I-Artificial Intelligence (AI) iya ngokuya ibaluleka futhi isivele ikhona ezicini eziningi zokuphila kwethu kwansuku zonke - kodwa ingabe i-AI ibona futhi icabange ngezwe ngendlela efanayo nabantu? Ukuze uphendule lo mbuzo, abacwaningi bakwa-Max Planck namalungu e-Justus Liebig University Giessen, Florian Mahner, Lukas Muttenthaler, kanye no-Martin Hebart, baphenya ukuthi i-AI iyazibona yini izinto ngendlela efanayo neyabantu. Bakhe indlela entsha ebavumela ukuthi bakhombe ngokucacile futhi baqhathanise ubukhulu obubalulekile abantu kanye ne-AI abayinaka lapho bebuka izinto.
"Lezi zilinganiso zimelela izici ezihlukahlukene zezinto, kusukela ezicini ezibonakalayo ezifana 'nendilinga' noma 'ezimhlophe' kuya kwezinye izici ze-semantic 'ezihlobene nesilwane' noma 'ezihlobene nomlilo,' ezinobukhulu obuningi obuqukethe kokubili izici ezibukwayo neze-semantic," kuchaza uFlorian Mahner, umbhali oholayo wocwaningo. "Imiphumela yethu yembula umehluko obalulekile: Nakuba abantu ngokuyinhloko begxila ebukhulu obuhlobene nencazelo-iyini into kanye nalokho esikwaziyo ngayo-amamodeli e-AI athembele kakhulu kubukhulu obuthwebula izici ezibonakalayo, njengokuma kwento noma umbala. Lesi senzo sibiza ngokuthi 'ukuchema kokubukwa' ku-AI. Ngisho nalapho i-AI ibonakala ibona izinto ngendlela ehlukile, ngokuvamile iphakamisa umehluko obalulekile womuntu njengoba isetshenziswa ngendlela eyisisekelo. ukuthi izinhlelo ze-AI, nakuba ziziphatha ngendlela efanayo kubantu, zingase zicabange futhi zenze izinqumo ngendlela ehluke kakhulu, okuthinta ukuthi singazethemba kangakanani.”
Ngokuziphatha komuntu, ososayensi basebenzise izahlulelo ezingaba yizigidi ezinhlanu ezitholakala esidlangalaleni ezithombeni ezihlukene eziyi-1.854. Isibonelo, umhlanganyeli waboniswa isithombe sesiginci, indlovu, nesihlalo futhi wabuza ukuthi iyiphi into engahambisani. Ososayensi babe sebesebenzisa amanethiwekhi ambalwa e-neural ajulile (ama-DNN), angakwazi ukubona izithombe ngokufana kubahlanganyeli abangabantu, futhi baqoqa izahlulelo zokufana zezithombe zezinto ezifanayo ezisetshenziselwa abantu. Babe sebesebenzisa i-algorithm efanayo ukuhlonza izici ezibalulekile zalezi zithombe-ososayensi abazibiza ngokuthi "ubukhulu" -ezingaphansi kwezahlulelo zangaphandle. Ukuphatha inethiwekhi ye-neural ngokufana nabantu kwaqinisekisa ukuqhathaniswa okuqondile phakathi kwalokhu kokubili. "Lapho siqala ukubheka izilinganiso esizitholile kumanethiwekhi ajulile we-neural, sasicabanga ukuthi zifana kakhulu nezabantu," kuchaza uMartin Hebart, umbhali ophezulu wocwaningo. "Kodwa-ke, lapho siqhathanisa ngokuqondile ubukhulu obufanayo phakathi kwabantu namanethiwekhi emizwa ajulile, sithole ukuthi inethiwekhi cishe ithwebule lobu bukhulu. Ngobukhulu obuhlobene nesilwane, izithombe eziningi zezilwane bezikhishiwe, futhi ngokufanayo, izithombe eziningi ezingezona nhlobo izilwane zafakwa. Lokhu kuyinto ebesingayigeja sisebenzisa amasu ajwayelekile." Ososayensi banethemba lokuthi ucwaningo lwesikhathi esizayo luzosebenzisa izindlela ezifanayo eziqhathanisa ngokuqondile abantu ne-AI ukuze baqonde kangcono ukuthi i-AI ibona umhlaba. “Ucwaningo lwethu lunikeza indlela ecacile neqondakalayo yokuphenya lo mehluko, okusisiza ukuthi siqonde kangcono ukuthi i-AI ilucubungula kanjani ulwazi uma iqhathaniswa nabantu,” kusho uMartin Hebart. "Lolu lwazi alukwazi ukusisiza kuphela ukuthuthukisa ubuchwepheshe be-AI kodwa futhi lunikeze nokuqonda okubalulekile ekuqapheliseni komuntu."
Ukushicilelwa kwangempela:
UFlorian P. Mahner, Lukas Muttenthaler, Umut Güçlü, Martin N. Hebart:
Ubukhulu obungaphansi kokuqondanisa kokumelela kwamanethiwekhi ajulile e-neural nabantu. Ubuhlakani bomshini wemvelo
Funda futhi
I-AI inikeza izinkampani amathuba okuvikela isimo sezulu | I-Pugnalom
