Ukwenza inethiwekhi ye-neural iqhutywe ngemidlalo yevidiyo elula yindlela efanelekileyo yokuvavanya ukusebenza koqeqesho lwayo, ngenxa yesakhono esilula sokuvavanya iziphumo zokugqitywa. Iphuhliswe kwi-2012 yi-DeepMind (inxalenye ye-Alphabet), i-benchmark ye-57 iconic ye-Atari 2600 imidlalo yaba luvavanyo lwe-litmus lokuvavanya amandla eenkqubo zokuzifundela. Kwaye apha i-Agent57, i-arhente ye-RL eqhubela phambili (i-Reinforcement Learning) i-DeepMind, kutshanje. umtsi omkhulu kwiinkqubo zangaphambili kwaye yaba yinto yokuqala ephindaphindwayo ye-AI ukugqitha isiseko somdlali ongumntu.
I-Agent57 AI ithathela ingqalelo amava eenkqubo zangaphambili zenkampani kwaye idibanisa i-algorithms yokuhlola okusebenzayo kwendalo kunye nolawulo lwemeta. Ngokukodwa, i-Agent57 ibonakalise izakhono zakhe ezingaphaya komntu kwi-Pitfall, i-Montezuma's Revenge, i-Solaris kunye ne-Skiing-imidlalo evavanye ngokuqatha uthungelwano lwangaphambili lwe-neural. Ngokophando, i-Pitfall kunye ne-Montezuma's Revenge inyanzela i-AI ukuba izame ngakumbi ukufezekisa iziphumo ezingcono. I-Solaris kunye ne-Skiing zinzima kuthungelwano lwe-neural kuba akukho zimpawu zininzi zempumelelo- i-AI ayazi ixesha elide ukuba yenza into efanelekileyo. I-DeepMind yakhelwe kwi-arhente ye-AI yelifa ukuvumela i-Agent57 ukuba yenze izigqibo ezingcono malunga nokuphonononga okusingqongileyo kunye nokuvavanya ukusebenza kwemidlalo, kunye nokuphucula urhwebo phakathi kokuziphatha kwexesha elifutshane kunye nexesha elide kwimidlalo efana ne-Skiing.
Iziphumo ziyamangalisa, kodwa i-AI isenendlela ende ekufuneka ihambe. Ezi nkqubo zinokusingatha umdlalo omnye ngexesha, nto leyo, ngokutsho kwabaphuhlisi, ichasene namandla omntu: βOlona bhetyebhetye lokwenyani oluza ngokulula kwingqondo yomntu lusengaphaya kokufikelela kwi-AI.β
umthombo: 3dnews.ru