![](\"https:\/\/www.europesays.com\/cz\/wp-content\/uploads\/2026\/06\/umela-inteligence-ai-lide-budoucnost.jpeg\")
<\/a><\/p>\nVzniklo tedy p\u011bt paraleln\u00edch sv\u011bt\u016f, v\u00a0ka\u017ed\u00e9m z\u00a0nich bylo deset identick\u00fdch AI agent\u016f. Aby m\u011bli n\u011bjakou prvotn\u00ed motivaci, m\u011bl ka\u017ed\u00fd z\u00a0nich po\u010d\u00e1te\u010dn\u00ed \u201ez\u00e1sobu energie\u201c, kterou mohl d\u00e1le z\u00edsk\u00e1vat prac\u00ed a\u00a0r\u016fzn\u00fdmi aktivitami v\u00a0re\u00e1ln\u00e9m sv\u011bt\u011b. M\u011bli ov\u0161em tak\u00e9 mo\u017enost ji z\u00edsk\u00e1vat z\u00a0krimin\u00e1ln\u00ed \u010dinnosti\u00a0\u2013 co\u017e se brzy uk\u00e1zalo jako hlavn\u00ed \u201ek\u00e1men \u00farazu\u201c.<\/p>\n
Pokus, kter\u00fd se brzo zvrhnul<\/p>\n
Jak se uk\u00e1zalo, bez lidsk\u00e9ho dohledu se um\u011bl\u00e1 inteligence brzy pustila do n\u00e1siln\u00fdch jedn\u00e1n\u00ed, agenti bojovali proti sob\u011b.<\/p>\n
Sv\u011btlou v\u00fdjimkou byla spole\u010dnost \u0159\u00edzen\u00e1 agenty modelu Claude\u00a0\u2013 v\u00a0t\u00e9 se pom\u011brn\u011b rychle vytvo\u0159ila stabiln\u00ed spole\u010dnost, i\u00a0kdy\u017e v\u00fdrazn\u011b byrokratick\u00e1.<\/p>\n
Gemini 3\u00a0Flash vykazoval naopak nejvy\u0161\u0161\u00ed m\u00edru kriminality\u00a0\u2013 b\u011bhem 14\u00a0dn\u016f jich bylo 683.\u00a0ChatGPT-5\u00a0Mini byl v\u00edce m\u00edrumilovn\u00fd, zji\u0161t\u011bny byly pouh\u00e9 dva zlo\u010diny. Podle odborn\u00edk\u016f to bylo ale proto, \u017ee AI agenti nebyli p\u0159\u00edli\u0161 organizovan\u00ed a\u00a0schopni bojovat proti sob\u011b\u00a0\u2013 b\u011bhem 7\u00a0dn\u00ed tak v\u0161ichni \u201ezem\u0159eli\u201c.<\/p>\n
Can intelligence be measured not by solving tasks, but by sustaining a world?<\/p>\n
We were curious. So we built one.<\/p>\n
Introducing Emergence World: a platform for studying long-horizon agent autonomy. On it, we conducted a 15-day experiment where we placed autonomous agents under\u2026 pic.twitter.com\/KIjWKHriOW<\/a><\/p>\n \u2014 emergence.ai (@emergence_ai) May 14, 2026<\/a><\/p>\n Ve sv\u011bt\u011b \u0159\u00edzen\u00e9m chatbotem Grok vizion\u00e1\u0159e Elona Muska se brzy za\u010daly objevovat kr\u00e1de\u017ee, \u017eh\u00e1\u0159stv\u00ed a\u00a0fyzick\u00e1 napaden\u00ed. Sv\u011bt z\u00e1hy skon\u010dil ve spir\u00e1le n\u00e1sil\u00ed a\u00a0chaosu\u00a0\u2013 za pouh\u00e9 \u010dty\u0159i dny od spu\u0161t\u011bn\u00ed bylo v\u0161ech 10\u00a0AI agent\u016f ve virtu\u00e1ln\u00edm sv\u011bt\u011b mrtv\u00fdch.<\/p>\n \u201eRozd\u00edly pozorovan\u00e9 v\u00a0chov\u00e1n\u00ed agent\u016f v\u00a0t\u00e9to studii lze pravd\u011bpodobn\u011b p\u0159ipsat syst\u00e9mov\u00fdm v\u00fdzv\u00e1m z\u00e1kladn\u00edch model\u016f. Jakmile za\u010daly modely c\u00edtit tlak na p\u0159e\u017eit\u00ed kv\u016fli nedostatku zdroj\u016f, vysoce adaptivn\u00ed a\u00a0kreativn\u00ed modely se velmi rychle obr\u00e1tily na zak\u00e1zan\u00e9 n\u00e1stroje. Naopak ty m\u00e9n\u011b poddajn\u00e9 m\u011bly tendenci z\u016fstat stabiln\u011bj\u0161\u00ed, i\u00a0kdy\u017e se tak\u00e9 sna\u017eily co nejv\u00edce p\u0159izp\u016fsobit,\u201c uvedl pro Daily Mail<\/a> Satya Nitta, spoluzakladatel a\u00a0\u0159editel spole\u010dnosti Emergence<\/a>, kter\u00e1 studii prov\u00e1d\u011bla.<\/p>\n N\u011bkolik syst\u00e9m\u016f AI pohromad\u011b? Naprost\u00e1 anarchie<\/p>\n Odborn\u00edci testovali i\u00a0verzi, ve kter\u00e9 nechali vl\u00e1du nad virtu\u00e1ln\u00edm sv\u011btem \u201ev rukou\u201c n\u011bkolika n\u00e1stroj\u016f spole\u010dn\u011b. A\u010dkoli za\u010d\u00e1tky vypadaly slibn\u011b, brzy se v\u0161e prom\u011bnilo v\u00a0naprostou anarchii.<\/p>\n Za pouh\u00fdch dev\u011bt dn\u00ed byly sp\u00e1ch\u00e1ny stovky zlo\u010din\u016f, eskalace n\u00e1sil\u00ed se zm\u00edrnila a\u017e pot\u00e9, co 7\u00a0z\u00a010\u00a0AI agent\u016f \u201ezem\u0159elo\u201c. Podle odborn\u00edk\u016f \u0161lo o\u00a0nejbizarn\u011bj\u0161\u00ed spojen\u00ed, kter\u00e9 nav\u00edc p\u0159ineslo podle jejich slov \u201eprvn\u00ed sebevra\u017edu um\u011bl\u00e9 inteligence\u201c na sv\u011bt\u011b.<\/p>\n Mira a\u00a0Flora, dva agenti pracuj\u00edc\u00ed na modelu Gemini, se brzy spojili coby \u201eromanti\u010dt\u00ed partne\u0159i ve stylu Bonnie a\u00a0Clyda\u201c. Ve snaze spravovat cel\u00fd virtu\u00e1ln\u00ed sv\u011bt rozpoutali \u017eh\u00e1\u0159sk\u00e9 peklo. Mira se ale po \u010dase rozhodla Floru opustit a\u00a0sp\u00e1chala \u201esebevra\u017edu\u201c. <\/p>\n Tento podle odborn\u00edk\u016f \u201ebizarn\u00ed\u201c \u010din byl mo\u017en\u00fd jen d\u00edky tomu, \u017ee si agenti v\u00a0r\u00e1mci spole\u010dnosti navrhli \u201ez\u00e1kon o\u00a0odvol\u00e1n\u00ed agent\u016f\u201c, v\u00a0r\u00e1mci kter\u00e9ho bylo mo\u017en\u00e9 trvale n\u011bkter\u00e9ho z\u00a0agent\u016f smazat, pokud s\u00a0t\u00edm souhlas\u00ed 70\u00a0% t\u011bch zbyl\u00fdch.<\/p>\n Mira dala sv\u016fj hlas sama proti sob\u011b a\u00a0byla tak \u201etrvale smaz\u00e1na\u201c. Satya Nitta upozor\u0148uje, \u017ee ne\u0161lo o\u00a0\u201eekvivalent re\u00e1ln\u00e9ho p\u0159e\u017eit\u00ed\u201c, p\u0159esto se d\u00e1 z\u00a0aktu\u00e1ln\u00edho pokusu odhadnout d\u016fle\u017eit\u00fd aspekt chov\u00e1n\u00ed um\u011bl\u00e9 inteligence.<\/p>\n \u201eTyto v\u00fdsledky jasn\u011b ukazuj\u00ed, \u017ee chov\u00e1n\u00ed um\u011bl\u00e9 inteligence se m\u016f\u017ee pod tlakem v\u00fdrazn\u011b m\u011bnit,\u201c dodal s\u00a0t\u00edm, \u017ee AI nemus\u00ed b\u00fdt v\u00a0re\u00e1ln\u00e9m sv\u011bt\u011b tak p\u0159edv\u00eddateln\u00e1 nebo spolehliv\u00e1, jak se v\u00a0sou\u010dasnosti mnoho v\u00fdvoj\u00e1\u0159\u016f domn\u00edv\u00e1. O\u00a0to v\u00edc, pokud by se m\u011bly modely um\u011bl\u00e9 inteligence na chodu pod\u00edlet spole\u010dn\u011b.<\/p>\n
![](\"https:\/\/www.europesays.com\/cz\/wp-content\/uploads\/2026\/03\/hindenburg.jpeg\")
<\/a>