{"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\/cs\/blog\/news\/o-predvzyatosti-iskusstvennogo-intellekta","title":{"rendered":"O zaujatosti um\u011bl\u00e9 inteligence","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"<p><img decoding=\"async\" alt=\"O zaujatosti um\u011bl\u00e9 inteligence\" src=\"\/wp-content\/uploads\/2019\/04\/05a731677e1231acceff953db8e1e26a.jpg\" style=\"display:block;margin: 0 auto;\" \/><\/p>\n<p><\/p>\n<h2 id=\"tldr\">shrnut\u00ed:<\/h2>\n<p><\/p>\n<ul>\n<li>Strojov\u00e9 u\u010den\u00ed hled\u00e1 vzory v datech. Av\u0161ak um\u011bl\u00e1 inteligence m\u016f\u017ee b\u00fdt \"p\u0159edpojat\u00e1\" \u2013 to znamen\u00e1, \u017ee m\u016f\u017ee nach\u00e1zet nespr\u00e1vn\u00e9 vzory. Nap\u0159\u00edklad syst\u00e9m detekce rakoviny k\u016f\u017ee na fotografii m\u016f\u017ee v\u011bnovat zvl\u00e1\u0161tn\u00ed pozornost sn\u00edmk\u016fm po\u0159\u00edzen\u00fdm v l\u00e9ka\u0159sk\u00e9 ordinaci. Strojov\u00e9 u\u010den\u00ed nedok\u00e1\u017ee <em>ch\u00e1pat<\/em>: jeho algoritmy pouze identifikuj\u00ed vzorce v \u010d\u00edslech, a pokud data nejsou reprezentativn\u00ed, takov\u00fd bude i v\u00fdsledek jejich zpracov\u00e1n\u00ed. Odhalit takov\u00e9 chyby m\u016f\u017ee b\u00fdt obt\u00ed\u017en\u00e9 kv\u016fli samotn\u00e9 mechanice strojov\u00e9ho u\u010den\u00ed. <noindex><a rel=\"nofollow\" name=\"habracut\"><\/a><\/noindex><\/li>\n<li>Nejz\u0159ejm\u011bj\u0161\u00edm a nejv\u00edce znepokojuj\u00edc\u00edm probl\u00e9mem je rozmanitost lid\u00ed. Existuje mnoho d\u016fvod\u016f, pro\u010d mohou \u00fadaje o lidech ztratit objektivitu u\u017e v etap\u011b sb\u011bru. Ale nemyslete si, \u017ee tento probl\u00e9m se t\u00fdk\u00e1 pouze lid\u00ed: podobn\u00e9 pot\u00ed\u017ee se objevuj\u00ed p\u0159i pokusu o detekci povodn\u011b ve skladu nebo poru\u0161en\u00e9 plynov\u00e9 turb\u00edny. N\u011bkter\u00e9 syst\u00e9my mohou m\u00edt p\u0159edsudky v\u016f\u010di barv\u011b pleti, jin\u00e9 budou p\u0159edpojat\u011b ch\u00e1pat senzory Siemens. <\/li>\n<li>Takov\u00e9 probl\u00e9my nejsou pro strojov\u00e9 u\u010den\u00ed nov\u00e9 a nejsou v\u00fdsadou jen jeho. Nespr\u00e1vn\u00e9 p\u0159edpoklady se d\u011blaj\u00ed v jak\u00fdchkoli slo\u017eit\u00fdch struktur\u00e1ch a pochopit, pro\u010d bylo p\u0159ijato ur\u010dit\u00e9 rozhodnut\u00ed, je v\u017edy obt\u00ed\u017en\u00e9. S t\u00edm je t\u0159eba bojovat komplexn\u011b: vytv\u00e1\u0159et n\u00e1stroje a procesy pro ov\u011b\u0159ov\u00e1n\u00ed \u2014 a vzd\u011bl\u00e1vat u\u017eivatele, aby neslep\u011b n\u00e1sledovali doporu\u010den\u00ed AI. Strojov\u00e9 u\u010den\u00ed skute\u010dn\u011b d\u011bl\u00e1 n\u011bkter\u00e9 v\u011bci mnohem l\u00e9pe ne\u017e my \u2014 ale psi, nap\u0159\u00edklad, jsou mnohem efektivn\u011bj\u0161\u00ed ne\u017e lid\u00e9 p\u0159i detekci drog, co\u017e ur\u010dit\u011b nen\u00ed d\u016fvod je povol\u00e1vat jako sv\u011bdky a vyn\u00e1\u0161et rozsudky na z\u00e1klad\u011b jejich v\u00fdpov\u011bd\u00ed. A psi jsou, mimochodem, mnohem chyt\u0159ej\u0161\u00ed ne\u017e jak\u00fdkoli syst\u00e9m strojov\u00e9ho u\u010den\u00ed. <\/li>\n<\/ul>\n<p>Strojov\u00e9 u\u010den\u00ed je dnes jedn\u00edm z nejd\u016fle\u017eit\u011bj\u0161\u00edch z\u00e1kladn\u00edch technologick\u00fdch trend\u016f. Je to jeden z hlavn\u00edch zp\u016fsob\u016f, jak\u00fdm technologie zm\u011bn\u00ed okoln\u00ed sv\u011bt v p\u0159\u00ed\u0161t\u00edm desetilet\u00ed. N\u011bkter\u00e9 aspekty t\u011bchto zm\u011bn vyvol\u00e1vaj\u00ed obavy. Nap\u0159\u00edklad potenci\u00e1ln\u00ed dopad strojov\u00e9ho u\u010den\u00ed na pracovn\u00ed trh nebo jeho vyu\u017e\u00edv\u00e1n\u00ed pro neetick\u00e9 \u00fa\u010dely (nap\u0159\u00edklad autorit\u00e1\u0159sk\u00fdmi re\u017eimy). Existuje je\u0161t\u011b jeden probl\u00e9m, kter\u00e9mu se tento p\u0159\u00edsp\u011bvek v\u011bnuje: <strong>p\u0159edpojatost um\u011bl\u00e9 inteligence<\/strong>. <\/p>\n<p><\/p>\n<p>Tohle je slo\u017eit\u00fd p\u0159\u00edb\u011bh. <\/p>\n<p><\/p>\n<p><img decoding=\"async\" alt=\"O zaujatosti um\u011bl\u00e9 inteligence\" src=\"\/wp-content\/uploads\/2019\/04\/d692fbfff47614758ab1d8e1516b2742.jpg\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<em>Um\u011bl\u00e1 inteligence od Googlu um\u00ed naj\u00edt ko\u010dky. Tato zpr\u00e1va z roku 2012 byla tehdy n\u011b\u010d\u00edm v\u00fdjime\u010dn\u00fdm.<\/em><\/p>\n<p><\/p>\n<h2 id=\"chto-takoe-predvzyatost-ii\">Co je to \"p\u0159edpojatost AI\"?<\/h2>\n<p><\/p>\n<blockquote><p><em>\"Surov\u00e1 data\" jsou z\u00e1rove\u0148 oxym\u00f3ron a \u0161patn\u00fd n\u00e1pad; data je t\u0159eba dob\u0159e a pe\u010dliv\u011b p\u0159ipravit.<\/em> -Jeffrey Boker <\/p><\/blockquote>\n<p>A\u017e do roku 2013 bylo pot\u0159eba popisovat logick\u00e9 kroky pro vytvo\u0159en\u00ed syst\u00e9mu, kter\u00fd by nap\u0159\u00edklad rozpozn\u00e1val ko\u010dky na fotografi\u00edch. Jak naj\u00edt na obr\u00e1zku rohy, rozpoznat o\u010di, analyzovat textury na p\u0159\u00edtomnost srsti, spo\u010d\u00edtat nohy a tak d\u00e1le. Pak shrom\u00e1\u017edit v\u0161echny komponenty \u2014 a zjistit, \u017ee to vlastn\u011b nefunguje. P\u0159ipom\u00edn\u00e1 to mechanick\u00e9ho kon\u011b \u2014 teoreticky ho lze vyrobit, ale prakticky je p\u0159\u00edli\u0161 slo\u017eit\u00e9 jeho pops\u00e1n\u00ed. Na v\u00fdstupu m\u00e1te stovky (nebo dokonce tis\u00edce) ru\u010dn\u011b psan\u00fdch pravidel. A \u017e\u00e1dn\u00fd funguj\u00edc\u00ed model. <\/p>\n<p><\/p>\n<p>S n\u00e1stupem strojov\u00e9ho u\u010den\u00ed jsme p\u0159estali pou\u017e\u00edvat \"ru\u010dn\u00ed\" pravidla pro rozpozn\u00e1v\u00e1n\u00ed ur\u010dit\u00e9ho objektu. M\u00edsto toho vezmeme tis\u00edc vzork\u016f \"toho\", X, tis\u00edc vzork\u016f \"jin\u00e9ho\", Y, a p\u0159im\u011bjeme po\u010d\u00edta\u010d, aby na z\u00e1klad\u011b jejich statistick\u00e9 anal\u00fdzy sestavil model. Pot\u00e9 d\u00e1me tomuto modelu n\u011bjak\u00fd p\u0159\u00edklad dat a on s ur\u010ditou p\u0159esnost\u00ed ur\u010d\u00ed, zda odpov\u00edd\u00e1 jednomu z t\u011bchto soubor\u016f. Strojov\u00e9 u\u010den\u00ed generuje model na z\u00e1klad\u011b dat, ne prost\u0159ednictv\u00edm \u010dlov\u011bka, kter\u00fd ho p\u00ed\u0161e. V\u00fdsledky jsou ohromuj\u00edc\u00ed, zejm\u00e9na v oblasti rozpozn\u00e1v\u00e1n\u00ed obraz\u016f a vzor\u016f, a proto se cel\u00e1 tech industrie nyn\u00ed p\u0159ech\u00e1z\u00ed na strojov\u00e9 u\u010den\u00ed (ML). <\/p>\n<p><\/p>\n<p>Ale nen\u00ed to tak jednoduch\u00e9. V re\u00e1ln\u00e9m sv\u011bt\u011b va\u0161e tis\u00edce p\u0159\u00edklad\u016f X nebo Y obsahuj\u00ed tak\u00e9 A, B, J, L, O, R a dokonce L. Mohou b\u00fdt nepravideln\u011b rozlo\u017eeny a n\u011bkter\u00e9 z nich se mohou vyskytovat tak \u010dasto, \u017ee syst\u00e9m na n\u011b zam\u011b\u0159\u00ed v\u00edce pozornosti ne\u017e na objekty, kter\u00e9 v\u00e1s zaj\u00edmaj\u00ed.<\/p>\n<p><\/p>\n<p>Co to znamen\u00e1 v praxi? M\u016fj obl\u00edben\u00fd p\u0159\u00edklad je, kdy\u017e syst\u00e9my rozpozn\u00e1v\u00e1n\u00ed obraz\u016f <noindex><a rel=\"nofollow\" href=\"http:\/\/aiweirdness.com\/post\/171451900302\/do-neural-nets-dream-of-electric-sheep\">d\u00edvaj\u00ed se na travnat\u00fd kopec a \u0159\u00edkaj\u00ed: \"ovce\"<\/a><\/noindex>. Je jasn\u00e9, pro\u010d: v\u011bt\u0161ina fotografi\u00ed-p\u0159\u00edklad\u016f \"ovce\" byla po\u0159\u00edzena na louk\u00e1ch, kde \u017eij\u00ed, a na t\u011bchto obr\u00e1zc\u00edch tr\u00e1va zab\u00edr\u00e1 mnohem v\u00edce m\u00edsta ne\u017e mal\u00e9 b\u00edl\u00e9 chlupat\u00e9 bytosti, a pr\u00e1v\u011b tr\u00e1vu syst\u00e9my pova\u017euj\u00ed za nejd\u016fle\u017eit\u011bj\u0161\u00ed. <\/p>\n<p><\/p>\n<p>Existuj\u00ed p\u0159\u00edklady a v\u00e1\u017en\u011bj\u0161\u00ed. Z ned\u00e1vna \u2014 jeden <noindex><a rel=\"nofollow\" href=\"https:\/\/www.medpagetoday.com\/dermatology\/generaldermatology\/70513\">projekt<\/a><\/noindex> o detekci rakoviny k\u016f\u017ee na fotografi\u00edch. Uk\u00e1zalo se, \u017ee dermatologov\u00e9 \u010dasto fotografuj\u00ed prav\u00edtko spolu s projevem rakoviny k\u016f\u017ee, aby zdokumentovali velikost \u00fatvar\u016f. U p\u0159\u00edklad\u016f fotografi\u00ed zdrav\u00e9 k\u016f\u017ee prav\u00edtka nejsou. Pro syst\u00e9m AI se tato prav\u00edtka (p\u0159esn\u011bji, pixely, kter\u00e9 my ozna\u010dujeme jako \u201eprav\u00edtko\u201c) stala jedn\u00edm z rozd\u00edl\u016f mezi sadami p\u0159\u00edklad\u016f, a n\u011bkdy d\u016fle\u017eit\u011bj\u0161\u00edmi ne\u017e mal\u00e9 vyr\u00e1\u017eky na k\u016f\u017ei. Tak syst\u00e9m, vytvo\u0159en\u00fd pro rozpozn\u00e1v\u00e1n\u00ed rakoviny k\u016f\u017ee, ob\u010das nam\u00edsto toho rozpozn\u00e1val prav\u00edtka. <\/p>\n<p><\/p>\n<p>Kl\u00ed\u010dov\u00fd bod zde je to, \u017ee syst\u00e9m nem\u00e1 semantick\u00e9 porozum\u011bn\u00ed tomu, na co se d\u00edv\u00e1. My se d\u00edv\u00e1me na sadu pixel\u016f a vid\u00edme v nich ovci, k\u016f\u017ei nebo prav\u00edtka, zat\u00edmco syst\u00e9m vid\u00ed pouze \u010d\u00edseln\u00fd \u0159et\u011bzec. Nevid\u00ed trojrozm\u011brn\u00fd prostor, nevid\u00ed ani objekty, ani textury, ani ovce. Vid\u00ed pouze vzory v datech. <\/p>\n<p><\/p>\n<p>Slo\u017eitost diagnostiky takov\u00fdch probl\u00e9m\u016f spo\u010d\u00edv\u00e1 v tom, \u017ee neuronov\u00e1 s\u00ed\u0165 (model generovan\u00fd va\u0161\u00edm syst\u00e9mem strojov\u00e9ho u\u010den\u00ed) se skl\u00e1d\u00e1 z tis\u00edc\u016f a set tis\u00edc\u016f uzl\u016f. Neexistuje jednoduch\u00fd zp\u016fsob, jak nahl\u00e9dnout do modelu a vid\u011bt, jak p\u0159ij\u00edm\u00e1 rozhodnut\u00ed. M\u00edt takov\u00fd zp\u016fsob by znamenalo, \u017ee proces je dostate\u010dn\u011b jednoduch\u00fd na to, aby bylo mo\u017en\u00e9 v\u0161echna pravidla popsat ru\u010dn\u011b, bez pou\u017eit\u00ed strojov\u00e9ho u\u010den\u00ed. Lid\u00e9 maj\u00ed obavy, \u017ee strojov\u00e9 u\u010den\u00ed se stalo jak\u00fdmsi \u201e\u010dern\u00fdm boxem\u201c. (Za chv\u00edli vysv\u011btl\u00edm, pro\u010d je to p\u0159irovn\u00e1n\u00ed p\u0159ehnan\u00e9.) <\/p>\n<p><\/p>\n<p>To, ve zjednodu\u0161en\u00e9 podob\u011b, je probl\u00e9m zaujatosti um\u011bl\u00e9 inteligence nebo strojov\u00e9ho u\u010den\u00ed: syst\u00e9m pro nalezen\u00ed vzor\u016f v datech m\u016f\u017ee nal\u00e9zt myln\u00e9 vzory a vy si toho nemus\u00edte v\u0161imnout. To je z\u00e1kladn\u00ed charakteristika technologie a je to z\u0159ejm\u00e9 v\u0161em, kdo s n\u00ed pracuj\u00ed ve v\u011bdeck\u00fdch kruz\u00edch a ve velk\u00fdch technologick\u00fdch spole\u010dnostech. Ale jej\u00ed n\u00e1sledky jsou slo\u017eit\u00e9 a na\u0161e mo\u017en\u00e1 \u0159e\u0161en\u00ed t\u011bchto n\u00e1sledk\u016f tak\u00e9. <\/p>\n<p><\/p>\n<p>Nejprve si promluvme o d\u016fsledc\u00edch. <\/p>\n<p><\/p>\n<p><img decoding=\"async\" alt=\"O zaujatosti um\u011bl\u00e9 inteligence\" src=\"\/wp-content\/uploads\/2019\/04\/43748412bd82ae53a9c2e47afec9c655.jpg\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<em>AI n\u00e1m m\u016f\u017ee implicitn\u011b vyb\u00edrat pro r\u016fzn\u00e9 kategorie lid\u00ed na z\u00e1klad\u011b velk\u00e9ho mno\u017estv\u00ed nepov\u0161imnuteln\u00fdch sign\u00e1l\u016f<\/em><\/p>\n<p><\/p>\n<h2 id=\"scenarii-predvzyatosti-ii\">Sc\u00e9n\u00e1\u0159e zaujatosti AI<\/h2>\n<p><\/p>\n<p>Nejsamoz\u0159ejm\u011bj\u0161\u00ed a nejd\u011bsiv\u011bj\u0161\u00ed je, \u017ee tento probl\u00e9m se m\u016f\u017ee projevit, kdy\u017e jde o lidskou rozmanitost. Ned\u00e1vno <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\">se \u0161\u00ed\u0159il zv\u011bst<\/a><\/noindex>, \u017ee Amazon se pokusil vybudovat syst\u00e9m strojov\u00e9ho u\u010den\u00ed pro prim\u00e1rn\u00ed v\u00fdb\u011br uchaze\u010d\u016f o zam\u011bstn\u00e1n\u00ed. Proto\u017ee mezi zam\u011bstnanci Amazonu je v\u00edce mu\u017e\u016f, p\u0159\u00edklady \"\u00fasp\u011b\u0161n\u00e9ho n\u00e1boru\" byly tak\u00e9 \u010dast\u011bji mu\u017esk\u00e9ho pohlav\u00ed, a v selekci \u017eivotopis\u016f, kterou syst\u00e9m navrhl, bylo v\u00edce mu\u017e\u016f. Amazon si toho v\u0161iml a syst\u00e9m neuv\u00e1d\u011bl do produkce. <\/p>\n<p><\/p>\n<p>Nejd\u016fle\u017eit\u011bj\u0161\u00ed na tomto p\u0159\u00edkladu je, \u017ee syst\u00e9m \u00fadajn\u011b up\u0159ednost\u0148oval uchaze\u010de mu\u017esk\u00e9ho pohlav\u00ed, a\u010dkoliv pohlav\u00ed nebylo v \u017eivotopise uvedeno.<strong> Syst\u00e9m vid\u011bl jin\u00e9 vzory v p\u0159\u00edkladech \"\u00fasp\u011b\u0161n\u00e9ho n\u00e1boru\": nap\u0159\u00edklad \u017eeny mohou pou\u017e\u00edvat specifick\u00e1 slova k popisu sv\u00fdch \u00fasp\u011bch\u016f, nebo m\u00edt specifick\u00e1 kon\u00ed\u010dky. Samoz\u0159ejm\u011b, \u017ee syst\u00e9m nev\u011bd\u011bl, co je \"hokej\", kdo jsou \"lid\u00e9\", ani co je \"\u00fasp\u011bch\" \u2014 jen prov\u00e1d\u011bl statistickou anal\u00fdzu textu.<\/strong> Ale vzory, kter\u00e9 vid\u011bl, by pravd\u011bpodobn\u011b z\u016fstaly nepozorov\u00e1ny \u010dlov\u011bkem, a n\u011bkter\u00e9 z nich (nap\u0159\u00edklad to, \u017ee lid\u00e9 r\u016fzn\u00fdch pohlav\u00ed r\u016fzn\u011b popisuj\u00ed \u00fasp\u011bch) bychom pravd\u011bpodobn\u011b i p\u0159i jejich pohledu t\u011b\u017eko post\u0159ehli. <\/p>\n<p><\/p>\n<p>D\u00e1l \u2014 h\u016f\u0159. Syst\u00e9m strojov\u00e9ho u\u010den\u00ed, kter\u00fd velmi dob\u0159e identifikuje rakovinu na sv\u011btl\u00e9 pleti, m\u016f\u017ee hor\u0161\u00ed pracovat s tmavou plet\u00ed, nebo naopak. Nen\u00ed to nutn\u011b kv\u016fli zaujatosti, ale proto, \u017ee pravd\u011bpodobn\u011b pot\u0159ebujete vybudovat pro jinou barvu pleti samostatn\u00fd model, vybrat jin\u00e9 charakteristiky.<strong> Syst\u00e9my strojov\u00e9ho u\u010den\u00ed nejsou zam\u011bniteln\u00e9 ani v tak \u00fazk\u00e9 oblasti, jako je rozpozn\u00e1v\u00e1n\u00ed obraz\u016f.<\/strong> Mus\u00edte syst\u00e9m nastavit, n\u011bkdy prost\u011b pomoc\u00ed pokus\u016f a omyl\u016f, abyste dob\u0159e post\u0159ehli rysy v datech, kter\u00e1 v\u00e1s zaj\u00edmaj\u00ed, dokud nedos\u00e1hnete po\u017eadovan\u00e9 p\u0159esnosti. Ale mo\u017en\u00e1 si nev\u0161imnete, \u017ee syst\u00e9m je v 98 % p\u0159\u00edpad\u016f p\u0159esn\u00fd p\u0159i pr\u00e1ci s jednou skupinou a pouze v 91 % (i kdy\u017e je to p\u0159esn\u011bj\u0161\u00ed ne\u017e anal\u00fdza proveden\u00e1 \u010dlov\u011bkem) \u2014 s jinou. <\/p>\n<p><\/p>\n<p>Doposud jsem pou\u017e\u00edval p\u0159edev\u0161\u00edm p\u0159\u00edklady t\u00fdkaj\u00edc\u00ed se lid\u00ed a jejich charakteristik. T\u00e9ma, na kter\u00e9 se tato diskuse v\u011bt\u0161inou zam\u011b\u0159uje. Ale je d\u016fle\u017eit\u00e9 pochopit, \u017ee zaujatost v\u016f\u010di lidem je pouze \u010d\u00e1st\u00ed probl\u00e9mu. Budeme pou\u017e\u00edvat strojov\u00e9 u\u010den\u00ed pro mnoho v\u011bc\u00ed, a chyba v\u00fdb\u011bru bude relevantn\u00ed pro v\u0161echny z nich. Na druhou stranu, pokud pracujete s lidmi, m\u016f\u017ee b\u00fdt zaujatost dat spojena nikoli s nimi. <\/p>\n<p><\/p>\n<p>Abychom to pochopili, vra\u0165me se k p\u0159\u00edkladu s rakovinou k\u016f\u017ee a zva\u017eme t\u0159i hypotetick\u00e9 mo\u017enosti selh\u00e1n\u00ed syst\u00e9mu. <\/p>\n<p><\/p>\n<ol>\n<li>Nepravideln\u00e1 distribuce lid\u00ed: nevyv\u00e1\u017een\u00fd po\u010det fotografi\u00ed pleti r\u016fzn\u00fdch t\u00f3n\u016f, co\u017e vede k fale\u0161n\u011b pozitivn\u00edm nebo fale\u0161n\u011b negativn\u00edm v\u00fdsledk\u016fm spojen\u00fdm s pigmentac\u00ed. <\/li>\n<li>Data, na kter\u00fdch se syst\u00e9m tr\u00e9nuje, obsahuj\u00ed \u010dasto se vyskytuj\u00edc\u00ed a nepravideln\u011b distribuovanou charakteristiku, kter\u00e1 nesouvis\u00ed s lidmi a nem\u00e1 diagnostickou hodnotu: prav\u00edtko na fotografi\u00edch projev\u016f rakoviny k\u016f\u017ee nebo tr\u00e1vu na fotografi\u00edch ovc\u00ed. V takov\u00e9m p\u0159\u00edpad\u011b se v\u00fdsledek li\u0161\u00ed, pokud syst\u00e9m na obr\u00e1zku najde pixely n\u011b\u010deho, co lidsk\u00e9 oko ur\u010d\u00ed jako \"prav\u00edtko\". <\/li>\n<li>Data obsahuj\u00ed ciz\u00ed charakteristiku, kterou \u010dlov\u011bk nem\u016f\u017ee vid\u011bt, i kdy\u017e ji bude hledat. <\/li>\n<\/ol>\n<p><\/p>\n<p>Co to znamen\u00e1? Apriori v\u00edme, \u017ee data mohou r\u016fzn\u011b reprezentovat r\u016fzn\u00e9 skupiny lid\u00ed, a minim\u00e1ln\u011b m\u016f\u017eeme napl\u00e1novat hled\u00e1n\u00ed podobn\u00fdch v\u00fdjimek. Jin\u00fdmi slovy, existuje spousta soci\u00e1ln\u00edch d\u016fvod\u016f p\u0159edpokl\u00e1dat, \u017ee data o skupin\u00e1ch lid\u00ed ji\u017e obsahuj\u00ed jist\u00e9 p\u0159edsudky. Kdy\u017e se pod\u00edv\u00e1me na foto s prav\u00edtkem, uvid\u00edme to prav\u00edtko \u2013 jednodu\u0161e jsme to d\u0159\u00edve ignorovali, v\u011bdouce, \u017ee to nen\u00ed d\u016fle\u017eit\u00e9, a zapomn\u011bli, \u017ee syst\u00e9mu nic nen\u00ed zn\u00e1mo.<\/p>\n<p><\/p>\n<p>Ale co kdy\u017e jsou v\u0161echny va\u0161e fotografie nezdrav\u00e9 k\u016f\u017ee po\u0159\u00edzen\u00e9 v kancel\u00e1\u0159i, kde se pou\u017e\u00edvaj\u00ed \u017e\u00e1rovky, a zdrav\u00e9 \u2013 p\u0159i fluorescen\u010dn\u00edm sv\u011btle? Co kdy\u017e, kdy\u017e jste dokon\u010dili sn\u00edm\u00e1n\u00ed zdrav\u00e9 k\u016f\u017ee, jste p\u0159ed po\u0159\u00edzen\u00edm nezdrav\u00e9 aktualizovali opera\u010dn\u00ed syst\u00e9m na telefonu a Apple nebo Google m\u00edrn\u011b zm\u011bnili algoritmus potla\u010den\u00ed \u0161umu? \u010clov\u011bk si toho nev\u0161imne, jak moc by se to sna\u017eil hledat. Ale syst\u00e9m strojov\u00e9ho u\u010den\u00ed to okam\u017eit\u011b uvid\u00ed a vyu\u017eije. Nic nev\u00ed. <\/p>\n<p><\/p>\n<p>Kdy\u017e jsme mluvili o fale\u0161n\u00fdch korelac\u00edch, m\u016f\u017ee se tak\u00e9 st\u00e1t, \u017ee data jsou p\u0159esn\u00e1 a v\u00fdsledky spr\u00e1vn\u00e9, ale z etick\u00fdch, pr\u00e1vn\u00edch nebo mana\u017eersk\u00fdch d\u016fvod\u016f je nechcete pou\u017e\u00edt. V n\u011bkter\u00fdch jurisdikc\u00edch nap\u0159\u00edklad nen\u00ed mo\u017en\u00e9 poskytovat \u017een\u00e1m slevu na poji\u0161t\u011bn\u00ed, i kdy\u017e \u017eeny mohou \u0159\u00eddit bezpe\u010dn\u011bji. M\u016f\u017eeme si snadno p\u0159edstavit syst\u00e9m, kter\u00fd p\u0159i anal\u00fdze historick\u00fdch dat p\u0159i\u0159ad\u00ed \u017eensk\u00fdm jm\u00e9n\u016fm ni\u017e\u0161\u00ed rizikov\u00fd koeficient. Dob\u0159e, odstran\u00edme jm\u00e9na z v\u00fdb\u011bru. Ale vzpome\u0148te si na p\u0159\u00edklad s Amazonem: syst\u00e9m m\u016f\u017ee ur\u010den\u00ed pohlav\u00ed na z\u00e1klad\u011b jin\u00fdch faktor\u016f (i kdy\u017e nev\u00ed, co je pohlav\u00ed, ani co je auto), a vy si toho nev\u0161imnete, dokud regul\u00e1tor zp\u011btn\u011b neanalyzuje va\u0161e nab\u00edzen\u00e9 tarify a neud\u011bl\u00ed v\u00e1m pokutu. <\/p>\n<p><\/p>\n<p>Nakonec se \u010dasto p\u0159edpokl\u00e1d\u00e1, \u017ee tyto syst\u00e9my budeme pou\u017e\u00edvat pouze pro projekty souvisej\u00edc\u00ed s lidmi a soci\u00e1ln\u00edmi interakcemi. To ale nen\u00ed pravda. Pokud vyr\u00e1b\u00edte plynov\u00e9 turb\u00edny, ur\u010dit\u011b budete cht\u00edt pou\u017e\u00edt strojov\u00e9 u\u010den\u00ed na telemetrii p\u0159en\u00e1\u0161enou des\u00edtkami nebo stovkami senzor\u016f na va\u0161em produktu (audio-, video-, teplotn\u00ed a dal\u0161\u00ed senzory generuj\u00ed data, kter\u00e1 lze velmi snadno p\u0159izp\u016fsobit k vytvo\u0159en\u00ed modelu strojov\u00e9ho u\u010den\u00ed). Hypoteticky m\u016f\u017eete \u0159\u00edci: \u201eTady jsou data o tis\u00edci porouchan\u00fdch turb\u00edn\u00e1ch, z\u00edskan\u00e1 p\u0159ed jejich poruchou, a tady jsou data z tis\u00edce turb\u00edn, kter\u00e9 neporouchaly. Vytvo\u0159te model, abyste \u0159ekli, jak\u00fd je mezi nimi rozd\u00edl.\u201c A te\u010f si p\u0159edstavte, \u017ee senzory Siemens jsou na 75 % \u0161patn\u00fdch turb\u00edn a pouze na 12 % dobr\u00fdch (p\u0159itom neexistuje \u017e\u00e1dn\u00e1 vazba na poruchy). Syst\u00e9m vytvo\u0159\u00ed model, aby na\u0161el turb\u00edny se senzory Siemens. Ups!<\/p>\n<p><\/p>\n<p><img decoding=\"async\" alt=\"O zaujatosti um\u011bl\u00e9 inteligence\" src=\"\/wp-content\/uploads\/2019\/04\/e8097e4913756fdf04826a6628bc6836.jpg\" style=\"display:block;margin: 0 auto;\" \/><br \/>\nObr\u00e1zek \u2014 Moritz Hardt, UC Berkeley<\/p>\n<p><\/p>\n<h2 id=\"upravlenie-predvzyatostyu-ii\">\u0158\u00edzen\u00ed p\u0159edpojatosti AI<\/h2>\n<p><\/p>\n<p>Co s t\u00edm m\u016f\u017eeme ud\u011blat? M\u016f\u017eeme se na probl\u00e9m pod\u00edvat ze t\u0159\u00ed stran:<\/p>\n<p><\/p>\n<ol>\n<li>Metodologick\u00e1 p\u0159\u00edsnost p\u0159i sb\u011bru a spr\u00e1v\u011b dat pro tr\u00e9nink syst\u00e9mu.<\/li>\n<li>Technick\u00e9 n\u00e1stroje pro anal\u00fdzu a diagnostiku chov\u00e1n\u00ed modelu.<\/li>\n<li>\u0160kolen\u00ed, vzd\u011bl\u00e1v\u00e1n\u00ed a opatrnost p\u0159i implementaci strojov\u00e9ho u\u010den\u00ed do produkt\u016f.<\/li>\n<\/ol>\n<p><\/p>\n<p>V knize \u201eM\u011b\u0161\u0165an v \u0161lecht\u011b\u201c od Moli\u00e8ra je vtip: jednomu mu\u017ei \u0159ekli, \u017ee literatura se d\u011bl\u00ed na pr\u00f3zu a poezii, a on s nad\u0161en\u00edm zji\u0161\u0165uje, \u017ee cel\u00fd \u017eivot mluvil pr\u00f3zou, ani\u017e by to v\u011bd\u011bl. Pravd\u011bpodobn\u011b se statistici dnes c\u00edt\u00ed n\u011bjak podobn\u011b: ani\u017e by si toho v\u0161imli, zasv\u011btili sv\u00e9 kari\u00e9ry um\u011bl\u00e9 inteligenci a chyb\u011b ve vzorkov\u00e1n\u00ed. Hled\u00e1n\u00ed chyby ve vzorkov\u00e1n\u00ed a tr\u00e1pen\u00ed se kv\u016fli n\u00ed \u2014 to nen\u00ed nov\u00fd probl\u00e9m, mus\u00edme k jeho \u0159e\u0161en\u00ed p\u0159istoupit systematicky. Jak ji\u017e bylo zm\u00edn\u011bno, v n\u011bkter\u00fdch p\u0159\u00edpadech je skute\u010dn\u011b jednodu\u0161\u0161\u00ed zkoumat probl\u00e9my spojen\u00e9 s daty o lidech. A priori p\u0159edpokl\u00e1d\u00e1me, \u017ee m\u016f\u017eeme m\u00edt p\u0159edsudky v\u016f\u010di r\u016fzn\u00fdm skupin\u00e1m lid\u00ed, ale p\u0159edsudek v\u016f\u010di senzor\u016fm Siemens si t\u011b\u017eko dok\u00e1\u017eeme p\u0159edstavit.<\/p>\n<p><\/p>\n<p>Nov\u00e9 na tom v\u0161em je, \u017ee lid\u00e9 se ji\u017e nezab\u00fdvaj\u00ed statistickou anal\u00fdzou p\u0159\u00edmo. Tu vykon\u00e1vaj\u00ed stroje, kter\u00e9 vytv\u00e1\u0159ej\u00ed velk\u00e9 slo\u017eit\u00e9 modely, kter\u00e9 jsou obt\u00ed\u017en\u00e9 k pochopen\u00ed. Ot\u00e1zka transparentnosti je jedn\u00edm z hlavn\u00edch aspekt\u016f probl\u00e9mu zaujatosti. Boj\u00edme se, \u017ee syst\u00e9m nen\u00ed jen zaujat\u00fd, ale \u017ee nen\u00ed \u017e\u00e1dn\u00e1 mo\u017enost odhalit jeho zaujatost, a \u017ee strojov\u00e9 u\u010den\u00ed se t\u00edmto li\u0161\u00ed od jin\u00fdch forem automatizace, kter\u00e9 se, jak se p\u0159edpokl\u00e1d\u00e1, skl\u00e1daj\u00ed z jasn\u00fdch logick\u00fdch krok\u016f, kter\u00e9 lze ov\u011b\u0159it. <\/p>\n<p><\/p>\n<p>Zde jsou dva probl\u00e9my. Mo\u017en\u00e1 bychom p\u0159ece jen mohli prov\u00e1d\u011bt ur\u010dit\u00fd audit syst\u00e9m\u016f strojov\u00e9ho u\u010den\u00ed. A audit jak\u00e9hokoli jin\u00e9ho syst\u00e9mu ve skute\u010dnosti nen\u00ed o nic snaz\u0161\u00ed.<\/p>\n<p><\/p>\n<p>Za prv\u00e9, jedn\u00edm z sm\u011br\u016f sou\u010dasn\u00e9ho v\u00fdzkumu v oblasti strojov\u00e9ho u\u010den\u00ed je hled\u00e1n\u00ed metod, jak identifikovat d\u016fle\u017eitou funk\u010dnost syst\u00e9m\u016f strojov\u00e9ho u\u010den\u00ed. Strojov\u00e9 u\u010den\u00ed (ve sv\u00e9m sou\u010dasn\u00e9m stavu) je zcela nov\u00e1 v\u011bdeck\u00e1 oblast, kter\u00e1 se rychle m\u011bn\u00ed, tak\u017ee nem\u00e1 smysl myslet si, \u017ee v\u011bci, kter\u00e9 dnes nejsou mo\u017en\u00e9, se nemohou brzy st\u00e1t docela re\u00e1ln\u00fdmi. Projekt <noindex><a rel=\"nofollow\" href=\"https:\/\/openai.com\/\">OpenAI<\/a><\/noindex> je zaj\u00edmav\u00fdm p\u0159\u00edkladem. <\/p>\n<p><\/p>\n<p>Za prv\u00e9, my\u0161lenka, \u017ee lze zkoumat a ch\u00e1pat proces rozhodov\u00e1n\u00ed v existuj\u00edc\u00edch syst\u00e9mech nebo organizac\u00edch, je teoreticky zaj\u00edmav\u00e1, ale v praxi to u\u017e tak skv\u011bl\u00e9 nen\u00ed. Pochopit, jak se rozhodnut\u00ed p\u0159ij\u00edmaj\u00ed ve velk\u00e9 organizaci, nen\u00ed v\u016fbec snadn\u00e9. I kdy\u017e existuje form\u00e1ln\u00ed proces rozhodov\u00e1n\u00ed, neodr\u00e1\u017e\u00ed to skute\u010dn\u00fd zp\u016fsob, jak\u00fdm lid\u00e9 interaguj\u00ed, a nav\u00edc \u010dasto sami nemaj\u00ed logick\u00fd syst\u00e9mov\u00fd p\u0159\u00edstup k p\u0159ij\u00edm\u00e1n\u00ed sv\u00fdch rozhodnut\u00ed. Jak \u0159ekl m\u016fj kolega <noindex><a rel=\"nofollow\" href=\"https:\/\/en.wikipedia.org\/wiki\/Vijay_S._Pande\">Vidhaj Pand\u00e9<\/a><\/noindex>, <strong>lid\u00e9 jsou tak\u00e9 \u010dern\u00e9 boxy<\/strong>. <\/p>\n<p><\/p>\n<p>Vezm\u011bte tis\u00edc lid\u00ed z n\u011bkolika p\u0159ekr\u00fdvaj\u00edc\u00edch se firem a instituc\u00ed, a probl\u00e9m se stane je\u0161t\u011b slo\u017eit\u011bj\u0161\u00edm. Po skute\u010dnosti v\u00edme, \u017ee \u201eSpace Shuttle\u201c m\u011blo rozpadnout na kusy p\u0159i n\u00e1vratu, a jednotlivci uvnit\u0159 NASA m\u011bli informace, kter\u00e9 je oprav\u0148ovaly k tomu, aby si mysleli, \u017ee by se mohlo st\u00e1t n\u011bco \u0161patn\u00e9ho, ale syst\u00e9m <em>toho obecn\u011b<\/em> nebyl v\u011bdom. NASA pr\u00e1v\u011b pro\u0161la podobn\u00fdm auditem, kdy ztratila p\u0159edchoz\u00ed Shuttle, a p\u0159esto ztratila dal\u0161\u00ed \u2014 ze velmi podobn\u00e9ho d\u016fvodu. Je snadn\u00e9 tvrdit, \u017ee organizace a lid\u00e9 dodr\u017euj\u00ed jasn\u00e1 logick\u00e1 pravidla, kter\u00e1 lze ov\u011b\u0159it, pochopit a zm\u011bnit \u2014 ale zku\u0161enosti dokazuj\u00ed opak. To je \"<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\">iluze Gosp pl\u00e1nov\u00e1n\u00ed<\/a><\/noindex>\u00bb.<\/p>\n<p><\/p>\n<p><strong>\u010casto srovn\u00e1v\u00e1m strojov\u00e9 u\u010den\u00ed s datab\u00e1zemi, zejm\u00e9na s rela\u010dn\u00edmi \u2014 novou z\u00e1kladn\u00ed technologi\u00ed, kter\u00e1 zm\u011bnila mo\u017enosti informatiky a sv\u011bt kolem n\u00ed, kter\u00e1 se stala sou\u010d\u00e1st\u00ed v\u0161eho, co neust\u00e1le pou\u017e\u00edv\u00e1me, ani\u017e bychom si to uv\u011bdomovali.<\/strong> I datab\u00e1z\u00edch tak\u00e9 existuj\u00ed probl\u00e9my, kter\u00e9 maj\u00ed podobnou povahu: syst\u00e9m m\u016f\u017ee b\u00fdt postaven na nespr\u00e1vn\u00fdch p\u0159edpokladech nebo \u0161patn\u00fdch datech, co\u017e v\u0161ak bude t\u011b\u017ek\u00e9 post\u0159ehnout, a lid\u00e9, kte\u0159\u00ed syst\u00e9m pou\u017e\u00edvaj\u00ed, budou d\u011blat to, co jim \u0159\u00edk\u00e1, ani\u017e by se ptali. Existuje spousta star\u00fdch vtip\u016f o da\u0148ov\u00fdch \u00fa\u0159edn\u00edc\u00edch, kte\u0159\u00ed jednou \u0161patn\u011b zapsali va\u0161e jm\u00e9no, a p\u0159esv\u011bd\u010dit je, aby chybu opravili, je mnohem obt\u00ed\u017en\u011bj\u0161\u00ed ne\u017e vlastn\u011b zm\u011bnit jm\u00e9no. O tom se m\u016f\u017ee p\u0159em\u00fd\u0161let r\u016fzn\u011b, ale nen\u00ed jasn\u00e9, co je lep\u0161\u00ed: pova\u017eovat to za technick\u00fd probl\u00e9m v SQL, za chybu v nov\u00e9m vyd\u00e1n\u00ed Oracle, nebo za selh\u00e1n\u00ed byrokratick\u00fdch instituc\u00ed? Jak obt\u00ed\u017en\u00e9 je naj\u00edt chybu v procesu, kter\u00fd vedl k tomu, \u017ee syst\u00e9m nem\u00e1 takovou funkci, jako je oprava p\u0159eklep\u016f? Mohlo se to pochopit je\u0161t\u011b p\u0159edt\u00edm, ne\u017e za\u010dali lid\u00e9 st\u011b\u017eovat? <\/p>\n<p><\/p>\n<p>Tento probl\u00e9m ilustruj\u00ed je\u0161t\u011b l\u00e9pe p\u0159\u00edb\u011bhy, kdy \u0159idi\u010di kv\u016fli zastaral\u00fdm dat\u016fm v navigaci sj\u00ed\u017ed\u011bj\u00ed do \u0159ek. Dob\u0159e, mapy by se m\u011bly neust\u00e1le aktualizovat. Ale do jak\u00e9 m\u00edry je TomTom vinen za to, \u017ee va\u0161e auto odn\u00e1\u0161\u00ed do mo\u0159e?<\/p>\n<p><\/p>\n<p>\u0158\u00edk\u00e1m to proto, \u017ee ano \u2014 zaujatost strojov\u00e9ho u\u010den\u00ed vytvo\u0159\u00ed probl\u00e9my. Ale tyto probl\u00e9my budou podobn\u00e9 t\u011bm, se kter\u00fdmi jsme se v minulosti setk\u00e1vali, a budou je moci post\u0159ehnout a vy\u0159e\u0161it (nebo ne) p\u0159ibli\u017en\u011b tak dob\u0159e, jak se n\u00e1m to da\u0159ilo v minulosti. Proto je nepravd\u011bpodobn\u00e9, \u017ee by se sc\u00e9n\u00e1\u0159, ve kter\u00e9m by zaujatost AI zp\u016fsobila \u0161kodu, stal s p\u0159edn\u00edmi v\u00fdzkumn\u00edky, kte\u0159\u00ed pracuj\u00ed ve velk\u00e9 organizaci. Sp\u00ed\u0161e n\u011bjak\u00fd nezaj\u00edmav\u00fd technologick\u00fd dodavatel nebo v\u00fdrobce softwaru nap\u00ed\u0161e n\u011bco z brucha, p\u0159i\u010dem\u017e pou\u017eije pro n\u011bj nepochopiteln\u00e9 open-source komponenty, knihovny a n\u00e1stroje. A ne\u0161\u0165astn\u00fd z\u00e1kazn\u00edk se nech\u00e1 un\u00e9st slovn\u00edm spojen\u00edm \u201eum\u011bl\u00e1 inteligence\u201c v popisu produktu a bez zbyte\u010dn\u00fdch ot\u00e1zek to p\u0159ed\u00e1 sv\u00fdm n\u00edzko placen\u00fdm zam\u011bstnanc\u016fm, kte\u0159\u00ed budou nuceni d\u011blat to, co \u0159ekne AI. To se pr\u00e1v\u011b stalo s datab\u00e1zemi. To nen\u00ed probl\u00e9m um\u011bl\u00e9 inteligence, ani probl\u00e9m softwaru. To je lidsk\u00fd faktor.<\/p>\n<p><\/p>\n<h2 id=\"zaklyuchenie\">Z\u00e1v\u011br<\/h2>\n<p><\/p>\n<blockquote><p><em>Strojov\u00e9 u\u010den\u00ed m\u016f\u017ee ud\u011blat v\u0161e, \u010demu m\u016f\u017eete nau\u010dit psa \u2014 ale nikdy si nem\u016f\u017eete b\u00fdt jisti, \u010demu jste toho psa skute\u010dn\u011b nau\u010dili.<\/em><\/p><\/blockquote>\n<p>\u010casto m\u00e1m dojem, \u017ee term\u00edn \u201eum\u011bl\u00e1 inteligence\u201c jen br\u00e1n\u00ed tomu, abychom se dostali k diskuz\u00edm jako je tato. Tento term\u00edn vytv\u00e1\u0159\u00ed fale\u0161n\u00fd dojem, \u017ee jsme jej skute\u010dn\u011b vytvo\u0159ili \u2014 tuto inteligenci. \u017de jsme na cest\u011b k HAL9000 nebo Skynet \u2014 k n\u011b\u010demu, co skute\u010dn\u011b <em>rozum\u00ed<\/em>. Ale ne. Jsou to jen stroje a je mnohem spr\u00e1vn\u011bj\u0161\u00ed je srovn\u00e1vat, \u0159ekn\u011bme, s pra\u010dkou. Ta mnohem l\u00e9pe zvl\u00e1d\u00e1 pran\u00ed ne\u017e \u010dlov\u011bk, ale pokud do n\u00ed d\u00e1te n\u00e1dob\u00ed m\u00edsto pr\u00e1dla, to jej\u2026 vypere. N\u00e1dob\u00ed bude dokonce \u010dist\u00e9. Ale nebude to to, co jste o\u010dek\u00e1vali, a k tomu dojde nejen proto, \u017ee syst\u00e9m m\u00e1 n\u011bjak\u00e9 p\u0159edsudky o n\u00e1dob\u00ed. Pra\u010dka nev\u00ed, co je n\u00e1dob\u00ed, ani co je oble\u010den\u00ed \u2014 je to jen p\u0159\u00edklad automatizace, kter\u00e1 se konceptu\u00e1ln\u011b neli\u0161\u00ed od toho, jak byly procesy automatizov\u00e1ny d\u0159\u00edve.<\/p>\n<p><\/p>\n<p>A\u0165 u\u017e je t\u00e9ma jak\u00e9koli \u2014 o stroj\u00edch, letadlech nebo datab\u00e1z\u00edch \u2014 tyto syst\u00e9my budou z\u00e1rove\u0148 velmi mocn\u00e9 a velmi omezen\u00e9. Bude to zcela z\u00e1visl\u00e9 na tom, jak lid\u00e9 tyto syst\u00e9my pou\u017e\u00edvaj\u00ed, zda maj\u00ed dobr\u00e9 nebo \u0161patn\u00e9 \u00famysly a jak dob\u0159e ch\u00e1pou jejich fungov\u00e1n\u00ed.<\/p>\n<p><\/p>\n<p>Proto tvrzen\u00ed, \u017ee \u201eum\u011bl\u00e1 inteligence je matematika, tud\u00ed\u017e nem\u016f\u017ee m\u00edt p\u0159edsudky\u201c, je naprosto nespr\u00e1vn\u00e9. Ale stejn\u011b tak \u0161patn\u00e9 je tvrdit, \u017ee strojov\u00e9 u\u010den\u00ed je \u201esubjektivn\u00ed ve sv\u00e9 podstat\u011b\u201c. Strojov\u00e9 u\u010den\u00ed nach\u00e1z\u00ed vzory v datech a jak\u00e9 vzory najde, z\u00e1vis\u00ed na datech, a ta z\u00e1vis\u00ed na n\u00e1s. Stejn\u011b jako na tom, co s nimi d\u011bl\u00e1me. Strojov\u00e9 u\u010den\u00ed skute\u010dn\u011b d\u011bl\u00e1 n\u011bkter\u00e9 v\u011bci mnohem l\u00e9pe ne\u017e my \u2014 ale psi jsou nap\u0159\u00edklad mnohem efektivn\u011bj\u0161\u00ed ne\u017e lid\u00e9 p\u0159i objevov\u00e1n\u00ed drog, co\u017e rozhodn\u011b nen\u00ed d\u016fvod k tomu je povol\u00e1vat jako sv\u011bdky a vyn\u00e1\u0161et rozsudky na z\u00e1klad\u011b jejich sv\u011bdectv\u00ed. A psi, mimochodem, jsou mnohem chyt\u0159ej\u0161\u00ed ne\u017e jak\u00fdkoli syst\u00e9m strojov\u00e9ho u\u010den\u00ed.<\/p>\n<p><strong>P\u0159eklad:<\/strong>\u00a0<noindex><a rel=\"nofollow\" href=\"https:\/\/habr.com\/en\/users\/757NF\/\">Diana Leck\u00e1<\/a><\/noindex>.<br \/>\n<strong>Editace:<\/strong>\u00a0<noindex><a rel=\"nofollow\" href=\"http:\/\/teleg.one\/ponchiknews\">Alexej Ivanov<\/a><\/noindex>.<br \/>\n<strong>Komunita:<\/strong> <noindex><a rel=\"nofollow\" href=\"http:\/\/teleg.one\/ponchiknews\">@PonchikNews<\/a><\/noindex>.<\/p>\n<p>Zdroj: <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 5.0.3 - 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\/cs\/blog\/news\/o-predvzyatosti-iskusstvennogo-intellekta\" \/>\n\t\t<meta name=\"generator\" content=\"All in One SEO (AIOSEO) 5.0.3\" \/>\n\t\t<meta property=\"og:locale\" content=\"cs_CZ\" \/>\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\/cs\/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\udd47O p\u0159edsudc\u00edch um\u011bl\u00e9 inteligence | ProHoster","description":"tl;dr: .","canonical_url":"https:\/\/prohoster.info\/cs\/blog\/news\/o-predvzyatosti-iskusstvennogo-intellekta","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"cs_CZ","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\/cs\/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\/cs\/wp-json\/wp\/v2\/posts\/32781","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/comments?post=32781"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/posts\/32781\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/media\/24558"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/media?parent=32781"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/categories?post=32781"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/cs\/wp-json\/wp\/v2\/tags?post=32781"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}