Yin hanyar sadarwa ta jijiyoyi da ke gudana ta hanyar wasanni masu sauƙi na bidiyo shine hanya mai kyau don gwada tasirin horo, godiya ga sauƙi mai sauƙi don kimanta sakamakon kammalawa. An haɓaka shi a cikin 2012 ta DeepMind (ɓangare na Alphabet), ma'auni na 57 gunkin wasannin Atari 2600 ya zama gwaji mai haske don gwada ƙarfin tsarin koyon kai. Kuma a nan Agent57, wakili na RL mai ci gaba (Ƙarfafa Koyo) DeepMind, kwanan nan babban tsalle daga tsarin da suka gabata kuma shine farkon fitowar AI don wuce tushen ɗan adam.
Agent57 AI yayi la'akari da kwarewar tsarin da kamfanin ya yi a baya kuma ya haɗu da algorithms don ingantaccen bincike na yanayi tare da sarrafa meta. Musamman ma, Agent57 ya tabbatar da ƙwarewar sa na ɗan adam a cikin Pitfall, Revenge na Montezuma, Solaris da Skiing - wasannin da suka gwada hanyoyin sadarwa na baya. Dangane da bincike, Pitfall da ramuwar gayya ta Montezuma sun tilasta AI don ƙarin gwaji don cimma sakamako mafi kyau. Solaris da Skiing suna da wahala ga hanyoyin sadarwar jijiyoyi saboda babu alamun nasara da yawa - AI bai sani ba na dogon lokaci ko yana yin abin da ya dace. DeepMind ya gina a kan abubuwan da aka bari na AI don ba da damar Agent57 don yanke shawara mafi kyau game da binciken yanayi da kuma kimanta ayyukan wasanni, da kuma inganta cinikayya tsakanin gajeren lokaci da kuma dogon lokaci a cikin wasanni kamar Skiing.
Sakamakon yana da ban sha'awa, amma AI har yanzu yana da doguwar hanya don tafiya. Waɗannan tsarin suna iya ɗaukar wasa ɗaya kawai a lokaci ɗaya, wanda, a cewar masu haɓakawa, ya saba wa iyawar ɗan adam: “Sausanin gaskiya da ke zuwa cikin sauƙi ga kwakwalwar ɗan adam har yanzu ba ta kai ga AI ba.”
source: 3dnews.ru