I-AI iyithinta kanjani imvelo?

I-AI iyithinta kanjani imvelo?

Impendulo emfushane: I-AI ithinta imvelo ikakhulukazi ngokusebenzisa ugesi ezikhungweni zedatha (kokubili ukuqeqeshwa kanye nokucabanga kwansuku zonke), kanye namanzi okupholisa, kanye nemiphumela evezwe ekukhiqizweni kwehadiwe kanye nokuchitha i-e. Uma ukusetshenziswa kufinyelela ezigidini zemibuzo, ukuphetha kungadlula ukuqeqeshwa; uma amagridi ehlanzekile futhi izinhlelo zisebenza kahle, imiphumela iyawa ngenkathi izinzuzo zingakhula.

Izinto ezibalulekile okufanele uzicabangele:

Ugesi : Landelela ukusetshenziswa kwekhompyutha; ukukhishwa kwegesi kwehla lapho umthwalo wemisebenzi usebenza kumagridi ahlanzekile.

Amanzi : Izinketho zokupholisa zishintsha imiphumela; izindlela ezisekelwe emanzini zibaluleke kakhulu ezindaweni ezingavamile.

Ihadiwe : Ama-chip namaseva anemiphumela ebalulekile; andisa isikhathi sokuphila futhi abeka phambili ukulungiswa.

Ukubuyela emuva : Ukusebenza kahle kungakhuphula isidingo esiphelele; kukala imiphumela, hhayi nje izinzuzo zomsebenzi ngamunye.

Izikhombisi zokusebenza : Amamodeli osayizi wesokudla, athuthukise ukuqagela, futhi abike izilinganiso zesicelo ngasinye ngokusobala.

I-AI iyithinta kanjani imvelo?

Izihloko ongase uthande ukuzifunda ngemva kwalesi:

🔗 Ingabe i-AI imbi emvelweni?
Hlola umthelela we-AI wekhabhoni, ukusetshenziswa kukagesi, kanye nezidingo zesikhungo sedatha.

🔗 Kungani i-AI ingalungile emphakathini?
Bheka ubandlululo, ukuphazamiseka kwemisebenzi, ulwazi olungalungile, kanye nokwanda kokungalingani kwezenhlalo.

🔗 Kungani i-AI imbi? Uhlangothi olumnyama lwe-AI
Qonda izingozi ezifana nokuqapha, ukukhohlisa, kanye nokulahlekelwa ukulawula kwabantu.

🔗 Ingabe i-AI idlulele kakhulu?
Izingxoxo mayelana nokuziphatha, imithethonqubo, kanye nalapho ukusungula izinto ezintsha kufanele kudwebe khona imingcele.

Indlela i-AI ethinta ngayo imvelo: isithombe esisheshayo ⚡🌱

Uma ukhumbula amaphuzu ambalwa kuphela, cabanga ngalawa:

Bese kuba khona ingxenye abantu abayikhohlwayo: isikali . Umbuzo owodwa we-AI ungaba mncane, kodwa izigidigidi zawo ziyisilwane esihluke ngokuphelele… njengebhola leqhwa elincane ngandlela thile eliba yiqhwa elingangesofa. (Leso sifaniso siphambene kancane, kodwa uyaqonda.) I-IEA: Amandla kanye ne-AI

Umthelela we-AI kwezemvelo akuyona into eyodwa - kuyisitaki 🧱🌎

Uma abantu bephikisana nge-AI kanye nokusimama, bavame ukuxoxa ngale komunye nomunye ngoba bakhomba ezingqimbeni ezahlukene:

1) Ugesi wekhompyutha

  • Ukuqeqesha amamodeli amakhulu kungadinga amaqoqo amakhulu asebenza kanzima isikhathi eside. I-IEA: Amandla kanye ne-AI

  • Ukuqagela (ukusetshenziswa kwansuku zonke) kungaba yisisekelo esikhulu ngokuhamba kwesikhathi ngoba kwenzeka njalo, yonke indawo. I-IEA: Amandla kanye ne-AI

2) Izindleko zesikhungo sedatha

3) Amanzi nokushisa

4) Uchungechunge lokuhlinzekwa kwehadiwe

5) Imiphumela yokuziphatha kanye nokubuyela emuva

Ngakho-ke uma umuntu ebuza ukuthi i-AI iyithinta kanjani iMvelo, impendulo eqondile ithi: kuncike ekutheni ulinganisa luphi ungqimba, nokuthi kusho ukuthini ukuthi “i-AI” kuleso simo.

Ukuqeqeshwa vs ukuphetha: umehluko oshintsha konke 🧠⚙️

Abantu bathanda ukukhuluma ngokuqeqeshwa ngoba kuzwakala kuyisimanga - "imodeli eyodwa isebenzise amandla e-X." Kodwa isiphetho yisona sikhulu esithule. I-IEA: Amandla kanye ne-AI

Ukuqeqeshwa (ukwakha okukhulu)

Ukuqeqeshwa kufana nokwakha ifektri. Ukhokha izindleko kusengaphambili: ukubala okunzima, izikhathi zokusebenza ezinde, ukugijima okuningi kokuzama nokuphutha (futhi yebo, ukuphindaphinda okuningi "oops okungazange kusebenze, zama futhi"). Ukuqeqeshwa kungathuthukiswa, kodwa kusengaba yinto enkulu. I-IEA: Amandla kanye ne-AI

Isiphetho (ukusetshenziswa kwansuku zonke)

Isiphetho sifana nefektri esebenza nsuku zonke, kuwo wonke umuntu, ngezinga:

  • Ama-Chatbot aphendula imibuzo

  • Ukukhiqizwa kwezithombe

  • Izinga lokusesha

  • Izincomo

  • Inkulumo ibe umbhalo

  • Ukutholwa kokukhwabanisa

  • Abashayeli bezindiza abakopishayo kumadokhumenti namathuluzi ekhodi

Ngisho noma isicelo ngasinye sincane kakhulu, umthamo wokusetshenziswa unganciphisa ukuqeqeshwa. Yisimo sakudala esithi “utshani olulodwa alulutho, utshani obuyisigidi buyinkinga”. I-IEA: Amandla kanye ne-AI

Inothi elincane - eminye imisebenzi ye-AI inzima kakhulu kuneminye. Ukukhiqiza izithombe noma amavidiyo amade kuvame ukudinga amandla amaningi kunokuhlela umbhalo omfushane. Ngakho ukufaka i-“AI” ebhakedeni elilodwa kufana nokuqhathanisa ibhayisikili nomkhumbi wezimpahla bese uzibiza zombili ngokuthi “ezokuthutha.” I-IEA: Amandla kanye ne-AI

Izikhungo zedatha: ugesi, ukupholisa, kanye naleyo ndaba yamanzi athule 💧🏢

Izikhungo zedatha azizintsha, kodwa i-AI ishintsha ubukhali. Ama-accelerator asebenza kahle angadonsa amandla amaningi ezindaweni eziqinile, okuphenduka ukushisa, okumele kulawulwe. I-LBNL (2024): Umbiko Wokusetshenziswa Kwamandla Esikhungo Sedatha sase-United States (PDF) IEA: Amandla kanye ne-AI

Izisekelo zokupholisa (ezilula, kodwa ezisebenzayo)

Yilokho okuphambene nalokho: ngezinye izikhathi unganciphisa ukusetshenziswa kukagesi ngokuthembela ekupholiseni okusekelwe emanzini. Kuye ngokuntuleka kwamanzi endawo, lokho kungaba kuhle… noma kungaba yinkinga yangempela. ULi et al. (2023): Ukwenza i-AI “Ingabi Nokoma” Kakhulu (PDF)

Futhi, ukuthambekela kwemvelo kuncike kakhulu ku:

Uma sikhuluma iqiniso: ingxoxo yomphakathi ivame ukuphatha “isikhungo sedatha” njengebhokisi elimnyama. Akukubi, akusilo imilingo. Kuyingqalasizinda. Kuziphatha njengengqalasizinda.

Ama-chips kanye ne-hardware: ingxenye abantu abayigwemayo ngoba ayithandeki kakhulu 🪨🔧

I-AI iphila ngehadiwe. Ihadiwe inomjikelezo wokuphila, futhi imiphumela yomjikelezo wokuphila ingaba mikhulu. I-US EPA: Imboni Ye-Semiconductor I-ITU: I-Global E-waste Monitor 2024

Lapho kuvela khona umthelela wemvelo

Amaseva e-e-waste kanye “asebenza kahle kakhulu”

Umonakalo omkhulu wemvelo awuveli kudivayisi eyodwa ekhona - ubangelwa ukuyishintsha kusenesikhathi ngoba ayisasebenzi kahle. I-AI iyasheshisa lokhu ngoba ukusebenza kahle kungaba kukhulu. Isilingo sokuvuselela ihadiwe singokoqobo. I-ITU: I-Global E-waste Monitor 2024

Iphuzu eliwusizo: ukukhulisa impilo yehadiwe, ukuthuthukisa ukusetshenziswa, kanye nokuvuselela kungabaluleka njenganoma yikuphi ukulungiswa kwemodeli okumangalisayo. Ngezinye izikhathi i-GPU eluhlaza kakhulu yileyo ongayithengi. (Lokho kuzwakala njengesiqubulo, kodwa futhi... kuyiqiniso ngandlela thile.)

Indlela i-AI ethinta ngayo imvelo: indlela yokuziphatha "abantu bayakhohlwa lokhu" 🔁😬

Nansi ingxenye yezenhlalo exakile: I-AI yenza izinto zibe lula, ukuze abantu benze izinto eziningi. Lokho kungaba kuhle kakhulu - ukukhiqiza okwengeziwe, ubuhlakani obuningi, ukufinyelela okwengeziwe. Kodwa kungasho futhi ukusetshenziswa kwezinsizakusebenza kabanzi. I-OECD (2012): Izinzuzo Eziningi Zokuthuthukiswa Kokusetshenziswa Kahle Kwamandla (PDF)

Izibonelo:

  • Uma i-AI yenza ukukhiqizwa kwamavidiyo kube lula, abantu bakhiqiza amavidiyo amaningi.

  • Uma i-AI yenza ukukhangisa kusebenze kangcono, izikhangiso eziningi ziyakhonzwa, ukuzibandakanya okuningi kuyajikeleza.

  • Uma i-AI yenza ukuthunyelwa kwempahla kube ngcono, ukuhweba nge-e-commerce kungakhula kakhulu.

Lesi akusona isizathu sokwesaba. Kuyisizathu sokulinganisa imiphumela, hhayi ukusebenza kahle nje kuphela.

Isifaniso esingaphelele kodwa esijabulisayo: Ukusebenza kahle kwe-AI kufana nokunikeza intsha ifriji enkulu - yebo, indawo yokugcina ukudla iyathuthuka, kodwa ngandlela thile ifriji isuke ingenalutho futhi ngosuku olulodwa. Akuyona isifaniso esiphelele, kodwa… ukubonile kwenzeka 😅

Okuhle: I-AI ingasiza ngempela imvelo (uma iqondiswe kahle) 🌿✨

Manje ngengxenye ethathwa kancane: I-AI inganciphisa ukukhishwa kwegesi kanye nemfucuza ezinhlelweni ezikhona... empeleni, ezingafaneleki. I-IEA: I-AI yokwenza ngcono amandla kanye nokusungula izinto ezintsha

Izindawo lapho i-AI ingasiza khona

Iphuzu elibalulekile: "Ukusiza" kwe-AI akuvimbeli ngokuzenzakalela ukunyatheliswa kwe-AI. Kuya ngokuthi i-AI isetshenzisiwe ngempela, isetshenziswa ngempela, nokuthi iholela ekunciphiseni kwangempela yini kunokuba kube amadeshibhodi angcono nje. Kodwa yebo, amandla akhona angokoqobo. I-IEA: I-AI yokwenza ngcono amandla kanye nokusungula izinto ezintsha

Yini eyenza inguqulo enhle ye-AI enobungani nemvelo? ✅🌍

Lesi yisigaba esithi “kulungile-ke kufanele senzeni”. Ukusethwa okuhle kwe-AI okunesibopho semvelo kuvame ukuba nalokhu:

  • Sula inani lokusetshenziswa : Uma imodeli ingashintshi izinqumo noma imiphumela, kumane nje kuyi-computing ekhangayo.

  • Ukulinganisa okufakwe ku : Amandla, ukulinganisa kwekhabhoni, ukusetshenziswa, kanye nezilinganiso zokusebenza kahle ezilandelwa njenganoma iyiphi enye i-KPI. Ikhodi yeKhabhoni: Indlela Yokusebenza

  • Amamodeli anosayizi ofanele : Sebenzisa amamodeli amancane uma amamodeli amancane esebenza. Akusikho ukwehluleka kokuziphatha ukusebenza kahle.

  • Umklamo wokuqonda osebenzayo : ukugcinwa kwesikhashana, ukuhlanganisa, ukulinganisa, ukubuyisa, kanye namaphethini amahle okukhuthaza. UGholami et al. (2021): Ucwaningo Lwezindlela Zokulinganisa (i-PDF) ULewis et al. (2020): Isizukulwane Esingeziwe Sokubuyisa

  • Ukuqwashisa ngehadiwe nendawo : sebenzisa imithwalo yemisebenzi lapho igridi ihlanzekile khona nengqalasizinda isebenza kahle (uma kungenzeka). I-Carbon Intensity API (GB)

  • Impilo ende yehadiwe : sebenzisa kahle, uyisebenzise kabusha, futhi uyilungisa kabusha. I-ITU: I-Global E-waste Monitor 2024

  • Ukubika okuqondile : gwema ulimi oluhlanza imvelo kanye nezimangalo ezingacacile ezifana nokuthi “i-AI enobungani nemvelo” ngaphandle kwezinombolo.

Uma usalandela ukuthi i-AI ithinta kanjani iMvelo, yileli phuzu lapho impendulo iyeka khona ukuba yifilosofi bese iqala ukusebenza: iyithinta ngokusekelwe ezinqumweni zakho.

Ithebula Lokuqhathanisa: amathuluzi nezindlela ezinciphisa umthelela 🧰⚡

Ngezansi kunethebula elisheshayo nelisebenzayo. Aliphelele, futhi yebo, amaseli ambalwa anemibono ethile… ngoba yindlela ukukhetha amathuluzi kwangempela okusebenza ngayo.

Ithuluzi / Indlela Izithameli Intengo Kungani kusebenza
Amalabhulali okulandelela i-carbon/amandla (izilinganiso zesikhathi sokusebenza) Amaqembu e-ML Mahhala Inikeza ukubonakala - okuyingxenye yempi, noma ngabe izilinganiso azicacile kahle.. IkhodiCarbon
Ukuqapha amandla ehadiwe (i-GPU/CPU telemetry) I-Infra + ML Mahhala Ilinganisa ukusetshenziswa kwangempela; ilungele ukulinganisa ukusebenza (ayikhazimuli kodwa igolide)
Ukuhluzwa kwemodeli Onjiniyela be-ML Mahhala (izindleko zesikhathi 😵) Amamodeli amancane abafundi avame ukufanisa ukusebenza nezindleko eziphansi kakhulu zokucabanga UHinton et al. (2015): Ukuhlanza Ulwazi Kunethiwekhi Yezinzwa
Ukulinganisa (ukuphetha okuphansi kokunemba) ML + umkhiqizo Mahhala Inciphisa ukubambezeleka kanye nokusetshenziswa kwamandla; ngezinye izikhathi ngezinguquko ezincane zekhwalithi, ngezinye izikhathi azikho UGholami et al. (2021): Ucwaningo Lwezindlela Zokulinganisa (PDF)
Ukugcina isikhashana + ukuphetha nge-batching Umkhiqizo + ipulatifomu Mahhala Kunciphisa ukubala okungadingekile; ikakhulukazi kuyasiza ezicelweni eziphindaphindwayo noma izicelo ezifanayo
Isizukulwane esingeziwe sokubuyiswa (i-RAG) Amaqembu ohlelo lokusebenza Kuxutshiwe Ilayisha "inkumbulo" ukuze ibuyiselwe; inganciphisa isidingo samafasitela amakhulu omongo ULewis et al. (2020): Isizukulwane Esithuthukisiwe Sokubuyiswa
Ukuhlela imithwalo yemisebenzi ngokuqina kwekhabhoni I-Infra/ops Kuxutshiwe Ishintsha imisebenzi eguquguqukayo ibe amafasitela kagesi ahlanzekile - kodwa idinga ukuhlanganiswa I-Carbon Intensity API (GB)
Ukugxila ekusebenzeni kahle kwesikhungo sedatha (ukusetshenziswa, ukuhlanganiswa) Ubuholi be-IT Kukhokhwa (ngokuvamile) Isivikelo esingesihle kangako, kodwa ngokuvamile esikhulu kunazo zonke - yeka ukusebenzisa izinhlelo ezingenalutho Igridi Eluhlaza: PUE
Amaphrojekthi okusebenzisa kabusha ukushisa Izinsiza Kuya ngesimo Kuguqula ukushisa okulahliwe kube yinani; akunakwenzeka ngaso sonke isikhathi, kodwa uma kwenzeka, kuba kuhle kancane
"Ingabe sidinga ngisho ne-AI lapha?" hlola Wonke umuntu Mahhala Kuvimbela ukubala okungenamsebenzi. Ukuthuthukisa okunamandla kakhulu ukuthi cha (ngezinye izikhathi)

Uyabona ukuthi yini engekho? “Thenga isitika esiluhlaza okomlingo.” Leso asikho 😬

Incwadi yokudlala ewusizo: ukunciphisa umthelela we-AI ngaphandle kokubulala umkhiqizo 🛠️🌱

Uma wakha noma uthenga izinhlelo ze-AI, nansi indlela engokoqobo esebenza ngayo:

Isinyathelo 1: Qala ngokulinganisa

  • Landelela ukusetshenziswa kwamandla noma ukulinganise njalo. IkhodiCarbon: Indlela Yokusebenza

  • Kala ngokugijima kokuqeqeshwa kanye nesicelo sokuphetha.

  • Ukusetshenziswa kwe-monitha - izinsiza ezingasebenzi zinendlela yokucasha endaweni esobala. I-Green Grid: PUE

Isinyathelo 2: Usayizi ofanele wemodeli emsebenzini

  • Sebenzisa amamodeli amancane ukuze uhlukanise, ukhiphe, futhi uqondise.

  • Londoloza imodeli esindayo yezikhwama ezinzima.

  • Cabanga "nge-cascade yemodeli": imodeli encane kuqala, imodeli enkulu kuphela uma kudingeka.

Isinyathelo 3: Lungiselela ukuphetha (lapha yilapho ukulunywa khona kwesikali)

  • Ukugcina isikhashana : gcina izimpendulo zemibuzo ephindaphindwayo (ngokulawula ubumfihlo ngokucophelela).

  • Ukuhlanganisa : izicelo zeqembu zokuthuthukisa ukusebenza kahle kwehadiwe.

  • Imiphumela emifushane : imiphumela emide ibiza kakhulu - ngezinye izikhathi awudingi indatshana.

  • Isiyalo esisheshayo : izeluleko ezingahlelekile zidala izindlela zokubala ezinde… futhi yebo, amathokheni engeziwe.

Isinyathelo 4: Thuthukisa ukuhlanzeka kwedatha

Lokhu kuzwakala kungahlobene, kodwa akuhlobene:

  • Amasethi edatha ahlanzekile anganciphisa ukuphindaphinda kokuqeqeshwa.

  • Umsindo omncane usho ukuthi izivivinyo ezimbalwa kanye nokugijima okunciphile okulahliwe.

Isinyathelo 5: Phatha ihadiwe njengempahla, hhayi into engasetshenziswa

  • Yandisa imijikelezo yokuvuselela lapho kungenzeka khona. I-ITU: I-Global E-waste Monitor 2024

  • Sebenzisa kabusha ihadiwe endala ukuze uthole imisebenzi elula.

  • Gwema ukunikeza "izinto ezisezingeni eliphezulu njalo".

Isinyathelo 6: Khetha ngokuhlakanipha ukusetha i-oda

  • Yenza imisebenzi eguquguqukayo lapho amandla ehlanzekile khona uma ungakwazi. I-Carbon Intensity API (GB)

  • Nciphisa ukuphindaphindwa okungadingekile.

  • Gcina imigomo yokubambezeleka ingokoqobo (ukubambezeleka okuphansi kakhulu kungaphoqa ukusethwa okuhlala kuvuliwe okungasebenzi kahle).

Futhi yebo... ngezinye izikhathi isinyathelo esingcono kakhulu simane nje: ungasebenzisi ngokuzenzakalelayo imodeli enkulu kunazo zonke isenzo somsebenzisi ngamunye. Lowo mkhuba ufana nendawo ezungezile nokushiya zonke izibani zikhanya ngoba ukuya eswishini kuyacasula.

Izinganekwane ezivamile (nokuthi yini eseduze neqiniso) 🧠🧯

Inganekwane: "I-AI ihlale imbi kakhulu kunesofthiwe yendabuko"

Iqiniso: I-AI ingaba nzima kakhulu ekubaleni, kodwa ingathatha indawo yezinqubo ezingasebenzi kahle ngesandla, inciphise imfucuza, futhi yenze ngcono izinhlelo. Kuyisimo. I-IEA: I-AI yokwenza ngcono amandla kanye nokusungula izinto ezintsha

Inganekwane: "Ukuqeqeshwa kuphela inkinga"

Iqiniso: Ukuqagela ngezinga kungabusa ngokuhamba kwesikhathi. Uma umkhiqizo wakho usetshenziswa kakhulu, lokhu kuba indaba eyinhloko. I-IEA: Amandla kanye ne-AI

Inganekwane: “Izinto ezivuselelekayo ziyakuxazulula ngokushesha”

Iqiniso: Ugesi ohlanzekile uyasiza kakhulu, kodwa awususi umkhondo wehadiwe, ukusetshenziswa kwamanzi, noma imiphumela yokubuyela emuva. Kodwa-ke kubalulekile. I-IEA: Amandla kanye ne-AI

Inganekwane: “Uma kusebenza kahle, kuyasimeme”

Iqiniso: Ukusebenza kahle ngaphandle kokulawulwa kwesidingo kusengandisa umthelela ophelele. Yilokho okubangela ukuphinda kubuye. I-OECD (2012): Izinzuzo Eziningi Zokuthuthukiswa Kokusetshenziswa Kahle Kwamandla (PDF)

Ukubusa, ukucaca, nokungadlali ngakho 🧾🌍

Uma uyinkampani, yilapho ukwethembana kwakheka khona noma kulahleka khona.

Lena yingxenye lapho abantu bebheka khona, kodwa kubalulekile. Ubuchwepheshe obunomthwalo wemfanelo abugcini nje ngobunjiniyela obuhlakaniphile. Buphinde bube mayelana nokungenzi sengathi akukho ukuhwebelana.

Isifinyezo sokuphetha: isifinyezo esincane sendlela i-AI ethinta ngayo iMvelo 🌎✅

Indlela i-AI ethinta ngayo iNdawo incike emthwalweni owengeziwe: ugesi, amanzi (ngezinye izikhathi), kanye nesidingo sehadiwe. I-IEA: Amandla kanye ne-AI Li et al. (2023): Ukwenza i-AI “Ingabi Nokoma” Kancane (PDF) Iphinde inikeze amathuluzi anamandla okunciphisa ukukhishwa kwegesi kanye nokulahlwa kwemfucuza kweminye imikhakha. I-IEA: I-AI yokwenza ngcono amandla kanye nokusungula izinto ezintsha Umphumela ophelele uncike esikalini, ukuhlanzeka kwegridi, ukukhetha ukusebenza kahle, nokuthi ngabe i-AI ixazulula izinkinga zangempela noma imane idala izinto ezintsha ngenxa yezinto ezintsha. I-IEA: Amandla kanye ne-AI

Uma ufuna ukuthatha okulula okusebenzayo:

  • Kala.

  • Usayizi wesokudla.

  • Lungiselela ukuphetha.

  • Nweba impilo yehadiwe.

  • Khuluma iqiniso mayelana nokushintshisana.

Futhi uma uzizwa ukhungathekile, nansi iqiniso eliduduzayo: izinqumo ezincane zokusebenza, eziphindaphindwa izikhathi eziyinkulungwane, ngokuvamile zidlula isitatimende esisodwa esikhulu sokuzinza. Kufana nokuxubha amazinyo akho. Akuyona into ekhangayo, kodwa kuyasebenza… 😄🪥

Imibuzo Evame Ukubuzwa

I-AI iyithinta kanjani imvelo ekusetshenzisweni kwansuku zonke, hhayi nje amalabhorethri amakhulu ocwaningo?

Iningi lezinyathelo ze-AI livela kugesi onika amandla izikhungo zedatha ezisebenzisa ama-GPU nama-CPU ngesikhathi sokuqeqeshwa kanye "nokucabanga" kwansuku zonke. Isicelo esisodwa singase sibe sincane, kodwa ngezinga lezo zicelo ziqoqana ngokushesha. Umthelela ubuye uncike ekutheni isikhungo sedatha sikhona kuphi, ukuthi igridi yendawo ihlanzekile kangakanani, nokuthi ingqalasizinda isebenza kahle kangakanani.

Ingabe ukuqeqesha imodeli ye-AI kubi kakhulu emvelweni kunokuyisebenzisa (ukucabanga)?

Ukuqeqeshwa kungaba umsebenzi omkhulu, oqhubekayo wokubala, kodwa ukuqagela kungaba yisisekelo esikhulu ngokuhamba kwesikhathi ngoba kusebenza njalo futhi ngezinga elikhulu. Uma ithuluzi lisetshenziswa izigidi zabantu nsuku zonke, izicelo eziphindaphindwayo zingadlula izindleko zokuqeqeshwa zesikhathi esisodwa. Yingakho ukwenza ngcono kuvame ukugxila ekusebenzeni kahle kokuqagela.

Kungani i-AI isebenzisa amanzi, futhi ingabe kuhlale kuyinkinga?

I-AI ingasebenzisa amanzi ikakhulukazi ngoba ezinye izikhungo zedatha zithembele ekupholiseni okusekelwe emanzini, noma ngoba amanzi asetshenziswa ngokungaqondile ngokukhiqizwa kukagesi. Ezimweni ezithile, ukupholisa okubangelwa ukuhwamuka kunganciphisa ukusetshenziswa kukagesi ngenkathi kukhulisa ukusetshenziswa kwamanzi, okudala ukuphikisana kwangempela. Ukuthi "kubi" kuncike ekushodeni kwamanzi endawo, ukwakheka kokupholisa, nokuthi ukusetshenziswa kwamanzi kuyalinganiswa futhi kuyaphathwa yini.

Yiziphi izingxenye zomthelela we-AI kwezemvelo ezivela ku-hardware kanye ne-e-waste?

I-AI incike kuma-chip, amaseva, imishini yokuxhumana, izakhiwo, kanye nezintambo zokuhlinzeka - okusho ukumbiwa kwezimayini, ukukhiqiza, ukuthunyelwa, kanye nokulahla ekugcineni. Ukukhiqizwa kwama-semiconductor kudinga amandla amaningi, futhi imijikelezo yokuthuthukisa okusheshayo ingakhuphula ukukhishwa okubonakalayo kanye nokulahlwa kwe-e. Ukwandisa impilo yehadiwe, ukuvuselela, kanye nokuthuthukisa ukusetshenziswa kunganciphisa kakhulu umthelela, ngezinye izikhathi kuqhathaniswa nezinguquko zezinga lemodeli.

Ingabe ukusebenzisa amandla avuselelekayo kuyayixazulula ithonya le-AI kwezemvelo?

Ugesi ohlanzekile unganciphisa ukukhishwa kwegesi okuvela ku-compute, kodwa awususi eminye imiphumela efana nokusetshenziswa kwamanzi, ukukhiqizwa kwehadiwe, kanye nokusetshenziswa kwe-e-waste. Futhi awuxazululi ngokuzenzakalelayo "imiphumela yokubuyela emuva," lapho ukubala okungabizi kakhulu kuholela ekusetshenzisweni okwengeziwe sekukonke. Izinto ezivuselelekayo ziyisivikelo esibalulekile, kodwa ziyingxenye eyodwa nje ye-footprint stack.

Iyini imiphumela yokubuyela emuva, futhi kungani kubalulekile ku-AI kanye nokusimama?

Umphumela wokubuyela emuva uba lapho izinzuzo zokusebenza kahle zenza into ibe lula noma ishibhile, ngakho abantu benza okuningi - ngezinye izikhathi baqeda ukonga. Nge-AI, ukukhiqiza noma ukwenza ngokuzenzakalelayo okushibhile kungandisa isidingo esiphelele sokuqukethwe, ukubala, kanye nezinsizakalo. Yingakho ukulinganisa imiphumela ekusebenzeni kubaluleke kakhulu kunokugubha ukusebenza kahle kukodwa.

Yiziphi izindlela ezisebenzayo zokunciphisa umthelela we-AI ngaphandle kokulimaza umkhiqizo?

Indlela evamile ukuqala ngokulinganisa (ukulinganisa kwamandla nekhabhoni, ukusetshenziswa), bese kuba amamodeli osayizi ofanele emsebenzini bese wenza ngcono ukuqagela ngokugcina isikhashana, ukuhlanganisa, kanye nemiphumela emifushane. Amasu anjengokulinganisa, ukuhluza, kanye nokukhiqiza okukhuliswe ngokubuyisa kunganciphisa izidingo zokubala. Izinketho zokusebenza - njengokuhlela umsebenzi ngamandla ekhabhoni kanye nokuphila isikhathi eside kwehadiwe - kuvame ukuletha impumelelo enkulu.

I-AI ingayisiza kanjani imvelo kunokuba iyilimaze?

I-AI inganciphisa ukukhishwa kwegesi kanye nemfucuza uma isetshenziswa ukuze kuthuthukiswe izinhlelo zangempela: ukubikezela kwegridi, impendulo yesidingo, ukwakha ukulawulwa kwe-HVAC, umzila wezokuthutha, ukulungiswa kokubikezela, kanye nokutholwa kokuvuza. Ingaphinde isekele ukuqapha imvelo njengezixwayiso zokugawulwa kwamahlathi kanye nokutholwa kwe-methane. Isihluthulelo ukuthi uhlelo luyazishintsha yini izinqumo futhi lukhiqize ukunciphisa okulinganiselwe, hhayi nje amadeshibhodi angcono.

Yiziphi izilinganiso okufanele izinkampani zibike ngazo ukuze zigweme izimangalo ze-AI "zokuhlanza imvelo"?

Kunengqondo kakhulu ukubika izilinganiso zomsebenzi ngamunye noma zesicelo ngasinye kunezinombolo ezinkulu kuphela, ngoba kubonisa ukusebenza kahle ezingeni leyunithi. Ukulandelela ukusetshenziswa kwamandla, ukulinganisa kwekhabhoni, ukusetshenziswa, kanye - lapho kufanele khona - imiphumela yamanzi kudala ukuziphendulela okucacile. Kubalulekile futhi: chaza imingcele (okufakwayo) futhi ugweme amalebula angacacile njenge-“AI enobungani nemvelo” ngaphandle kobufakazi obulinganiselwe.

Izinkomba

  1. I-International Energy Agency (IEA) - Amandla kanye ne-AI - iea.org

  2. I-International Energy Agency (IEA) - I-AI yokwenza ngcono amandla kanye nokusungula izinto ezintsha - iea.org

  3. I-International Energy Agency (IEA) - Ukwenziwa Kwedijithali - iea.org

  4. I-Lawrence Berkeley National Laboratory (LBNL) - Umbiko Wokusetshenziswa Kwamandla Esikhungo Sedatha sase-United States (2024) (PDF) - lbl.gov

  5. ULi nabanye - Ukwenza i-AI “Ingabi Nokoma” Kakhulu (2023) (PDF) - arxiv.org

  6. I-ASHRAE (TC 9.9) - Ukuvela Nokwanda Kokupholisa Uketshezi Ezikhungweni Zedatha Ezijwayelekile (PDF) - ashrae.org

  7. Igridi Eluhlaza - PUE-Ukuhlolwa Okuphelele Kwe-Metric - thegreengrid.org

  8. UMnyango Wezamandla wase-US (DOE) - FEMP - Amathuba Okusebenzisa Kahle Amanzi Okupholisa Ezikhungo Zedatha Zikahulumeni - energy.gov

  9. UMnyango Wezamandla wase-US (DOE) - FEMP - Ukusebenza Kahle Kwamandla Ezikhungweni Zedatha - energy.gov

  10. I-Ejensi Yokuvikela Imvelo yase-US (i-EPA) - Imboni Yezikhombisi-ndlela - epa.gov

  11. I-International Telecommunication Union (ITU) - I-Global E-waste Monitor 2024 - itu.int

  12. I-OECD - Izinzuzo Eziningi Zokuthuthukiswa Kokusetshenziswa Kahle Kwamandla (2012) (PDF) - oecd.org

  13. I-Carbon Intensity API (GB) - carbonintensity.org.uk

  14. i-imec - Ukunciphisa umthelela kwezemvelo ekukhiqizweni kwama-chip - imec-int.com

  15. I-UNEP - Indlela i-MARS esebenza ngayo - unep.org

  16. Ukuqapha Amahlathi Omhlaba Wonke - Izexwayiso ze-GLAD zokugawulwa kwamahlathi - globalforestwatch.org

  17. I-Alan Turing Institute - I-AI kanye nezinhlelo ezizimele zokuhlola ukuhlukahluka kwezinto eziphilayo kanye nempilo ye-ecosystem - turing.ac.uk

  18. IkhodiCarbon - Indlela Yokwenza - mlco2.github.io

  19. UGholami nabanye - Ucwaningo Lwezindlela Zokulinganisa (2021) (PDF) - arxiv.org

  20. ULewis nabanye - Isizukulwane Esithuthukisiwe Sokubuyiswa (2020) - arxiv.org

  21. UHinton nabanye - Ukuhlanza Ulwazi Kunethiwekhi Yezinzwa (2015) - arxiv.org

  22. IkhodiCarbon - codecarbon.io

Thola i-AI Yakamuva Esitolo Esisemthethweni Somsizi we-AI

Mayelana NATHI

Buyela kubhulogi