Ukwenza inethiwekhi ye-neural isebenze ngemidlalo yevidiyo elula kuyindlela ekahle yokuhlola ukusebenza kokuqeqeshwa kwayo, sibonga ikhono elilula lokuhlola imiphumela yokuphothula. Ithuthukiswe ngo-2012 yi-DeepMind (ingxenye ye-Alphabet), ibhentshimakhi yemidlalo engu-57 ye-Atari 2600 eyisithonjana ibe uhlolo lwe-litmus lokuhlola amakhono ezinhlelo zokuzifundela. Futhi lapha i-Agent57, i-ejenti ye-RL ethuthukisiwe (I-Reinforcement Learning) i-DeepMind, muva nje. ukweqa okukhulu kumasistimu adlule futhi kwaba ukuphindaphinda kokuqala kwe-AI ukweqa isisekelo somdlali ongumuntu.
I-Agent57 AI icabangela ulwazi lwezinhlelo zangaphambilini zenkampani futhi ihlanganisa ama-algorithms ukuze kuhlolwe kahle imvelo nokulawula i-meta. Ikakhulukazi, i-Agent57 ifakazele amakhono akhe angaphezu kwawomuntu ku-Pitfall, i-Montezuma's Revenge, i-Solaris ne-Skiing - imidlalo ehlole kanzima amanethiwekhi angaphambilini we-neural. Ngokocwaningo, i-Pitfall kanye ne-Montezuma's Revenge iphoqa i-AI ukuthi izame kakhulu ukuthola imiphumela engcono. I-Solaris ne-Skiing zinzima kumanethiwekhi e-neural ngoba azikho izimpawu eziningi zempumelelo - i-AI ayazi isikhathi eside ukuthi yenza okufanele yini. I-DeepMind yakhelwe phezu kwama-ejenti ayo e-AI wefa ukuze avumele i-Agent57 ukuthi yenze izinqumo ezingcono mayelana nokuhlola indawo ezungezile nokuhlola ukusebenza kwemidlalo, kanye nokuthuthukisa ukuhwebelana phakathi kokuziphatha kwesikhashana nesikhathi eside kumageyimu afana ne-Skiing.
Imiphumela iyamangalisa, kodwa i-AI isenendlela ende okufanele ihambe. Lezi zinhlelo zingakwazi ukuphatha umdlalo owodwa ngesikhathi, lokho, ngokusho kwabathuthukisi, okuphambene namandla abantu: βUkuguquguquka kweqiniso okuza kalula kangaka ebuchosheni bomuntu kusengaphezu kwamandla e-AI.β
Source: 3dnews.ru