Lo mbhalo awusosiphumo sophando lwenzululwazi, kodwa sesinye sezimvo ezininzi malunga nophuhliso lwethu lwethekhinoloji lwangoku. Kwaye kwangaxeshanye isimemo kwingxoxo.
UGary Marcus, uprofesa kwiYunivesithi yaseNew York, ukholelwa ukuba ukufunda okunzulu kudlala indima ebalulekileyo ekuphuhliseni i-AI. Kodwa ukwakholelwa ukuba ukuzondelela okugqithisileyo kobu buchule kunokukhokelela ekuhleni kwayo.
Kwincwadi yakhe Ukuqalisa kwakhona i-AI: Ukwakha ubukrelekrele bokwenziwa esinokumthemba UMarcus, ingcali ye-neuroscientist ngoqeqesho oluye wakha ikhondo lomsebenzi kuphando lwe-AI, ujongana nemiba yobugcisa kunye neyeenqobo ezisesikweni. Ngokwembono yetekhnoloji, ukufunda nzulu kunokulinganisa ngempumelelo imisebenzi yokuqonda eyenziwa bubuchopho bethu, njengomfanekiso okanye ukuqondwa kwentetho. Kodwa kweminye imisebenzi, efana nokuqonda iincoko okanye ukumisela ubudlelwane besizathu-kunye nesiphumo, ukufunda nzulu akufanelekanga. Ukwenza oomatshini abakrelekrele abaphucuke ngakumbi abanokusombulula uluhlu olubanzi lweengxaki—ezidla ngokubizwa ngokuba bubukrelekrele bokuzenzela—ukufunda ngokunzulu kufuneka kudityaniswe nobunye ubugcisa.
Ukuba inkqubo ye-AI ayiyiqondi ngokwenene imisebenzi yayo okanye ihlabathi elijikelezileyo, oku kunokukhokelela kwimiphumo eyingozi. Kwanolona tshintsho luncinane olungalindelekanga kwimekobume yenkqubo lunokukhokelela ekuziphatheni okuphosakeleyo. Sele kukho imizekelo emininzi enjalo: izigxininiso zamabinzana angafanelekanga ekulula ukuwakhohlisa; iinkqubo zokukhangela umsebenzi ezihlala zicalula; iimoto ezingaqhubiyo ezintlithekayo kwaye ngamanye amaxesha zibulale umqhubi okanye umhambi ngeenyawo. Ukudala ubukrelekrele jikelele bokwenziwa ayisiyongxaki nje yophando enomdla, inezicelo ezininzi ezisebenzayo.
Kwincwadi yabo, uMarcus kunye nombhali-mbhali u-Ernest Davis baxoxa ngendlela eyahlukileyo. Bakholelwa ukuba sisekude ekudaleni i-AI ngokubanzi, kodwa baqinisekile ukuba kungekudala okanye kamva kuya kwenzeka ukuyidala.
Kutheni sifuna i-AI ngokubanzi? Iinguqulelo ezikhethekileyo sele zenziwe kwaye zizisa iinzuzo ezininzi.
Kulungile, kwaye kuya kubakho iinzuzo ezingakumbi. Kodwa zininzi iingxaki ezikhethekileyo ze-AI ezingenakukwazi ukuzicombulula. Umzekelo, ukuqonda intetho eqhelekileyo, okanye uncedo jikelele kwihlabathi virtual, okanye irobhothi enceda ukucoca nokupheka. Imisebenzi enjalo ingaphaya kwamandla e-AI ekhethekileyo. Omnye umbuzo onomdla osebenzayo: ngaba kunokwenzeka ukwenza imoto ekhuselekileyo yokuziqhuba usebenzisa i-AI ekhethekileyo? Amava abonisa ukuba i-AI enjalo iseneengxaki ezininzi zokuziphatha kwiimeko ezingaqhelekanga, naxa uqhuba, nto leyo eyenza nzima kakhulu imeko.
Ndicinga ukuba sonke singathanda ukuba ne-AI enokusinceda senze izinto ezintsha ezifunyenweyo kwezamayeza. Akukacaci ukuba itekhnoloji yangoku ikulungele na oku, kuba ibhayoloji yinkalo entsonkothileyo. Kufuneka ukulungele ukufunda iincwadi ezininzi. Iingcali zenzululwazi ziqonda ubudlelwane besizathu-kunye-nesiphumo ekusebenzisaneni kothungelwano kunye neemolekyuli, zinokuphuhlisa iithiyori malunga neeplanethi, njalo njalo. Nangona kunjalo, nge-AI ekhethekileyo, asikwazi ukwenza oomatshini abanakho ukufunyanwa okunjalo. Kwaye nge-AI ngokubanzi, sinokuguqula isayensi, itekhnoloji kunye namayeza. Ngokombono wam, kubaluleke kakhulu ukuqhubeka nokusebenza ekudaleni i-AI ngokubanzi.
Kuvakala ngathi "ngokubanzi" uthetha i-AI eyomeleleyo?
Ngokuthi "jikelele" ndithetha ukuba i-AI iya kukwazi ukucinga kunye nokusombulula iingxaki ezintsha kwi-fly. Ngokungafaniyo, yithi, Yiya, apho ingxaki ingatshintshanga kwiminyaka engama-2000 edlulileyo.
I-AI ngokubanzi kufuneka ikwazi ukwenza izigqibo kwezopolitiko kunye namayeza. Oku kufana nobuchule bomntu; nawuphi na umntu obhadlileyo unokwenza lukhulu. Uthatha abafundi abanamava kwaye ngeentsuku ezimbalwa basebenze phantse kuyo nantoni na, ukusuka kwingxaki yezomthetho ukuya kwingxaki yezonyango. Oku kungenxa yokuba banokuqonda ngokubanzi ihlabathi kwaye bayakwazi ukufunda, kwaye ke ngoko banokuba negalelo kuluhlu olubanzi kakhulu lwemisebenzi.
Ubudlelwane phakathi kobukrelekrele obunjalo kunye nobukrelekrele obunamandla kukuba ubukrelekrele obungenamandla ngokuqinisekileyo abuyi kukwazi ukusombulula iingxaki ngokubanzi. Ukudala into eyomeleleyo ngokwaneleyo yokujongana nehlabathi elihlala litshintsha, unokufuna ukuba ubuncinci usondele kubukrelekrele ngokubanzi.
Kodwa ngoku sikude kakhulu kule nto. I-AlphaGo inokudlala ngokugqibeleleyo kwibhodi ye-19x19, kodwa kufuneka iphinde iqeqeshwe ukuze idlale kwibhodi yoxande. Okanye thatha i-avareji yenkqubo yokufunda nzulu: iyakwazi ukubona indlovu ukuba ikhanyise kakuhle kwaye ukubonakala kwayo kwesikhumba kuyabonakala. Yaye ukuba kubonakala nje umzobo wendlovu, loo nkqubo ayinakukwazi ukuwubona.
Kwincwadi yakho, ukhankanya ukuba ukufunda okunzulu akunako ukufikelela kwizakhono ze-AI ngokubanzi kuba ayikwazi ukuqonda okunzulu.
Kwinzululwazi yokuqonda bathetha ngokuqulunqwa kwemizekelo eyahlukeneyo yokuqonda. Ndihleli kwigumbi lehotele kwaye ndiyaqonda ukuba kukho igumbi, kukho ibhedi, kukho iTV exhonywe ngendlela engaqhelekanga. Ndiyazazi zonke ezi zinto, andizichongi nje. Ndiyayiqonda nendlela ezinxibelelana ngayo. Ndinezimvo malunga nokusebenza kwehlabathi elindingqongileyo. Abafezekanga. Zisenokuba ziphosakele, kodwa zilungile. Kwaye ngokusekelwe kuzo, ndenza izigqibo ezininzi eziba zizikhokelo zezenzo zam zemihla ngemihla.
Olunye ugqithiso lwaluyinto efana nenkqubo yomdlalo we-Atari eyakhiwe yi-DeepMind, apho ikhumbule oko kufuneka iyenze xa ibona iipikseli kwiindawo ezithile kwisikrini. Ukuba ufumana idatha eyaneleyo, unokucinga ukuba unokuqonda, kodwa ngokwenene kuphezulu kakhulu. Ubungqina boku kukuba ukuba uhambisa izinto ngeepikseli ezintathu, i-AI idlala kakubi kakhulu. Utshintsho luyamxaka. Oku kuchasene nokuqonda okunzulu.
Ukusombulula le ngxaki, ucebisa ukubuyela kwi-AI yakudala. Ziziphi iingenelo esifanele sizame ukuzisebenzisa?
Kukho iingenelo ezininzi.
Okokuqala, i-classical AI eneneni sisikhokelo sokudala iimodeli zokuqonda zehlabathi, ezisekwe phezu kwezigqibo ezinokuthathwa.
Okwesibini, i-AI yeklasikhi ihambelana ngokugqibeleleyo nemithetho. Kukho umkhwa ongaqhelekanga wokufunda nzulu ngoku apho iingcali zizama ukuphepha imithetho. Bafuna ukwenza yonke into kwiinethiwekhi ze-neural kwaye bangenzi nantoni na ekhangeleka ngathi yinkqubo yeklasikhi. Kodwa kukho iingxaki eziye zasonjululwa ngokuzolileyo ngolu hlobo, kwaye akukho mntu wanikela ingqalelo kuyo. Umzekelo, iindlela zokwakha kwiiMephu zikaGoogle.
Enyanisweni, sifuna zombini iindlela. Ukufunda ngoomatshini kulungile ekufundeni kwidatha, kodwa kubi kakhulu ekuboniseni ukuthatyathwa okuyinkqubo yekhompyuter. I-Classic AI isebenza kakuhle ngokutsalwa, kodwa kufuneka icwangciswe ngokupheleleyo ngesandla, kwaye kukho ulwazi oluninzi kakhulu emhlabeni ukuze lucwangcise zonke. Ngokucacileyo kufuneka sidibanise zombini iindlela.
Oku kunxulumana nesahluko othetha kuso ngoko sinokukufunda kwingqondo yomntu. Kwaye okokuqala, malunga nengcamango esekelwe kwingcamango ekhankanywe ngasentla yokuba ukuqonda kwethu kuqulethe iinkqubo ezininzi ezahlukeneyo ezisebenza ngeendlela ezahlukeneyo.
Ndicinga ukuba enye indlela yokuchaza oku kukuba inkqubo nganye yokuqonda esinayo ngokwenene isombulula ingxaki eyahlukileyo. Iinxalenye ezifanayo ze-AI kufuneka ziyilwe ukusombulula iingxaki ezahlukeneyo ezineempawu ezahlukeneyo.
Ngoku sizama ukusebenzisa itekhnoloji ye-All-in-one ukusombulula iingxaki ezahluke kakhulu kwenye. Ukuqonda isivakalisi akufani kwaphela nokuqaphela into. Kodwa abantu bazama ukusebenzisa ukufunda okunzulu kuzo zombini iimeko. Ngokwembono yokuqonda, le yimisebenzi eyahlukileyo ngokomgangatho. Ndimangalisiwe yindlela oluncinci ngayo uxabiso olukhoyo lwe-AI yakudala kuluntu lokufunda nzulu. Kutheni ulinde ukuba kuvele imbumbulu yesilivere? Ayinakwenzeka, kwaye ukukhangela okungenaziqhamo akusivumeli ukuba siqonde ubunzima obupheleleyo bomsebenzi wokudala i-AI.
Ukwakhankanya ukuba iinkqubo ze-AI ziyafuneka ukuze uqonde ubudlelwane besizathu kunye nesiphumo. Ngaba ucinga ukuba ukufunda okunzulu, i-AI yakudala, okanye into entsha ngokupheleleyo iya kusinceda kule nto?
Le yenye indawo apho ukufunda nzulu kungafanelanga kakuhle. Ayichazi izizathu zeziganeko ezithile, kodwa ibala ukuba nokwenzeka kwesiganeko phantsi kweemeko ezithile.
Sithetha ngantoni? Ubukela iimeko ezithile, kwaye uyasiqonda isizathu sokuba oku kusenzeka nokuba yintoni enokwenzeka ukuba ezinye iimeko ziyatshintsha. Ndiyakwazi ukujonga kwi-stand ehlala kuyo i-TV kwaye ndicinge ukuba ukuba ndinqumle imilenze yayo, i-stand iya kugqitywa kwaye i-TV iya kuwa. Lo ngunobangela kunye nobudlelwane besiphumo.
I-Classic AI isinika izixhobo ezithile kule nto. Unokucinga, umzekelo, yintoni inkxaso kunye nokuba yintoni ukuwa. Kodwa andiyi kuncoma kakhulu. Ingxaki kukuba i-AI yakudala ixhomekeke kakhulu kulwazi olupheleleyo malunga nento eyenzekayo, kwaye ndafikelela kwisigqibo nje ngokujonga kwi-stand. Ndiyakwazi ngandlela thile ukwenza ngokubanzi, cinga iinxalenye zestand ezingabonakali kum. Asikabi nazo izixhobo zokusebenzisa lo mhlaba.
Kwakhona uthi abantu banolwazi lwendalo. Oku kungaphunyezwa njani kwi-AI?
Ngexesha lokuzalwa, ingqondo yethu sele iyinkqubo ecokisekileyo. Ayilungiswanga; indalo yakha idrafti yokuqala. Kwaye ke ukufunda kuyasinceda ukuba sihlaziye oluyilo kubomi bethu bonke.
Idrafti yobuchopho sele inezakhono ezithile. Ibhokhwe yasezintabeni esandul’ ukuzalwa iyakwazi ukuhla intaba ngeeyure nje ezimbalwa. Kucacile ukuba sele enolwazi lwesithuba esine-dimensional emithathu, umzimba wakhe kunye nobudlelwane phakathi kwabo. Inkqubo enzima kakhulu.
Kungenxa yoko le nto ndikholelwa ukuba sifuna imixube. Kunzima ukucinga ukuba umntu angayenza njani irobhothi esebenza kakuhle kwihlabathi elingenalo ulwazi olufanayo lokuba ungaqala phi, kunokuba uqale ngesileyiti esingenanto kwaye ufunde kumava amade, amakhulu.
Ngokuphathelele abantu, ulwazi lwethu lwangaphakathi luvela kwi-genome yethu, eye yavela ixesha elide. Kodwa ngeenkqubo ze-AI kuya kufuneka sihambe ngendlela eyahlukileyo. Inxalenye yale inokuba yimithetho yokwakha i-algorithms yethu. Inxalenye yale nto inokuba yimithetho yokudala izakhiwo zedatha ezisetyenziswa yi-algorithms. Kwaye inxalenye yale nto inokuba lulwazi ukuba siya kutyala ngokuthe ngqo koomatshini.
Kuyathakazelisa ukuba kwincwadi uzisa umbono wokuthembela kunye nokudalwa kweenkqubo zokuthembela. Kutheni ukhethe lo mgaqo ukhethekileyo?
Ndiyakholelwa ukuba namhlanje yonke le nto ngumdlalo webhola. Kum kubonakala ngathi siphila kwixesha elingaqhelekanga kwimbali, sithembela kwisoftware eninzi engathembekanga. Ndicinga ukuba amaxhala esinawo namhlanje awasayi kuhlala ngonaphakade. Kwiminyaka elikhulu, i-AI iya kulungelelanisa ukuthembela kwethu, kwaye mhlawumbi kungekudala.
Kodwa namhlanje i-AI iyingozi. Akunjalo kwimeko yokuba u-Elon Musk uloyiko, kodwa ngendlela yokuba iinkqubo zodliwano-ndlebe lomsebenzi zicalula abafazi, kungakhathaliseki ukuba yintoni eyenziwa ngabaprogram, kuba izixhobo zabo zilula kakhulu.
Ndinqwenela ukuba sibe ne-AI engcono. Andifuni ukubona "ubusika be-AI" apho abantu baqonda ukuba i-AI ayisebenzi kwaye iyingozi nje kwaye ayifuni ukuyilungisa.
Ngandlela thile, incwadi yakho ibonakala inethemba elikhulu. Ucinga ukuba kunokwenzeka ukwakha i-AI ethembekileyo. Kufuneka nje sijonge kwelinye icala.
Kunjalo, le ncwadi ayinathemba kakhulu kwixesha elifutshane kwaye inethemba elikhulu kwixesha elide. Sikholelwa ukuba zonke iingxaki esizichazileyo zinokusonjululwa ngokujonga ngokubanzi ukuba zeziphi iimpendulo ezichanekileyo. Kwaye sicinga ukuba ukuba oku kuyenzeka, ihlabathi liya kuba yindawo engcono.
umthombo: www.habr.com