I-EU's Regulatory Wall Dibana ne-Meta's Decentralized Imihlakulo: AI's Local, Regulated, and Paradoxical future.
Ngelixa iBrussels yayixakekile ibhala incwadi yolawulo yekamva le-AI, iMeta ithe cwaka yatyibilika iimodeli ezinamandla ze-AI zesikali esibalulekileyo kwifomu elingana nedesika yakho. Qikelela ukuba yeyiphi eneneni eya kubuyisela umhlaba ngokukhawuleza? Umthetho we-EU's AI Act, ngoku uphila ngokusemthethweni, lirhamncwa elibalulekileyo, eliburhulumente, ubungqina bomnqweno welizwekazi lokuthomalalisa intshona yasendle yobukrelekrele bokwenziwa. Kodwa njengokuba i-inki isoma kwisakhelo sayo esibanzi, iintshukumo zamva nje ze-Meta zisongela ukunika i-swathes enkulu yayo inqabileyo, ukuba ayiphelelwa lixesha, ngokwenza i-AI ephucukileyo ifikeleleke ngokubanzi ngeendlela abalawuli bebengazilindelanga.
Masicace: Umthetho we-EU AI awukho uhlaziyo oluncinci lokuthotyelwa. Lutshintsho olusisiseko, ileviathan elawulayo eyilelwe ukuyalela ukuba iisistim ze-AI ziyilwe, ziphuhliswe, kwaye zisetyenziswe njani kwihlabathi jikelele [1]. Ukuba inkampani yakho isebenza ngaphakathi okanye isebenzisana ne-EU, ngoku ukwikhonkco. Oku akukho malunga neebhokisi zokuphawula; imalunga nokuyila kwakhona ipotifoliyo yakho ye-AI. Uhlela iinkqubo ngomngcipheko - ukusuka "kungamkelekanga" (cinga ngokufumana amanqaku entlalo) ukuya "kumngcipheko ophezulu" (izixhobo zonyango, iziseko ezingundoqo) - kunye nokuphumeza ulawulo oluqinileyo, ukungafihli, kunye nemilinganiselo yomgangatho wedatha [2]. Usilele ukuthobela, kwaye ujonge izohlwayo ukuya kuthi ga kwi-7% yengeniso yonyaka yehlabathi. Oko asikompama esihlahleni; oko kukunqunyulwa kweshishini.
Injongo intle, kunjalo. I-EU ifuna ukukhuthaza i-AI ethembekileyo, egxile ebantwini. Bafuna ukuthintela i-algorithmic bias, ukuqinisekisa ukuphendula, kunye nokukhusela amalungelo asisiseko. Le "i-Brussels Effect" ithetha ukuba imimiselo yabo ngokuqinisekileyo iya kuba ngumgangatho wehlabathi jikelele, inyanzelela ingqwalasela yeenqobo ezisesikweni kumbindi wophuhliso kuyo nayiphi na inkampani eneenjongo zehlabathi. Kwii-CTOs, oku kuthetha ukuba umaleko omtsha, obizayo, nontsokothileyo wolawulo lwe-AI kunye nobuchule bomthetho ngoku akukho ngxoxo. Iimaphu zendlela zemveliso yakho ziye zantsonkotha ngakumbi, kwaye iqela lakho lezomthetho liye laba namandla ngakumbi.
Ibhombu yoKwandisa amagunya
Kodwa ke kukho iMeta. Ngelixa oomatshini abalawulayo besenza isenzo, iMeta isandula ukuwisa iimodeli ezinamandla ze-AI, njengothotho lwayo lweLlama, olunokuphuculwa ukuze lusebenze ngokufanelekileyo kwihardware yomgangatho wabathengi, kubandakanya iGPU enye [3]. Yiyeke ingene ngaphakathi. Iimodeli zesikali esibalulekileyo, ezisebenza kwi-GPU enye. Le asiyompumelelo yobugcisa kuphela; yinyikima yeqhinga. Kangangeminyaka, ubulumko obukhoyo kukuba i-AI inamandla ngokwenene, ngakumbi iimodeli ezinkulu zolwimi kunye neearhente, ziya kuhlala ziyisizinda esikhethekileyo se-hyperscaler, evalelwe kumaziko amakhulu edatha, alambele amandla, afikeleleka kuphela ngeefowuni ze-API kunye namatyala amafu amakhulu. I-Meta ivele yavuthela loo modeli kwi-smithereens.
Oku akukho malunga nokuncitshiswa kweendleko kuphela, nangona oko kungumba omkhulu. Imalunga nokufikelela kwidemokhrasi kwi-AI ephucukileyo. Khawucinge ngesizukulwana esitsha sezixhobo zokusebenza ze-AI ezingasoloko zifowunela ekhaya kwilifu. Cinga ngabancedisi bakho be-AI abakuqonda ngokwenene ngaphandle kokuthumela idatha yakho esondeleyo kwifama yeseva eVirginia. Iinkqubo ezizinzisiweyo ezenza ixesha lokwenyani, izigqibo ezinzima ngaphandle kokubambezeleka. Izisombululo zeshishini apho ubumfihlo bedatha bubaluleke kakhulu, kwaye i-AI ayikhe ishiye indawo. Le ayisiyiyo nje "i-AI edge"; yi-AI* eyongamileyo yenkitha yabantu.
Iimpembelelo zokwenziwa kwezinto ziyamangalisa. Abaphuhlisi, abaqalisi, kunye nabaqaphi ngokuzimeleyo banokukha kwaye basebenzise iiarhente ezinamandla ze-AI ngaphandle kokufuna isifuba semfazwe eyinkunzi okanye umgca othe ngqo kwi-AWS. Olu tshintsho lunciphisa ukulibaziseka, lunciphisa iindleko zokusebenza, kwaye luguqula ngokusisiseko icalculus yabucala. Idatha ihlala kwindawo. Ulawulo luhlala kwindawo. Amandla aguquguqukayo ukusuka kubaboneleli belifu eliphakathi ukuya kwisixhobo, umsebenzisi, inethiwekhi yendawo.
IParadoksi yoLawulo
Apha kulapho ezi ngxelo zimbini zingqubana khona, zenza umxhelo owoyikekayo. Umthetho we-EU AI uyilwe ngobunono kwihlabathi apho i-AI iphuhliswe kakhulu kwaye isasazwe ngamaqumrhu amakhulu, abonakalayo, ahlala efini, apho ukuhamba kwedatha kunokulandeleka, kwaye ukuthotyelwa kunokunyanzeliswa kwiindawo eziminxekileyo. Yenzelwe ilizwe lolawulo olusembindini.
Kodwa kwenzeka ntoni xa iiarhente ezinamandla ze-AI zisebenza kwizigidi zee-GPUs, kwiilaptops, iinkqubo ezifakwe ngaphakathi, kunye neeseva zasekhaya, ezinokulungiswa, zilungiswe kakuhle, okanye zivuleleke kubasebenzisi? Uyiphicotha njani imodeli ephucukileyo esebenza kwi-GPU enye kwiofisi yamashishini amancinci angasemva? Unyanzelisa njani iimfuno zokungafihli xa umnombo wemodeli unokuba ngumsebenzi odibeneyo wamacandelo avulelekileyo kunye nohlengahlengiso lwendawo? Iindidi "ezinomngcipheko omkhulu" ngequbuliso ziba nzima kakhulu ukujonga nokulawula xa i-AI ingahlali kwindawo elawulwayo yamafu.
Isakhelo se-EU, ngelixa i-groundbreaking, ivakalelwa ngathi izama ukulawula umlambo ngokwakha amadama kwindawo yawo, kanye njengoko loo mlambo uza kuqhekeza iindonga zawo kwaye ukhukulise yonke indawo. I-"Brussels Effect" isenokuthi ibambe intambo phezu kokuthunyelwa kweenkampani ezinkulu, kodwa i-"Meta Effect" yokufikeleleka ngokubanzi, i-AI enamandla inokudala uqoqosho lwesithunzi esikhulu, esingaphantsi kokulawulwa kwezicelo ze-AI.
ICTO's Tightrope Walk
Kwii-CTOs kunye neenkokeli zobunjineli, oku kubonisa ukuhamba ngentambo enkulu. Kwelinye icala, umthetho we-EU AI ufuna ukuthotyelwa okungqongqo, okuxabisa kakhulu, kwaye kuhlala kucotha. Kufuneka * utyale imali kulawulo lwe-AI, uvavanyo lomngcipheko, kunye nezikhokelo zokuziphatha [4]. Amaqela akho asemthethweni kunye nokuthobela ngoku abalulekile kwisicwangciso sakho se-AI. Ukungahoyi oku kukuzibulala ngokwemali.
Kwelinye icala, iMeta isanda kuvula amasango ezikhukula kwizinto ezintsha ezingazange zibonwe ngaphambili emphethweni. Ithuba lokwakha ngokukhawuleza, ngexabiso eliphantsi, labucala ngakumbi, kwaye linamandla ngakumbi izicelo ze-AI likhulu. Inzuzo yokukhuphisana ye-AI yengingqi enamandla inokuba yinguqu [5]. Khawube nomfanekiso welizwe apho iirobhothi zakho zemizi-mveliso zifunda kwaye zilungelelanise kumgangatho wefektri ngaphandle kokuthumela idatha yokusebenza ebuthathaka kumntu wesithathu, okanye apho inkonzo yakho yabathengi i-AI ijongana nemibuzo entsonkothileyo ngokupheleleyo kwizakhiwo, ikhusela ubumfihlo bomsebenzisi ngokuyila.
Umceli mngeni kukudibanisa le mikhosi mibini. Uwamkela njani amandla asasaziweyo e-Edge AI ngelixa ngaxeshanye uqinisekisa ukuthotyelwa kwemithetho engqongqo, kwaye ihlala iphakathi kwamafu? Oku akuyongxaki yobugcisa kuphela; yeyobuchule. Ifuna uhlobo olutsha lomakhi we-AI-oqondayo kungekuphela nje ukwenziwa kwemodeli kunye nokusasazwa, kodwa kunye nezikhokelo zolawulo, ulongamo lwedatha, kunye nefuthe leenqobo ezisesikweni kwigranular, inqanaba lesixhobo.
Ikamva le-AI alikho malunga neemodeli ezinkulu okanye iitshiphusi ezikhawulezayo. Imalunga nokungavisisani okusisiseko phakathi kolawulo oluphakathi kunye negunya lokwabela amagunya. I-EU ifuna ukwakha igadi ekhuselekileyo, enokuxelwa kwangaphambili ye-AI. I-Meta isanda kunika wonke umntu imbewu yehlathi. Ukuhamba kulo mhlaba mtsha kuya kuchaza ishumi leminyaka elizayo lobunkokeli betekhnoloji.
Imithombo
[1] IPalamente yaseYurophu, iBhunga leManyano yaseYurophu - Ummiselo (EU) 2024/1689 kwi-Artificial Intelligence (AI Act) - https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A32024R1689 [2] PwC — Umthetho we-EU AI: Isikhokelo esisebenzayo sokuthotyelwa — https://www.pwc.com/gx/en/issues/data-privacy/eu-ai-act.html [3] I-Meta AI — Llama 2: Vula iSiseko esiVulekileyo kunye neeModeli zeNgxoxo eziLungelwe kakuhle — https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/ [4] Gartner — IiNdlela eziPhakamileyo zobuChwepheshe boBuchule ngo-2024: I-AI Trust, iRisk and Security Management (AI TRiSM) — https://www.gartner.com/en/articles/top-strategic-technology-trends-for-2024 [5] UMcKinsey kunye neNkampani - Amandla oqoqosho okuvelisa i-AI: Umda olandelayo wemveliso - https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier