I-EU's Regulatory Wall Ihlangabezana Nezimpophoma Zakwa-Meta Ezabelwe: Ikusasa Lendawo, Elilawulwayo, kanye ne-Paradoxical ye-AI
Ngenkathi i-Brussels imatasatasa ibhala i-rulebook yekusasa le-AI, i-Meta yashelela buthule amamodeli e-AI anamandla esikali esibalulekile efomini elilingana etafuleni lakho. Qagela ukuthi iyiphi ngempela ezolungisa kabusha umhlaba ngokushesha? Umthetho we-EU's AI Act, manje osebenza ngokusemthethweni, uyisilo esiyisikhumbuzo, esisemthethweni, ubufakazi besifiso sezwekazi sokulawula intshonalanga yasendle yobuhlakani bokwenziwa. Kodwa njengoba u-inki oma ohlakeni lwayo olubanzi, ukunyakaza kwakamuva kwe-Meta kusongela ukwenza ama-swathes amakhulu abe mahle, uma engasasebenzi, ngokwenza i-AI eyinkimbinkimbi ifinyeleleke kabanzi ngezindlela abalawuli ababengazilindele.
Masicacise: Umthetho we-EU AI awusona isibuyekezo esincane sokuthobela. Uguquko oluyisisekelo, ileviathan elawulayo eklanyelwe ukusho ukuthi amasistimu e-AI aklanywa kanjani, athuthukiswa kanjani, futhi asetshenziswe kanjani emhlabeni jikelele [1]. Uma inkampani yakho isebenza ngaphakathi noma isebenzisana ne-EU, manje usuyabamba iqhaza. Lokhu akuphathelene nokumaka amabhokisi; imayelana nokuqamba kabusha iphothifoliyo yakho ye-AI. Uhlela amasistimu ngezigaba ngengcuphe - kusukela "kokungamukeleki" (cabanga ngokuthola amaphuzu komphakathi) kuya "kubungozi obukhulu" (amadivayisi ezokwelapha, ingqalasizinda ebalulekile) - nokusebenzisa ukubusa okuqinile, ukungafihli, nezilinganiso zekhwalithi yedatha [2]. Yehlulekile ukuthobela, futhi ubheke izinhlawulo ezingafika ku-7% wenzuzo yonyaka yomhlaba wonke. Lokho akusikho impama esihlakaleni; lokho kuwukunqunywa kwezitho zenkampani.
Inhloso yinhle impela. I-EU ifuna ukukhuthaza i-AI ethembekile, egxile kumuntu. Bafuna ukuvimbela ukuchema kwe-algorithmic, ukuqinisekisa ukuziphendulela, nokuvikela amalungelo ayisisekelo. Lokhu "Brussels Effect" kusho imithetho yabo nakanjani izoba indinganiso yomhlaba wonke, ephoqelela ukucatshangelwa kwezimiso zokuziphatha emnyombeni wentuthuko yanoma iyiphi inkampani enezifiso zamazwe ngamazwe. Kuma-CTO, lokhu kusho ungqimba olusha, olubizayo, noluyinkimbinkimbi lokubusa kwe-AI nobungcweti bezomthetho manje akusaxoxiswana. Imephu yemigwaqo yomkhiqizo wakho isiye yaba nzima kakhulu, futhi ithimba lakho lezomthetho lisanda kuba namandla kakhulu.
Ibhomu Lokwehlukaniswa Kwamazwe
Kodwa-ke kukhona iMeta. Ngenkathi umshini olawulayo usasebenza, i-Meta isanda kwehlisa amamodeli e-AI anamandla, njengochungechunge lwayo lwe-Llama, olungathuthukiswa ukuze lusebenze kahle ku-hardware yezinga lomthengi, okuhlanganisa ne-GPU eyodwa [3]. Vumela lokho kungene phakathi. Amamodeli esikali esibalulekile, asebenza nge-GPU eyodwa. Lokhu akuyona nje impumelelo yobuchwepheshe; kuwukuzamazama komhlaba okunamasu. Sekuyiminyaka, ukuhlakanipha okukhona kube ukuthi i-AI enamandla ngempela, ikakhulukazi amamodeli nama-ejenti wolimi amakhulu, izohlala iyisizinda esikhethekile sama-hyperscaler, evalelwe ezindaweni ezinkulu zedatha, ezilambele amandla, ezifinyeleleka kuphela ngezingcingo ze-API nezikweletu ezinkulu zamafu. I-Meta isanda kushaya leyo modeli kuma-smithereens.
Lokhu akukhona nje mayelana nokunciphisa izindleko, nakuba lokho kuyisici esikhulu. Imayelana nokwenza intando yeningi ukufinyelela ku-AI esezingeni eliphezulu. Cabanga ngesizukulwane esisha sezinhlelo zokusebenza ze-AI ezikudivayisi ezingahlali zishayela ekhaya kumafu. Cabanga ngabasizi be-AI bomuntu siqu abakuqonda ngempela ngaphandle kokuthumela idatha yakho eseduze kakhulu epulazini leseva eVirginia. Amasistimu ashumekiwe enza izinqumo zesikhathi sangempela, eziyinkimbinkimbi ngaphandle kokubambezeleka. Izixazululo zebhizinisi lapho ubumfihlo bedatha bubaluleke kakhulu, futhi i-AI ayilokothi ishiye izakhiwo. Lokhu akuwona nje "unqenqema lwe-AI"; iyi-AI ebusayo yoquqaba.
Imiphumela yokuqamba izinto ezintsha iyamangalisa. Onjiniyela, abaqalayo, ngisho nabaqambi ngabanye manje sebengakwazi ukwakha futhi basebenzise ama-agent anamandla e-AI ngaphandle kokudinga isifuba sempi yebhizinisi noma umugqa oqondile ku-AWS. Lokhu kushintsha kunciphisa ukubambezeleka, kunciphisa izindleko zokusebenza, futhi kuguqule ngokuyisisekelo isibalo sobumfihlo. Idatha ihlala isendaweni. Ukulawula kuhlala kusendaweni. Amandla aguqukayo asuka kubahlinzeki bamafu amaphakathi aye kudivayisi, umsebenzisi, inethiwekhi yendawo.
Indida Yokulawula
Nakhu lapho kungqubuzana khona lezi zindaba ezimbili, kudala indida ethakazelisayo, eyesabekayo. Umthetho we-EU AI uklanywe ngobunono ukuze wenze umhlaba lapho i-AI ithuthukiswa kakhulu futhi isetshenziswe amabhizinisi amakhulu, angakhonjwa, ngokuvamile emafini, lapho ukugeleza kwedatha kulandeleka, futhi ukuthobela kungaphoqelelwa ezindaweni eziminyanisayo. Yakhelwe umhlaba wokulawula okumaphakathi.
Kodwa kwenzekani lapho ama-ejenti e-AI anamandla esebenza ezigidini zama-GPU ngamanye, kumakhompyutha aphathekayo, amasistimu ashumekiwe, namaseva endawo, okungenzeka alungiswe, ashunwe kahle, noma avuleleke ngabasebenzisi? Uyihlola kanjani imodeli esezingeni eliphezulu esebenza ku-GPU eyodwa ehhovisi elingemuva lebhizinisi elincane? Uziphoqelela kanjani izimfuneko zokubonisa ngale lapho uhlu lwemodeli lungase lube isiqephu sezingxenye zomthombo ovulekile kanye nokujwayela kwendawo? Izigaba "ezinobungozi obukhulu" ngokuzumayo ziba nzima kakhulu ukuqapha nokulawula lapho i-AI ingahlali endaweni elawulwayo yamafu.
Uhlaka lwe-EU, ngenkathi lusaqala, luzwakala sengathi luzama ukulawula umfula ngokwakha amadamu emthonjeni wawo, njengoba nje lowo mfula usuzoqhuma usebe futhi ukhukhule yonke indawo. "I-Brussels Effect" ingahle ibambe amandla phezu kokuthunyelwa kwezinkampani ezinkulu, kodwa "i-Meta Effect" ye-AI efinyeleleka kabanzi, enamandla ingase idale umnotho omkhulu, ongalawulwa kancane wezinhlelo ze-AI.
I-CTO's Tightrope Walk
Kuma-CTO nabaphambili bonjiniyela, lokhu kuveza uhambo olukhulu lwentambo eqinile. Ngakolunye uhlangothi, uMthetho we-EU AI ufuna ukulandelwa okuqinile, okumba eqolo, futhi okuvamise ukunensa. Kumelwe * utshale imali ekubuseni kwe-AI, ukuhlola ubungozi, nezinhlaka zokuziphatha [4]. Amaqembu akho ezomthetho nokuthobela manje aseyingxenye yesu lakho le-AI. Ukuziba lokhu ukuzibulala ngokwezimali.
Ngakolunye uhlangothi, i-Meta isanda kuvula amasango ezikhukhula emisha engakaze ibonwe emaphethelweni. Ithuba lokwakha izinhlelo zokusebenza ze-AI ezisheshayo, ezishibhile, eziyimfihlo, neziqine kakhudlwana likhulu kakhulu. Inzuzo yokuncintisana yokusebenzisa i-AI yendawo enamandla ingaba yinguquko [5]. Cabanga ngezwe lapho amarobhothi ezimboni zakho afunda khona futhi azivumelanise nezimo endaweni yemboni ngaphandle kokuthumela idatha yokusebenza ebucayi enkampanini yangaphandle, noma lapho isevisi yakho yamakhasimende i-AI isingatha khona imibuzo eyinkimbinkimbi yonke indawo, ivikela ubumfihlo bomsebenzisi ngedizayini.
Inselele iwukuhlanganisa la mabutho amabili. Uwamukela kanjani amandla asabalalisiwe onqenqema lwe-AI kuyilapho ngasikhathi sinye uqinisekisa ukuthobela imithetho eqinile, futhi evamise ukuhlanganisa amafu? Lena akuyona nje inkinga yobuchwepheshe; ingeyamasu. Idinga uhlobo olusha lomakhi we-AI - ongaqondi nje ukwenziwa ngcono kwemodeli nokusetshenziswa, kodwa futhi nezinhlaka zokulawula, ubukhosi bedatha, nemithelela yezimiso zokuziphatha ezingeni eliyimbudumbudu, ledivayisi.
Ikusasa le-AI aligcini nje ngamamodeli amakhulu noma ama-chips asheshayo. Kumayelana nokungezwani okuyisisekelo phakathi kokulawula okumaphakathi namandla ahlukaniswe. I-EU ifuna ukwakha ingadi ephephile, ebikezelwa ye-AI. I-Meta isanda kunikeza wonke umuntu imbewu yehlathi. Ukuzulazula kulo mkhakha omusha kuzochaza ishumi leminyaka elilandelayo lobuholi bezobuchwepheshe.
Imithombo
[1] I-European Parliament, Council of the European Union — Regulation (EU) 2024/1689 on Artificial Intelligence (AI Act) — https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A32024R1689 [2] I-PwC — Umthetho we-EU AI: Umhlahlandlela osebenzayo wokuhambisana — https://www.pwc.com/gx/en/issues/data-privacy/eu-ai-act.html [3] I-Meta AI — Llama 2: Isisekelo Esivulekile kanye Nezinhlobo Zengxoxo Eshuniwe Kahle — https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/ [4] U-Gartner — Amathrendi Aphezulu Obuchwepheshe Amaqhinga ka-2024: I-AI Trust, Ukulawulwa Kwengozi Nokuphepha (AI TRiSM) — https://www.gartner.com/en/articles/top-strategic-technology-trends-for-2024 [5] McKinsey & Company — Amandla omnotho we-AI ekhiqizayo: Umngcele olandelayo wokukhiqiza — https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier