Ukuqondanisa kwe-AI: Ukufanisa imigomo yobuhlakani bokwenziwa kufana namagugu omuntu

• About the Author:
• Igama lomlobi

  I-Quantumrun Foresight
• January 25, 2023

Ukuqondanisa kobuhlakani bokwenziwa (AI) kulapho imigomo yesistimu ye-AI ifana namanani abantu. Izinkampani ezifana ne-OpenAI, i-DeepMind, ne-Anthropic zinamaqembu abacwaningi okuwukuphela kwabo okugxilwe kuwo ukutadisha ama-guardrails ezimo ezihlukahlukene lapho lokhu kungase kwenzeke khona.

Umongo wokuqondanisa we-AI

Ngokocwaningo locwaningo lwango-2021 lweNyuvesi yaseCornell, izifundo ezimbalwa zibonise ukuthi amathuluzi noma amamodeli adalwe ngama-algorithms abonisa ukuchema okuvela kudatha aqeqeshwe kuyo. Isibonelo, ekucutshungulweni kolimi lwemvelo (NLP), khetha amamodeli e-NLP aqeqeshelwe amasethi edatha alinganiselwe abhalwe phansi enza izibikezelo ezisekelwe emibonweni eyingozi yobulili kwabesifazane. Ngokufanayo, olunye ucwaningo luthole ukuthi ama-algorithms aqeqeshelwe kusethi yedatha ephazanyisiwe abangele izincomo ezicwasa ngokwebala, ikakhulukazi emsebenzini wamaphoyisa.

Kunezibonelo eziningi lapho izinhlelo zokufunda zomshini zenze okubi kakhulu kwabancane noma amaqembu ahlushwa ukonakala okuningi. Ikakhulukazi, ukuhlaziywa kobuso okuzenzakalelayo kanye nokuxilongwa kwezempilo ngokuvamile akusebenzi kahle kwabesifazane nakubantu bemibala. Uma amasistimu abucayi okufanele asekelwe emaqinisweni nasekucabangeni esikhundleni semizwa esetshenziswa ezimweni ezifana nokwaba ukunakekelwa kwezempilo noma imfundo, angenza umonakalo omkhulu ngokwenza kube nzima ukuhlonza ukucabanga ngalezi zincomo.

Ngenxa yalokho, amafemu ezobuchwepheshe adala amaqembu wokuqondanisa i-AI ukuze agxile ekugcineni ama-algorithms alungile futhi anobuntu. Ucwaningo lubalulekile ekuqondeni ukuqondiswa kwezinhlelo ze-AI ezithuthukisiwe, kanye nezinselelo esingase sibhekane nazo njengoba amakhono e-AI ekhula.

Umthelela ophazamisayo

Ngokuka-Jan Leike, inhloko yokuqondanisa kwe-AI kwa-OpenAI (2021), uma kubhekwa ukuthi izinhlelo ze-AI sezikwazi ukusebenza ngeminyaka yawo-2010s, kuyaqondakala ukuthi ucwaningo oluningi lokuqondanisa kwe-AI belunzima kakhulu. Lapho amasistimu e-AI anamandla kakhulu eqondaniswe, enye yezinselelo abantu ababhekana nayo ukuthi le mishini ingase idale izixazululo eziyinkimbinkimbi kakhulu ukuthi zingabuyekezwa futhi zihlolwe uma zinengqondo ngokwezimiso zokuziphatha.

U-Leike usungule isu le-recursive reward modelling (RRM) ukuze alungise le nkinga. Nge-RRM, ama-AI "omsizi" ambalwa afundiswa ukusiza umuntu ukuthi ahlole ukuthi i-AI eyinkimbinkimbi isebenza kahle kangakanani. Unethemba lokuthi kungenzeka enze into ayibiza ngokuthi "i-MVP yokuqondanisa." Ngokwamagama okuqalisa, i-MVP (noma umkhiqizo omncane osebenzayo) iwumkhiqizo olula ongakha inkampani ukuze ihlole umbono. Ithemba ukuthi ngolunye usuku, i-AI ifanisa ukusebenza komuntu ekucwaningeni i-AI kanye nokuhambisana kwayo namanani ngenkathi isebenza futhi.

Yize intshisekelo ekhulayo yokuqondanisa i-AI iyinto enhle, abahlaziyi abaningi kulo mkhakha bacabanga ukuthi umsebenzi omningi "wezimiso zokuziphatha" kumalebhu aholayo e-AI uwubudlelwane bomphakathi obuklanyelwe ukwenza izinkampani zobuchwepheshe zibukeke zizinhle futhi zigweme ukubonakala okungekuhle. Laba bantu abalindele ukuthi izinqubo zokuthuthukiswa kokuziphatha zibe yinto eza kuqala kulezi zinkampani maduze.

Lokhu kuqaphelisisa kugqamisa ukubaluleka kwezindlela zemikhakha ehlukene emizamweni yokuqondanisa ukubaluleka, njengoba lena kuyindawo entsha uma kuqhathaniswa yophenyo lokuziphatha kanye nobuchwepheshe. Amagatsha ahlukene olwazi kufanele abe yingxenye ye-ajenda yocwaningo olubandakanya wonke umuntu. Lolu hlelo luphinde lukhombise isidingo sokuthi ochwepheshe nabakhi bezinqubomgomo bahlale besazi isimo sabo senhlalo nababambe iqhaza, njengoba izinhlelo ze-AI zithuthuka kakhulu.

Imithelela yokuqondanisa kwe-AI

Imithelela ebanzi yokuqondanisa kwe-AI ingase ihlanganise: 

• Amalebhu obuhlakani bokwenziwa aqasha amabhodi okuziphatha ahlukahlukene ukuze angamele amaphrojekthi futhi agcwalise imihlahlandlela ye-AI yokuziphatha. 
• Ohulumeni benza imithetho edinga izinkampani ukuthi zilethe uhlaka lwazo lwe-AI olunesibopho nokuthi zihlela kanjani ukuqhubeka nokuthuthukisa amaphrojekthi azo e-AI.
• Ukwenyuka kwezingxabano ekusetshenzisweni kwama-algorithms ekuqasheni abantu, ukubhekwa komphakathi, kanye nokugcinwa komthetho.
• Abacwaningi baxoshwa kumalebhu amakhulu e-AI ngenxa yokungqubuzana kwezintshisekelo phakathi kwezimiso zokuziphatha nezinjongo zebhizinisi.
• Ingcindezi eyengeziwe yokuthi ohulumeni balawule amasistimu e-AI athuthukile womabili anamandla amangalisayo kodwa angahle ephule amalungelo abantu.

Imibuzo ongaphawula ngayo

• Amafemu angenziwa kanjani aphendule ngezinhlelo ze-AI eziwadalayo?
• Yiziphi ezinye izingozi ezingaba khona uma kunokungalungi kahle kwe-AI?