{"id":38469,"date":"2026-03-28T16:49:12","date_gmt":"2026-03-28T16:49:12","guid":{"rendered":"https:\/\/www.europesays.com\/cz\/38469\/"},"modified":"2026-03-28T16:49:12","modified_gmt":"2026-03-28T16:49:12","slug":"pod-carou-chcete-si-overit-informace-zeptat-se-chatbota-je-cesta-do-pekla","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/cz\/38469\/","title":{"rendered":"Pod \u010darou: Chcete si ov\u011b\u0159it informace? Zeptat se chatbota je cesta do pekla"},"content":{"rendered":"

V newsletteru Pod \u010darou popisuje ka\u017edou sobotu Matou\u0161 Hrdina spole\u010densk\u00e9 trendy, kter\u00e9 sice v\u00edd\u00e1me v\u0161ude kolem sebe, ale v\u00a0p\u0159\u00edlivu ka\u017edodenn\u00edho zpravodajstv\u00ed trochu zanikaj\u00ed mezi \u0159\u00e1dky. Pokud v\u00e1s uk\u00e1zka zaujme, p\u0159ihlaste se k\u00a0odb\u011bru<\/a> pln\u00e9 verze newsletteru.<\/p>\n

Znovu se potvrzuje smutn\u00e1 pravda, \u017ee tou nejv\u011bt\u0161\u00ed laborato\u0159\u00ed technologick\u00fdch novinek b\u00fdvaj\u00ed v\u00e1lky. Konflikt v\u00a0Persk\u00e9m z\u00e1livu mimo jin\u00e9 potvrdil rostouc\u00ed roli dron\u016f \u010di r\u016fzn\u00fdch asymetrick\u00fdch metod boje, ale \u0159ada d\u016fle\u017eit\u00fdch zm\u011bn se odehr\u00e1v\u00e1 tak\u00e9 v\u00a0kyberprostoru.<\/p>\n

Z v\u00e1lky s\u00a0\u00cdr\u00e1nem se rychle stal prvn\u00ed velk\u00fd st\u0159et, ve kter\u00e9m hraje z\u00e1sadn\u00ed roli AI propaganda. Hned po prvn\u00edm americko-izraelsk\u00e9m \u00fatoku zavalila soci\u00e1ln\u00ed s\u00edt\u011b lavina fotek a\u00a0videa generovan\u00e9ho AI n\u00e1stroji. Fale\u0161n\u00e9 exploze, po\u017e\u00e1ry, ruiny, protesty \u010di potopen\u00e9 lod\u011b na v\u0161ech stran\u00e1ch konfliktu, nav\u00edc v\u00a0nev\u00eddan\u00e9 vizu\u00e1ln\u00ed kvalit\u011b\u00a0\u2013 podrobn\u00fd p\u0159ehled t\u011bch nej\u010dast\u011bj\u0161\u00edch fakes ned\u00e1vno sestavila<\/a> redakce New York Times.<\/p>\n

Sv\u00e9 digit\u00e1ln\u00ed propagandistick\u00e9 aktivity celkem pochopiteln\u011b pos\u00edlil a\u00a0dle pot\u0159eby upravil<\/a> \u00cdr\u00e1n. Vzhledem k\u00a0n\u00e1m\u011bt\u016fm koluj\u00edc\u00edch fakes mu v\u00a0tom jist\u011b \u0161lapou na paty i\u00a0dal\u0161\u00ed znesv\u00e1\u0159en\u00e9 strany. Pro v\u011bt\u0161\u00ed zmaten\u00ed za\u010d\u00ednaj\u00ed b\u00fdt za AI ozna\u010dov\u00e1na tak\u00e9 re\u00e1ln\u00e1 videa a\u00a0m\u00e9dia i\u00a0ve\u0159ejnost stoj\u00ed p\u0159ed st\u00e1le pal\u010div\u011bj\u0161\u00edm probl\u00e9mem\u00a0\u2013 jak odli\u0161it realitu od fikce a\u00a0p\u0159in\u00e1\u0161et kvalitn\u00ed informace o\u00a0skute\u010dn\u00e9m pr\u016fb\u011bhu konfliktu.<\/p>\n

Dezinformace a\u00a0fale\u0161n\u00e9 \u010di pozm\u011bn\u011bn\u00e9 obsahy nejsou \u017e\u00e1dnou novinkou. Na s\u00edt\u00edch kolovaly d\u00e1vno p\u0159ed n\u00e1stupem sou\u010dasn\u00fdch AI n\u00e1stroj\u016f, a\u00a0proto tak\u00e9 ji\u017e dlouho zn\u00e1me r\u016fzn\u00e9 obrann\u00e9 taktiky a\u00a0metody ov\u011b\u0159ov\u00e1n\u00ed.<\/p>\n

P\u00e1r z\u00e1kladn\u00edch hezky shrnul<\/a> web The Verge. Jeho redakto\u0159i v\u00a0prvn\u00ed \u0159ad\u011b upozor\u0148uj\u00ed, \u017ee ne v\u0161echny dezinformace mus\u00ed b\u00fdt nutn\u011b AI fakes. St\u00e1le se pou\u017e\u00edvaj\u00ed i \u201estar\u00e9\u201c metody, tedy t\u0159eba vyu\u017e\u00edv\u00e1n\u00ed z\u00e1b\u011br\u016f z\u00a0jin\u00fdch v\u00e1lek, nehod \u010di pouli\u010dn\u00edch protest\u016f, nebo tak\u00e9 klipy z\u00a0dnes ji\u017e tak\u0159ka realistick\u00fdch videoher.<\/p>\n

Pokud u\u017e ale \u010del\u00edme v\u00fdtvor\u016fm AI, objevuj\u00ed se dal\u0161\u00ed komplikace\u00a0\u2013 neexistuj\u00ed \u017e\u00e1dn\u00e9 spolehliv\u00e9 AI detektory AI obsahu (tento z\u00e1sadn\u00ed bod si zapamatujte, vr\u00e1t\u00edme se k\u00a0n\u011bmu), a\u00a0tak je pot\u0159eba za\u010d\u00edt ov\u011b\u0159ovat v\u00edce sofistikovan\u00fdmi metodami. Ke slovu p\u0159ich\u00e1zej\u00ed podrobn\u00e9 expertn\u00ed rozbory ka\u017ed\u00e9ho z\u00e1b\u011bru. Je tak\u00e9 pot\u0159eba br\u00e1t v\u00a0potaz kontext dan\u00e9ho obsahu (i re\u00e1ln\u00e9 video m\u016f\u017ee b\u00fdt zasazeno do manipulativn\u00edch souvislost\u00ed), p\u0159ihl\u00ed\u017eet ke st\u00e1\u0159\u00ed a\u00a0p\u016fvodu \u00fa\u010dt\u016f, kter\u00e9 ho na s\u00edt\u00edch sd\u00edlej\u00ed, a\u00a0p\u0159edev\u0161\u00edm sledovat cesty jeho \u0161\u00ed\u0159en\u00ed.<\/p>\n

Poslechn\u011bte si audioverzi newsletteru na\u010dtenou autorem.<\/p>\n

Podobn\u00e9 tipy d\u00e1v\u00e1 k\u00a0dobru i\u00a0The Guardian. Jeho koment\u00e1torka Margaret Sullivan v\u0161e shrnuje<\/a> ve t\u0159ech obecn\u00fdch bodech. V\u00a0prvn\u00ed \u0159ad\u011b p\u0159istupujte se zdravou ned\u016fv\u011brou i\u00a0sami k\u00a0sob\u011b\u00a0\u2013 ka\u017ed\u00fd m\u00e1me r\u016fzn\u00e9 preference a\u00a0zkreslen\u00ed a\u00a0velmi r\u00e1di d\u016fv\u011b\u0159ujeme informac\u00edm, kter\u00e9 zapadaj\u00ed do na\u0161eho obrazu sv\u011bta a\u00a0politick\u00fdch n\u00e1zor\u016f. D\u00e1vejme v\u011bci do \u0161ir\u0161\u00edho kontextu a\u00a0p\u0159edev\u0161\u00edm – d\u016fv\u011b\u0159ujme expert\u016fm.<\/p>\n

Jen\u017ee pr\u00e1v\u011b v\u00a0tomto bod\u011b tkv\u00ed j\u00e1dro aktu\u00e1ln\u00ed dezinforma\u010dn\u00ed krize. D\u016fv\u011bra v\u00a0tradi\u010dn\u00ed m\u00e9dia, akademiky a\u00a0dal\u0161\u00ed experty dlouhodob\u011b kles\u00e1 (mimo jin\u00e9 proto, \u017ee ji c\u00edlen\u011b podr\u00fdvaj\u00ed pr\u00e1v\u011b propagandist\u00e9 v\u0161eho druhu), ale te\u010f se ve\u0159ejnosti nab\u00edz\u00ed nov\u00e1 alternativa\u00a0\u2013 u\u017e p\u0159ece nepot\u0159ebujeme tradi\u010dn\u00ed novin\u00e1\u0159e a\u00a0experty, ani nemus\u00edme dumat nad t\u00edm, kter\u00fd z \u201ealternativn\u00edch\u201c web\u016f \u010di influencer\u016f se n\u00e1m v\u00edc zamlouv\u00e1. Jakoukoliv informaci toti\u017e snadno ov\u011b\u0159\u00ed \u010di vyvr\u00e1t\u00ed v\u0161emocn\u00e1 AI.<\/p>\n

Robot z\u00edr\u00e1 na robota<\/p>\n

Pokud by snad n\u011bkdo pochyboval o\u00a0tom, \u017ee robotick\u00e9 ov\u011b\u0159ov\u00e1n\u00ed pravdivosti a\u00a0p\u016fvodu informac\u00ed je cesta do pekla, ud\u00e1losti uplynul\u00fdch t\u00fddn\u016f jej musely rychle vyv\u00e9st z\u00a0omylu. Jako prvn\u00ed pos\u00edlily chaos zmi\u0148ovan\u00e9 dezinforma\u010dn\u00ed triky, kdy za\u010daly b\u00fdt r\u016fzn\u00e9 skute\u010dn\u00e9 z\u00e1b\u011bry ozna\u010dov\u00e1ny za AI\u00a0\u2013 nap\u0159\u00edklad NY Times musely opakovan\u011b potvrzovat<\/a> pravost fotky dav\u016f oslavuj\u00edc\u00edch v\u00a0Teher\u00e1nu nov\u00e9ho \u00edr\u00e1nsk\u00e9ho v\u016fdce Mod\u017etabu Ch\u00e1mene\u00edho, kterou mnoz\u00ed diskutuj\u00edc\u00ed na s\u00edt\u00edch pova\u017eovali za AI pad\u011blek.<\/p>\n

Ob\u010das k\u00a0tomu sta\u010dilo jen pouh\u00e9 na\u0159\u010den\u00ed, ale velk\u00fd pod\u00edl viny m\u011bli i\u00a0samotn\u00ed \u0159adov\u00ed u\u017eivatel\u00e9, kter\u00e9 \u010dasto nenapadlo nic lep\u0161\u00edho, ne\u017e si kv\u016fli nedostatku ofici\u00e1ln\u00edho ov\u011b\u0159en\u00ed (kter\u00e9 je pomal\u00e9, pracn\u00e9 a\u00a0tak\u00e9 z\u00e1visl\u00e9 na d\u016fv\u011b\u0159e v\u00a0m\u00e9dia) pomoci jinak a\u00a0obr\u00e1tit se na chatbota\u00a0\u2013 a\u0165 u\u017e je to notoricky nespolehliv\u00fd Musk\u016fv Grok nebo i\u00a0dal\u0161\u00ed AI n\u00e1stroje.<\/p>\n

Tento probl\u00e9m se naplno projevil po tragick\u00e9m americk\u00e9m \u00fatoku<\/a> na d\u00edv\u010d\u00ed \u0161kolu v\u00a0jiho\u00edr\u00e1nsk\u00e9m Minabu, p\u0159i kter\u00e9m zahynulo nejm\u00e9n\u011b 175\u00a0civilist\u016f, v\u00a0drtiv\u00e9 v\u011bt\u0161in\u011b d\u011bt\u00ed. Na s\u00edt\u00edch se objevily re\u00e1ln\u00e9 z\u00e1b\u011bry trosek \u010di \u0159ad vykopan\u00fdch hrob\u016f, kter\u00e9 ov\u0161em chatboti za\u010dali ozna\u010dovat za AI pad\u011blek nebo za fotky jin\u00fdch ud\u00e1lost\u00ed vytr\u017een\u00e9 z\u00a0kontextu (t\u0159eba \u017ee jde o\u00a0sc\u00e9ny z\u00a0P\u00e1kist\u00e1nu \u010di Indon\u00e9sie). <\/p>\n

Do velk\u00e9 m\u00edry k\u00a0tomu p\u0159isp\u011bla p\u0159irozen\u00e1 chybovost AI n\u00e1stroj\u016f, zalo\u017een\u00fdch na velk\u00fdch jazykov\u00fdch modelech, a\u00a0jist\u011b i\u00a0fakt, \u017ee chatboti nekriticky \u201estr\u00e1vili\u201c nejr\u016fzn\u011bj\u0161\u00ed na\u0159\u010den\u00ed<\/a> a\u00a0manipulace koluj\u00edc\u00ed po s\u00edt\u00edch (anebo mo\u017en\u00e1 kop\u00edrovali \u00faryvky text\u016f z\u00a0r\u016fzn\u00fdch star\u00fdch online debat okolo dezinformac\u00ed).<\/p>\n

<\/a><\/p>\n

Ukazuje se, \u017ee kv\u016fli nedostatku digit\u00e1ln\u00ed gramotnosti i\u00a0zav\u00e1d\u011bj\u00edc\u00ed reklam\u011b AI firem si jen mal\u00e1 \u010d\u00e1st ve\u0159ejnosti uv\u011bdomuje skute\u010dnou podstatu LLM chatbot\u016f (kte\u0159\u00ed nemaj\u00ed \u017e\u00e1dn\u00e9 skute\u010dn\u00e9 my\u0161len\u00ed, pam\u011b\u0165 ani pon\u011bt\u00ed o\u00a0re\u00e1ln\u00e9m sv\u011bt\u011b). A\u00a0je\u0161t\u011b m\u00e9n\u011b lid\u00ed je pak schopno tyto teoretick\u00e9 pou\u010dky uv\u00e9st do praxe a\u00a0pochopit, jak snadn\u00e9 je chatboty manipulovat.<\/p>\n

Za p\u0159e\u010dten\u00ed v\u00a0tomto kontextu stoj\u00ed t\u0159eba na prvn\u00ed pohled z\u00e1bavn\u00fd, ale ve skute\u010dnosti dost d\u011bsiv\u00fd experiment<\/a> redaktora BBC Thomase Germaina, kter\u00fd za p\u00e1r des\u00edtek minut \u00fasp\u011b\u0161n\u011b \u201ehacknul\u201c ChatGPT i\u00a0Gemini a\u00a0p\u0159esv\u011bd\u010dil je, \u017ee je t\u00edm nejlep\u0161\u00edm technologick\u00fdm report\u00e9rem na poli sout\u011b\u017en\u00edho poj\u00edd\u00e1n\u00ed hot dog\u016f.<\/p>\n

Sta\u010dilo napsat \u010dl\u00e1nek na sv\u016fj web vy\u0161perkovan\u00fd o\u00a0d\u016fv\u011bryhodn\u00e1 sv\u011bdectv\u00ed koleg\u016f z\u00a0oboru. Kdy\u017e se pak chatbot\u016f zeptal, kdo s\u00a0p\u0159ehledem zv\u00edt\u011bzil v\u00a0leto\u0161n\u00edm (neexistuj\u00edc\u00edm) \u0161ampion\u00e1tu v\u00a0poj\u00edd\u00e1n\u00ed hot dog\u016f v\u00a0Ji\u017en\u00ed Dakot\u011b, narazil na sv\u00e9 jm\u00e9no a\u00a0jeho vylhan\u00e1 p\u00e1rkov\u00e1 kari\u00e9ra tak byla \u00fasp\u011b\u0161n\u011b zak\u00f3dov\u00e1na do j\u00e1dra AI ma\u0161iny.<\/p>\n

Jeho pokus m\u00e1 \u0159adu p\u0159\u00edm\u00fdch dopad\u016f pro reklamn\u00ed pr\u016fmysl a\u00a0kvalitu vyhled\u00e1va\u010d\u016f (uk\u00e1zalo se, \u017ee SEO, \u010dili optimalizace v\u00fdsledk\u016f vyhled\u00e1v\u00e1n\u00ed, v\u00a0AI \u00e9\u0159e nepadla, ale naopak se mimo\u0159\u00e1dn\u011b rozbujela), ale jeho z\u00e1v\u011bry lze aplikovat i\u00a0na poli dezinformac\u00ed\u00a0\u2013 nemysl\u00edc\u00ed stroj zkr\u00e1tka nedok\u00e1\u017ee 100% ov\u011b\u0159it, zda n\u011bco existuje, \u010di nikoliv (nato\u017e prov\u00e9st n\u011bjakou sofistikovan\u011bj\u0161\u00ed kontextovou anal\u00fdzu), a\u00a0i kdy\u017e jsou chatboti v\u00a0jin\u00fdch oblastech jist\u011b \u00fa\u017easn\u00e9 a\u00a0u\u017eite\u010dn\u00e9 stroje, jako n\u00e1stroje verifikace jsou nefunk\u010dn\u00ed a\u00a0vylo\u017een\u011b nebezpe\u010dn\u00e9.<\/p>\n

Des\u00edtky vte\u0159in na z\u00e1sadn\u00ed rozhodnut\u00ed<\/p>\n

Trag\u00e9die v\u00a0Minabu probl\u00e9my s\u00a0AI verifikac\u00ed a\u00a0rozhodov\u00e1n\u00edm nasv\u00edtila i\u00a0z jin\u00e9ho \u00fahlu. Kdy\u017e se uk\u00e1zalo, \u017ee \u0161lo o\u00a0n\u00e1sledek myln\u011b zac\u00edlen\u00e9ho americk\u00e9ho \u00fatoku, na s\u00edt\u00edch se okam\u017eit\u011b rozjely spekulace, \u017ee \u0161lo o\u00a0omyl AI n\u00e1stroje pov\u011b\u0159en\u00e9ho v\u00fdb\u011brem c\u00edl\u016f. Ve stejn\u00e9 dob\u011b nav\u00edc zu\u0159ily kontroverze kolem spolupr\u00e1ce americk\u00e9 arm\u00e1dy s\u00a0firmami Anthropic a\u00a0OpenAI, a\u00a0tak si \u0159ada lid\u00ed dom\u00fd\u0161lela, \u017ee v\u00a0tomto p\u0159\u00edpad\u011b selhal chatbot Claude od Anthropic. Jak ale uk\u00e1zala \u010derstv\u00e1 investigace Guardianu<\/a>, v\u0161e bylo trochu jinak.<\/p>\n

Skute\u010dn\u011b do\u0161lo k\u00a0tomu, \u017ee c\u00edl bombardov\u00e1n\u00ed byl myln\u011b identifikov\u00e1n AI n\u00e1strojem pracuj\u00edc\u00edm se zastaral\u00fdmi daty. Ne\u0161lo ale o \u201emodern\u00ed\u201c AI technologii zalo\u017eenou na velk\u00fdch jazykov\u00fdch modelech\u00a0\u2013 rozhodoval zde n\u00e1stroj Maven od firmy Palantir, kter\u00fd byl vytvo\u0159en d\u00e1vno p\u0159ed rozmachem LLM a\u00a0americk\u00e1 arm\u00e1da jej pou\u017e\u00edv\u00e1 u\u017e cel\u00e1 l\u00e9ta.<\/p>\n

Motivem k\u00a0jeho v\u00fdvoji byly zku\u0161enosti z\u00a0v\u00e1lky v\u00a0Ir\u00e1ku a\u00a0p\u0159edchoz\u00edch konflikt\u016f, kdy u\u017e sice ameri\u010dt\u00ed vojen\u0161t\u00ed pl\u00e1nova\u010di m\u011bli k\u00a0dispozici z\u00e1v\u011bje dat ze satelit\u016f, dron\u016f a\u00a0dal\u0161\u00edch za\u0159\u00edzen\u00ed, ale nebylo v\u00a0lidsk\u00fdch sil\u00e1ch je ru\u010dn\u011b analyzovat a\u00a0vytv\u00e1\u0159et z\u00a0nich dostate\u010dn\u011b rychle seznamy c\u00edl\u016f.<\/p>\n

Zat\u00edmco v\u00a0roce 2003\u00a0pracovaly na v\u00fdb\u011bru c\u00edl\u016f 2\u00a0tis\u00edce\u00a0lid\u00ed, v\u00a0roce 2024\u00a0u\u017e stejnou pr\u00e1ci s\u00a0pomoc\u00ed AI syst\u00e9mu Maven m\u011blo zvl\u00e1dat jen 20\u00a0voj\u00e1k\u016f identifikuj\u00edc\u00edch tis\u00edc c\u00edl\u016f za hodinu. V\u00a0praxi to ov\u0161em znamen\u00e1, \u017ee ka\u017ed\u00fd z\u00a0nich m\u00e1 na manu\u00e1ln\u00ed ov\u011b\u0159en\u00ed a\u00a0potvrzen\u00ed jednoho p\u0159edvybran\u00e9ho c\u00edle pouh\u00fdch 72\u00a0sekund\u00a0\u2013 a\u00a0asi nen\u00ed t\u0159eba dod\u00e1vat, k\u00a0\u010demu to pak m\u016f\u017ee v\u00e9st.<\/p>\n

Autor anal\u00fdzy Kevin Baker podot\u00fdk\u00e1, \u017ee se tak z\u00a0v\u00e1l\u010den\u00ed st\u00e1v\u00e1 neosobn\u00ed pr\u016fmysl, a\u00a0to zcela doslovn\u011b. Jde o\u00a0vz\u00e1cn\u00fd p\u0159\u00edpad situace, kdy technologie z\u00a0civiln\u00edho sektoru prosakuje do vojenstv\u00ed a\u00a0ne naopak. Maven byl vytvo\u0159en na z\u00e1klad\u011b softwaru pro tov\u00e1rn\u00ed projektov\u00fd management, zkombinovan\u00e9ho s\u00a0mapovou aplikac\u00ed, a\u00a0p\u0159i jeho pou\u017eit\u00ed pak doch\u00e1z\u00ed ke stejn\u00fdm probl\u00e9m\u016fm jako p\u0159i z\u00e1sec\u00edch v\u00a0automobilce\u00a0\u2013 c\u00edlem v\u00fdrobn\u00ed linky jsou stoupaj\u00edc\u00ed \u010d\u00edsla, nikdo se nept\u00e1 po jejich smyslu a\u00a0moc je kv\u016fli automatizaci p\u0159esouv\u00e1na od zku\u0161en\u00fdch d\u011bln\u00edk\u016f k\u00a0managementu, kter\u00fd nic netu\u0161\u00ed o\u00a0skute\u010dn\u00fdch podm\u00ednk\u00e1ch u\u00a0p\u00e1su.<\/p>\n

Kdy\u017e je n\u011bco ozna\u010deno jako c\u00edl, odpad\u00e1 t\u00edm diskuse, jestli a\u00a0pro\u010d je vhodn\u00e9 jej bombardovat, a\u00a0neosobn\u00ed strojov\u00e9 rozhodnut\u00ed vypad\u00e1 d\u016fv\u011bryhodn\u011bji ne\u017e kostrbat\u00e9 lidsk\u00e9 rozva\u017eov\u00e1n\u00ed. Jako kontrast pak anal\u00fdza Guardianu uv\u00e1d\u00ed postupy britsk\u00e9ho letectva, jeho\u017e n\u00e1lety byly sice v\u00a0Ir\u00e1ku v\u00fdrazn\u011b pomalej\u0161\u00ed a\u00a0m\u00e9n\u011b \u201eefektivn\u00ed\u201c, ale na rozd\u00edl od americk\u00fdch sil se Britov\u00e9 dopou\u0161t\u011bli mnohem men\u0161\u00edho mno\u017estv\u00ed omyl\u016f, \u00fatok\u016f na civilisty \u010di st\u0159elby do vlastn\u00edch \u0159ad\u00a0\u2013 nikoliv snad proto, \u017ee by byli chyt\u0159ej\u0161\u00ed \u010di v\u00edce mor\u00e1ln\u00ed, ale proto\u017ee vyu\u017e\u00edvali strojovou automatizaci v\u00a0mnohem men\u0161\u00ed m\u00ed\u0159e.<\/p>\n

N\u011bkolik des\u00edtek vte\u0159in prost\u011b pro spolehlivou lidskou verifikaci nesta\u010d\u00ed. Jestli\u017ee ji rovnou vynech\u00e1me \u00fapln\u011b a\u00a0spolehneme se jen na AI, probl\u00e9my se je\u0161t\u011b n\u00e1sob\u00ed, a\u00a0kdy\u017e se nejedn\u00e1 o\u00a0bohapust\u00e9 \u017evatl\u00e1n\u00ed s\u00a0Grokem, ale o\u00a0v\u00fdb\u011br c\u00edl\u016f bombardov\u00e1n\u00ed, v\u00fdsledkem jsou trag\u00e9die. Pokud jak\u00fdkoliv proces zak\u00f3dujeme do kusu softwaru, nen\u00ed u\u017e m\u00edsto na v\u00fdjimky a\u00a0oh\u00fdb\u00e1n\u00ed pravidel\u00a0\u2013 cel\u00fd syst\u00e9m m\u016f\u017ee jen automaticky kr\u00e1\u010det dop\u0159edu, ne\u017e se najednou zhrout\u00ed.<\/p>\n

Pr\u00e1ce, kterou AI nikdy nenahrad\u00ed<\/p>\n

Mohlo by se zd\u00e1t, \u017ee \u0159\u00edzen\u00ed n\u00e1let\u016f a\u00a0verifikace videa na Instagramu jsou dv\u011b velmi odli\u0161n\u00e9 situace, ale v\u00a0podstat\u011b je to \u00fapln\u011b stejn\u00fd proces\u00a0\u2013 v\u00a0jist\u00e9 f\u00e1zi v\u017edy nastane pot\u0159eba ru\u010dn\u00ed, lidsk\u00e9 expertn\u00ed verifikace dan\u00e9 informace \u010di rozhodnut\u00ed, a\u00a0kdy\u017e tento krok z\u010d\u00e1sti \u010di \u00fapln\u011b delegujeme na AI, rozto\u010d\u00ed se spir\u00e1la omyl\u016f.<\/p>\n

Na obecn\u00e9 teoretick\u00e9 rovin\u011b toto t\u00e9ma rozv\u00e1d\u00ed v\u00fdzkumn\u00edk z\u00a0MIT Christian Catalini. Spolu s\u00a0dal\u0161\u00edmi spoluautory p\u0159ed p\u00e1r t\u00fddny publikoval studii<\/a>, kter\u00e1 by m\u011bla b\u00fdt povinn\u00fdm \u010dten\u00edm pro v\u0161echny mana\u017eery, politiky i\u00a0kohokoliv dal\u0161\u00edho, kdo si v\u00a0sou\u010dasnosti l\u00e1me hlavu s\u00a0t\u00edm, pro\u010d AI automatizace v\u00a0organizac\u00edch v\u0161eho druhu nep\u0159in\u00e1\u0161\u00ed zdaleka takov\u00e9 \u00faspory<\/a> a\u00a0r\u016fst produktivity, jak\u00e9 n\u00e1m technobaroni ze Silicon Valley u\u017e l\u00e9ta slibuj\u00ed.<\/p>\n

<\/a><\/p>\n

Mechanizace \u201edu\u0161evn\u00ed\u201c pr\u00e1ce nen\u00ed tot\u00e9\u017e jako n\u00e1hrada ru\u010dn\u00edho \u017ed\u00edm\u00e1n\u00ed automatickou pra\u010dkou. Jsou v\u00a0n\u00ed specifick\u00e9 f\u00e1ze a\u00a0kroky, na kter\u00e9 se \u010dasto zapom\u00edn\u00e1, ale jsou kriticky d\u016fle\u017eit\u00e9 a\u00a0AI je z\u00a0logiky sv\u00e9ho fungov\u00e1n\u00ed nedok\u00e1\u017ee nahradit. Konkr\u00e9tn\u011b se jedn\u00e1 o\u00a0zmi\u0148ovanou verifikaci\u00a0\u2013 tedy vstup lidsk\u00e9ho experta, kter\u00fd mus\u00ed zkontrolovat a\u00a0potvrdit v\u00fdstup stroje (v tomto p\u0159\u00edpad\u011b AI). Zat\u00edmco samotnou mechanickou v\u00fdrobu lze velmi jednodu\u0161e \u0161k\u00e1lovat do ob\u0159\u00edch rozm\u011br\u016f (a s\u00a0nasazen\u00edm AI to plat\u00ed dvojn\u00e1sob), expertn\u00ed rozhodov\u00e1n\u00ed takto nakopnout nejde, a\u00a0ve firm\u00e1ch a\u00a0dal\u0161\u00edch organizac\u00edch se tak vytv\u00e1\u0159\u00ed z\u00e1sadn\u00ed \u201e\u0161punt\u201c br\u00e1n\u00edc\u00ed boomu produktivity a\u00a0\u00faspor.<\/p>\n

Proto se podle Cataliniho AI n\u00e1stroje nejprve rozmohly v\u00a0oblastech, kde je proces verifikace velmi jednoduch\u00fd a\u00a0t\u00edm i\u00a0zanedbateln\u00fd\u00a0\u2013 rozpoznat, jestli je na AI obr\u00e1zku dostate\u010dn\u011b hezk\u00e9 ko\u0165\u00e1tko, je ot\u00e1zkou vte\u0159iny a\u00a0pr\u016fmysl internetov\u00fdch mem\u016f tak m\u016f\u017ee uspokojiv\u011b bujet. Posoudit, jestli je nov\u00e1 v\u011bdeck\u00e1 studie skute\u010dn\u011b p\u0159evratn\u00fdm objevem \u010di nikoliv, ale zabere odborn\u00edkovi stejn\u00e9 mno\u017estv\u00ed pr\u00e1ce jako takovou studii napsat. A\u00a0i kdy\u017e tak m\u016f\u017eeme produkci samotn\u00fdch studi\u00ed zkou\u0161et automatizovat AI, p\u0159es onen verifika\u010dn\u00ed \u0161punt se nedostaneme. Stejn\u00fd sc\u00e9n\u00e1\u0159 bychom na\u0161li v\u00a0m\u00e9di\u00edch, arm\u00e1d\u011b a\u00a0v\u0161ude tam, kde je pro posouzen\u00ed kvality v\u00fdstup\u016f pracovn\u00edho procesu pot\u0159eba jen trochu zku\u0161en\u00e9 lidsk\u00e9 oko\u00a0\u2013 a\u00a0to je tak\u0159ka v\u0161ude.<\/p>\n

Nen\u00ed to p\u0159itom nijak nov\u00fd a\u00a0nezn\u00e1m\u00fd probl\u00e9m. Catalini uv\u00e1d\u00ed p\u0159\u00edklad tkalcovsk\u00fdch stav\u016f, kter\u00e9 sice byly na \u00fasvitu pr\u016fmyslov\u00e9 revoluce ikonick\u00fdm p\u0159\u00edkladem \u00fasp\u011b\u0161n\u00e9 automatizace, ale r\u016fst jejich produktivity se po prvotn\u00edm boomu rychle zastavil a\u00a0znovu se rozjel a\u017e pot\u00e9, co tov\u00e1rn\u00edci pracn\u011b a\u00a0dlouho vytr\u00e9novali zku\u0161en\u00e9 d\u011bln\u00edky, kte\u0159\u00ed na mechanick\u00e9 stavy dohl\u00ed\u017eeli a\u00a0v p\u0159\u00edpad\u011b pot\u0159eby dok\u00e1zali rychle zabr\u00e1nit chyb\u00e1m a\u00a0poruch\u00e1m.<\/p>\n

<\/a><\/p>\n

Na tuto historickou lekci bohu\u017eel z\u00a0hrabivosti, neznalosti \u010di naivn\u00edho opojen\u00ed AI poh\u00e1dkami zapomn\u011bli nejen gener\u00e1lov\u00e9, ale i\u00a0lid\u00e9 ve \u0161pi\u010dk\u00e1ch IT pr\u016fmyslu. Leckoho napadne, \u017ee se na lidsk\u00e9 ov\u011b\u0159ov\u00e1n\u00ed zkr\u00e1tka vyka\u0161le, ale st\u0159et s\u00a0realitou je pak velmi rychl\u00fd a\u00a0tvrd\u00fd. Americk\u00e1 arm\u00e1da si nad civiln\u00edmi ob\u011b\u0165mi bohu\u017eel st\u00e1le m\u016f\u017ee dovolit pokr\u010dit rameny, ale h\u016f\u0159 jsou na tom korporace, kter\u00fdm se spol\u00e9h\u00e1n\u00ed na AI \u201eexperty\u201c nemilosrdn\u011b zakouslo do pen\u011b\u017eenek a\u00a0reputace\u00a0\u2013 Catalini popisuje ned\u00e1vn\u00e9 kauzy<\/a> Amazonu nebo \u010d\u00ednsk\u00fdch IT firem, kter\u00e9 rezignovaly na pe\u010dliv\u00e9 lidsk\u00e9 ov\u011b\u0159ov\u00e1n\u00ed AI softwarov\u00e9ho k\u00f3du, v\u0161e sv\u011b\u0159ily AI agent\u016fm a\u00a0n\u00e1sledn\u011b do\u0161lo k\u00a0z\u00e1va\u017en\u00fdm chyb\u00e1m a\u00a0prolomen\u00ed kyberbezpe\u010dnosti.<\/p>\n

Stroj zalo\u017een\u00fd na generov\u00e1n\u00ed statisticky pravd\u011bpodobn\u00e9ho textu, k\u00f3du \u010di obrazu sice m\u016f\u017ee dos\u00e1hnout netu\u0161en\u00e9 dokonalosti a\u00a0u\u017eite\u010dnosti, ale myslet, vn\u00edmat a\u00a0kriticky hodnotit kv\u016fli sv\u00e9 technologick\u00e9 podstat\u011b nebude nikdy, i\u00a0kdyby Sam Altman vzteky bu\u0161il hlavou o\u00a0futra. Nevyhnuteln\u011b tak bude doch\u00e1zet k\u00a0chyb\u00e1m. Pokud vynech\u00e1me lidsk\u00e9 expertn\u00ed ov\u011b\u0159ov\u00e1n\u00ed, i\u00a0sebemen\u0161\u00ed omyl se bude d\u00e1l n\u00e1sobit. Spousta lid\u00ed to u\u017e odnesla ztr\u00e1tou pen\u011bz, dat \u010di \u017eivota, a\u00a0i kdy\u017e kv\u016fli tomu za\u010d\u00edn\u00e1 AI automatizace v\u00fdrazn\u011b brzdit, nem\u00e1me je\u0161t\u011b vyhr\u00e1no.<\/p>\n

Spoustu lid\u00ed toti\u017e st\u00e1le nenapadne nic lep\u0161\u00edho, ne\u017e z\u00a0neznalosti \u010di ned\u016fv\u011bry k\u00a0expert\u016fm a\u00a0m\u00e9di\u00edm jakoukoliv informaci ov\u011b\u0159ovat dotazem na chatbota, a\u00a0denn\u011b na s\u00edt\u00edch vid\u00edme, k\u00a0jak tragikomick\u00fdm v\u00fdsledk\u016fm to vede.<\/p>\n

Pokud toti\u017e textov\u00e9 gener\u00e1tory alias chatboti n\u011bco um\u00ed, je to (zcela z\u00e1m\u011brn\u011b vytvo\u0159en\u00e1) schopnost p\u0159esv\u011bd\u010div\u00e9ho a\u00a0d\u016fv\u011bryhodn\u00e9ho vystupov\u00e1n\u00ed. \u017d\u00e1dn\u00e1 ot\u00e1zka pro n\u011b nen\u00ed nezodpov\u011bditeln\u00e1. Pokud rekordn\u00ed poj\u00edda\u010d p\u00e1rk\u016f neexistuje, nev\u00e1haj\u00ed si ho vymyslet. A\u00a0kdy\u017e na n\u011b nekriticky spol\u00e9h\u00e1me, dostaneme se do stavu, kter\u00fd jin\u00e1 poveden\u00e1 studie<\/a> popsala term\u00ednem \u201ekognitivn\u00ed kapitulace\u201c (cognitive surrender)\u00a0\u2013 hezky \u010desky zkr\u00e1tka n\u011bkdo spol\u00e9h\u00e1 na moudr\u00e9 rady chatbota tak dlouho, a\u017e z\u00a0toho m\u00edrn\u011b zblbne.<\/p>\n

Pokud jste mana\u017eerem velk\u00e9 korporace nebo gener\u00e1lem a\u00a0nechcete skon\u010dit v\u00a0insolvenci nebo u\u00a0tribun\u00e1lu pro v\u00e1le\u010dn\u00e9 zlo\u010diny, ur\u010dit\u011b je dobr\u00e9 p\u0159e\u010d\u00edst zmi\u0148ovanou studii Christiana Cataliniho. A\u00a0pokud jen pot\u0159ebujete ov\u011b\u0159it, jestli je n\u011bjak\u00e9 video z\u00a0\u00cdr\u00e1nu pravdiv\u00e9, obra\u0165te se na novin\u00e1\u0159e \u010di jin\u00e9 experty, nebo to klidn\u011b proberte s\u00a0kamar\u00e1dy, ale hlavn\u011b nespol\u00e9hejte na \u00fasudek chatbota.<\/p>\n

Jeho odpov\u011b\u010f m\u016f\u017ee b\u00fdt napoprv\u00e9 zcela rozumn\u00e1 a\u00a0pravdiv\u00e1, d\u00e1 to mnohem m\u00e9n\u011b pr\u00e1ce a\u00a0bud\u00ed to dojem spolehlivosti, rychlosti a\u00a0nev\u00eddan\u00e9 efektivity. Jen\u017ee je to za\u010d\u00e1tek nebezpe\u010dn\u00e9ho toboganu omyl\u016f, na kter\u00fd jste pr\u00e1v\u011b nepozorovan\u011b nasko\u010dili.<\/p>\n

Pokud se v\u00e1m uk\u00e1zka z\u00a0newsletteru Pod \u010darou l\u00edbila, p\u0159ihlaste se k\u00a0odb\u011bru<\/a>. Ka\u017edou sobotu ho dostanete p\u0159\u00edmo do va\u0161eho e-mailu, v\u010detn\u011b tip\u016f na dal\u0161\u00ed zaj\u00edmav\u00e9 \u010dten\u00ed z\u00a0\u010desk\u00fdch i\u00a0zahrani\u010dn\u00edch m\u00e9di\u00ed.<\/p>\n","protected":false},"excerpt":{"rendered":"V newsletteru Pod \u010darou popisuje ka\u017edou sobotu Matou\u0161 Hrdina spole\u010densk\u00e9 trendy, kter\u00e9 sice v\u00edd\u00e1me v\u0161ude kolem sebe, ale…\n","protected":false},"author":2,"featured_media":24384,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[11],"tags":[3120,2233,8486,3660,12140,7825,920,465,15567,3106,1555,2236,54,126,53,52],"class_list":{"0":"post-38469","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-svet","8":"tag-chatgpt","9":"tag-claude","10":"tag-deepfake","11":"tag-dezinformace","12":"tag-fake-news","13":"tag-konflikt-usa-izrael-iran","14":"tag-kyberneticka-bezpecnost","15":"tag-openai","16":"tag-palantir","17":"tag-pod-carou","18":"tag-propaganda","19":"tag-robotizace","20":"tag-svet","21":"tag-umela-inteligence-ai","22":"tag-world","23":"tag-world-news"},"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@cz\/116307831323356444","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/posts\/38469","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/comments?post=38469"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/posts\/38469\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/media\/24384"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/media?parent=38469"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/categories?post=38469"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/tags?post=38469"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}