{"id":6945,"date":"2024-07-22T13:27:09","date_gmt":"2024-07-22T12:27:09","guid":{"rendered":"https:\/\/sbunker.org\/?p=6945"},"modified":"2024-08-22T13:05:29","modified_gmt":"2024-08-22T12:05:29","slug":"halucinacionet-e-inteligjences-artificiale-dhe-dezinformimi-sintetik","status":"publish","type":"post","link":"https:\/\/sbunker.org\/sr\/disinfo\/halucinacionet-e-inteligjences-artificiale-dhe-dezinformimi-sintetik\/","title":{"rendered":"Halucinacionet e Inteligjenc\u00ebs Artificiale dhe dezinformimi sintetik"},"content":{"rendered":"

Mjetet gjeneruese t\u00eb IA-s\u00eb, krijojn\u00eb p\u00ebrmbajtje si tekst, imazhe, audio dhe media t\u00eb tjera bazuar n\u00eb t\u00eb dh\u00ebna nga interneti. Megjithat\u00eb, ato ngjallin shqet\u00ebsime p\u00ebr k\u00ebrc\u00ebnimet potenciale. Problemet lindin kur p\u00ebrmbajtjet e gjeneruara nga IA p\u00ebrdoren\u00a0 p\u00ebr t\u00eb mashtruar njer\u00ebzit, duke p\u00ebrhapur dezinformim dhe keqinformim me kosto t\u00eb ul\u00ebt. Ekspert\u00ebt paralajm\u00ebrojn\u00eb se k\u00ebto p\u00ebrmbajtje mund t\u00eb mashtrojn\u00eb publikun m\u00eb leht\u00eb se ato t\u00eb krijuara nga njer\u00ebzit.\u00a0<\/span><\/p>\n

Mjetet gjeneruese t\u00eb IA-s\u00eb ngren\u00eb dy shqet\u00ebsime n\u00eb lidhje me dezinformimin: Ato mund t\u00eb \u201chalucinojn\u00eb\u201d ose t\u00eb krijojn\u00eb tekste dhe imazhe q\u00eb ting\u00ebllojn\u00eb dhe duken t\u00eb besueshme, por q\u00eb devijojn\u00eb nga realiteti ose nuk kan\u00eb baz\u00eb n\u00eb fakt, dhe t\u00eb cilat p\u00ebrdoruesit e pakujdessh\u00ebm ose t\u00eb pavet\u00ebdijsh\u00ebm mund t’i p\u00ebrhapin pa dashje si fakt. Ato mund t\u00eb leht\u00ebsojn\u00eb krijimin dhe p\u00ebrhapjen e q\u00ebllimshme t\u00eb dezinformatave bind\u00ebse, duke p\u00ebrfshir\u00eb “falsifikime t\u00eb thella (deepfake)”, ose audio, video dhe fotografi t\u00eb ngjarjeve q\u00eb nuk kan\u00eb ndodhur kurr\u00eb.<\/span><\/p>\n

Halucionacionet e IA-s\u00eb si shkak i keqinformimit<\/b><\/p>\n

Keqinformimi nuk \u00ebsht\u00eb gjithmon\u00eb i q\u00ebllimsh\u00ebm. Ne shpesh presupozojm\u00eb se marrim p\u00ebrgjigjet t\u00eb sakta nga chatbot-\u00ebt e drejtuar nga IA. Megjithat\u00eb, algoritmet e tyre ndonj\u00ebher\u00eb mund t\u00eb prodhojn\u00eb p\u00ebrgjigje q\u00eb nuk bazohen n\u00eb t\u00eb dh\u00ebnat e m\u00ebsuara dhe nuk ndjekin ndonj\u00eb model identifikues. K\u00ebto \u201chalucinacione\u201d gjat\u00eb gjenerimit t\u00eb p\u00ebrmbajtjeve.<\/span><\/p>\n

Si p\u00ebrkufizim <\/span>konceptual<\/span><\/a>, halucinacioni \u00ebsht\u00eb perceptimi i objekteve apo ngjarjeve q\u00eb nuk kan\u00eb nj\u00eb burim t\u00eb jasht\u00ebm real. P\u00ebr shembull, d\u00ebgjimi i emrit t\u00eb nj\u00eb personi q\u00eb thirret nga nj\u00eb z\u00eb q\u00eb askush tjet\u00ebr nuk e d\u00ebgjon. <\/span>Halucinacioni i inteligjenc\u00ebs artificiale<\/span><\/a> ndodh kur nj\u00eb model i IA prodhon rezultate t\u00eb pamjaftueshme ose mashtruese gjat\u00eb analizimit t\u00eb t\u00eb dh\u00ebnave. Ky fenomen mund t\u00eb shkaktohet nga t\u00eb dh\u00ebna t\u00eb pamjaftueshme, kontradiktore ose nga kompleksiteti i modelit.<\/span><\/p>\n

Modele t\u00eb m\u00ebdha gjuh\u00ebsore (LLM) <\/span>jan\u00eb programuar p\u00ebr t\u2019iu p\u00ebrgjigjur<\/span><\/a> pyetjeve, jo p\u00ebr t\u2019u p\u00ebrgjigjur sakt\u00eb, por vet\u00ebm p\u00ebr t\u2019iu p\u00ebrgjigjur. Ato prodhojn\u00eb p\u00ebrgjigje bazuar n\u00eb probabilitetin e fjal\u00ebve q\u00eb vijn\u00eb nj\u00ebra pas tjetr\u00ebs dhe modelin midis fjal\u00ebve. M\u00ebnyra se si fjal\u00ebt jan\u00eb rreshtuar nj\u00ebra pas tjetr\u00ebs n\u00eb grupet e t\u00eb dh\u00ebnave q\u00eb p\u00ebrdoren p\u00ebr t\u00eb trajnuar IA-n\u00eb \u00ebsht\u00eb vendimtare. N\u00ebse fjala me nj\u00eb probabilitet t\u00eb lart\u00eb \u00ebsht\u00eb zgjedhja e duhur e fjal\u00ebs sipas kontekstit, vet\u00ebm at\u00ebher\u00eb modelet e m\u00ebdha gjuh\u00ebsore mund t\u00eb japin informacionin e sakt\u00eb.<\/span><\/p>\n

N\u00eb nj\u00eb<\/span> studim<\/span><\/a> q\u00eb teston sakt\u00ebsin\u00eb e p\u00ebrgjigjeve t\u00eb ChatGPT-s\u00eb p\u00ebr vendlindjet e politikan\u00ebve amerikan\u00eb, u zbulua se ChatGPT 3.5 ofroi informacione t\u00eb pav\u00ebrteta p\u00ebr vendlindjet e disa politikan\u00ebve si Hillary Clinton dhe John F. Kennedy Jr. Gjithashtu, n\u00eb janar t\u00eb vitit t\u00eb kaluar, platforma <\/span>NewsGuard<\/span><\/a>, zbuloi q\u00eb chatboti gjeneroi narrativa t\u00eb rreme dhe mashtruese p\u00ebr tema t\u00eb r\u00ebnd\u00ebsishme si Covid-19 dhe agresioni rus n\u00eb Ukrain\u00eb etj. S\u00eb fundmi, <\/span>Reuters Institute<\/span><\/a>, testoi p\u00ebrgjigjet e chatbot\u00ebve (ChatGPT-4o, Google Gemini dhe Perplexity.ai ) ndaj pyetjeve baz\u00eb rreth zgjedhjeve t\u00eb Parlamentit Evropian t\u00eb vitit 2024. Mes p\u00ebrgjigjeve t\u00eb sakta kishte edhe t\u00eb atilla q\u00eb ishin pjes\u00ebrisht t\u00eb sakta apo edhe t\u00ebr\u00ebsisht t\u00eb rreme. Prandaj, ChatGPT paralajm\u00ebrimon: \u201c<\/span>ChatGPT-ja mund t\u00eb b\u00ebj\u00eb gabime. Kontrollo informacionet e r\u00ebnd\u00ebsishme<\/span><\/i>\u201d. <\/span><\/p>\n

Halucinacione t\u00eb ngjashme jan\u00eb shfaqur edhe gjat\u00eb gjenerimit t\u00eb imazheve me inteligjenc\u00eb artificiale. Me nisjen e konfliktit Hamas-Izrael, n\u00eb rrjet filloi shp\u00ebrndarje e imazheve q\u00eb pretendonin t\u00eb paraqisnin pamje nga ky konflikt, p\u00ebrfshi edhe imazhe t\u00eb gjeneruara nga IA-ja. N\u00eb nj\u00eb nga imazhet e gjeneruara shihen <\/span>dy mjek\u00eb duke kryer nj\u00eb operacion n\u00eb nj\u00eb objekt t\u00eb shkat\u00ebrruar<\/span><\/a>, ku v\u00ebrehen shtremb\u00ebrime dhe me nj\u00eb kontekst t\u00eb palogjiksh\u00ebm i fotografis\u00eb. <\/span><\/p>\n

Por ky nuk \u00ebsht\u00eb imazhi i vet\u00ebm i shp\u00ebrndar\u00eb q\u00eb \u00ebsht\u00eb gjeneruar me Inteligjenc\u00eb Artificiale, e q\u00eb p\u00ebrmbajtur halucinacione. Shembuj t\u00eb tjer\u00eb p\u00ebrfshijn\u00eb fotografin\u00eb e <\/span>dy f\u00ebmij\u00ebve duke u lutur p\u00ebr n\u00ebn\u00ebn e tyre t\u00eb vdekur<\/span><\/a>, <\/span>nj\u00eb burr\u00eb me pes\u00eb f\u00ebmij\u00eb mes g\u00ebrmadhash<\/span><\/a>, <\/span>imazhe t\u00eb f\u00ebmij\u00ebve<\/span><\/a> n\u00eb <\/span>mes t\u00eb rr\u00ebnojave<\/span><\/a>. K\u00ebto imazhe jan\u00eb publikuar n\u00eb mediat sociale m\u00eb p\u00ebrmbajtje n\u00eb gjuh\u00ebn shqipe, duke p\u00ebrdorur elementin emocional, e duke i b\u00ebr\u00eb t\u00eb duken t\u00eb besueshme dhe t\u00eb nxisin shp\u00ebrndarjen e tyre. <\/span><\/p>\n

\"\"<\/p>\n

\"\"<\/p>\n

Dezinformimi sintetik: Deepfake<\/b>\u00a0<\/span><\/p>\n

Nd\u00ebrsa IA-ja mjegullon vij\u00ebn midis faktit dhe trillimit, v\u00ebrehet nj\u00eb rritje e fushatave t\u00eb dezinformimit me video deepfake q\u00eb mund t\u00eb manipulojn\u00eb opinionin publik dhe t\u00eb ndikojn\u00eb negativisht n\u00eb proceset demokratike.<\/span><\/p>\n

Nj\u00eb nga aplikimet kryesore t\u00eb Deep Neural Networks (Deep Learning) \u00ebsht\u00eb prodhimi i t\u00eb dh\u00ebnave sintetike. \u00c7do dit\u00eb, modeleve t\u00eb reja t\u00eb IA-s\u00eb u shtohen aft\u00ebsi p\u00ebr t\u00eb krijuar sintetikisht imazhe, video, tinguj dhe muzik\u00eb. K\u00ebto modele krijojn\u00eb mund\u00ebsin\u00eb e gjenerimit t\u00eb t\u00eb dh\u00ebnave sintetike q\u00eb jan\u00eb t\u00eb pamundura p\u00ebr t’u dalluar nga realiteti.<\/span><\/p>\n

Inteligjenca artificiale ka hapur dimensione t\u00eb reja n\u00eb krijimin e materialeve t\u00eb rreme audio dhe video. <\/span>Deepfake<\/span><\/a>, si <\/span>media sintetike<\/span><\/a>, \u00ebsht\u00eb nj\u00eb teknik\u00eb e p\u00ebrdorur p\u00ebr t\u00eb z\u00ebvend\u00ebsuar fytyrat, z\u00ebrat dhe shprehjet e njer\u00ebzve n\u00eb video, duke i b\u00ebr\u00eb ata t\u00eb duken sikur po thon\u00eb ose b\u00ebjn\u00eb di\u00e7ka q\u00eb nuk e kan\u00eb b\u00ebr\u00eb kurr\u00eb. Kjo teknologji mund\u00ebson gjithashtu gjenerimin e audio dhe videove q\u00eb duken dhe ting\u00ebllojn\u00eb autentike.<\/span><\/p>\n

Fotografit\u00eb <\/span>e Pap\u00ebs me palto t\u00eb bardh\u00eb<\/span><\/a> dhe ato t\u00eb <\/span>arrestimit t\u00eb Donald Trump<\/span><\/a> ishin nd\u00ebr rastet e para q\u00eb paralajm\u00ebruan p\u00ebr fenomenin e gjenerimit t\u00eb imazheve me IA. N\u00eb mediat sociale jan\u00eb publikuar disa p\u00ebrmbajtje deepfake, sikurse videoja e kryeministrit t\u00eb Kosov\u00ebs, <\/span>Albin Kurti duke folur n\u00eb gjuh\u00ebn arabe<\/span><\/a>, apo e <\/span>Elon Musk-ut duke folur n\u00eb shqip<\/span><\/a>. K\u00ebto u publikuan kryesisht n\u00eb TikTok me p\u00ebrmbajtje n\u00eb gjuh\u00ebn shqipe. <\/span><\/p>\n

\"\"<\/p>\n

Keqinformimi i krijuar nga IA priret t\u00eb ket\u00eb nj\u00eb t\u00ebrheqje m\u00eb t\u00eb madhe emocionale, sikurse q\u00eb ishte rasti me publikimin e videos me <\/span>z\u00eb t\u00eb sintetizuar t\u00eb figur\u00ebs s\u00eb Liridona Murseli<\/span><\/a>t me nj\u00eb tekst emocional t\u00eb p\u00ebrgatitur  K\u00ebto p\u00ebrmbajtje kan\u00eb ndikuar n\u00eb form\u00ebsimin e opinionit publik p\u00ebr shkak t\u00eb nd\u00ebrveprimit q\u00eb ngjall\u00ebn.<\/span><\/p>\n

P\u00ebr t\u2019i dalluar halucionacionet e IA-s\u00eb, duhet pasur parasysh aspektin e p\u00ebrgjithsh\u00ebm t\u00eb tekstit t\u00eb shkruar nga IA-ja, si p\u00ebrdorimi i fjal\u00ebve t\u00eb \u00e7uditshme, p\u00ebrfshirja pa r\u00ebnd\u00ebsi ose pa p\u00ebrshtatje t\u00eb plot\u00eb me kontekstin. Me imazhet dhe videot, v\u00ebmendja duhet t\u00eb p\u00ebrqendrohet n\u00eb ndryshimet e papritura n\u00eb ndri\u00e7im, l\u00ebvizjet e pazakonta t\u00eb fytyr\u00ebs, ose p\u00ebrzierjet e \u00e7uditshme t\u00eb sfondit, q\u00eb mund t\u00eb tregojn\u00eb se ato jan\u00eb krijuar p\u00ebrmes IA. Kur shqyrtohet burimi, verifikohet n\u00ebse ky vjen nga nj\u00eb burim i besuesh\u00ebm medial, ose nga nj\u00eb faqe e panjohur. Gjithashtu, n\u00ebse nj\u00eb postim duket shum\u00eb i \u00e7uditsh\u00ebm p\u00ebr t\u00eb qen\u00eb i v\u00ebrtet\u00eb, k\u00ebrkoni n\u00eb Google p\u00ebr t\u00eb par\u00eb n\u00ebse \u00ebsht\u00eb e v\u00ebrtet\u00eb apo \u00ebsht\u00eb vet\u00ebm nj\u00eb p\u00ebrmbajtje IA q\u00eb \u00ebsht\u00eb b\u00ebr\u00eb virale.<\/span><\/p>\n

<\/b>Roli i mediave n\u00eb parandalimin e dezinformimit<\/b><\/p>\n

Mediat informative mund t\u00eb luajn\u00eb nj\u00eb rol t\u00eb r\u00ebnd\u00ebsish\u00ebm n\u00eb <\/span>pre-bunking<\/span><\/i> (demaskim paraprak), nj\u00eb teknik\u00eb e krijuar p\u00ebr t\u00eb ndihmuar njer\u00ebzit t\u00eb identifikojn\u00eb dhe t\u2019i rezistojn\u00eb p\u00ebrmbajtjeve manipuluese. Duke paralajm\u00ebruar dhe pajisur publikun me aft\u00ebsi p\u00ebr t\u00eb dalluar dhe p\u00ebr t\u00eb hedhur posht\u00eb mashtrimet, k\u00ebto mesazhe ndihmojn\u00eb njer\u00ebzit t\u00eb zhvillojn\u00eb rezistenc\u00eb ndaj manipulimeve n\u00eb t\u00eb ardhmen.<\/span><\/p>\n

Sipas  <\/span>udh\u00ebzuesit t\u00eb KMSHK-s\u00eb<\/span><\/a>, mediat mund t\u00eb p\u00ebrdorin inteligjenc\u00ebn artificiale p\u00ebr t\u00eb krijuar p\u00ebrmbajtje mediatike, por duhet ta b\u00ebjn\u00eb k\u00ebt\u00eb n\u00eb m\u00ebnyr\u00eb t\u00eb p\u00ebrgjegjshme. N\u00eb procesin e p\u00ebrdorimit t\u00eb IA-s\u00eb, mediat duhet t\u00eb mbrojn\u00eb t\u00eb dh\u00ebnat personale, sigurin\u00eb e informacionit dhe t\u00eb shmangin diskriminimin. P\u00ebrgjegj\u00ebsia editoriale p\u00ebr p\u00ebrmbajtjen e krijuar me IA mbetet tek mediat dhe nuk i liron gazetar\u00ebt dhe redaktor\u00ebt nga standardet e redaktimit. <\/span><\/p>\n

Mediat duhet t\u00eb informojn\u00eb lexuesit n\u00eb m\u00ebnyr\u00eb t\u00eb qart\u00eb p\u00ebr p\u00ebrdorimin e IA-s\u00eb n\u00eb krijimin e p\u00ebrmbajtjes, duke specifikuar segmentin dhe llojin e IA-s\u00eb q\u00eb \u00ebsht\u00eb p\u00ebrdorur dhe m\u00ebnyr\u00ebn e funksionimit t\u00eb saj. P\u00ebrdorimi i IA-s\u00eb duhet t\u00eb respektoj\u00eb t\u00eb drejtat e autorit. Lexuesit duhet t\u00eb informohen paraprakisht n\u00ebse nd\u00ebrveprimet e tyre jan\u00eb me IA-n\u00eb apo me njer\u00ebz. P\u00ebr vendimet e marra nga IA-ja q\u00eb mund t\u00eb shqet\u00ebsojn\u00eb lexuesit, mediat duhet t’u ofrojn\u00eb mund\u00ebsin\u00eb p\u00ebr komunikim me nj\u00eb person fizik dhe t\u00eb drejt\u00ebn e kund\u00ebrshtimit t\u00eb vendimit t\u00eb marra.<\/span><\/p>\n

Ky artikull u p\u00ebrgatit nga Sbunker n\u00eb kuad\u00ebr t\u00eb projektit \u201cFuqizimi i q\u00ebndrueshm\u00ebris\u00eb s\u00eb komunitetit kund\u00ebr dezinformimit\u201d, mb\u00ebshtetur p\u00ebrmes Programit t\u00eb Aktivizmit Dixhital nga TechSoup Global.<\/p>","protected":false},"excerpt":{"rendered":"

Mjetet gjeneruese t\u00eb IA-s\u00eb, krijojn\u00eb p\u00ebrmbajtje si tekst, imazhe, audio dhe media t\u00eb tjera bazuar n\u00eb t\u00eb dh\u00ebna nga interneti. Megjithat\u00eb, ato ngjallin shqet\u00ebsime p\u00ebr k\u00ebrc\u00ebnimet potenciale. Problemet lindin kur p\u00ebrmbajtjet e gjeneruara nga IA p\u00ebrdoren\u00a0 p\u00ebr t\u00eb mashtruar njer\u00ebzit, duke p\u00ebrhapur dezinformim dhe keqinformim me kosto t\u00eb ul\u00ebt. Ekspert\u00ebt paralajm\u00ebrojn\u00eb se k\u00ebto p\u00ebrmbajtje mund […]<\/p>","protected":false},"author":445,"featured_media":6964,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[363],"tags":[662,1020,1021,989,624],"ppma_author":[1022],"class_list":["post-6945","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-disinfo","tag-deepfake","tag-dezinformimi","tag-dezinformimi-sintetik","tag-inteligjenca-artificiale","tag-kryesore"],"authors":[{"term_id":1022,"user_id":445,"is_guest":0,"slug":"festim-rizanaj","display_name":"Festim Rizanaj","avatar_url":{"url":"https:\/\/sbunker.org\/wp-content\/uploads\/2024\/07\/festim.png","url2x":"https:\/\/sbunker.org\/wp-content\/uploads\/2024\/07\/festim.png"},"user_url":"","last_name":"Rizanaj","first_name":"Festim","description":"Festim Rizanaj \u00ebsht\u00eb hulumtues n\u00eb hibrid.info."}],"_links":{"self":[{"href":"https:\/\/sbunker.org\/sr\/wp-json\/wp\/v2\/posts\/6945","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/sbunker.org\/sr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/sbunker.org\/sr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/sbunker.org\/sr\/wp-json\/wp\/v2\/users\/445"}],"replies":[{"embeddable":true,"href":"https:\/\/sbunker.org\/sr\/wp-json\/wp\/v2\/comments?post=6945"}],"version-history":[{"count":10,"href":"https:\/\/sbunker.org\/sr\/wp-json\/wp\/v2\/posts\/6945\/revisions"}],"predecessor-version":[{"id":7264,"href":"https:\/\/sbunker.org\/sr\/wp-json\/wp\/v2\/posts\/6945\/revisions\/7264"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/sbunker.org\/sr\/wp-json\/wp\/v2\/media\/6964"}],"wp:attachment":[{"href":"https:\/\/sbunker.org\/sr\/wp-json\/wp\/v2\/media?parent=6945"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/sbunker.org\/sr\/wp-json\/wp\/v2\/categories?post=6945"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/sbunker.org\/sr\/wp-json\/wp\/v2\/tags?post=6945"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/sbunker.org\/sr\/wp-json\/wp\/v2\/ppma_author?post=6945"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}