dar ce se \u00eent\u00e2mpl\u0103 dac\u0103 toate fotografiile cu piele nes\u0103n\u0103toas\u0103 au fost f\u0103cute \u00eentr-un birou unde se utilizeaz\u0103 becuri incandescente, iar cele cu piele s\u0103n\u0103toas\u0103 \u2014 sub luminile fluorescente? Ce se \u00eent\u00e2mpl\u0103 dac\u0103, dup\u0103 ce ai terminat de fotografiat pielea s\u0103n\u0103toas\u0103, \u00eenainte de a fotografia pielea nes\u0103n\u0103toas\u0103, ai actualizat sistemul de operare al telefonului t\u0103u, iar Apple sau Google au modificat pu\u021bin algoritmul de reducere a zgomotului? Un om nu va observa asta, oric\u00e2t de mult ar c\u0103uta asemenea caracteristici. \u00cens\u0103 sistemul de utilizare a ma\u0219inilor va observa imediat \u0219i va folosi acest lucru. El nu \u0219tie nimic. <\/p>\n<p><\/p>\n<p>De\u0219i vorbeam despre corela\u021bii false, este posibil ca datele s\u0103 fie precise \u0219i rezultatele s\u0103 fie corecte, dar s\u0103 nu dori\u021bi s\u0103 le folosi\u021bi din motive etice, legale sau de management. \u00cen unele jurisdic\u021bii, de exemplu, nu se poate oferi reduceri la asigurare femeilor, chiar dac\u0103 femeile ar putea conduce mai sigur. Ne putem imagina cu u\u0219urin\u021b\u0103 un sistem care, analiz\u00e2nd date istorice, va atribui numelui femeilor un coeficient de risc mai mic. Bine, s\u0103 elimin\u0103m numele din e\u0219antion. Dar aminti\u021bi-v\u0103 exemplul cu Amazon: sistemul poate determina sexul pe baza altor factori (de\u0219i el nu \u0219tie ce este sexul, nici ce este ma\u0219ina), iar dumneavoastr\u0103 nu ve\u021bi observa asta p\u00e2n\u0103 c\u00e2nd un regulator nu va analiza retroactiv tarifele pe care le-a\u021bi propus \u0219i nu v\u0103 va impune o amend\u0103. <\/p>\n<p><\/p>\n<p>\u00cen cele din urm\u0103, se presupune adesea c\u0103 vom folosi astfel de sisteme doar pentru proiecte legate de oameni \u0219i de interac\u021biuni sociale. Asta nu este adev\u0103rat. Dac\u0103 produce\u021bi turbine cu gaz, cu siguran\u021b\u0103 ve\u021bi dori s\u0103 aplica\u021bi \u00eenv\u0103\u021barea automat\u0103 pe telemetria transmis\u0103 de zeci sau sute de senzori de pe produsul dumneavoastr\u0103 (senzorii audio, video, de temperatur\u0103 \u0219i orice al\u021bi senzori genereaz\u0103 date care pot fi foarte u\u0219or adaptate pentru a crea un model de \u00eenv\u0103\u021bare automat\u0103). Ipoteza este c\u0103 pute\u021bi spune: \u201eIat\u0103 datele despre o mie de turbinii defecte ob\u021binute \u00eenainte de defectarea lor \u0219i iat\u0103 datele de la o mie de turbine care nu s-au stricat. Construie\u0219te un model pentru a ar\u0103ta \u00een ce const\u0103 diferen\u021ba dintre ele.\u201d Acum imagina\u021bi-v\u0103 c\u0103 senzorii Siemens sunt monta\u021bi pe 75% din turbinele defecte \u0219i doar pe 12% din cele bune (f\u0103r\u0103 leg\u0103tur\u0103 cu defectele \u00een sine). Sistemul va construi un model pentru a g\u0103si turbinii cu senzori Siemens. Ups!<\/p>\n<p><\/p>\n<p><img decoding=\"async\" alt=\"Despre biasul inteligen\u021bei artificiale\" src=\"\/wp-content\/uploads\/2019\/04\/e8097e4913756fdf04826a6628bc6836.jpg\" style=\"display:block;margin: 0 auto;\" \/><br \/>\nImagine \u2013 Moritz Hardt, UC Berkeley<\/p>\n<p><\/p>\n<h2 id=\"upravlenie-predvzyatostyu-ii\">Gestionarea prejudec\u0103\u021bii \u00een AI<\/h2>\n<p><\/p>\n<p>Ce putem face \u00een leg\u0103tur\u0103 cu asta? Putem aborda problema din trei unghiuri:<\/p>\n<p><\/p>\n<ol>\n<li>Rigor metodologic \u00een colectarea \u0219i gestionarea datelor pentru antrenarea sistemului.<\/li>\n<li>Instrumente tehnice pentru analizarea \u0219i diagnosticarea comportamentului modelului.<\/li>\n<li>Instruire, educa\u021bie \u0219i pruden\u021b\u0103 \u00een implementarea \u00eenv\u0103\u021b\u0103rii automate \u00een produse.<\/li>\n<\/ol>\n<p><\/p>\n<p>\u00cen cartea \u201eBurghezul gentilom\u201d de Moli\u00e8re exist\u0103 o glum\u0103: unui b\u0103rbat i s-a spus c\u0103 literatura se \u00eemparte \u00een proz\u0103 \u0219i poezie, iar el descoper\u0103 cu uimire c\u0103 a vorbit toat\u0103 via\u021ba proz\u0103, f\u0103r\u0103 s\u0103 \u0219tie. Probabil c\u0103 statisticienii se simt cam la fel ast\u0103zi: f\u0103r\u0103 s\u0103-\u0219i dea seama, \u0219i-au dedicat carierele inteligen\u021bei artificiale \u0219i erorii de e\u0219antionare. C\u0103utarea erorii de e\u0219antionare \u0219i \u00eengrijorarea din cauza ei nu este o problem\u0103 nou\u0103, trebuie doar s\u0103 abord\u0103m sistematic solu\u021bia. A\u0219a cum am men\u021bionat mai sus, \u00een unele cazuri este cu adev\u0103rat mai simplu s\u0103 studiem problemele legate de datele despre oameni. Noi presupunem din start c\u0103 putem avea prejudec\u0103\u021bi fa\u021b\u0103 de diferite grupuri de oameni, dar prejudecata cu privire la senzorii Siemens este greu de imaginat.<\/p>\n<p><\/p>\n<p>Nou \u00een toat\u0103 aceasta, desigur, este c\u0103 oamenii nu mai fac analiza statistic\u0103 direct. Acest lucru este realizat de ma\u0219ini, care creeaz\u0103 modele mari \u0219i complexe, greu de \u00een\u021beles. Problema transparen\u021bei este unul dintre aspectele fundamentale ale problemei prejudec\u0103\u021bii. Ne este fric\u0103 c\u0103 sistemul nu este doar p\u0103rtinitor, ci c\u0103 nu exist\u0103 nicio posibilitate de a detecta p\u0103rtinirea sa, \u0219i c\u0103 \u00eenv\u0103\u021barea automat\u0103 se deosebe\u0219te de alte forme de automatizare, care, a\u0219a cum se presupune, constau \u00een pa\u0219i logici clari, care pot fi verifica\u021bi. <\/p>\n<p><\/p>\n<p>Aici exist\u0103 dou\u0103 probleme. Poate c\u0103 putem efectua \u00centr-un fel un audit al sistemelor de \u00eenv\u0103\u021bare automat\u0103. Iar auditul oric\u0103rui alt sistem, de fapt, nu este deloc mai u\u0219or.<\/p>\n<p><\/p>\n<p>\u00cen primul r\u00e2nd, una dintre direc\u021biile actuale de cercetare \u00een domeniul \u00eenv\u0103\u021b\u0103rii automate este c\u0103utarea de metode pentru identificarea func\u021bionalit\u0103\u021bilor importante ale sistemelor de \u00eenv\u0103\u021bare automat\u0103. \u00cen acest context, \u00eenv\u0103\u021barea automat\u0103 (\u00een starea sa actual\u0103) este un domeniu complet nou al \u0219tiin\u021bei, care se schimb\u0103 rapid, a\u0219a c\u0103 nu ar trebui s\u0103 credem c\u0103 lucruri imposibile ast\u0103zi nu pot deveni \u00een cur\u00e2nd complet reale. Proiectul <noindex><a rel=\"nofollow\" href=\"https:\/\/openai.com\/\">OpenAI<\/a><\/noindex> este un exemplu interesant \u00een acest sens. <\/p>\n<p><\/p>\n<p>\u00cen al doilea r\u00e2nd, ideea c\u0103 putem analiza \u0219i \u00een\u021belege procesul de luare a deciziilor \u00een sistemele sau organiza\u021biile existente este bun\u0103 \u00een teorie, dar nu foarte util\u0103 \u00een practic\u0103. A \u00een\u021belege cum se iau deciziile \u00eentr-o organiza\u021bie mare nu este deloc simplu. Chiar dac\u0103 exist\u0103 un proces formal de luare a deciziilor, acesta nu reflect\u0103 modul \u00een care oamenii interac\u021bioneaz\u0103 cu adev\u0103rat, iar ei \u00een\u0219i\u0219i adesea nu au o abordare logic\u0103 \u0219i sistematic\u0103 \u00een cadrul deciziilor lor. A\u0219a cum a spus colegul meu <noindex><a rel=\"nofollow\" href=\"https:\/\/en.wikipedia.org\/wiki\/Vijay_S._Pande\">Vijay Pande<\/a><\/noindex>, <strong>oamenii sunt, de asemenea, cutii negre<\/strong>. <\/p>\n<p><\/p>\n<p>Lua\u021bi o mie de oameni din mai multe companii \u0219i institu\u021bii intersec\u021bionate, iar problema devine \u0219i mai complex\u0103. \u0218tim, \u00een retrospectiv\u0103, c\u0103 naveta spa\u021bial\u0103 'Space Shuttle' a fost destinat\u0103 s\u0103 se dezintegreze la \u00eentoarcere, iar anumite persoane din cadrul NASA aveau informa\u021bii care le d\u0103deau motive s\u0103 cread\u0103 c\u0103 ar putea avea loc ceva r\u0103u, dar sistemul <em>\u00een ansamblu<\/em> nu \u0219tia acest lucru. NASA tocmai trecuse printr-un audit similar, dup\u0103 ce pierduse naveta anterioar\u0103, \u0219i totu\u0219i a pierdut una nou\u0103 - dintr-un motiv foarte asem\u0103n\u0103tor. E simplu s\u0103 afirmi c\u0103 organiza\u021biile \u0219i oamenii urmeaz\u0103 reguli logice clare, care pot fi verificate, \u00een\u021belese \u0219i schimbate - dar experien\u021ba demonstreaz\u0103 contrariul. Aceasta este '<noindex><a rel=\"nofollow\" href=\"https:\/\/ru.wikipedia.org\/wiki\/%D0%93%D0%BE%D1%81%D0%BF%D0%BB%D0%B0%D0%BD_%D0%A1%D0%A1%D0%A1%D0%A0\">iluzia Gosplan<\/a><\/noindex>\u00bb.<\/p>\n<p><\/p>\n<p><strong>Adesea compar \u00eenv\u0103\u021barea automat\u0103 cu bazele de date, \u00een special cu cele rela\u021bionale - o tehnologie fundamental\u0103 nou\u0103, care a revolu\u021bionat capacit\u0103\u021bile informaticii \u0219i lumea din jurul ei, care a devenit parte din tot ceea ce facem \u0219i pe care o folosim constant f\u0103r\u0103 s\u0103 ne d\u0103m seama.<\/strong> Bazele de date au \u0219i ele probleme, iar acestea sunt de o natur\u0103 similar\u0103: sistemul poate fi construit pe presupuneri gre\u0219ite sau pe date eronate, dar acest lucru va fi greu de observat, iar persoanele care folosesc sistemul vor face ceea ce le spune el, f\u0103r\u0103 s\u0103 pun\u0103 \u00eentreb\u0103ri. Exist\u0103 o mul\u021bime de glume vechi despre angaja\u021bii fiscului care au notat odat\u0103 gre\u0219it numele t\u0103u, \u0219i este mult mai dificil s\u0103-i convingi s\u0103 corecteze eroarea dec\u00e2t s\u0103-\u021bi schimbi efectiv numele. Poate fi g\u00e2ndit diferit, dar nu este clar care este abordarea corect\u0103: ca o problem\u0103 tehnic\u0103 \u00een SQL, ca o eroare \u00een lansarea Oracle sau ca o defec\u021biune a institu\u021biilor birocratice? C\u00e2t de greu este s\u0103 g\u0103se\u0219ti o eroare \u00een procesul care a dus la faptul c\u0103 sistemul nu are o func\u021bie precum corectarea gre\u0219elilor de tipar? A fost posibil s\u0103 se \u00een\u021beleag\u0103 asta \u00eenainte ca oamenii s\u0103 \u00eenceap\u0103 s\u0103 se pl\u00e2ng\u0103? <\/p>\n<p><\/p>\n<p>O problem\u0103 asem\u0103n\u0103toare este ilustrat\u0103 \u0219i de pove\u0219tile \u00een care \u0219oferii, din cauza datelor \u00eenvechite din navigator, ajung \u00een r\u00e2uri. E clar c\u0103 h\u0103r\u021bile trebuie actualizate constant. Dar p\u00e2n\u0103 la ce punct este TomTom vinovat c\u0103 ma\u0219ina ta este dus\u0103 \u00een mare?<\/p>\n<p><\/p>\n<p>Spun asta pentru c\u0103, da, p\u0103rtinirea \u00een \u00eenv\u0103\u021barea automat\u0103 va crea probleme. Dar aceste probleme vor fi similare cu cele cu care ne-am confruntat \u00een trecut, \u0219i le putem observa \u0219i rezolva (sau nu) cam la fel de bine cum am reu\u0219it \u00een trecut. Prin urmare, scenariul \u00een care p\u0103rtinirea IA va provoca daune este pu\u021bin probabil s\u0103 se \u00eent\u00e2mple cu cercet\u0103torii de frunte care lucreaz\u0103 \u00eentr-o organiza\u021bie mare. Cel mai probabil, un contractor sau un v\u00e2nz\u0103tor de software minor scrie ceva f\u0103r\u0103 a \u00een\u021belege, folosind componente open-source, biblioteci \u0219i instrumente de care nu \u00eei pas\u0103. \u0218i clientul nefericit se va l\u0103sa p\u0103c\u0103lit de expresia \u201einteligen\u021b\u0103 artificial\u0103\u201d din descrierea produsului \u0219i, f\u0103r\u0103 s\u0103 pun\u0103 \u00eentreb\u0103ri, va oferi acest produs angaja\u021bilor s\u0103i slab pl\u0103ti\u021bi, oblig\u00e2ndu-i s\u0103 fac\u0103 ceea ce le spune IA. Aceasta este situa\u021bia cu bazele de date. Nu este o problem\u0103 a inteligen\u021bei artificiale, \u0219i nici m\u0103car o problem\u0103 a software-ului. Este un factor uman.<\/p>\n<p><\/p>\n<h2 id=\"zaklyuchenie\">Concluzie<\/h2>\n<p><\/p>\n<blockquote><p><em>\u00cenv\u0103\u021barea automat\u0103 poate face tot ceea ce po\u021bi \u00eenv\u0103\u021ba un c\u00e2ine \u2014 dar nu vei putea fi niciodat\u0103 sigur ce anume l-ai \u00eenv\u0103\u021bat pe acel c\u00e2ine.<\/em><\/p><\/blockquote>\n<p>Adesea \u00eemi pare c\u0103 termenul \u201einteligen\u021b\u0103 artificial\u0103\u201d doar \u00eempiedic\u0103 discu\u021biile de acest gen. Acest termen creeaz\u0103 o impresie fals\u0103 c\u0103 noi, de fapt, am creat-o \u2014 aceast\u0103 inteligen\u021b\u0103. C\u0103 suntem pe calea c\u0103tre HAL9000 sau Skynet \u2014 c\u0103tre ceva ce, \u00een realitate, <em>\u00een\u021belege<\/em>. Dar nu. Sunt doar ma\u0219ini, \u0219i este mult mai corect s\u0103 le compar\u0103m, spre exemplu, cu o ma\u0219in\u0103 de sp\u0103lat. Aceasta se descurc\u0103 mult mai bine dec\u00e2t omul cu sp\u0103larea, dar dac\u0103 pui \u00een ea vase \u00een loc de rufe, ea le va\u2026 sp\u0103la. Vasele vor fi chiar curate. Dar nu va fi ceea ce te a\u0219tep\u021bi, iar acest lucru nu se va \u00eent\u00e2mpla pentru c\u0103 sistemul are anumite prejudec\u0103\u021bi fa\u021b\u0103 de vase. Ma\u0219ina de sp\u0103lat nu \u0219tie ce sunt vasele sau rufele \u2014 este doar un exemplu de automatizare, conceptual nu diferit de modul \u00een care procesele au fost automatizate mai devreme.<\/p>\n<p><\/p>\n<p>Indiferent de subiect \u2014 despre ma\u0219ini, avioane sau baze de date \u2014 aceste sisteme vor fi \u00een acela\u0219i timp foarte puternice \u0219i foarte limitate. Vor depinde complet de modul \u00een care oamenii folosesc aceste sisteme, de inten\u021biile lor bune sau rele \u0219i de c\u00e2t de bine \u00een\u021beleg cum func\u021bioneaz\u0103 acestea.<\/p>\n<p><\/p>\n<p>Prin urmare, a spune c\u0103 \u201einteligen\u021ba artificial\u0103 este matematic\u0103, deci nu poate avea prejudec\u0103\u021bi\u201d este complet gre\u0219it. Dar la fel de gre\u0219it este s\u0103 afirmi c\u0103 \u00eenv\u0103\u021barea automat\u0103 \u201eeste subiectiv\u0103 prin natura sa\u201d. \u00cenv\u0103\u021barea automat\u0103 g\u0103se\u0219te modele \u00een date, iar ce modele va g\u0103si depinde de date, iar datele depind de noi. La fel \u0219i ceea ce facem cu ele. \u00cenv\u0103\u021barea automat\u0103 face \u00eentr-adev\u0103r unele lucruri mult mai bine dec\u00e2t noi \u2014 dar c\u00e2inii, de exemplu, sunt mult mai eficien\u021bi dec\u00e2t oamenii \u00een detectarea drogurilor, ceea ce nu este deloc un motiv pentru a-i considera martori \u0219i a emite verdicturi pe baza m\u0103rturiilor lor. \u0218i c\u00e2inii, ca o parantez\u0103, sunt mult mai inteligen\u021bi dec\u00e2t orice sistem de \u00eenv\u0103\u021bare automat\u0103.<\/p>\n<p><strong>Traducere:<\/strong>\u00a0<noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/en\/users\/757NF\/\">Diana Le\u021bca<\/a><\/noindex>.<br \/>\n<strong>Revizuire:<\/strong>\u00a0<noindex><a rel=\"nofollow\" href=\"http:\/\/teleg.one\/ponchiknews\">Alexey Ivanov<\/a><\/noindex>.<br \/>\n<strong>Comunitate:<\/strong> <noindex><a rel=\"nofollow\" href=\"http:\/\/teleg.one\/ponchiknews\">@PonchikNews<\/a><\/noindex>.<\/p>\n<p>Sursa: <a content=\"nofollow\" rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/post\/449224\/\">habr.com<\/a><\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"<p>tl;dr: \u041c\u0430\u0448\u0438\u043d\u043d\u043e\u0435 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u0435 \u0438\u0449\u0435\u0442 \u0437\u0430\u043a\u043e\u043d\u043e\u043c\u0435\u0440\u043d\u043e\u0441\u0442\u0438 \u0432 \u0434\u0430\u043d\u043d\u044b\u0445. \u041d\u043e \u0438\u0441\u043a\u0443\u0441\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u0439 \u0438\u043d\u0442\u0435\u043b\u043b\u0435\u043a\u0442 \u043c\u043e\u0436\u0435\u0442 \u0431\u044b\u0442\u044c \u00ab\u043f\u0440\u0435\u0434\u0432\u0437\u044f\u0442\u00bb \u2014\u00a0\u0442\u043e \u0435\u0441\u0442\u044c, \u043d\u0430\u0445\u043e\u0434\u0438\u0442\u044c \u043d\u0435\u0432\u0435\u0440\u043d\u044b\u0435 \u043f\u0430\u0442\u0442\u0435\u0440\u043d\u044b. \u041a \u043f\u0440\u0438\u043c\u0435\u0440\u0443, \u0441\u0438\u0441\u0442\u0435\u043c\u0430 \u043e\u0431\u043d\u0430\u0440\u0443\u0436\u0435\u043d\u0438\u044f \u0440\u0430\u043a\u0430 \u043a\u043e\u0436\u0438 \u043f\u043e \u0444\u043e\u0442\u043e\u0433\u0440\u0430\u0444\u0438\u0438 \u043c\u043e\u0436\u0435\u0442 \u043e\u0431\u0440\u0430\u0449\u0430\u0442\u044c \u043e\u0441\u043e\u0431\u043e\u0435 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u0435 \u043d\u0430 \u0441\u043d\u0438\u043c\u043a\u0438, \u0441\u0434\u0435\u043b\u0430\u043d\u043d\u044b\u0435 \u0432\u043e \u0432\u0440\u0430\u0447\u0435\u0431\u043d\u043e\u043c \u043a\u0430\u0431\u0438\u043d\u0435\u0442\u0435. \u041c\u0430\u0448\u0438\u043d\u043d\u043e\u0435 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u0435 \u043d\u0435 \u0443\u043c\u0435\u0435\u0442 \u043f\u043e\u043d\u0438\u043c\u0430\u0442\u044c: \u0435\u0433\u043e \u0430\u043b\u0433\u043e\u0440\u0438\u0442\u043c\u044b \u043b\u0438\u0448\u044c \u0432\u044b\u044f\u0432\u043b\u044f\u044e\u0442 \u0437\u0430\u043a\u043e\u043d\u043e\u043c\u0435\u0440\u043d\u043e\u0441\u0442\u0438 \u0432 \u0447\u0438\u0441\u043b\u0430\u0445, \u0438 \u0435\u0441\u043b\u0438 \u0434\u0430\u043d\u043d\u044b\u0435 \u043d\u0435 \u0440\u0435\u043f\u0440\u0435\u0437\u0435\u043d\u0442\u0430\u0442\u0438\u0432\u043d\u044b, \u0442\u0430\u043a\u0438\u043c \u0431\u0443\u0434\u0435\u0442 [&hellip;]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":1,"featured_media":24558,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[702],"tags":[],"class_list":["post-32781","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news"],"aioseo_notices":[],"aioseo_head":"\n\t\t<!-- All in One SEO 5.0.2.1 - aioseo.com -->\n\t<meta name=\"description\" content=\"tl;dr: .\" \/>\n\t<meta name=\"robots\" content=\"max-image-preview:large\" \/>\n\t<meta name=\"author\" content=\"Yuri Gagarin\"\/>\n\t<link rel=\"canonical\" href=\"https:\/\/prohoster.info\/ro\/blog\/news\/o-predvzyatosti-iskusstvennogo-intellekta\" \/>\n\t<meta name=\"generator\" content=\"All in One SEO (AIOSEO) 5.0.2.1\" \/>\n\t\t<meta property=\"og:locale\" content=\"ro_RO\" \/>\n\t\t<meta property=\"og:site_name\" content=\"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b\" \/>\n\t\t<meta property=\"og:type\" content=\"article\" \/>\n\t\t<meta property=\"og:title\" content=\"\ud83e\udd47\u041e \u043f\u0440\u0435\u0434\u0432\u0437\u044f\u0442\u043e\u0441\u0442\u0438 \u0438\u0441\u043a\u0443\u0441\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u0433\u043e \u0438\u043d\u0442\u0435\u043b\u043b\u0435\u043a\u0442\u0430 | ProHoster\" \/>\n\t\t<meta property=\"og:description\" content=\"tl;dr: .\" \/>\n\t\t<meta property=\"og:url\" content=\"https:\/\/prohoster.info\/ro\/blog\/news\/o-predvzyatosti-iskusstvennogo-intellekta\" \/>\n\t\t<meta property=\"og:image\" content=\"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg\" \/>\n\t\t<meta property=\"og:image:secure_url\" content=\"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg\" \/>\n\t\t<meta property=\"og:image:width\" content=\"350\" \/>\n\t\t<meta property=\"og:image:height\" content=\"350\" \/>\n\t\t<meta property=\"article:published_time\" content=\"2019-10-31T18:48:54+00:00\" \/>\n\t\t<meta property=\"article:modified_time\" content=\"2019-10-31T18:48:54+00:00\" \/>\n\t\t<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/prohoster\" \/>\n\t\t<meta property=\"article:author\" content=\"https:\/\/www.facebook.com\/prohoster\" \/>\n\t\t<!-- All in One SEO -->\n\n","aioseo_head_json":{"title":"\ud83e\udd47Despre prejudec\u0103\u021bile inteligen\u021bei artificiale | ProHoster","description":"tl;dr: .","canonical_url":"https:\/\/prohoster.info\/ro\/blog\/news\/o-predvzyatosti-iskusstvennogo-intellekta","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"ro_RO","og:site_name":"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b","og:type":"article","og:title":"\ud83e\udd47\u041e \u043f\u0440\u0435\u0434\u0432\u0437\u044f\u0442\u043e\u0441\u0442\u0438 \u0438\u0441\u043a\u0443\u0441\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u0433\u043e \u0438\u043d\u0442\u0435\u043b\u043b\u0435\u043a\u0442\u0430 | ProHoster","og:description":"tl;dr: .","og:url":"https:\/\/prohoster.info\/ro\/blog\/news\/o-predvzyatosti-iskusstvennogo-intellekta","og:image":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:secure_url":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:width":350,"og:image:height":350,"article:published_time":"2019-10-31T18:48:54+00:00","article:modified_time":"2019-10-31T18:48:54+00:00","article:publisher":"https:\/\/www.facebook.com\/prohoster","article:author":"https:\/\/www.facebook.com\/prohoster"},"aioseo_meta_data":{"post_id":"32781","title":null,"description":null,"keywords":null,"keyphrases":null,"primary_term":null,"canonical_url":null,"og_title":null,"og_description":null,"og_object_type":"default","og_image_type":"default","og_image_url":null,"og_image_width":null,"og_image_height":null,"og_image_custom_url":null,"og_image_custom_fields":null,"og_video":null,"og_custom_url":null,"og_article_section":null,"og_article_tags":null,"twitter_use_og":false,"twitter_card":"default","twitter_image_type":"default","twitter_image_url":null,"twitter_image_custom_url":null,"twitter_image_custom_fields":null,"twitter_title":null,"twitter_description":null,"schema":{"blockGraphs":[],"customGraphs":[],"default":{"data":{"Article":[],"Course":[],"Dataset":[],"FAQPage":[],"Movie":[],"Person":[],"Product":[],"ProductReview":[],"Car":[],"Recipe":[],"Service":[],"SoftwareApplication":[],"WebPage":[]},"graphName":"","isEnabled":true},"graphs":[]},"schema_type":null,"schema_type_options":null,"pillar_content":false,"robots_default":true,"robots_noindex":false,"robots_noarchive":false,"robots_nosnippet":false,"robots_nofollow":false,"robots_noimageindex":false,"robots_noodp":false,"robots_notranslate":false,"robots_max_snippet":null,"robots_max_videopreview":null,"robots_max_imagepreview":"large","priority":null,"frequency":null,"local_seo":null,"seo_analyzer_scan_date":"2026-01-21 12:30:22","breadcrumb_settings":null,"limit_modified_date":false,"reviewed_by":null,"ai":null,"created":"2021-03-01 02:52:26","updated":"2026-01-21 12:30:22","focus_keyword":null,"additional_keywords":null,"truseo_locale":null},"gt_translate_keys":[{"key":"link","format":"url"}],"_links":{"self":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts\/32781","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/comments?post=32781"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts\/32781\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/media\/24558"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/media?parent=32781"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/categories?post=32781"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/tags?post=32781"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}