I-Neuro-symbolic AI: Umshini ongagcina ubambe kokubili okunengqondo nokufunda

• About the Author:
• Igama lomlobi

  I-Quantumrun Foresight
• April 13, 2023

Ukufunda ngomshini (ML) bekulokhu kuwubuchwepheshe obuthembisayo nezinselele zabo ezihlukile, kodwa abacwaningi babheke ukudala isistimu esekelwe kumqondo eyeqa idatha enkulu. Amasistimu asekelwe enengqondo aklanyelwe ukusebenza ngezethulo ezingokomfanekiso nokucabanga, okunganikeza indlela esobala neqondakalayo yokuqonda inqubo yokuthatha izinqumo yesistimu. 

I-Neuro-symbolic AI umongo

I-Neuro-symbolic AI (ebuye ibizwe nge-composite AI) ihlanganisa amagatsha amabili obuhlakani bokwenziwa (AI). Okokuqala i-AI engokomfanekiso, esebenzisa izimpawu ukuqonda ubudlelwano nemithetho (okungukuthi, umbala nokuma kwento). Ukuze i-AI engokomfanekiso isebenze, isisekelo solwazi kufanele sinembile, sinemininingwane, futhi siphelele. Lesi sidingo sisho ukuthi alikwazi ukuzifundela ngokwalo futhi kuncike elwazini lomuntu ukuze liqhubeke nokuvuselela isisekelo solwazi. 

Enye ingxenye ye-Neuro-symbolic AI amanethiwekhi emizwa ajulile (amanethi ajulile) noma ukufunda okujulile (DL). Lobu buchwepheshe busebenzisa izendlalelo eziningi zamanodi ezilingisa ama-neuron obuchopho bomuntu ukuze sizifundele ukucubungula amasethi edatha amakhulu. Isibonelo, amanetha ajulile angadlula ezithombeni ezahlukene zamakati nezinja ukuze abone kahle ukuthi yikuphi, futhi athuthuka ngokuhamba kwesikhathi. Nokho, lokho amanetha ajulile angakwazi ukukwenza ukucubungula ubudlelwano obuyinkimbinkimbi. Ngokuhlanganisa i-AI engokomfanekiso namanetha ajulile, abacwaningi basebenzisa i-DL ukuze bahlanganise amanani amakhulu edatha kusisekelo solwazi, ngemva kwalokho i-AI engokomfanekiso ingakwazi ukunquma noma ukukhomba imithetho nobudlelwano. Le nhlanganisela ivumela ukutholwa kolwazi olusebenza kahle futhi olunembile nokwenza izinqumo.

Enye indawo amakheli e-Neuro-symbolic AI iyinqubo yokuqeqesha ebizayo yenetha. Ukwengeza, amanetha ajulile angazwela izinguquko ezincane zokufakwayo, okuholela emaphutheni okuhlukanisa. Baphinde babhekane nobunzima bokucabanga okungaqondakali kanye nokuphendula imibuzo ngaphandle kwedatha eminingi yokuqeqeshwa. Ngaphezu kwalokho, ukusebenza kwangaphakathi kwala manethiwekhi kuyinkimbinkimbi futhi kunzima ukuba abantu bakuqonde, okwenza kube inselele ukuhumusha ukucabanga okusemuva kokubikezela kwabo.

Umthelela ophazamisayo

Abacwaningi abavela eNyuvesi yaseStanford benze izifundo zokuqala ze-AI eyinhlanganisela besebenzisa izithombe ezingu-100,000 zomumo we-3D oyisisekelo (izikwele, ama-sphere, amasilinda, njll.) Babe sebesebenzisa imibuzo ehlukene ukuze baqeqeshe i-hybrid ukucubungula idatha kanye nobudlelwano be-infer (isb, ingabe amakhyubhu abomvu? ). Bathole ukuthi i-Neuro-symbolic AI ingaphendula le mibuzo ngendlela efanele ngamaphesenti angama-98.9 wesikhathi. Ukwengeza, i-hybrid idinga kuphela amaphesenti angu-10 wedatha yokuqeqeshwa ukuze kuthuthukiswe izixazululo. 

Njengoba izimpawu noma imithetho ilawula amanetha ajulile, abacwaningi bangabona kalula ukuthi “bafunda” kanjani nokuthi ukuwohloka kwenzeka kuphi. Ngaphambilini, lokhu kube obunye bobuthakathaka bamanetha ajulile, ukungakwazi ukulandelwa ngenxa yezendlalelo nezingqimba zamakhodi ayinkimbinkimbi kanye ne-algorithms. I-Neuro-symbolic AI iyahlolwa ezimotweni ezizishayelayo ukuze ibone izinto ezisemgwaqeni nanoma yiziphi izinguquko emvelweni. Bese iqeqeshelwa ukusabela ngendlela efanele kulezi zici zangaphandle. 

Kodwa-ke, kunemibono eyahlukene mayelana nokuthi inhlanganisela ye-AI engokomfanekiso namanetha ajulile iyindlela engcono kakhulu ebheke ku-AI ethuthuke kakhulu. Abanye abacwaningi, njengalabo abavela eNyuvesi yaseBrown, bakholelwa ukuthi le ndlela exubile ingase ingafani nezinga lokucabanga okungaqondakali elizuzwa izingqondo zabantu. Umqondo womuntu ungakha imidwebo engokomfanekiso yezinto futhi wenze izinhlobo ezahlukene zokucabanga usebenzisa lezi zimpawu, usebenzisa amanethiwekhi e-biological neural, ngaphandle kokudinga ingxenye engokomfanekiso ezinikele. Abanye ochwepheshe baphikisa ngokuthi ezinye izindlela, njengokwengeza izici kumanethi ajulile alingisa amakhono abantu, zingase zisebenze kangcono ekuthuthukiseni amakhono e-AI.

Izicelo ze-Neuro-symbolic AI

Ezinye izinhlelo zokusebenza ze-Neuro-symbolic AI zingafaka:

• Amabhothi, njengama-chatbots, angaqonda kangcono imiyalo yabantu nokugqugquzela, akhiqize izimpendulo nezinsiza ezinembe kakhudlwana.
• Isebenza ezimeni eziyinkimbinkimbi nezibucayi zokuxazulula izinkinga ezifana nokuxilongwa kwezokwelapha, ukuhlela ukwelashwa, nokuthuthukiswa kwezidakamizwa. Ubuchwepheshe bungasetshenziswa futhi ukusheshisa ucwaningo lwesayensi nobuchwepheshe emikhakheni efana nezokuthutha, amandla, kanye nokukhiqiza. 
• Ukuzenzakalela kwezinqubo zokwenza izinqumo okwamanje ezidinga ukwahlulela komuntu. Ngenxa yalokho, izinhlelo zokusebenza ezinjalo zingase ziholele ekulahlekelweni kozwela nokuziphendulela emikhakheni ethile efana nesevisi yamakhasimende.
• Imishini ehlakaniphile enembile kanye nabasizi ababonakalayo abangacubungula izimo ezihlukene, njengokulondoloza ugesi ngokuqhubekayo nokusebenzisa izinyathelo zokuphepha.
• Imibuzo emisha yezimiso zokuziphatha nezomthetho, njengezindaba eziphathelene nobumfihlo, ubunikazi, nokuzibophezela.
• Ukwenziwa kwezinqumo okuthuthukisiwe kuhulumeni nakwezinye izimo zezepolitiki. Lobu buchwepheshe bungasetshenziswa futhi ukuze kuthonye umbono womphakathi ngokukhangisa okuqondiswe kakhulu kanye nokukhiqizwa kwezikhangiso eziqondene nawe kakhulu nemidiya.

Imibuzo okufanele icatshangelwe

• Ucabanga ukuthi i-AI ye-neuro-symbolic izothinta kanjani impilo yethu yansuku zonke?
• Lobu buchwepheshe bungasetshenziswa kanjani kwezinye izimboni?