Kwenzekeni
I-Google Research yethule i-Planetary Prediction Engine yokuhlola, isistimu ye-AI ephendula imibuzo ye-geospatial yolimi lwemvelo ibe ukuhamba komsebenzi okuqagelayo. Le nkampani ithi uhlelo luzitholela futhi luhlanze idatha ngokuzenzakalelayo, izici zonjiniyela, luqeqeshe futhi luhlole izinhlobo eziningi zamamodeli, futhi lukhiqize umbiko. I-Google ibika imiphumela ethuthukisiwe uma iqhathaniswa namapayipi okuqhathanisa kuzo zonke izinkomba zezempilo yomphakathi zase-U.S. nezemvelo, ukwehla kokuvikeleka kokudla kwaseNigeria, kanye nokuqubuka kwe-Ebola manje eDemocratic Republic of the Congo.
Ucwaningo lwe-Google lumemezele Injini Yokubikezela Iplanethi, noma i-PPE, ngo-Agasti 27, 2026, njengamandla okuhlola ngaphakathi kohlelo lwayo lwe-Earth AI. Uhlelo lwamukela umbuzo wokubikezela we-geospatial obhalwe ngolimi lwemvelo futhi ludizayinelwe ukusebenzisa ukugeleza komsebenzi kusukela ekutholweni kwedatha ngokuqeqeshwa okuyimodeli, ukuhlola nokukhiqizwa kwemibiko. I-Google ichaza lokhu njengohlelo lwe-AI oluzimele, olunamamodeli amakhulu olimi ahlela isigaba ngasinye. Isimemezelo sithi inqubo inganciphisa imigqa yesikhathi yokwakha imodeli kusukela emavikini ahlanganisa ubunjiniyela bedatha okwenziwa ngesandla kuya kumaminithi, nakuba umthombo unganikezi isifundo sesikhathi esiqinisekisiwe ngokuzimele noma ulwazi olujwayelekile lokutholakala.
Isigaba sokuqala sihumusha umbuzo ube yizingqinamba zendawo nezesikhashana, okuhlanganisa ubumbudumbudu bendawo, okhiye bokujoyina kanye nobubanzi besikhathi. I-PPE ibe sesha amakhosombe amisiwe njengeDatha Commons kanye ne-Google Earth Engine ukuze uthole okuguquguqukayo okufanelekile. Uma leyo mithombo ingenayo isignali edingekayo, i-Google ithi uhlelo luthola bukhoma iwebhu evulekile kuwo wonke amaphothali kahulumeni nasezinqolobaneni zezemfundo. Iphinde yakhe ama-hypotheses esizinda futhi ibheke amasiginali we-proxy aqondile neyimbangela ethi asekelwa izincwadi ezishicilelwe. Umthombo awuchazi ukuthi lokho kuhlolwa kwezincwadi kucwaningwa kanjani noma ukuthi ukungezwani phakathi kwemithombo kuxazululwa kanjani.
Esigabeni sesibili, i-PPE ihlanganisa okuguquguqukayo kwezibalo okukhethiwe nokushumeka okuvela kumamodeli esisekelo se-geospatial. I-Google iqamba amamodeli esisekelo se-Population Dynamics wezifundazwe ezifihlekile zezenhlalo yabantu kanye ne-AlphaEarth ye-semantics yesithombe sesathelayithi. Uhlelo olubizwa ngokuthi Isango Lesici luhloselwe ukunciphisa ukuvuza okuhlosiwe ngokuhlunga izingxenye ezincane zezibalo, idatha yocwaningo olwabiwe, imiphumela yembangela ephansi kanye nedatha yesikhashana yesikhathi esizayo. Esigabeni sokugcina, i-PPE isesha phakathi kwamamodeli ajwayelekile ajwayelekile, izihlahla zezinqumo ezithuthukisiwe kanye nama-perceptron ezinhlaka eziningi. I-Overfitting Guard Protocol ichazwa njengokuhlola ubungozi bedathasethi nokusebenzisa iluphu yokuzilungisa lapho kutholwa ukwehluleka kokwenza okuvamile. Isimemezelo asivezi imikhawulo yephrothokholi noma izinga lokuhluleka.
I-Google ibika imiphumela kuyo yonke imisebenzi embalwa. Ezinkomba zezempilo ze-U.S. Centers for Disease Control and Prevention ezingu-21, ibika isilinganiso esingu-R² esingu-76.8%, uma siqhathaniswa no-60.0% wepayipi lochwepheshe abazenzelayo. Ngezinkomba ze-FEMA kazwelonke zobungozi, ibika u-64.9% uma kuqhathaniswa no-60.0%, futhi ku-Social Vulnerability Index, u-66.2% uma uqhathaniswa no-58.6%. E-Nigeria, i-Google ithi i-PPE ithuthukise ukwehliswa kokuvikeleka kokudla kusuka kudatha yesifundazwe ye-ADM1 kuya ezindaweni zasendaweni ze-ADM2, ibika u-R² wama-66.1% uma uqhathaniswa no-31.5%. Ezibikezelweni zamasonto onke ngesikhathi sokuqubuka kwe-ebolavirus ye-Bundibugyo ka-2026 eDemocratic Republic of the Congo, kubika i-Recall@10 yama-83.3%, ekhomba izindawo zezempilo ezisanda kuhlaselwa eziyi-15 kwezingu-18 kuzo zonke izibikezelo ezinhlanu, uma kuqhathaniswa cishe nama-73% wesisekelo esishicilelwe saseBayesian. Lezi zibalo ziyizimangalo ezivela emthonjeni wenkampani futhi azisungulwa ngokuzimele lapha.
Imininingwane yomthombo: research.google ↗
Kungani kubalulekile
Uma iqinisekiswa ngokuzimela, isistimu inganciphisa umsebenzi wobunjiniyela obukhethekile odingekayo ukuze kwakhiwe amamodeli e-geospatial futhi yenze ukuhlaziya okuzwela isikhathi kufinyeleleke kakhulu ezinhlanganweni ezisiza abantu, abacwaningi kanye nabakhi benqubomgomo. Imiphumela ebikiwe iphinde ibonise ukuthi amasistimu e-AI angahlanganisa kanjani okuguquguqukayo kwezibalo okujwayelekile nokuboniswa okuvela kumamodeli esisekelo se-geospatial. Kodwa-ke, ubufakazi buvela ku-akhawunti yocwaningo ye-Google uqobo, futhi uhlelo luhlala luyiphrojekthi yokuhlola yesigaba sangaphambi kwesikhathi kunenkonzo yomphakathi ebonisiwe.
Ukubikezela kwe-Geospatial ngokuvamile kuncike ekujoyineni amasethi edatha anemingcele ehlukene, izikhathi zesikhathi, amasistimu okulinganisa namaleveli emininingwane. Isimangalo esimaphakathi se-Google ukuthi i-PPE ingakwazi ukuzenzela ingxenye enkulu yalowo msebenzi kuyilapho ilondoloza ukuxhumana nombuzo wezinga eliphezulu lomsebenzisi. Uma le ndlela isebenza ngokuthembekile, amaqembu amancane ocwaningo noma izinhlangano ezisabela ezinkingeni zingase zikwazi ukukhiqiza izilinganiso zasendaweni ngaphandle kokuhlanganisa iqembu elikhulu lonjiniyela bedatha layo yonke iphrojekthi. Inani elingokoqobo lingancika ekutheni isistimu ingakwazi yini ukusebenza ngekhwalithi yedatha, uxhumo, ukubusa nobungcweti besizinda obutholakala ekusetshenzisweni kwangempela.
Umphumela obikiwe waseNigeria ubonisa ukusetshenziswa okungenzeka kwenzuzo yomphakathi okuchazwe yi-Google. Izibalo zesifundazwe zokuvikeleka kokudla zingafihla umehluko phakathi kwezindawo zasendaweni, kuyilapho uhlelo luhloselwe ukuhlanganisa ukushaqeka kwemakethe, okudidayo kwentengo yokudla kanye nezinkomba ze-microclimate ukuze kukhiqizwe amamephu anemininingwane eminingi yokuba sengozini. Amamephu anjalo angasiza amaqembu ezenhlalakahle aqondise ukuhlolwa noma izinsiza. Umthombo awubonisi ukuthi noma iyiphi inhlangano esebenzise i-PPE ukwenza isinqumo sangempela sokwabiwa, futhi ayiqinisekisi ukuthi izibikezelo zithuthukise imiphumela. Ngakho-ke isimemezelo sisekela isimangalo mayelana nokusebenza kokumodela, hhayi umthelela wobuntu.
Isibonelo se-Ebola yase-DRC sikhomba inzuzo ehlukile: isivinini. I-Google ithi i-PPE ikhombe izindawo eziningi zezempilo ezisanda kuhlaselwa kuzibikezelo zeviki ezinhlanu ezilandelanayo futhi yeqa isisekelo esikhonjiwe sase-Bayesia ngamaphesenti angu-10.3. Ukumodela okusheshayo kungaba nendaba lapho amaqembu ezempilo yomphakathi edinga ukubeka phambili ukubhekwa noma impendulo. Kepha amaphuzu akhumbulayo awasho ngokwawo ukuthi mangaki amanga akhiqizwe yisistimu, ukuthi izibikezelo zifike kusenesikhathi ukuze zishintshe isenzo, noma ukuthi imodeli yahlala ithembekile njengoba ukuqubuka nezimo zokubika zishintsha. Leyo mibuzo yokusebenza yinto engaziwa.
I-Google ifaka izinzuzo ekuhlanganisweni kwezindlela eziningi nokukhetha idatha ehlakaniphile. Ama-covariate akhiwe anganikeza amasignali asobala, kuyilapho ukushumeka kwemodeli yesisekelo kungase kufake amaphethini okunzima ukuwamela ngesandla. Umthombo uthi ucwaningo lokwehliswa kwezimali luthole ngokungaguquki ukuthi ukuhlanganisa izindlela zokwenza kudlule izindlela eziyisisekelo, okuphakamisa ukuphelelisana esikhundleni sokungadingeki. Kodwa isimemezelo asihlinzeki ngamathebula okukhipha imali, izikhawu zokuzethemba, amasampula osayizi, ukuhlukaniswa kwendawo noma ukuphindaphinda okuzimele. Ngenxa yokuthi unjiniyela wesistimu uphinde abike ukuhlaziya, abafundi kufanele baphathe ukuthuthukiswa kwezinombolo njengobufakazi obuthembisayo kodwa besikhashana.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
crm_get_transaction(id='4092').A route planner searches possible journeys using explicit rules. What does this illustrate about AI?
Ongakubuka ngokulandelayo
Imibuzo ebalulekile iwukuthi ingabe izinzuzo ezibikiwe zibambelela kumadathasethi azimele, kwezinye izifunda kanye naphakathi nemicimbi yesikhathi esizayo; ukuthi isistimu iyisingatha kanjani ngokuthembekile idatha engaphelele, echemile noma eshintshayo; nokuthi ochwepheshe babantu bangakwazi yini ukuhlola ukukhetha kwayo kwedatha kanye nokuqagela okuyimbangela. Ukucutshungulwa okwengeziwe kufanele kuhlole ukwakheka kwebhentshimakhi yenkampani, izisekelo, izilinganiso zokungaqiniseki, amacala okuhluleka, izindleko zokusebenza nezimo zokufinyelela. I-Google futhi ithi ihlela ukwengeza imithombo eminingi ye-geospatial kanye nokushumeka kwemodeli yesisekelo se-multimodal, kodwa ayinikezi imininingwane yesikhathi noma yokutholakala.
Okubalulekile kokuqala ukuphindaphinda okuzimele. Abahloli kufanele bahlole i-PPE kuzindawo ezimisiwe, izinkathi zesikhathi nokubikezela okuhlosiwe okukhethiwe ngaphandle kokubandakanyeka kwe-Google, kuyilapho beyiqhathanisa namapayipi ochwepheshe bendawo aqinile kunezisekelo ezishiwo esimemezelweni kuphela. Kufanele futhi babike ukungaqiniseki, ukulinganiswa, imibono engamanga kanye nokuphikisa okungamanga, hhayi kuphela u-R² noma u-Recall@10. Lokhu kubaluleke kakhulu ezinhlelweni zokusebenza zomphakathi nezokusiza abantu, lapho imodeli ingakwazi ukuthola amaphuzu amahle lapho ingasebenzi kahle emiphakathini ethile noma phakathi nemicimbi engajwayelekile.
Ukutholakala kwedatha kanye nokubusa kuzoba okuphakathi. I-PPE ingasesha amaqoqo omphakathi kanye newebhu evuliwe ngesikhathi sokubikezela, kodwa umthombo awucacisi ukuthi isingatha kanjani amalayisense, izinguquko zedatha, amanani ashodayo, izilinganiso ezingqubuzanayo noma ulwazi olubucayi lwezepolitiki. Ukusetshenziswa kwayo kwezethulo zenhlalo yabantu kanye nesathelayithi nakho kuphakamisa imibuzo mayelana nokuchema kwezindawo kanye nokuthi ingabe ukuguquguquka kwama-proxy kukhiqiza kabusha ukungalingani okungokomlando. Amadokhumenti esikhathi esizayo kufanele abonise ukuthi yikuphi okokufaka okukhethiwe, okunqatshwe Isango Lesici, nokuthi ababuyekezi abangabantu bangayibekela kanjani inselele ukukhetha kwedatha noma ukuqagela kwembangela.
Ukuzimela kwesistimu kudinga ukubhekwa ngokucophelela. I-Google ithi izigaba ezihlukene zishintshanisa ama-artifact ngezibambo ezingacacile esikhundleni sokubeka idatha ye-serialized ekwazisweni kwemodeli yolimi, okusiza ukugwema imikhawulo yewindi lomongo. Lowo mklamo ungase unciphise ibhodlela elilodwa lobuchwepheshe, kodwa wona ngokwawo awuqinisekisi ukuthi izinqumo zesistimu ziyaqondakala noma ziyakhiqizeka. Abasebenzisi abasebenzayo bazodinga amalogi osesho, ukuguqulwa, ukukhetha amamodeli, ukuhlukaniswa kokuhlola nokulungiswa. Bazophinda badinge izinqubo ezicacile zokumisa noma zokweqa ukuhamba komsebenzi lapho isistimu ihlangabezana nenkinga entsha, umthombo ongathembekile noma isibikezelo esingaphandle kobubanzi bayo obuqinisekisiwe.
I-Google ihlonza i-PPE njengephrojekthi yocwaningo lwesigaba sangaphambi kwesikhathi futhi ithi ihlose ukwandisa ububanzi bedatha ye-geospatial kanye nokushumeka kwemodeli yesisekelo. Isimemezelo asisho ukuthi uhlelo luzokhululwa yini, ubani ongalusebenzisa, yiziphi izinsiza zekhompyutha oludingayo, noma ukuthi amamodeli alo ayisisekelo kanye nekhodi kuzofinyeleleka yini. Lokho okungaziwa kuzonquma ukuthi ukuncishiswa okumangalelwe kwesithiyo sezobuchwepheshe kufinyelela ezinhlanganweni ezingaphandle kwengqalasizinda ye-Google. Kuze kube yilapho imininingwane enjalo kanye nobufakazi obuzimele butholakala, i-PPE iqondwa kangcono njengomboniso ophawulekayo wocwaningo onenani elingaba khona emphakathini, hhayi ukumiselela okuqinisekisiwe kochwepheshe besizinda noma izinhlelo ezisunguliwe zokuhlaziya izimo eziphuthumayo.